Start Printed Page 42986
Environmental Protection Agency and National Highway Traffic Safety Administration.
Notice of proposed rulemaking.
The National Highway Traffic Safety Administration (NHTSA) and the Environmental Protection Agency (EPA) are proposing the “Safer Affordable Fuel-Efficient (SAFE) Vehicles Rule for Model Years 2021-2026 Passenger Cars and Light Trucks” (SAFE Vehicles Rule). The SAFE Vehicles Rule, if finalized, would amend certain existing Corporate Average Fuel Economy (CAFE) and tailpipe carbon dioxide emissions standards for passenger cars and light trucks and establish new standards, all covering model years 2021 through 2026. More specifically, NHTSA is proposing new CAFE standards for model years 2022 through 2026 and amending its 2021 model year CAFE standards because they are no longer maximum feasible standards, and EPA is proposing to amend its carbon dioxide emissions standards for model years 2021 through 2025 because they are no longer appropriate and reasonable in addition to establishing new standards for model year 2026. The preferred alternative is to retain the model year 2020 standards (specifically, the footprint target curves for passenger cars and light trucks) for both programs through model year 2026, but comment is sought on a range of alternatives discussed throughout this document. Compared to maintaining the post-2020 standards set forth in 2012, current estimates indicate that the proposed SAFE Vehicles Rule would save over 500 billion dollars in societal costs and reduce highway fatalities by 12,700 lives (over the lifetimes of vehicles through MY 2029). U.S. fuel consumption would increase by about half a million barrels per day (2-3 percent of total daily consumption, according to the Energy Information Administration) and would impact the global climate by 3/1000th of one degree Celsius by 2100, also when compared to the standards set forth in 2012.
Comments: Comments are requested on or before October 23, 2018. Under the Paperwork Reduction Act, comments on the information collection provisions must be received by the Office of Management and Budget (OMB) on or before October 23, 2018. See the SUPPLEMENTARY INFORMATION section on “Public Participation,” below, for more information about written comments.
Public Hearings: NHTSA and EPA will jointly hold three public hearings in Washington, DC; the Detroit, MI area; and in the Los Angeles, CA area. The agencies will announce the specific dates and addresses for each hearing location in a supplemental Federal Register notice. The agencies will accept oral and written comments to the rulemaking documents, and NHTSA will also accept comments to the Draft Environmental Impact Statement (DEIS) at these hearings. The hearings will start at 10 a.m. local time and continue until everyone has had a chance to speak. See the SUPPLEMENTARY INFORMATION section on “Public Participation,” below, for more information about the public hearings.
You may send comments, identified by Docket No. EPA-HQ-OAR-2018-0283 and/or NHTSA-2018-0067, by any of the following methods:
Federal eRulemaking Portal:
http://www.regulations.gov. Follow the instructions for sending comments.
EPA: (202) 566-9744; NHTSA: (202) 493-2251.
○ EPA: Environmental Protection Agency, EPA Docket Center (EPA/DC), Air and Radiation Docket, Mail Code 28221T, 1200 Pennsylvania Avenue NW, Washington, DC 20460, Attention Docket ID No. EPA-HQ-OAR-2018-0283. In addition, please mail a copy of your comments on the information collection provisions for the EPA proposal to the Office of Information and Regulatory Affairs, Office of Management and Budget (OMB), Attn: Desk Officer for EPA, 725 17th St. NW, Washington, DC 20503.
○ NHTSA: Docket Management Facility, M-30, U.S. Department of Transportation, West Building, Ground Floor, Rm. W12-140, 1200 New Jersey Avenue SE, Washington, DC 20590.
○ EPA: Docket Center (EPA/DC), EPA West, Room B102, 1301 Constitution Avenue NW, Washington, DC, Attention Docket ID No. EPA-HQ-OAR-2018-0283. Such deliveries are only accepted during the Docket's normal hours of operation, and special arrangements should be made for deliveries of boxed information.
○ NHTSA: West Building, Ground Floor, Rm. W12-140, 1200 New Jersey Avenue SE, Washington, DC 20590, between 9 a.m. and 4 p.m. Eastern Time, Monday through Friday, except Federal holidays.
Instructions: All submissions received must include the agency name and docket number or Regulatory Information Number (RIN) for this rulemaking. All comments received will be posted without change to http://www.regulations.gov, including any personal information provided. For detailed instructions on sending comments and additional information on the rulemaking process, see the “Public Participation” heading of the SUPPLEMENTARY INFORMATION section of this document.
Docket: For access to the dockets to read background documents or comments received, go to http://www.regulations.gov, and/or:
Start Further Info
For EPA: EPA Docket Center (EPA/DC), EPA West, Room 3334, 1301 Constitution Avenue NW, Washington, DC 20460. The Public Reading Room is open from 8:30 a.m. to 4:30 p.m., Monday through Friday, excluding legal holidays. The telephone number for the Public Reading Room is (202) 566-1744.
For NHTSA: Docket Management Facility, M-30, U.S. Department of Transportation, West Building, Ground Floor, Rm. W12-140, 1200 New Jersey Avenue SE, Washington, DC 20590. The Docket Management Facility is open between 9 a.m. and 4 p.m. Eastern Time, Monday through Friday, except Federal holidays.
FOR FURTHER INFORMATION CONTACT:
EPA: Christopher Lieske, Office of Transportation and Air Quality, Assessment and Standards Division, Environmental Protection Agency, 2000 Traverwood Drive, Ann Arbor, MI 48105; telephone number: (734) 214-4584; fax number: (734) 214-4816; email address: firstname.lastname@example.org, or contact the Assessment and Standards Division, email address: email@example.com. NHTSA: James Tamm, Office of Rulemaking, Fuel Economy Division, National Highway Traffic Safety Administration, 1200 New Jersey Avenue SE, Washington, DC 20590; telephone number: (202) 493-0515.
End Further Info
Start Supplemental Information
Start Printed Page 42987
I. Overview of Joint NHTSA/EPA Proposal
II. Technical Foundation for NPRM Analysis
III. Proposed CAFE and CO2 Standards for MYs 2021-2026
IV. Alternative CAFE and GHG Standards Considered for MYs 2021/22-2026
V. Proposed Standards, the Agencies' Statutory Obligations, and Why the Agencies Propose To Choose Them Over the Alternatives
VI. Preemption of State and Local Laws
VII. Impacts of the Proposed CAFE and CO2 Standards
VIII. Impacts of Alternative CAFE and CO2 Standards Considered for MYs 2021/22-2026
IX. Vehicle Classification
X. Compliance and Enforcement
XI. Public Participation
XII. Regulatory Notices and Analyses
I. Overview of Joint NHTSA/EPA Proposal
A. Executive Summary
In this notice, the National Highway Traffic Safety Administration (NHTSA) and the Environmental Protection Agency (EPA) (collectively, “the agencies”) are proposing the “Safer Affordable Fuel-Efficient (SAFE) Vehicles Rule for Model Years 2021-2026 Passenger Cars and Light Trucks” (SAFE Vehicles Rule). The proposed SAFE Vehicles Rule would set Corporate Average Fuel Economy (CAFE) and carbon dioxide (CO2) emissions standards, respectively, for passenger cars and light trucks manufactured for sale in the United States in model years (MYs) 2021 through 2026.
CAFE and CO2 standards have the power to transform the vehicle fleet and affect Americans' lives in significant, if not always immediately obvious, ways. The proposed SAFE Vehicles Rule seeks to ensure that government action on these standards is appropriate, reasonable, consistent with law, consistent with current and foreseeable future economic realities, and supported by a transparent assessment of current facts and data.
The agencies must act to propose and finalize these standards and do not have discretion to decline to regulate. Congress requires NHTSA to set CAFE standards for each model year.
Congress also requires EPA to set emissions standards for light-duty vehicles if EPA has made an “endangerment finding” that the pollutant in question—in this case, CO2—“cause[s] or contribute[s] to air pollution which may reasonably be anticipated to endanger public health or welfare.” 
NHTSA and EPA are proposing these standards concurrently because tailpipe CO2 emissions standards are directly and inherently related to fuel economy standards,
and if finalized, these rules would apply concurrently to the same fleet of vehicles. By working together to develop these proposals, the agencies reduce regulatory burden on industry and improve administrative efficiency.
Consistent with both agencies' statutes, this proposal is entirely de novo, based on an entirely new analysis reflecting the best and most up-to-date information available to the agencies at the time of this rulemaking. The agencies worked together in 2012 to develop CAFE and CO2 standards for MYs 2017 and beyond; in that rulemaking action, EPA set CO2 standards for MYs 2017-2025, while NHTSA set final CAFE standards for MYs 2017-2021 and also put forth “augural” CAFE standards for MYs 2022-2025, consistent with EPA's CO2 standards for those model years. EPA's CO2 standards for MYs 2022-2025 were subject to a “mid-term evaluation,” by which EPA bound itself through regulation to re-evaluate the CO2 standards for those model years and to undertake to develop new CO2 standards through a regulatory process if it concluded that the previously finalized standards were no longer appropriate. EPA regulations on the mid-term evaluation process required EPA to issue a Final Determination no later than April 1, 2018 on whether the GHG standards for MY 2022-2025 light-duty vehicles remain appropriate under section 202(a) of the Clean Air Act.
The regulations also required the issuance of a draft Technical Assessment Report (TAR) by November 15, 2017, an opportunity for public comment on the draft TAR, and, before making a Final Determination, an opportunity for public comment on whether the GHG standards for MY 2022-2025 remain appropriate. In July 2016, the draft TAR was issued for public comment jointly by the EPA, NHTSA, and the California Air Resources Board (CARB).
Following the draft TAR, EPA published a Proposed Determination for public comment on December 6, 2016 and provided less than 30 days for public comments over major holidays.
EPA published the January 2017 Determination on EPA's website and regulations.gov finding that the MY 2022-2025 standards remained appropriate.
On March 15, 2017, President Trump announced a restoration of the original mid-term review timeline. The President made clear in his remarks, “[i]f the standards threatened auto jobs, then commonsense changes” would be made in order to protect the economic viability of the U.S. automotive industry.” 
In response to the President's direction, EPA announced in a March 22, 2017, Federal Register notice, its intention to reconsider the Final Determination of the mid-term evaluation of GHGs emissions standards for MY 2022-2025 light-duty vehicles.
The Administrator stated that EPA would coordinate its reconsideration with the rulemaking process to be undertaken by NHTSA regarding CAFE standards for cars and light trucks for the same model years.
On August 21, 2017, EPA published a notice in the Federal Register announcing the opening of a 45-day public comment period and inviting stakeholders to submit any additional comments, data, and information they believed were relevant to the Administrator's reconsideration of the
Start Printed Page 42988
January 2017 Determination.
EPA held a public hearing in Washington DC on September 6, 2017.
EPA received more than 290,000 comments in response to the August 21, 2017 notice.
EPA has since concluded, based on more recent information, that those standards are no longer appropriate.
NHTSA's “augural” CAFE standards for MYs 2022-2025 were not final in 2012 because Congress prohibits NHTSA from finalizing new CAFE standards for more than five model years in a single rulemaking.
NHTSA was therefore obligated from the beginning to undertake a new rulemaking to set CAFE standards for MYs 2022-2025.
The proposed SAFE Vehicles Rule begins the rulemaking process for both agencies to establish new standards for MYs 2022-2025 passenger cars and light trucks. Standards are concurrently being proposed for MY 2026 in order to provide regulatory stability for as many years as is legally permissible for both agencies together.
Separately, the proposed SAFE Vehicles Rule includes revised standards for MY 2021 passenger cars and light trucks. The information now available and the current analysis suggest that the CAFE standards previously set for MY 2021 are no longer maximum feasible, and the CO2 standards previously set for MY 2021 are no longer appropriate. Agencies always have authority under the Administrative Procedure Act to revisit previous decisions in light of new facts, as long as they provide notice and an opportunity for comment, and it is plainly the best practice to do so when changed circumstances so warrant.
Thus, the proposed SAFE Vehicles Rule would maintain the CAFE and CO2 standards applicable in MY 2020 for MYs 2021-2026, while taking comment on a wide range of alternatives, including different stringencies and retaining existing CO2 standards and the augural CAFE standards.
Table I-4 below presents those alternatives. We note further that prior to MY 2021, CO2 targets include adjustments reflecting the use of automotive refrigerants with reduced global warming potential (GWP) and/or the use of technologies that reduce the refrigerant leaks, and optionally offsets for nitrous oxide and methane emissions. In the interests of harmonizing with the CAFE program, EPA is proposing to exclude air conditioning refrigerants and leakage, and nitrous oxide and methane emissions for compliance with CO2 standards after model year 2020 but seeks comment on whether to retain these element, and reinsert A/C leakage offsets, and remain disharmonized with the CAFE program. EPA also seeks comment on whether to change existing methane and nitrous oxide standards that were finalized in the 2012 rule. Specifically, EPA seeks information from the public on whether those existing standards are appropriate, or whether they should be revised to be less stringent or more stringent based on any updated data.
While actual requirements will ultimately vary for automakers depending upon their individual fleet mix of vehicles, many stakeholders will likely be interested in the current estimate of what the MY 2020 CAFE and CO2 curves would translate to, in terms of miles per gallon (mpg) and grams per mile (g/mi), in MYs 2021-2026. These estimates are shown in the following tables.
Start Printed Page 42989
Start Printed Page 42990
In the tables above, estimated required CO2 increases between MY 2020 and MY 2021 because, again, EPA is proposing to exclude CO2-equivalent emission improvements associated with air conditioning refrigerants and leakage (and, optionally, offsets for nitrous oxide and methane emissions) after model year 2020.
As explained above, the agencies are taking comment on a wide range of alternatives and have specifically modeled eight alternatives (including the proposed alternative) and the current requirements (i.e., baseline/no-action). The modeled alternatives are provided below:
Summary of Rationale
Since finalizing the agencies' previous joint rulemaking in 2012 titled “Final Rule for Model Year 2017 and Later Light-Duty Vehicle Greenhouse Gas Emission and Corporate Average Fuel Economy Standards,” and even since EPA's 2016 and early 2017 “mid-term evaluation” process, the agencies have gathered new information, and have performed new analysis. That new information and analysis has led the Start Printed Page 42991agencies to the tentative conclusion that holding standards constant at MY 2020 levels through MY 2026 is maximum feasible, for CAFE purposes, and appropriate, for CO2 purposes.
Technologies have played out differently in the fleet from what the agencies assumed in 2012.
The technology to improve fuel economy and reduce CO2 emissions has not changed dramatically since prior analyses were conducted: A wide variety of technologies are still available to accomplish the goals of the programs, and a wide variety of technologies would likely be used by industry to accomplish these goals. There remains no single technology that the majority of vehicles made by the majority of manufacturers can implement at low cost without affecting other vehicle attributes that consumers value more than fuel economy and CO2 emissions. Even when used in combination, technologies that can improve fuel economy and reduce CO2 emissions still need to (1) actually work together and (2) be acceptable to consumers and avoid sacrificing other vehicle attributes while also avoiding undue increases in vehicle cost. Optimism about the costs and effectiveness of many individual technologies, as compared to recent prior rounds of rulemaking, is somewhat tempered; a clearer understanding of what technologies are already on vehicles in the fleet and how they are being used, again as compared to recent prior rounds of rulemaking, means that technologies that previously appeared to offer significant “bang for the buck” may no longer do so. Additionally, in light of the reality that vehicle manufacturers may choose the relatively cost-effective technology option of vehicle lightweighting for a wide array of vehicles and not just the largest and heaviest, it is now recognized that as the stringency of standards increases, so does the likelihood that higher stringency will increase on-road fatalities. As it turns out, there is no such thing as a free lunch.
Technology that can improve both fuel economy and/or performance may not be dedicated solely to fuel economy.
As fleet-wide fuel efficiency has improved over time, additional improvements have become both more complicated and more costly. There are two primary reasons for this phenomenon. First, as discussed, there is a known pool of technologies for improving fuel economy and reducing CO2 emissions. Many of these technologies, when actually implemented on vehicles, can be used to improve other vehicle attributes such as “zero to 60” performance, towing, and hauling, etc., either instead of or in addition to improving fuel economy and reducing CO2 emissions. As one example, a V6 engine can be turbocharged and downsized so that it consumes only as much fuel as an inline 4-cylinder engine, or it can be turbocharged and downsized so that it consumes less fuel than it would originally have consumed (but more than the inline 4-cylinder would) while also providing more low-end torque. As another example, a vehicle can be lightweighted so that it consumes less fuel than it would originally have consumed, or so that it consumes the same amount of fuel it would originally have consumed but can carry more content, like additional safety or infotainment equipment. Manufacturers employing “fuel-saving/emissions-reducing” technologies in the real world make decisions regarding how to employ that technology such that fewer than 100% of the possible fuel-saving/emissions-reducing benefits result. They do this because this is what consumers want, and more so than exclusively fuel economy improvements.
This makes actual fuel economy gains more expensive.
Thus, even though the technologies may be largely the same, previous assumptions about how much fuel can be saved or how much emissions can be reduced by employing various technologies may not have played out as prior analyses suggested, meaning that previous assumptions about how much it would cost to save that much fuel or reduce that much in emissions fall correspondingly short. For example, the agencies assumed in the 2010 final rule that dual clutch transmissions would be widely used to improve fuel economy due to expectations of strong effectiveness and very low cost: In practice, dual clutch transmissions had significant customer acceptance issues, and few manufacturers employ them in the U.S. market today.
The agencies included some “technologies” in the 2012 final rule analysis that were defined ambiguously and/or in ways that precluded observation in the known (MYs 2008 and 2010) fleets, likely leading to double counting in cases where the known vehicles already reflected the assumed efficiency improvement. For example, the agencies assumed that transmission “shift optimizers” would be available and fairly widely used in MYs 2017-2025, but involving software controls, a “technology” not defined in a way that would be observed in the fleet (unlike, for example, a dual clutch transmission).
To be clear, this is no one's “fault”—the CAFE and CO2 standards do not require manufacturers to use particular technologies in particular ways, and both agencies' past analyses generally sought to illustrate technology paths to compliance that were assumed to be as cost-effective as possible. If manufacturers choose different paths for reasons not accounted for in regulatory analysis, or choose to use technologies differently from what the agencies previously assumed, it does not necessarily mean that the analyses were unreasonable when performed. It does mean, however, that the fleet ought to be reflected as it stands today, with the technology it has and as that technology has been used, and consider what technology remains on the table at this point, whether and when it can realistically be available for widespread use in production, and how much it would cost to implement.
Incremental additional fuel economy benefits are subject to diminishing returns.
As fleet-wide fuel efficiency improves and CO2 emissions are reduced, the incremental benefit of continuing to improve/reduce inevitably decreases. This is because, as the base level of fuel economy improves, fewer gallons are saved from subsequent incremental improvements. Put simply, a one mpg increase for vehicles with low fuel economy will result in far greater savings than an identical 1 mpg increase for vehicles with higher fuel economy, and the cost for achieving a one-mpg increase for low fuel economy vehicles is far less than for higher fuel economy vehicles. This means that improving fuel economy is subject to diminishing returns. Annual fuel consumption can be calculated as follows:
Start Printed Page 42992
For purposes of illustration, assume a vehicle owner who drives a light vehicle 15,000 miles per year (a typical assumption for analytical purposes).
If that owner trades in a vehicle with fuel economy of 15 mpg for one with fuel economy of 20 mpg, the owner's annual fuel consumption would drop from 1,000 gallons to 750 gallons—saving 250 gallons annually. If, however, that owner were to trade in a vehicle with fuel economy of 30 mpg for one with fuel economy of 40 mpg, the owner's annual gasoline consumption would drop from 500 gallons/year to 375 gallons/year—only 125 gallons even though the mpg improvement is twice as large. Going from 40 to 50 mpg would save only 75 gallons/year. Yet, each additional fuel economy improvement becomes much more expensive as the low-hanging fruit of low-cost technological improvement options are picked.
Automakers, who must nonetheless continue adding technology to improve fuel economy and reduce CO2 emissions, will either sacrifice other performance attributes or raise the price of vehicles—neither of which is attractive to most consumers.
If fuel prices are high, the value of those gallons may be enough to offset the cost of further fuel economy improvements, but (1) the most recent reference case projections in the Energy Information Administration's (EIA's) Annual Energy Outlook (AEO 2017 and AEO 2018) do not indicate particularly high fuel prices in the foreseeable future, given underlying assumptions,
and (2) as the baseline level of fuel economy continues to increase, the marginal cost of the next gallon saved similarly increases with the cost of the technologies required to meet the savings. The following figure illustrates the fact that fuel savings and corresponding avoided costs diminish with increasing fuel economy, showing the same basic pattern as a 2014 illustration developed by EIA.
Start Printed Page 42993
This effect is mathematical in nature and long-established, but when combined with relatively low fuel prices potentially through 2050, and the likelihood that a large majority of American consumers could consequently continue to place a higher value on vehicle attributes other than fuel economy, it makes manufacturers' ability to sell light vehicles with ever-higher fuel economy and ever-lower carbon dioxide emissions increasingly difficult. Put more simply, if gas is cheap and each additional improvement saves less gas anyway, most consumers would rather spend their money on attributes other than fuel economy when they are considering a new vehicle purchase, whether that is more safety technology, a better infotainment package, a more powerful powertrain, or other features (or, indeed, they may prefer to spend the savings on something other than automobiles). Manufacturers trying to sell consumers more fuel economy in such circumstances may convince consumers who place weight on efficiency and reduced carbon emissions, but consumers decide for themselves what attributes are worth to them. And while some contend that consumers do not sufficiently consider or value future fuel savings when making vehicle purchasing decisions,
information regarding the benefits of higher fuel economy has never been made more readily available than today, with a host of online tools and mandatory prominent disclosures on new vehicles on the Monroney label showing fuel savings compared to average vehicles. This is not a question of “if you build it, they will come.” Despite the widespread availability of fuel economy information, and despite manufacturers building and marketing vehicles with higher fuel economy and increasing their offerings of hybrid and electric vehicles, in the past several years as gas prices have remained low, consumer preferences have shifted markedly away from higher-fuel-economy smaller and midsize passenger vehicles toward crossovers and truck-based utility vehicles.
Some consumers plainly value fuel economy and low CO2 emissions above other attributes, and thanks in part to CAFE and CO2 standards, they have a plentiful selection of high-fuel economy and low CO2-emitting vehicles to choose from, but those consumers represent a relatively small percentage of buyers.
Changed petroleum market has supported a shift in consumer preferences.
In 2012, the agencies projected fuel prices would rise significantly, and the United States would continue to rely heavily upon imports of oil, subjecting the country to heightened risk of price shocks.
Things have changed significantly since 2012, with fuel prices significantly lower than anticipated, and projected to remain low through 2050. Furthermore, the global petroleum market has shifted dramatically with the United States taking advantage of its own oil supplies through technological advances that allow for cost-effective extraction of shale oil. The U.S. is now the world's largest oil producer and expected to become a net petroleum exporter in the next decade.
At least partially in response to lower fuel prices, consumers have moved more heavily into crossovers, sport utility vehicles and pickup trucks, than anticipated at the time of the last rulemaking. Because standards are based on footprint and specified separately for passenger cars and light trucks, these shifts do not necessarily pose compliance challenges by themselves, but they tend to reduce the overall average fuel economy rates and increase the overall average CO2 emission rates of the new vehicle fleet. Consumers are also demonstrating a preference for more powerful engines and vehicles with higher seating positions and ride height (and accompanying mass increase relative to footprint) 
—all of which present challenges for achieving increased fuel economy levels and lower CO2 emission rates.
The Consequence of Unreasonable Fuel Economy and CO2Standards: Increased vehicle prices keep consumers in older, dirtier, and less safe vehicles.
Consumers tend to avoid purchasing things that they neither want or need. The analysis in today's proposal moves closer to being able to represent this fact through an improved model for vehicle scrappage rates. While neither this nor a sales response model, also included in today's analysis, nor the combination of the two, are consumer choice models, today's analysis illustrates market-wide impacts on the sale of new vehicles and the retention of used vehicles. Higher vehicle prices, which result from more-stringent fuel economy standards, have an effect on consumer purchasing decisions. As prices increase, the market-wide incentive to extract additional travel from used vehicles increases. The average age of the in-service fleet has been increasing, and when fleet turnover slows, not only does it take longer for fleet-wide fuel economy and CO2 emissions to improve, but also safety improvements, criteria pollutant emissions improvements, many other vehicle attributes that also provide societal benefits take longer to be reflected in the overall U.S. fleet as well because of reduced turnover. Raising vehicle prices too far, too fast, such as through very stringent fuel economy and CO2 emissions standards (especially considering that, on a fleet-wide basis, new vehicle sales and turnover do not appear strongly responsive to fuel economy), has effects beyond simply a slowdown in sales. Improvements over time have better longer-term effects simply by not alienating consumers, as compared to great leaps forward that drive people out of the new car market or into vehicles that do not meet their needs. The industry has achieved tremendous gains in fuel economy over the past decade, and these increases will continue at least through 2020.
Along with these gains, there have also been tremendous increases in vehicle prices, as new vehicles become increasingly unaffordable—with the average new vehicle transaction price Start Printed Page 42994recently exceeding $36,000—up by more than $3,000 since 2014 alone.
In fact, a recent independent study indicated that the average new car price is unaffordable to median-income families in every metropolitan region in the United States except one: Washington, DC.
That analysis used the historically accepted approach that consumers should make a down-payment of at least 20% of a vehicle's purchase price, finance for no longer than four years, and make payments of 10% or less of the consumer's annual income to car payments and insurance. But the market looks nothing like that these days, with average financing terms of 68 months, and an increasing proportion exceeding 72 or even 84 months.
Longer financing terms may allow a consumer to keep their monthly payment affordable but can have serious potential financial consequences. Longer-term financing leads (generally) to higher interest rates, larger finance charges and total consumer costs, and a longer period of time with negative equity. In 2012, the agencies expected prices to increase under the standards announced at that time. The agencies estimated that, compared to a continuation of the model year 2016 standards, the standards issued through model year 2025 would eventually increase average prices by about $1,500-$1,800.  
Circumstances have changed, the analytical methods and inputs have been updated (including updates to address issues still present in analyses published in 2016, 2017, and early 2018), and today, the analysis suggests that, compared to the proposed standards today, the previously-issued standards would increase average vehicle prices by about $2,100. While today's estimate is similar in magnitude to the 2012 estimate, it is relative to a baseline that includes increases in stringency between MY 2016 and MY 2020. Compared to leaving vehicle technology at MY 2016 levels, today's analysis shows the previously-issued standards through model year 2025 could eventually increase average vehicle prices by approximately $2,700. A pause in continued increases in fuel economy standards, and cost increases attributable thereto, is appropriate.
Start Printed Page 42995
For all of these reasons, the agencies are proposing to maintain the MY 2020 fuel economy and CO2 emissions standards for MYs 2021-2026. Our goal is to establish standards that promote both energy conservation and safety, in light of what is technologically feasible and economically practicable, as directed by Congress.
EPCA requires that NHTSA, when determining the maximum feasible levels of CAFE standards, consider the need of the Nation to conserve energy. However, EPCA also requires that NHTSA consider other factors, such as technological feasibility and economic practicability. The analysis suggests that, compared to the standards issued previously for MYs 2021-2025, today's proposed rule will eventually (by the early 2030s) increase U.S. petroleum consumption by about 0.5 million barrels per day—about two to three percent of projected total U.S. consumption. While significant, this additional petroleum consumption is, from an economic perspective, dwarfed by the cost savings also projected to result from today's proposal, as indicated by the consideration of net benefits appearing below.
Safety Benefits From Preferred Alternative
Today's proposed rule is anticipated to prevent more than 12,700 on-road fatalities 
and significantly more injuries as compared to the standards set forth in the 2012 final rule over the lifetimes of vehicles as more new, safer vehicles are purchased than the current (and augural) standards. A large portion of these safety benefits will come from improved fleet turnover as more consumers will be able to afford newer and safer vehicles.
Recent NHTSA analysis shows that the proportion of passengers killed in a vehicle 18 or more model years old is nearly double that of a vehicle three model years old or newer.
As the average car on the road is approaching 12 years old, apparently the oldest in our history,
major safety benefits will occur by reducing fleet age. Other safety benefits will occur from other areas such as avoiding the increased driving Start Printed Page 42996that would otherwise result from higher fuel efficiency (known as the rebound effect) and avoiding the mass reductions in passenger cars that might otherwise be required to meet the standards established in 2012.
Together these and other factors lead to estimated annual fatalities under the proposed standards that are significantly reduced 
relative to those that would occur under current (and augural) standards.
The Preferred Alternative Would Have Negligible Environmental Impacts on Air Quality
Improving fleet turnover will result in consumers getting into newer and cleaner vehicles, accelerating the rate at which older, more-polluting vehicles are removed from the roadways. Also, reducing fuel economy (relative to levels that would occur under previously-issued standards) would increase the marginal cost of driving newer vehicles, reducing mileage accumulated by those vehicles, and reducing corresponding emissions. On the other hand, increasing fuel consumption would increase emissions resulting from petroleum refining and related “upstream” processes. Our analysis shows that none of the regulatory alternatives considered in this proposal would noticeably impact net emissions of smog-forming or other “criteria” or toxic air pollutants, as illustrated by the following graph. That said, the resultant tailpipe emissions reductions should be especially beneficial to highly trafficked corridors.
Climate Change Impacts From Preferred Alternative
The estimated effects of this proposal in terms of fuel savings and CO2 emissions, again perhaps somewhat counter-intuitively, is relatively small as compared to the 2012 final rule.
NHTSA's Environmental Impact Statement performed for this rulemaking shows that the preferred alternative would result in 3/1,000ths of a degree Celsius increase in global average temperatures by 2100, relative to the standards finalized in 2012. On a net CO2 basis, the results are similarly minimal. The following graph compares the estimated atmospheric CO2 concentration (789.76 ppm) in 2100 under the proposed standards to the estimated level (789.11 ppm) under the standards set forth in 2012—or an 8/100ths of a percentage increase:
Start Printed Page 42997
Net Benefits From Preferred Alternative
Maintaining the MY 2020 curves for MYs 2021-2026 will save American consumers, the auto industry, and the public a considerable amount of money as compared to if EPA retained the previously-set CO2 standards and NHTSA finalized the augural standards. This was identified as the preferred alternative, in part, because it maximizes net benefits compared to the other alternatives analyzed, recognizing the statutory considerations for both agencies. Comment is sought on whether this is an appropriate basis for selection.
Start Printed Page 42998
These estimates, reported as changes relative to impacts under the standards issued in 2012, account for impacts on vehicles produced during model years 2016-2029, as well as (through changes in utilization) vehicles produced in earlier model years, throughout those vehicles' useful lives. Reported values are in 2016 dollars, and reflect three-percent and seven-percent discount rates. Under CAFE standards, costs are estimated to decrease by $502 billion overall at a three-percent discount rate ($335 billion at a seven-percent discount rate); benefits are estimated to decrease by $326 billion at a three-percent discount rate ($204 billion at a seven-percent discount rate). Thus, net benefits are estimated to increase by $176 billion at a three-percent discount rate and $132 billion at a seven-percent discount rate. The estimated impacts under CO2 standards are similar, with net benefits estimated to increase by $201 billion at a three-percent discount rate and $141 billion at a seven-percent discount rate.
This proposal also seeks comment on a variety of changes to NHTSA's and EPA's compliance programs for CAFE and CO2 as well as related programs. Compliance flexibilities can generally be grouped into two categories. The first category are those compliance flexibilities that reduce unnecessary compliance costs and provide for a more efficient program. The second category of compliance flexibilities are those that distort the market—such as by incentivizing the implementation of one type of technology by providing credit for compliance in excess of real-world fuel savings.
Both programs provide for the generation of credits based upon fleet-wide over-compliance, provide for adjustments to the test measured value of each individual vehicle based upon the implementation of certain fuel saving technologies, and provide additional incentives for the implementation of certain preferred technologies (regardless of actual fuel savings). Auto manufacturers and others have petitioned for a host of additional adjustment- and incentive-type flexibilities, where there is not always consumer interest in the technologies to be incentivized nor is there necessarily clear fuel-saving and emissions-reducing benefit to be derived from that incentivization. The agencies seek comment on all of those requests as part of this proposal.
Over-compliance credits, which can be built up in part through use of the above-described per-vehicle adjustments and incentives, can be saved and either applied retroactively to accounts for previous non-compliance, or carried forward to mitigate future non-compliance. Such credits can also be traded to other automakers for cash or for other credits for different fleets. But such trading is not pursued openly. Under the CAFE program, the public is not made aware of inter-automaker trades, nor are shareholders. And even the agencies are not informed of the price of credits. With the exception of statutorily-mandated credits, the agencies seek comment on all aspects of the current system. The agencies are particularly interested in comments on flexibilities that may distort the market. Start Printed Page 42999The agencies seek comment as to whether some adjustments and non-statutory incentives and other provisions should be eliminated and stringency levels adjusted accordingly. In general, well-functioning banking and trading provisions increase market efficiency and reduce the overall costs of compliance with regulatory objectives. The agencies request comment on whether the current system as implemented might need improvements to achieve greater efficiencies. We seek comment on specific programmatic changes that could improve compliance with current standards in the most efficient way, ranging from requiring public disclosure of some or all aspects of credit trades, to potentially eliminating credit trading in the CAFE program. We request commenters to provide any data, evidence, or existing literature to help agency decision-making.
One National Standard
Setting appropriate and maximum feasible fuel economy and tailpipe CO2 emissions standards requires regulatory efficiency. This proposal addresses a fundamental and unnecessary complication in the currently-existing regulatory framework, which is the regulation of GHG emissions from passenger cars and light trucks by the State of California through its GHG standards and Zero Emission Vehicle (ZEV) mandate and subsequent adoption of these standards by other States. Both EPCA and the CAA preempt State regulation of motor vehicle emissions (in EPCA's case, standards that are related to fuel economy standards). The CAA gives EPA the authority to waive preemption for California under certain circumstances. EPCA does not provide for a waiver of preemption under any circumstances. In short, the agencies propose to maintain one national standard—a standard that is set exclusively by the Federal government.
Proposed Withdrawal of California's Clean Air Act Preemption Waiver
EPA granted a waiver of preemption to California in 2013 for its “Advanced Clean Car” regulations, composed of its GHG standards, its “Low Emission Vehicle (LEV)” program and the ZEV program,
and, as allowed under the CAA, a number of other States adopted California's standards.
The CAA states that EPA shall not grant a waiver of preemption if EPA finds that California's determination that its standards are, in the aggregate, at least as protective of public health and welfare as applicable Federal standards, is arbitrary and capricious; that California does not need its own standards to meet compelling or extraordinary conditions; or that such California standards and accompanying enforcement procedures are not consistent with Section 202(a) of the CAA. In this proposal, EPA is proposing to withdraw the waiver granted to California in 2013 for the GHG and ZEV requirements of its Advanced Clean Cars program, in light of all of these factors.
Attempting to solve climate change, even in part, through the Section 209 waiver provision is fundamentally different from that section's original purpose of addressing smog-related air quality problems. When California was merely trying to solve its air quality issues, there was a relatively-straightforward technology solution to the problems, implementation of which did not affect how consumers lived and drove. Section 209 allowed California to pursue additional reductions to address its notorious smog problems by requiring more stringent standards, and allowed California and other States that failed to comply with Federal air quality standards to make progress toward compliance. Trying to reduce carbon emissions from motor vehicles in any significant way involves changes to the entire vehicle, not simply the addition of a single or a handful of control technologies. The greater the emissions reductions are sought, the greater the likelihood that the characteristics and capabilities of the vehicle currently sought by most American consumers will have to change significantly. Yet, even decades later, California continues to be in widespread non-attainment with Federal air quality standards.
In the past decade, California has disproportionately focused on GHG emissions. Parts of California have a real and significant local air pollution problem, but CO2 is not part of that local problem.
California's Tailpipe CO2 Emissions Standards and ZEV Mandate Conflict With EPCA
Moreover, California regulation of tailpipe CO2 emissions, both through its GHG standards and ZEV program, conflicts directly and indirectly with EPCA and the CAFE program. EPCA expressly preempts State standards related to fuel economy. Tailpipe CO2 standards, whether in the form of fleet-wide CO2 limits or in the form of requirements that manufacturers selling vehicles in California sell a certain number of low- and no-tailpipe-CO2 emissions vehicles as part of their overall sales, are unquestionably related to fuel economy standards. Standards that control tailpipe CO2 emissions are de facto fuel economy standards because CO2 is a direct and inevitable byproduct of the combustion of carbon-based fuels to make energy, and the vast majority of the energy that powers passenger cars and light trucks comes from carbon-based fuels.
Improving fuel economy means getting the vehicle to go farther on a gallon of gas; a vehicle that goes farther on a gallon of gas produces less CO2 per unit of distance; therefore, improving fuel economy necessarily reduces tailpipe CO2 emissions, and reducing CO2 emissions necessarily improves fuel economy. EPCA therefore necessarily preempts California's Advanced Clean Cars program to the extent that it regulates or prohibits tailpipe CO2 emissions. Section VI of this proposal, below, discusses the CAA waiver and EPCA preemption in more detail.
Eliminating California's regulation of fuel economy pursuant to Congressional direction will provide benefits to the American public. The automotive industry will, appropriately, deal with fuel economy standards on a national basis—eliminating duplicative regulatory requirements. Further, elimination of California's ZEV program will allow automakers to develop such vehicles in response to consumer demand instead of regulatory mandate. This regulatory mandate has required automakers to spend tens of billions of dollars to develop products that a significant majority of consumers have not adopted, and consequently to sell such products at a loss. All of this is paid for through cross subsidization by increasing prices of other vehicles not just in California and other States that have adopted California's ZEV mandate, but throughout the country.
Request for Comment
The agencies look forward to all comments on this proposal, and wish to emphasize that obtaining public input is extremely important to us in selecting from among the alternatives in a final rule. While the agencies and the Administration met with a variety of stakeholders prior to issuance of this proposal, those meetings have not resulted in a predetermined final rule outcome. The Administrative Procedure Act requires that agencies provide the Start Printed Page 43000public with adequate notice of a proposed rule followed by a meaningful opportunity to comment on the rule's content. The agencies are committed to following that directive.
II. Technical Foundation for NPRM Analysis
A. Basics of CAFE and CO 2 Standards Analysis
The agencies' analysis of CAFE and CO2 standards involves two basic elements: first, estimating ways each manufacturer could potentially respond to a given set of standards in a manner that considers potential consumer response; and second, estimating various impacts of those responses. Estimating manufacturers' potential responses involves simulating manufacturers' decision-making processes regarding the year-by-year application of fuel-saving technologies to specific vehicles. Estimating impacts involves calculating resultant changes in new vehicle costs, estimating a variety of costs (e.g., for fuel) and effects (e.g., CO2 emissions from fuel combustion) occurring as vehicles are driven over their lifetimes before eventually being scrapped, and estimating the monetary value of these effects. Estimating impacts also involves consideration of the response of consumers—e.g., whether consumers will purchase the vehicles and in what quantities. Both of these basic analytical elements involve the application of many analytical inputs.
The agencies' analysis uses the CAFE model to estimate manufacturers' potential responses to new CAFE and CO2 standards and to estimate various impacts of those responses. The model makes use of many inputs, values of which are developed outside of the model and not by the model. For example, the model applies fuel prices; it does not estimate fuel prices. The model does not determine the form or stringency of the standards; instead, the model applies inputs specifying the form and stringency of standards to be analyzed and produces outputs showing effects of manufacturers working to meet those standards, which become the basis for comparing between different potential stringencies.
DOT's Volpe National Transportation Systems Center (often simply referred to as the “Volpe Center”) develops, maintains, and applies the model for NHTSA. NHTSA has used the CAFE model to perform analyses supporting every CAFE rulemaking since 2001, and the 2016 rulemaking regarding heavy-duty pickup and van fuel consumption and GHG emissions also used the CAFE model for analysis.
DOT recently arranged for a formal peer review of the model. In general, reviewers' comments strongly supported the model's conceptual basis and implementation, and commenters provided several specific recommendations. DOT staff agreed with many of these recommendations and have worked to implement them wherever practicable. Implementing some of them would require considerable further research, development, and testing, and will be considered going forward. For a handful of other recommendations, DOT staff disagreed, often finding the recommendations involved considerations (e.g., other policies, such as those involving fuel taxation) beyond the model itself or were based on concerns with inputs rather than how the model itself functioned. A report available in the docket for this rulemaking presents peer reviewers' detailed comments and recommendations, and provides DOT's detailed responses.
The agencies also use four DOE and DOE-sponsored models to develop inputs to the CAFE model, including three developed and maintained by DOE's Argonne National Laboratory. The agencies use the DOE Energy Information Administration's (EIA's) National Energy Modeling System (NEMS) to estimate fuel prices,
and used Argonne's Greenhouse gases, Regulated Emissions, and Energy use in Transportation (GREET) model to estimate emissions rates from fuel production and distribution processes.
DOT also sponsored DOE/Argonne to use their Autonomie full-vehicle simulation system to estimate the fuel economy impacts for roughly a million combinations of technologies and vehicle types. 
EPA developed two models after 2009, referred to as the “ALPHA” and “OMEGA” models, which provide some of the same capabilities as the Autonomie and CAFE models. EPA applied the OMEGA model to conduct analysis of GHG standards promulgated in 2010 and 2012, and the ALPHA and OMEGA models to conduct analysis discussed in the above-mentioned 2016 Draft TAR and Proposed and Final Determinations regarding standards beyond 2021. In an August 2017 notice, the agencies requested comments on, among other things, whether EPA should use alternative methodologies and modeling, including DOE/Argonne's Autonomie full-vehicle simulation tool and DOT's CAFE model.
Having reviewed comments on the subject and having considered the matter fully, the agencies have determined it is reasonable and appropriate to use DOE/Argonne's model for full-vehicle simulation, and to use DOT's CAFE model for analysis of regulatory alternatives. EPA interprets Section 202(a) of the CAA as giving the agency broad discretion in how it develops and sets GHG standards for light-duty vehicles. Nothing in Section 202(a) mandates that EPA use any specific model or set of models for analysis of potential CO2 standards for light-duty vehicles. EPA weighs many factors when determining appropriate levels for CO2 standards, including the cost of compliance (see Section 202(a)(2)), lead time necessary for compliance (also Section 202(a)(2)), safety (see NRDC v. EPA, 655 F.2d 318, 336 n. 31 (D.C. Cir. 1981) and other impacts on consumers,
and energy impacts associated with use of the technology.
Using the CAFE model Start Printed Page 43001allows consideration of the following factors: the CAFE model explicitly evaluates the cost of compliance for each manufacturer, each fleet, and each model year; it accounts for lead time necessary for compliance by directly incorporating estimated manufacturer production cycles for every vehicle in the fleet, ensuring that the analysis does not assume vehicles can be redesigned to incorporate more technology without regard to lead time considerations; it provides information on safety effects associated with different levels of standards and information about many other impacts on consumers, and it calculates energy impacts (i.e., fuel saved or consumed) as a primary function, besides being capable of providing information about many other factors within EPA's broad CAA discretion to consider.
Because the CAFE model simulates a wide range of actual constraints and practices related to automotive engineering, planning, and production, such as common vehicle platforms, sharing of engines among different vehicle models, and timing of major vehicle redesigns, the analysis produced by the CAFE model provides a transparent and realistic basis to show pathways manufacturers could follow over time in applying new technologies, which helps better assess impacts of potential future standards. Furthermore, because the CAFE model also accounts fully for regulatory compliance provisions (now including CO2 compliance provisions), such as adjustments for reduced refrigerant leakage, production “multipliers” for some specific types of vehicles (e.g., PHEVs), and carried-forward (i.e., banked) credits, the CAFE model provides a transparent and realistic basis to estimate how such technologies might be applied over time in response to CAFE or CO2 standards.
There are sound reasons for the agencies to use the CAFE model going forward in this rulemaking. First, the CAFE and CO2 fact analyses are inextricably linked. Furthermore, the analysis produced by the CAFE model and DOE/Argonne's Autonomie addresses several analytical needs. The CAFE model provides an explicit year-by-year simulation of manufacturers' application of technology to their products in response to a year-by-year progression of CAFE standards and accounts for sharing of technologies and the implications for timing, scope, and limits on the potential to optimize powertrains for fuel economy. In the real world, standards actually are specified on a year-by-year basis, not simply some single year well into the future, and manufacturers' year-by-year plans involve some vehicles “carrying forward” technology from prior model years and some other vehicles possibly applying “extra” technology in anticipation of standards in ensuing model years, and manufacturers' planning also involves applying credits carried forward between model years. Furthermore, manufacturers cannot optimize the powertrain for fuel economy on every vehicle model configuration—for example, a given engine shared among multiple vehicle models cannot practicably be split into different versions for each configuration of each model, each with a slightly different displacement. The CAFE model is designed to account for these real-world factors.
Considering the technological heterogeneity of manufacturers' current product offerings, and the wide range of ways in which the many fuel economy-improving/CO2 emissions-reducing technologies included in the analysis can be combined, the CAFE model has been designed to use inputs that provide an estimate of the fuel economy achieved for many tens of thousands of different potential combinations of fuel-saving technologies. Across the range of technology classes encompassed by the analysis fleet, today's analysis involves more than a million such estimates. While the CAFE model requires no specific approach to developing these inputs, the National Academy of Sciences (NAS) has recommended, and stakeholders have commented, that full-vehicle simulation provides the best balance between realism and practicality. DOE/Argonne has spent several years developing, applying, and expanding means to use distributed computing to exercise its Autonomie full-vehicle simulation tool over the scale necessary for realistic analysis of CAFE or average CO2 standards. This scalability and related flexibility (in terms of expanding the set of technologies to be simulated) makes Autonomie well-suited for developing inputs to the CAFE model.
Additionally, DOE/Argonne's Autonomie also has a long history of development and widespread application by a much wider range of users in government, academia, and industry. Many of these users apply Autonomie to inform funding and design decisions. These real-world exercises have contributed significantly to aspects of Autonomie important to producing realistic estimates of fuel economy levels and CO2 emission rates, such as estimation and consideration of performance, utility, and driveability metrics (e.g., towing capability, shift business, frequency of engine on/off transitions). This steadily increasing realism has, in turn, steadily increased confidence in the appropriateness of using Autonomie to make significant investment decisions. Notably, DOE uses Autonomie for analysis supporting budget priorities and plans for programs managed by its Vehicle Technologies Office (VTO). Considering the advantages of DOE/Argonne's Autonomie model, it is reasonable and appropriate to use Autonomie to estimate fuel economy levels and CO2 emission rates for different combinations of technologies as applied to different types of vehicles.
Commenters have also suggested that the CAFE model's graphical user interface (GUI) facilitates others' ability to use the model quickly—and without specialized knowledge or training—and to comment accordingly.
For today's proposal, DOT has significantly expanded and refined this GUI, providing the ability to observe the model's real-time progress much more closely as it simulates year-by-year compliance with either CAFE or CO2 standards.
Although the model's ability to produce realistic results is independent of the model's GUI, it is anticipated the CAFE model's GUI will facilitate stakeholders' meaningful review and comment during the comment period.
Beyond these general considerations, several specific related technical comments and considerations underlie the agencies' decision in this area, as discussed, where applicable, in the remainder of this Section.
Other commenters expressed a number of concerns with whether DOT's CAFE model could be used for CAA analysis. Many of these concerns focused on inputs used by the CAFE model for prior rulemaking analyses.  
Because inputs are Start Printed Page 43002exogenous to any model, they do not determine whether it would be reasonable and appropriate for EPA to use DOT's model for analysis. Other concerns focused on characteristics of the CAFE model that were developed to better align the model with EPCA and EISA; the model has been revised to accommodate both EPCA/EISA and CAA analysis, as explained further below. Some commenters also argued that use of any models other than ALPHA and OMEGA for CAA analysis would constitute an arbitrary and capricious delegation of EPA's decision-making authority to DOT, if DOT models are used for analysis instead. These comments were made prior to the development of the CAA analysis function in the CAFE model, and, moreover, appear to conflate the analytical tool used to inform decision-making with the action of making the decision. As explained elsewhere in this document and as made repeatedly clear over the past several rulemakings, the CAFE model neither sets standards nor dictates where and how to set standards; it simply informs as to the effects of setting different levels of standards. In this rulemaking, EPA will be making its own decisions regarding what CO2 standards would be appropriate under the CAA. The CAA does not require EPA to create a specific model or use a specific model of its own creation in setting GHG standards. The fact EPA's decision may be informed by non-EPA-created models does not, in any way, constitute a delegation of its statutory power to set standards or decision-making authority.
Arguing to the contrary would suggest, for example, that EPA's decision would be invalid because it relied on EIA's Annual Energy Outlook for fuel prices rather than developing its own model for estimating future trends in fuel prices. Yet, all Federal agencies that have occasion to use forecasts of future fuel prices regularly (and appropriately) defer to EIA's expertise in this area and rely on EIA's NEMS-based analysis in the AEO, even when those same agencies are using EIA's forecasts to inform their own decision-making.
Moreover, DOT's CAFE model with inputs from DOE/Argonne's Autonomie model has produced analysis supporting rulemaking under the CAA. In 2015, EPA proposed new GHG standards for MY 2021-2027 heavy-duty pickups and vans, finalizing standards in 2016. Supporting the NPRM and final rule, EPA relied on analysis implemented by DOT using DOT's CAFE model, and DOT used inputs developed by DOE/Argonne using DOE/Argonne's Autonomie model.
The following sections provide a brief technical overview of the CAFE model, including changes NHTSA made to the model since 2012, before discussing inputs to the model and then diving more deeply into how the model works. For more information on the latter topic, see the CAFE model documentation July 2018 draft, available in the docket for this rulemaking and on NHTSA's website.
1. Brief Technical Overview of the Model
The CAFE model is designed to simulate compliance with a given set of CAFE or CO2 standards for each manufacturer selling vehicles in the United States. The model begins with a representation of the current (for today's analysis, model year 2016) vehicle model offerings for each manufacturer that includes the specific engines and transmissions on each model variant, observed sales volumes, and all fuel economy improvement technology that is already present on those vehicles. From there it adds technology, in response to the standards being considered, in a way that minimizes the cost of compliance and reflects many real-world constraints faced by automobile manufacturers. After simulating compliance, the model calculates impacts of the simulated standard: technology costs, fuel savings (both in gallons and dollars), CO2 reductions, social costs and benefits, and safety impacts.
Today's analysis reflects several changes made to the CAFE model since 2012, when NHTSA used the model to estimate the effects, costs, and benefits of final CAFE standards for light-duty vehicles produced during MYs 2017-2021 and augural standards for MYs 2022-2025. Key changes relevant to this analysis include the following:
- Expansion of model inputs, procedures, and outputs to accommodate technologies not included in prior analyses,
- Updated approach to estimating the combined effect of fuel-saving technologies using large scale simulation modeling,
- Modules that dynamically estimate new vehicle sales and existing vehicle scrappage in response to changes to new vehicle prices that result from manufacturers' compliance actions,
- A safety module that estimates the changes in light-duty traffic fatalities resulting from changes to vehicle exposure, vehicle retirement rates, and reductions in vehicle mass to improve fuel economy,
- Disaggregation of each manufacturer's fleet into separate “domestic” passenger car and “import” passenger car fleets to better represent the statutory requirements of the CAFE program,
- Changes to the algorithm used to apply technologies, enabling more explicit accounting of shared vehicle components (engines, transmissions, platforms) and “inheritance” of major technology within or across powertrains and/or platforms over time,
- An industry labor quantity module, which estimates net changes in the amount of U.S. automobile labor for dealerships, Tier 1 and 2 supplier companies, and original equipment manufacturers (OEMs),
- Cost estimation of batteries for electrification technologies incorporates an updated version of Argonne National Laboratory's BatPAC (battery) model for hybrid electric vehicles (HEVs), plug-in Start Printed Page 43003hybrid electric vehicles (PHEVs), and battery electric vehicles (BEVs), consistent with how we estimate effectiveness for those values,
- Expanded accounting for CAFE credits carried over from years prior to those included in the analysis (a.k.a. “banked” credits) and application to future CAFE deficits to better evaluate anticipated manufacturer responses to proposed standards,
- The ability to represent a manufacturer's preference for fine payment (rather than achieving full compliance exclusively through fuel economy improvements) on a year-by-year basis,
- Year-by-year simulation of how manufacturers could comply with EPA's CO2 standards, including
○ Calculation of vehicle models' CO2 emission rates before and after application of fuel-saving (and, therefore, CO2-reducing) technologies,
○ Calculation of manufacturers' fleet average CO2 emission rates,
○ Calculation of manufacturers' fleet average CO2 emission rates under attribute-based CO2 standards,
○ Accounting for adjustments to average CO2 emission rates reflecting reduction of air conditioner refrigerant leakage,
○ Accounting for the treatment of alternative fuel vehicles for CO2 compliance,
○ Accounting for production “multipliers” for PHEVs, BEVs, compressed natural gas (CNG) vehicles, and fuel cell vehicles (FCVs),
○ Accounting for transfer of CO2 credits between regulated fleets,
○ Accounting for carried-forward (a.k.a. “banked”) CO2 credits, including credits from model years earlier than modeled explicitly.
2. Sensitivity Cases and Why We Examine Them
Today's notice presents estimated impacts of the proposed CAFE and CO2 standards defining the proposals, relative to a baseline “no action” regulatory alternative under which the standards announced in 2012 remain in place through MY 2025 and continue unchanged thereafter. Relative to this same baseline, today's notice also presents analysis estimating impacts under a range of other regulatory alternatives the agencies are considering. All but one involve different standards, and three involve a gradual discontinuation of CAFE and GHG adjustments reflecting the application of technologies that improve air conditioner efficiency or, in other ways, improve fuel economy under conditions not represented by long-standing fuel economy test procedures. Like the baseline no action alternative, all of these alternatives are more stringent than the preferred alternative. Section III and Section IV describe the preferred and other regulatory alternatives, respectively.
These alternatives were examined because they will be considered as options for the final rule. The agencies seek comment on these alternatives, seek any relevant data and information, and will review responses. That review could lead to the selection of one of the other regulatory alternatives for the final rule or some combination of the other regulatory alternatives (e.g., combining passenger cars standards from one alternative with light truck standards from a different alternative).
Because outputs depend on inputs (e.g., the results of the analysis in terms of quantities and kinds of technologies required to meet different levels of standards, and the societal and private benefits associated with manufacturers meeting different levels of standards depend on input data, estimates, and assumptions), the analysis also explores the sensitivity of results to many of these inputs. For example, the net benefits of any regulatory alternative will depend strongly on fuel prices well beyond 2025. Fuel prices a decade and more from now are not knowable with certainty. The sensitivity analysis involves repeating the “central” or “reference case” analysis under alternative inputs (e.g., higher fuel prices in one case, lower fuel prices in another case), and exploring changes in analytical results, which is discussed further in the agencies' Preliminary Regulatory Impact Analysis (PRIA) accompanying today's notice.
B. Developing the Analysis Fleet for Assessing Costs, Benefits, and Effects of Alternative CAFE Standards
The following sections describe what the analysis fleet is and why it is used, how it was developed for this NPRM, and the analysis-fleet-related topics on which comment is sought.
1. Purpose of Developing and Using an Analysis Fleet
The starting point for the evaluation of the potential feasibility of different stringency levels for future CAFE and CO2 standards is the analysis fleet, which is a snapshot of the recent vehicle market. The analysis fleet provides a snapshot to project what vehicles will exist in future model years covered by the standards and what technologies they will have, and then evaluate what additional technologies can feasibly be applied to those vehicles in a cost-effective way to raise their fuel economy and lower their CO2 emission levels.
Part of reflecting what vehicles will exist in future model years is knowing which vehicles are produced by which manufacturers, how many of each are sold, and whether they are passenger cars or light trucks. This is important because it improves our understanding of the overall impacts of different levels of CAFE and CO2 standards; overall impacts result from industry's response to standards, and industry's response is made up of individual manufacturer responses to the standards in light of the overall market and their individual assessment of consumer acceptance. Having an accurate picture of manufacturers' existing fleets (and the vehicle models in them) that will be subject to future standards helps us better understand individual manufacturer responses to those future standards in addition to potential changes in those standards.
Another part of reflecting what vehicles will exist in future model years is knowing what technologies are already on those vehicles. Accounting for technologies already being on vehicles helps avoid “double-counting” the value of those technologies, by assuming they are still available to be applied to improve fuel economy and reduce CO2 emissions. It also promotes more realistic determinations of what additional technologies can feasibly be applied to those vehicles: if a manufacturer has already started down a technological path to fuel economy or performance improvements, we do not assume it will completely abandon that path because that would be unrealistic and would not accurately represent manufacturer responses to standards. Each vehicle model (and configurations of each model) in the analysis fleet, therefore, has a comprehensive list of its technologies, which is important because different configurations may have different technologies applied to Start Printed Page 43004them.
Additionally, the analysis accounts for platforms within manufacturers' fleets, recognizing platforms will share technologies, and the vehicles that make up that platform should receive (or not receive) additional technological improvements together. The specific engineering characteristics of each model/configuration are available in the aforementioned input file.
For the regulatory alternatives considered in today's proposal, estimates of rates at which various technologies might be expected to penetrate manufacturers' fleets (and the overall market) are summarized below in Sections VI and VII, and in Chapter 6 of the accompanying PRIA and in detailed model output files available at NHTSA's website. A solid characterization of a recent model year as an analytical starting point helps to realistically estimate ways manufacturers could potentially respond to different levels of standards, and the modeling strives to realistically simulate how manufacturers could progress from that starting point. Nevertheless, manufacturers can respond in many ways beyond those represented in the analysis (e.g., applying other technologies, shifting production volumes, changing vehicle footprint), such that it is impossible to predict with any certainty exactly how each manufacturer will respond. Therefore, recent trends in manufacturer performance and technology application, although of interest in terms of understanding manufacturers' current compliance positions, are not in themselves innately indicative of future potential.
Yet, another part of reflecting what vehicles will exist in future model years is having reasonable real-world assumptions about when certain technologies can be applied to vehicles. Each vehicle model/configuration in the analysis fleet also has information about its redesign schedule, i.e., the last year it was redesigned and when the agencies expect it to be redesigned again. Redesign schedules are a key part of manufacturers' business plans, as each new product can cost more than $1.0B and involve a significant portion of a manufacturer's scarce research, development, and manufacturing and equipment budgets and resources.
Manufacturers have repeatedly told the agencies that sustainable business plans require careful management of resources and capital spending, and that the length of time each product remains in production is crucial to recouping the upfront product development and plant/equipment costs, as well as the capital needed to fund the development and manufacturing equipment needed for future products. Because the production volume of any given vehicle model varies within a manufacturer's product line and also varies among different manufacturers, redesign schedules typically vary for each model and manufacturer. Some (relatively few) technological improvements are small enough they can be applied in any model year; others are major enough they can only be cost-effectively applied at a vehicle redesign, when many other things about the vehicle are already changing. Ensuring the CAFE model makes technological improvements to vehicles only when it is feasible to do so also helps the analysis better represent manufacturer responses to different levels of standards.
A final important aspect of reflecting what vehicles will exist in future model years and potential manufacturer responses to standards is estimating how future sales might change in response to different potential standards. If potential future standards appear likely to have major effects in terms of shifting production from cars to trucks (or vice versa), or in terms of shifting sales between manufacturers or groups of manufacturers, that is important for the agencies to consider. For previous analyses, the CAFE model used a static forecast contained in the analysis fleet input file, which specified changes in production volumes over time for each vehicle model/configuration. This approach yielded results that, in terms of production volumes, did not change between scenarios or with changes in important model inputs. For example, very stringent standards with very high technology costs would result in the same estimated production volumes as less stringent standards with very low technology costs.
New for today's proposal, the CAFE model begins with the first-year production volumes (i.e., MY 2016 for today's analysis) and adjusts ensuing sales mix year by year (between cars and trucks, and between manufacturers) endogenously as part of the analysis, rather than using external forecasts of future car/truck split and future manufacturer sales volumes. This leads the model to produce different estimates of future production volumes under different standards and in response to different inputs, reflecting the expectation that regulatory standards and other external factors will, in fact, impact the market.
The input file for the CAFE model characterizing the analysis fleet 
includes a large amount of data about vehicle models/configurations, their technological characteristics, the manufacturers and fleets to which they belong, and initial prices and production volumes which provide the starting points for projection (by the sales model) to ensuing model years. The following sections discuss aspects of how the analysis fleet was built for this proposal and seek comment on those topics.
2. Source Data for Building the Analysis Fleet
The source data for the vehicle models/configurations in the analysis fleet and their technologies is a central input for the analysis. The sections below discuss pros and cons of different potential sources and what was used for this proposal.
(a) Use of Confidential Business Information Versus Publicly-Releasable Sources
Since 2001, CAFE analysis has used either confidential, forward-estimating product plans from manufacturers, or publicly available data on vehicles already sold, as a starting point for determining what technologies can be applied to what vehicles in response to potential different levels of standards. These two sources present a tradeoff. Confidential product plans comprehensively represent what vehicles a manufacturer expects to produce in coming years, accounting for plans to introduce new vehicles and fuel-saving technologies and, for example, plans to discontinue other vehicles and even brands. This information can be very thorough and can improve the accuracy of the analysis, but for competitive reasons, most of this information must be redacted prior to publication with rulemaking documentation. This makes it difficult for public commenters to reproduce the analysis for themselves as Start Printed Page 43005they develop their comments. Some non-industry commenters have also expressed concern manufacturers would have an incentive in the submitted plans to (deliberately or not) underestimate their future fuel economy capabilities and overstate their expectations about, for example, the levels of performance of future vehicle models in order to affect the analysis. Since 2010, EPA and NHTSA have based analysis fleets almost exclusively on information from commercial and public sources, starting with CAFE compliance data and adding information from other sources.
An analysis fleet based primarily on public sources can be released to the public, solving the issue of commenters being unable to reproduce the overall analysis when they want to. However, industry commenters have argued such an analysis fleet cannot accurately reflect manufacturers actual plans to apply fuel-saving technologies (e.g., manufacturers may apply turbocharging to improve not just fuel economy, but also to improve vehicle performance) or manufacturers' plans to change product offerings by introducing some vehicles and brands and discontinuing other vehicles and brands, precisely because that information is typically confidential business information (CBI). A fully-publicly-releasable analysis fleet holds vehicle characteristics unchanged over time and arguably lacks some level of accuracy when projected into the future. For example, over time, manufacturers introduce new products and even entire brands. On the other hand, plans announced in press releases do not always ultimately bear out, nor do commercially-available third-party forecasts. Assumptions could be made about these issues to improve the accuracy of a publicly-releasable analysis fleet, but concerns include that this information would either be largely incorrect, or information would be released that manufacturers would consider CBI. We seek comment on how to address this issue going forward, recognizing the competing interests involved and also recognizing typical timeframes for CAFE and CO2 standards rulemakings.
(b) Use of MY 2016 CAFE Compliance Data Versus Other Starting Points
Based on the assumption that a publicly-available analysis fleet continues to be desirable, for this NPRM, an analysis fleet was constructed starting with CAFE compliance information from manufacturers.
Information from MY 2016 was chosen as the foundation for today's analysis fleet because, at the time the rulemaking analysis was initiated, the 2016 fleet represented the most up-to-date information available in terms of individual vehicle models and configurations, production technology levels, and production volumes. If MY 2017 data had been used while this analysis was being developed, the agencies would have needed to use product planning information that could not be made available to the public until a later date.
The analysis fleet was initially developed with 2016 mid-model year compliance data because final compliance data was not available at that time, and the timing provided manufacturers the opportunity to review and comment on the characterization of their vehicles in the fleet. With a view toward developing an accurate characterization of the 2016 fleet to serve as an analytical starting point, corrections and updates to mid-year data (e.g., to production estimates) were sought, in addition to corroboration or correction of technical information obtained from commercial and other sources (to the extent that information was not included in compliance data), although future product planning information from manufacturers (e.g., future product offerings, products to be discontinued) was not requested, as most manufacturers view such information as CBI. Manufacturers offered a range of corrections to indicate engineering characteristics (e.g., footprint, curb weight, transmission type) of specific vehicle model/configurations, as well as updates to fuel economy and production volume estimates in mid-year reporting. After following up on a case-by-case basis to investigate significant differences, the analysis fleet was updated.
Sales, footprint, and fuel economy values with final compliance data were also updated if that data was available. In a few cases, final production and fuel economy values may be slightly different for specific model year 2016 vehicle models and configurations than are indicated in today's analysis; however, other vehicle characteristics (e.g., footprint, curb weight, technology content) important to the analysis should be accurate. While some commenters have, in the past, raised concerns that non-final CAFE compliance data is subject to change, the potential for change is likely not significant enough to merit using final data from an earlier model year reflecting a more outdated fleet. Moreover, even ostensibly final CAFE compliance data can sometimes be subject to later revision (e.g., if errors in fuel economy tests are discovered), and the purpose of today's analysis is not to support enforcement actions but rather to provide a realistic assessment of manufacturers' potential responses to future standards.
Manufacturers integrated a significant amount of new technology in the MY 2016 fleet, and this was especially true for newly-designed vehicles launched in MY 2016. While subsequent fleets will involve even further application of technology, using available data for MY 2016 provides the most realistic detailed foundation for analysis that can be made available publicly in full detail, allowing stakeholders to independently reproduce the analysis presented in this proposal. Insofar as future product offerings are likely to be more similar to vehicles produced in 2016 than to vehicles produced in earlier model years, using available data regarding the 2016 model year provides the most realistic, publicly releasable foundation for constructing a forecast of the future vehicle market for this proposal.
A number of comments to the Draft TAR, EPA's Proposed Determination, and EPA's 2017 Request for Comment 
stated that the most up-to-date analysis fleet possible should be used, because a more up-to-date analysis fleet will better capture how manufacturers apply technology and will account better for vehicle model/configuration introductions and deletions.
On the other hand, some commenters suggested that because manufacturers continue improving vehicle performance and utility over time, an older analysis fleet should be used to estimate how the fleet could have evolved had manufacturers applied all technological potential to Start Printed Page 43006fuel economy rather than continuing to improve vehicle performance and utility.
Because manufacturers change and improve product offerings over time, conducting analysis with an older analysis fleet (or with a fleet using fuel economy levels and CO2 emissions rates that have been adjusted to reflect an assumed return to levels of performance and utility typical of some past model year) would miss this real-world trend. While such an analysis could demonstrate what industry could do if, for example, manufacturers devoted all technological improvements toward raising fuel economy and reducing CO2 emissions (and if consumers decided to purchase these vehicles), we do not believe it would be consistent with a transparent examination of what effects different levels of standards would have on individual manufacturers and the fleet as a whole.
Generally, all else being equal, using a newer analysis fleet will produce more realistic estimates of impacts of potential new standards than using an outdated analysis fleet. However, among relatively current options, a balance must be struck between, on one hand, inputs' freshness, and on the other, inputs' completeness and accuracy.
During assembly of the inputs for today's analysis, final compliance data was available for the MY 2015 model year but not in a few cases for MY 2016. However, between mid-year compliance information and manufacturers' specific updates discussed above, a robust and detailed characterization of the MY 2016 fleet was developed. However, while information continued to develop regarding the MY 2017 and, to a lesser extent MY 2018 and even MY 2019 fleets, this information was—even in mid-2017—too incomplete and inconsistent to be assembled with confidence into an analysis fleet for modeling supporting deliberations regarding today's proposal.
In short, the 2016 fleet was, in fact, the most up-to-date fleet that could be produced for this NPRM. Moreover, during late 2016 and early 2017, nearly all manufacturers provided comments on the characterization of their vehicles in the analysis fleet, and many provided specific feedback about their vehicles, including aerodynamic drag coefficients, tire rolling resistance values, transmission efficiencies, and other information used in the analysis. NHTSA worked with manufacturers to clarify and correct some information and integrated the information into a single input file for use in the CAFE model. Accordingly, the current analysis fleet is reasonable to use for purposes of the NPRM analysis.
As always, however, ways to improve the analysis fleet used for subsequent modeling to evaluate potential new CAFE and CO2 standards will undergo continuous consideration. As described above, the compliance data is only the starting point for developing the analysis fleet; much additional information comes directly from manufacturers (such as details about technologies, platforms, engines, transmissions, and other vehicle information, that may not be present in compliance data), and other information must come from commercial and public sources (for example, fleet-wide information like market share, because individual manufacturers do not provide this kind of information). If newer compliance data (i.e., MY 2017) becomes available and can be analyzed during the pendency of this rulemaking, and if all of the other necessary steps can be performed, the analysis fleet will be updated, as feasible, and made publicly available. The agencies seek comment on the option used today and any other options, as well as on the tradeoffs between, on one hand, fidelity with manufacturers' actual plans and, on the other, the ability to make detailed analysis inputs and outputs publicly available.
(c) Observed Technology Content of 2016 Fleet
As explained above, the analysis fleet is defined not only by the vehicle models/configurations it contains but also by the technologies on those vehicles. Each vehicle model/configuration in the analysis fleet has an associated list of observed technologies and equipment that can improve fuel economy and reduce CO2 emissions.
With a portfolio of descriptive technologies arranged by manufacturer and model, the analysis fleet can be summarized and project how vehicles in that fleet may improve over time via the application of additional technology.
In many cases, vehicle technology is clearly observable from the 2016 compliance data (e.g., compliance data indicates clearly which vehicles have turbochargers and which have continuously variable transmissions), but in some cases technology levels are less observable. For the latter, like levels of mass reduction, the analysis categorized levels of technology already used in a given vehicle. Similarly, engineering judgment was used to determine if higher mass reduction levels may be used practicably and safely in a given vehicle.
Either in mid-year compliance data for MY 2016 or, separately and at the agencies' invitation (as discussed above), most manufacturers identified most of the technology already present in each of their MY 2016 vehicle model/configurations. This information was not as complete for all manufacturers' products as needed for today's analysis, so in some cases, information was supplemented with publicly available data, typically from manufacturer media sites. In limited cases, manufacturers did not supply information, and information from commercial and publicly available sources was used.
(d) Mapping Technology Content of 2016 Fleet to Argonne Technology Effectiveness Simulation Work
While each vehicle model/configuration in the analysis fleet has its list of observed technologies and equipment, the ways in which manufacturers apply technologies and equipment do not always coincide perfectly with how the analysis characterizes the various technologies that improve fuel economy and reduce CO2 emissions. To improve how the observed vehicle fleet “fits into” the analysis, each vehicle model/configuration is “mapped” to the full-Start Printed Page 43007vehicle simulation modeling 
by Argonne National Laboratory that is used to estimate the effectiveness of the fuel economy-improving/CO2 emissions-reducing technologies considered. Argonne produces full-vehicle simulation modeling for many combinations of technologies, on many types of vehicles, but it did not simulate literally every single vehicle model/configuration in the analysis fleet because it would be impractical to assemble the requisite detailed information—much of which would likely only be provided on a confidential basis—specific to each vehicle model/configuration and because the scale of the simulation effort would correspondingly increase by at least two orders of magnitude. Instead, Argonne simulated 10 different vehicle types, corresponding to the “technology classes” generally used in CAFE analysis over the past several rulemakings (e.g., small car, small performance car, pickup truck, etc.). Each of those 10 different vehicle types was assigned a set of “baseline characteristics,” to which Argonne added combinations of fuel-saving technologies and then ran simulations to determine the fuel economy achieved when applying each combination of technologies to that vehicle type given its baseline characteristics. These inputs, discussed at greater length in Sections II.D and II.G, provide the basis for the CAFE model's estimation of fuel economy levels and CO2 emission rates.
In the analysis fleet, inputs assign each specific vehicle model/configuration to a technology class, and once there, map to the simulation within that technology class most closely matching the combination of observed technologies and equipment on that vehicle.
This mapping to a specific simulation result most closely representing a given vehicle model/configuration's initial technology “state” enables the CAFE model to estimate the same vehicle model/configuration's fuel economy after application of some other combination of technologies, leading to an alternative technology state.
Start Printed Page 43008
Start Printed Page 43009
Start Printed Page 43010
Start Printed Page 43011
(e) Shared Vehicle Platforms, Engines, and Transmissions
Another aspect of characterizing vehicle model/configurations in the analysis fleet is based on whether they share a “platform” with other vehicle model/configurations. A “platform” refers to engineered underpinnings shared on several differentiated products. Manufacturers share and standardize components, systems, tooling, and assembly processes within their products (and occasionally with the products of another manufacturer) to cost-effectively maintain vibrant portfolios.
Vehicle model/configurations derived from the same platform are so identified in the analysis fleet. Many manufacturers' use of vehicle platforms is well documented in the public record and widely recognized among the vehicle engineering community. Engineering knowledge, information from trade publications, and feedback from manufacturers and suppliers was also used to assign vehicle platforms in the analysis fleet.
When the CAFE model is deciding where and how to add technology to vehicles, if one vehicle on the platform receives new technology, other vehicles on the platform also receive the technology as part of their next major redesign or refresh.
Similar to vehicle platforms, manufacturers create engines that share parts.
One engine family may appear on many vehicles on a platform, and changes to that engine may or may not carry through to all the vehicles. Some engines are shared across a range of different vehicle platforms. Vehicle model/configurations in the analysis fleet that share engines belonging to the same platform are also identified as such.
It is important to note that manufacturers define common engines differently. Some manufacturers consider engines as “common” if the engines shared an architecture, components, or manufacturing processes. Other manufacturers take a narrower definition, and only assume “common” engines if the parts in the engine assembly are the same. In some cases, manufacturers designate each engine in each application as a unique powertrain.
Engine families for each manufacturer were tabulated and assigned 
based on data-driven criteria. If engines shared a common cylinder count and configuration, displacement, valvetrain, and fuel type, those engines may have been considered together. Additionally, if the compression ratio, horsepower, and displacement of engines were only slightly different, those engines were considered to be the same for the purposes of redesign and sharing. Vehicles in the analysis fleet with the same engine family will therefore adopt engine technology in a coordinated fashion.
By grouping engines together, the CAFE model controls future engine families to retain reasonable powertrain complexity.
Like with engines, manufacturers often use transmissions that are the same or similar on multiple vehicles.
To reflect this reality, shared transmissions were considered for manufacturers as appropriate. To define common transmissions, the agencies considered transmission type (manual, automatic, dual-clutch, continuously variable), number of gears, and vehicle architecture (front-wheel-drive, rear-wheel-drive, all-wheel-drive based on a front-wheel-drive platform, or all-wheel-drive based on a rear-wheel-drive platform). If vehicles shared these attributes, these transmissions were grouped for the analysis. Vehicles in the analysis fleet with the same transmission configuration 
will adopt transmission technology together, as described above.
Having all vehicles that share a platform (or engines that are part of a family) adopt fuel economy-improving/CO2 emissions-reducing technologies together, subject to refresh/redesign constraints, reflects the real-world considerations described above but also overlooks some decisions manufacturers might make in the real world in response to market pull, meaning that even though the analysis fleet is incredibly complex, it is also over-simplified in some respects compared to the real world. For example, the CAFE model does not currently attempt to simulate the potential for a manufacturer to shift the application of technologies to improve performance rather than fuel economy. Therefore, the model's representation of the “inheritance” of technology can lead to estimates a manufacturer might eventually exceed fuel economy Start Printed Page 43012standards as technology continues to propagate across shared platforms and engines. In the past, there were some examples of extended periods during which some manufacturers exceeded one or both of the CAFE and/or GHG standards, but in plenty of other examples, manufacturers chose to introduce (or even reintroduce) technological complexity into their vehicle lineups in response to buyer preferences. Going forward, and recognizing the recent trend for consolidating platforms, it seems likely manufacturers will be more likely to choose efficiency over complexity in this regard; therefore, the potential should be lower that today's analysis turns out to be over-simplified compared to the real world.
Options will be considered to further refine the representation of sharing and inheritance of technology, possibly including model revisions to account for tradeoffs between fuel economy and performance when applying technology. Please provide comments on the sharing and inheritance-related aspects of the analysis fleet and the CAFE model along with information that would support refinement of the current approach or development and implementation of alternative approaches.
(f) Estimated Product Design Cycles
Another aspect of characterizing vehicle model/configurations in the analysis fleet is based on when they can next be refreshed or redesigned. Redesign schedules play an important role in determining when new technologies may be applied. Many technologies that improve fuel economy and reduce CO2 emissions may be difficult to incorporate without a major product redesign. Therefore, each vehicle model in the analysis fleet has an associated redesign schedule, and the CAFE model uses that schedule to restrict significant advances in some technologies (like major mass reduction) to redesign years, while allowing manufacturers to include minor advances (such as improved tire rolling resistance) during a vehicle “refresh,” or a smaller update made to a vehicle, which can happen between redesigns. In addition to refresh and redesign schedules associated with vehicle model/configurations, vehicles that share a platform subsequently have platform-wide refresh and redesign schedules for mass reduction technologies.
To develop the refresh/redesign cycles used for the MY 2016 vehicles in the analysis fleet, information from commercially available sources was used to project redesign cycles through MY 2022, as for NHTSA's analysis for the Draft TAR published in 2016.
Commercially available sources' estimates through MY 2022 are generally supported by detailed consideration of public announcements plus related intelligence from suppliers and other sources, and recognize that uncertainty increases considerably as the forecasting horizon is extended. For MYs 2023-2035, in recognition of that uncertainty, redesign schedules were extended considering past pacing for each product, estimated schedules through MY 2022, and schedules for other products in the same technology classes. As mentioned above, potentially confidential forward-looking information was not requested from manufacturers; nevertheless, all manufacturers had an opportunity to review the estimates of product-specific redesign schedules, a few manufacturers provided related forecasts and, for the most part, that information corroborated the estimates.
Some commenters suggested supplanting these estimated redesign schedules with estimates applying faster cycles (e.g., four to five years), and this approach was considered for the analysis.
Some manufacturers tend to operate with faster redesign cycles and may continue to do so, and manufacturers tend to redesign some products more frequently than others. However, especially considering that information presented by manufacturers largely supports estimates discussed above, applying a “one size fits all” acceleration of redesign cycles would likely not improve the analysis; instead, doing so would likely reduce consistency with the real world, especially for light trucks. Moreover, if some manufacturers accelerate redesigns in response to new standards, doing so would likely involve costs (greater levels of stranded capital, reduced opportunity to benefit from “learning”-related cost reductions) greater than reflected in other inputs to the analysis. However, a wider range of technologies can practicably be applied during mid-cycle “freshenings” than has been represented by past analyses, and this part of the analysis has been expanded, as discussed below in Section II.D.
Also, in the sensitivity analysis supporting today's proposal and presented in Chapter 13 of the PRIA, one case involving faster redesign schedules and one involving slower redesign schedules has been analyzed.
Manufacturers use diverse strategies with respect to when, and how often they update vehicle designs. While most vehicles have been redesigned sometime in the last five years, many vehicles have not. In particular, vehicles with lower annual sales volumes tend to be redesigned less frequently, perhaps giving manufacturers more time to amortize the investment needed to bring the product to market. In some cases, manufacturers continue to produce and sell vehicles designed more than a decade ago.
Start Printed Page 43013
Each manufacturer may use different strategies throughout their product portfolio, and a component of each strategy may include the timing of refresh and redesign cycles. Table II-3 below summarizes the average time between redesigns, by manufacturer, by vehicle technology class.
Dashes mean the manufacturer has no volume in that vehicle technology class in the MY 2016 analysis fleet. Across the industry, manufacturers average 6.5 years between product redesigns.
Start Printed Page 43014
There are a few notable observations from this table. Pick-up trucks have much longer redesign schedules (7.8 years on average) than small cars (5.5 years on average). Some manufacturers redesign vehicles often (every 5.2 years in the case of Hyundai), while other manufacturers redesign vehicles less often (FCA waits on average 8.6 years between vehicle redesigns). Across the industry, light-duty vehicle designs last for about 6.5 years.
Even if two manufacturers have similar redesign cadence, the model years in which the redesigns occur may still be different and dependent on where each of the manufacturer's products are in their life cycle.
Table II-4 summarizes the average age of manufacturers' offering by vehicle technology class. A value of “0.0” means that every vehicle for a manufacturer in that vehicle technology class, represented in the MY 2016 analysis fleet was new in MY 2016. Across the industry, manufacturers redesigned MY 2016 vehicles an average of 3.2 years earlier.
Start Printed Page 43015
Based on historical observations and refresh/redesign schedule forecasts, careful consideration to redesign cycles for each manufacturer and each vehicle is important. Simply assuming every vehicle is redesigned by 2021 and by 2025 is not appropriate, as this would misrepresent both the likely timing of redesigns and the likely time between redesigns in most cases.
C. Development of Footprint-Based Curve Shapes
As in the past four CAFE rulemakings, the most recent two of which included related standards for CO2 emissions, NHTSA and EPA are proposing to set attribute-based CAFE standards that are defined by a mathematical function of vehicle footprint, which has observable correlation with fuel economy and vehicle emissions. EPCA, as amended by EISA, expressly requires that CAFE standards for passenger cars and light trucks be based on one or more vehicle attributes related to fuel economy and be expressed in the form of a mathematical function.
While the CAA includes no specific requirements regarding GHG regulation, EPA has chosen to adopt standards consistent with the EPCA/EISA requirements in the interest of simplifying compliance for the industry since 2010.
Section II.C.1 describes the advantages of attribute standards, generally. Section II.C.2 explains the agencies' specific decision to use vehicle footprint as the attribute over which to vary stringency for past and current rules. Section II.C.3 discusses the policy considerations in selecting the specific mathematical function. Section II.C.4 discusses the methodologies used to develop current attribute-based standards, and the agencies' current proposal to continue to do so for MYs 2022-2026. Section II.C.5 discusses the methodologies used to reconsider the mathematical function for the proposed standards.
1. Why attribute-based standards, and what are the benefits?
Under attribute-based standards, every vehicle model has fuel economy and CO2 targets, the levels of which depend on the level of that vehicle's determining attribute (for this proposed rule, footprint is the determining attribute, as discussed below). The manufacturer's fleet average performance is calculated by the harmonic production-weighted average of those targets, as defined below:
Start Printed Page 43016
Here, i represents a given model 
in a manufacturer's fleet, Productioni represents the U.S. production of that model, and Targeti represents the target as defined by the attribute-based standards. This means no vehicle is required to meet its target; instead, manufacturers are free to balance improvements however they deem best within (and, given credit transfers, at least partially across) their fleets.
The idea is to select the shape of the mathematical function relating the standard to the fuel economy-related attribute to reflect the trade-offs manufacturers face in producing more of that attribute over fuel efficiency (due to technological limits of production and relative demand of each attribute). If the shape captures these trade-offs, every manufacturer is more likely to continue adding fuel efficient technology across the distribution of the attribute within their fleet, instead of potentially changing the attribute—and other correlated attributes, including fuel economy—as a part of their compliance strategy. Attribute-based standards that achieve this have several advantages.
First, assuming the attribute is a measurement of vehicle size, attribute-based standards reduce the incentive for manufacturers to respond to CAFE standards by reducing vehicle size in ways harmful to safety.
Larger vehicles, in terms of mass and/or crush space, generally consume more fuel, but are also generally better able to protect occupants in a crash.
Because each vehicle model has its own target (determined by a size-related attribute), properly fitted attribute-based standards provide little, if any, incentive to build smaller vehicles simply to meet a fleet-wide average, because smaller vehicles are subject to more stringent compliance targets.
Second, attribute-based standards, if properly fitted, better respect heterogeneous consumer preferences than do single-valued standards. As discussed above, a single-valued standard encourages a fleet mix with a larger share of smaller vehicles by creating incentives for manufacturers to use downsizing the average vehicle in their fleet (possibly through fleet mixing) as a compliance strategy, which may result in manufacturers building vehicles for compliance reasons that consumers do not want. Under a size-related, attribute-based standard, reducing the size of the vehicle is a less viable compliance strategy because smaller vehicles have more stringent regulatory targets. As a result, the fleet mix under such standards is more likely to reflect aggregate consumer demand for the size-related attribute used to determine vehicle targets.
Third, attribute-based standards provide a more equitable regulatory framework across heterogeneous manufacturers who may each produce different shares of vehicles along attributes correlated with fuel economy.
A single, industry-wide CAFE standard imposes disproportionate cost burden and compliance challenges on manufacturers who produce more vehicles with attributes inherently correlated with lower fuel economy—i.e. manufacturers who produce, on average, larger vehicles. As discussed above, retaining the ability for manufacturers to produce vehicles which respect heterogeneous market preferences is an important consideration. Since manufacturers may target different markets as a part of their business strategy, ensuring that these manufacturers do not incur a disproportionate share of the regulatory cost burden is an important part of conserving consumer choices within the market.
2. Why footprint as the attribute?
It is important that the CAFE and CO2 standards be set in a way that does not encourage manufacturers to respond by selling vehicles that are less safe. Vehicle size is highly correlated with vehicle safety—for this reason, it is important to choose an attribute correlated with vehicle size (mass or some dimensional measure). Given this consideration, there are several policy and technical reasons why footprint is considered to be the most appropriate attribute upon which to base the standards, even though other vehicle size attributes (notably, curb weight) are more strongly correlated with fuel economy and emissions.
First, mass is strongly correlated with fuel economy; it takes a certain amount of energy to move a certain amount of mass. Footprint has some positive correlation with frontal surface area, likely a negative correlation with aerodynamics, and therefore fuel economy, but the relationship is less deterministic. Mass and crush space (correlated with footprint) are both important safety considerations. As discussed below and in the accompanying PRIA, NHTSA's research of historical crash data indicates that holding footprint constant, and decreasing the mass of the largest vehicles, will have a net positive safety impact to drivers overall, while holding footprint constant and decreasing the mass of the smallest vehicles will have a net decrease in fleetwide safety. Properly fitted footprint-based standards provide little, if any, incentive to build smaller vehicles to meet CAFE and CO2 standards, and therefore help minimize the impact of standards on overall fleet safety.
Second, it is important that the attribute not be easily manipulated in a manner that does not achieve the goals of EPCA or other goals, such as safety. Although weight is more strongly correlated with fuel economy than footprint, there is less risk of manipulation (changing the attribute(s) to achieve a more favorable target) by increasing footprint under footprint-based standards than there would be by increasing vehicle mass under weight-based standards. It is relatively easy for a manufacturer to add enough weight to a vehicle to decrease its applicable fuel economy target a significant amount, as compared to increasing vehicle Start Printed Page 43017footprint, which is a much more complicated change that typically takes place only with a vehicle redesign.
Further, some commenters on the MY 2011 CAFE rulemaking were concerned that there would be greater potential for gaming under multi-attribute standards, such as those that also depend on weight, torque, power, towing capability, and/or off-road capability. As discussed in NHTSA's MY 2011 CAFE final rule,
it is anticipated that the possibility of gaming is lowest with footprint-based standards, as opposed to weight-based or multi-attribute-based standards. Specifically, standards that incorporate weight, torque, power, towing capability, and/or off-road capability in addition to footprint would not only be more complex, but by providing degrees of freedom with respect to more easily-adjusted attributes, they could make it less certain that the future fleet would actually achieve the projected average fuel economy and CO2 levels. This is not to say that a footprint-based system will eliminate gaming, or that a footprint-based system will eliminate the possibility that manufacturers will change vehicles in ways that compromise occupant protection, but footprint-based standards achieve the best balance among affected considerations. Please provide comments on whether vehicular footprint is the most suitable attribute upon which to base standards.
3. What mathematical function should be used to specify footprint-based standards?
In requiring NHTSA to “prescribe by regulation separate average fuel economy standards for passenger and non-passenger automobiles based on 1 or more vehicle attributes related to fuel economy and express each standard in the form of a mathematical function”, EPCA/EISA provides ample discretion regarding not only the selection of the attribute(s), but also regarding the nature of the function. The CAA provides no specific direction regarding CO2 regulation, and EPA has continued to harmonize this aspect of its CO2 regulations with NHTSA's CAFE regulations. The relationship between fuel economy (and GHG emissions) and footprint, though directionally clear (i.e., fuel economy tends to decrease and CO2 emissions tend to increase with increasing footprint), is theoretically vague, and quantitatively uncertain; in other words, not so precise as to a priori yield only a single possible curve.
The decision of how to specify this mathematical function therefore reflects some amount of judgment. The function can be specified with a view toward achieving different environmental and petroleum reduction goals, encouraging different levels of application of fuel-saving technologies, avoiding any adverse effects on overall highway safety, reducing disparities of manufacturers' compliance burdens, and preserving consumer choice, among other aims. The following are among the specific technical concerns and resultant policy tradeoffs the agencies have considered in selecting the details of specific past and future curve shapes:
- Flatter standards (i.e., curves) increase the risk that both the size of vehicles will be reduced, potentially compromising highway safety, and reducing any utility consumers would have gained from a larger vehicle.
- Steeper footprint-based standards may create incentives to upsize vehicles, potentially oversupplying vehicles of certain footprints beyond what consumers would naturally demand, and thus increasing the possibility that fuel savings and CO2 reduction benefits will be forfeited artificially.
- Given the same industry-wide average required fuel economy or CO2 standard, flatter standards tend to place greater compliance burdens on full-line manufacturers.
- Given the same industry-wide average required fuel economy or CO2 standard, dramatically steeper standards tend to place greater compliance burdens on limited-line manufacturers (depending of course, on which vehicles are being produced).
- If cutpoints are adopted, given the same industry-wide average required fuel economy, moving small-vehicle cutpoints to the left (i.e., up in terms of fuel economy, down in terms of CO2 emissions) discourages the introduction of small vehicles, and reduces the incentive to downsize small vehicles in ways that could compromise overall highway safety.
- If cutpoints are adopted, given the same industry-wide average required fuel economy, moving large-vehicle cutpoints to the right (i.e., down in terms of fuel economy, up in terms of CO2 emissions) better accommodates the design requirements of larger vehicles — especially large pickups — and extends the size range over which downsizing is discouraged.
4. What mathematical functions have been used previously, and why?
Notwithstanding the aforementioned discretion under EPCA/EISA, data should inform consideration of potential mathematical functions, but how relevant data is defined and interpreted, and the choice of methodology for fitting a curve to that data, can and should include some consideration of specific policy goals. This section summarizes the methodologies and policy concerns that were considered in developing previous target curves (for a complete discussion see the 2012 FRIA).
As discussed below, the MY 2011 final curves followed a constrained logistic function defined specifically in the final rule.
The MYs 2012-2021 final standards and the MYs 2022-2025 augural standards are defined by constrained linear target functions of footprint, as shown below: 
Here, Target is the fuel economy target applicable to vehicles of a given footprint in square feet (Footprint). The upper asymptote, a, and the lower asymptote, b, are specified in mpg; the reciprocal of these values represent the lower and upper asymptotes, respectively, when the curve is instead specified in gallons per mile (gpm). The Start Printed Page 43018slope, c, and the intercept, d, of the linear portion of the curve are specified as gpm per change in square feet, and gpm, respectively.
The min and max functions will take the minimum and maximum values within their associated parentheses. Thus, the max function will first find the maximum of the fitted line at a given footprint value and the lower asymptote from the perspective of gpm. If the fitted line is below the lower asymptote it is replaced with the floor, which is also the minimum of the floor and the ceiling by definition, so that the target in mpg space will be the reciprocal of the floor in mpg space, or simply, a. If, however, the fitted line is not below the lower asymptote, the fitted value is returned from the max function and the min function takes the minimum value of the upper asymptote (in gpm space) and the fitted line. If the fitted value is below the upper asymptote, it is between the two asymptotes and the fitted value is appropriately returned from the min function, making the overall target in mpg the reciprocal of the fitted line in gpm. If the fitted value is above the upper asymptote, the upper asymptote is returned is returned from the min function, and the overall target in mpg is the reciprocal of the upper asymptote in gpm space, or b.
In this way curves specified as constrained linear functions are specified by the following parameters:
a = upper limit (mpg)
b = lower limit (mpg)
c = slope (gpm per sq. ft.)
d = intercept (gpm)
The slope and intercept are specified as gpm per sq. ft. and gpm instead of mpg per sq. ft. and mpg because fuel consumption and emissions appear roughly linearly related to gallons per mile (the reciprocal of the miles per gallon).
(a) NHTSA in MY 2008 and MY 2011 CAFE (Constrained Logistic)
For the MY 2011 CAFE rule, NHTSA estimated fuel economy levels by footprint from the MY 2008 fleet after normalization for differences in technology,
but did not make adjustments to reflect other vehicle attributes (e.g., power-to-weight ratios). Starting with the technology-adjusted passenger car and light truck fleets, NHTSA used minimum absolute deviation (MAD) regression without sales weighting to fit a logistic form as a starting point to develop mathematical functions defining the standards. NHTSA then identified footprints at which to apply minimum and maximum values (rather than letting the standards extend without limit) and transposed these functions vertically (i.e., on a gallons-per-mile basis, uniformly downward) to produce the promulgated standards. In the preceding rule, for MYs 2008-2011 light truck standards, NHTSA examined a range of potential functional forms, and concluded that, compared to other considered forms, the constrained logistic form provided the expected and appropriate trend (decreasing fuel economy as footprint increases), but avoided creating “kinks” the agency was concerned would provide distortionary incentives for vehicles with neighboring footprints.
(b) MYs 2012-2016 Standards (Constrained Linear)
For the MYs 2012-2016 rule, potential methods for specifying mathematical functions to define fuel economy and CO2 standards were reevaluated. These methods were fit to the same MY 2008 data as the MY 2011 standard. Considering these further specifications, the constrained logistic form, if applied to post-MY 2011 standards, would likely contain a steep mid-section that would provide undue incentive to increase the footprint of midsize passenger cars.
A range of methods to fit the curves would have been reasonable, and a minimum absolute deviation (MAD) regression without sales weighting on a technology-adjusted car and light truck fleet was used to fit a linear equation. This equation was used as a starting point to develop mathematical functions defining the standards. Footprints were then identified at which to apply minimum and maximum values (rather than letting the standards extend without limit). Finally, these constrained/piecewise linear functions were transposed vertically (i.e., on a gpm or CO2 basis, uniformly downward) by multiplying the initial curve by a single factor for each MY standard to produce the final attribute-based targets for passenger cars and light trucks described in the final rule.
These transformations are typically presented as percentage improvements over a previous MY target curve.
(c) MYs 2017 and Beyond Standards (Constrained Linear)
The mathematical functions finalized in 2012 for MYs 2017 and beyond changed somewhat from the functions for the MYs 2012-2016 standards. These changes were made to both address comments from stakeholders, and to further consider some of the technical concerns and policy goals judged more preeminent under the increased uncertainty of the impacts of finalizing and proposing standards for model years further into the future.
Recognizing the concerns raised by full-line OEMs, it was concluded that continuing increases in the stringency of the light truck standards would be more feasible if the light truck curve for MYs 2017 and beyond was made steeper than the MY 2016 truck curve and the right (large footprint) cut-point was extended only gradually to larger footprints. To accommodate these considerations, the 2012 final rule finalized the slope fit to the MY 2008 fleet using a sales-weighted, ordinary least-squares regression, using a fleet that had technology applied to make the technology application across the fleet more uniform, and after adjusting the data for the effects of weight-to-footprint. Information from an updated MY 2010 fleet was also considered to support this decision. As the curve was vertically shifted (with fuel economy specified as mpg instead of gpm or CO2 emissions) upwards, the right cutpoint was progressively moved for the light truck curves with successive model years, reaching the final endpoint for MY 2021; this is further discussed and shown in Chapter 4.3 of the PRIA.
5. Reconsidering the Mathematical Functions for This Proposal
(a) Why is it important to reconsider the mathematical functions?
By shifting the developed curves by a single factor, it is assumed that the underlying relationship of fuel consumption (in gallons per mile) to vehicle footprint does not change significantly from the model year data used to fit the curves to the range of model years for which the shifted curve shape is applied to develop the standards. However, it must be recognized that the relationship Start Printed Page 43019between vehicle footprint and fuel economy is not necessarily constant over time; newly developed technologies, changes in consumer demand, and even the curves themselves could, if unduly susceptible to gaming, influence the observed relationships between the two vehicle characteristics. For example, if certain technologies are more effective or more marketable for certain types of vehicles, their application may not be uniform over the range of vehicle footprints. Further, if market demand has shifted between vehicle types, so that certain vehicles make up a larger share of the fleet, any underlying technological or market restrictions which inform the average shape of the curves could change. That is, changes in the technology or market restrictions themselves, or a mere re-weighting of different vehicles types, could reshape the fit curves.
For the above reasons, the curve shapes were reconsidered using the newest available data, from MY 2016. With a view toward corroboration through different techniques, a range of descriptive statistical analyses were conducted that do not require underlying engineering models of how fuel economy and footprint might be expected to be related, and a separate analysis that uses vehicle simulation results as the basis to estimate the relationship from a perspective more explicitly informed by engineering theory was conducted as well. Despite changes in the new vehicle fleet both in terms of technologies applied and in market demand, the underlying statistical relationship between footprint and fuel economy has not changed significantly since the MY 2008 fleet used for the 2012 final rule; therefore, it is proposed to continue to use the curve shapes fit in 2012. The analysis and reasoning supporting this decision follows.
(b) What statistical analyses did NHTSA consider?
In considering how to address the various policy concerns discussed above, data from the MY 2016 fleet was considered, and a number of descriptive statistical analyses (i.e., involving observed fuel economy levels and footprints) using various statistical methods, weighting schemes, and adjustments to the data to make the fleets less technologically heterogeneous were performed. There were several adjustments to the data that were common to all of the statistical analyses considered.
With a view toward isolating the relationship between fuel economy and footprint, the few diesels in the fleet were excluded, as well as the limited number of vehicles with partial or full electric propulsion; when the fleet is normalized so that technology is more homogenous, application of these technologies is not allowed. This is consistent with the methodology used in the 2012 final rule.
The above adjustments were applied to all statistical analyses considered, regardless of the specifics of each of the methods, weights, and technology level of the data, used to view the relationship of vehicle footprint and fuel economy. Table II-5, below, summarizes the different assumptions considered and the key attributes of each. The analysis was performed considering all possible combinations of these assumptions, producing a total of eight footprint curves.
Start Printed Page 43020
(1) Current Technology Level Curves
The “current technology” level curves exclude diesels and vehicles with electric propulsion, as discussed above, but make no other changes to each model year fleet. Comparing the MY 2016 curves to ones built under the same methodology from previous model year fleets shows whether the observed curve shape has changed significantly over time as standards have become more stringent. Importantly, these curves will include any market forces which make technology application variable over the distribution of footprint. These market forces will not be present in the “maximum technology” level curves: By making technology levels homogenous, this variation is removed. The current technology level curves built using both regression types and both regression weight methodologies from the MY 2008, MY 2010, and MY 2016 fleets, shown in more detail in Chapter 220.127.116.11 of the PRIA, support the curve slopes finalized in the 2012 final rule. The curves built from most methodologies using each fleet generally shift, but remain very similar in slope. This suggests that the relationship of footprint to fuel economy, including both technology and market limits, has not significantly changed.
(2) Maximum Technology Level Curves
As in prior rulemakings, technology differences between vehicle models were considered to be a significant factor producing uncertainty regarding the relationship between fuel consumption and footprint. Noting that attribute-based standards are intended to encourage the application of additional technology to improve fuel efficiency and reduce CO2 emissions across the distribution of footprint in the fleet, approaches were considered in which technology application is simulated for purposes of the curve fitting analysis in order to produce fleets that are less varied in technology content. This approach helps reduce “noise” (i.e., dispersion) in the plot of vehicle footprints and fuel consumption levels and identify a more technology-neutral relationship between footprint and fuel consumption. The results of updated analysis for maximum technology level curves are also shown in Chapter 18.104.22.168 of the PRIA. Especially if vehicles progress over time toward more similar size-specific efficiency, further removing variation in technology application both better isolates the relationship between fuel consumption and footprint and further supports the curve slopes finalized in the 2012 final rule.
(c) What other methodologies were considered?
The methods discussed above are descriptive in nature, using statistical analysis to relate observed fuel economy levels to observed footprints for known vehicles. As such, these methods are clearly based on actual data, answering the question “how does fuel economy appear to be related to footprint?” However, being independent of explicit engineering theory, they do not answer the question “how might one expect fuel economy to be related to footprint?” Therefore, as an alternative to the above methods, an alternative methodology was also developed and applied that, using full-vehicle simulation, comes closer to answer the second question, providing a basis to either corroborate answers to the first, or suggest that further investigation could be important.
As discussed in the 2012 final rule, several manufacturers have confidentially shared with the agencies what they described as “physics-based” curves, with each OEM showing significantly different shapes for the footprint-fuel economy relationships. This variation suggests that manufacturers face different curves given the other attributes of the vehicles in their fleets (i.e., performance characteristics) and/or that their curves reflected different levels of technology application. In reconsidering the shapes of the proposed MYs 2021-2026 standards, a similar estimation of physics-based curves leveraging third-party simulation work form Argonne National Laboratories (ANL) was developed. Estimating physics-based curves better ensures that technology and performance are held constant for all footprints; augmenting a largely statistical analysis with an analysis that more explicitly incorporates engineering theory helps to corroborate that the relationship between fuel economy and footprint is in fact being characterized.
Tractive energy is the amount of energy it will take to move a vehicle.
Here, tractive energy effectiveness is defined as the share of the energy content of fuel consumed which is converted into mechanical energy and used to move a vehicle—for internal combustion engine (ICE) vehicles, this will vary with the relative efficiency of specific engines. Data from ANL simulations suggest that the limits of tractive energy effectiveness are approximately 25% for vehicles with internal combustion engines which do not possess ISG, other hybrid, plug-in, pure electric, or fuel cell technology.
A tractive energy prediction model was also developed to support today's proposal. Given a vehicle's mass, frontal area, aerodynamic drag coefficient, and rolling resistance as inputs, the model will predict the amount of tractive energy required for the vehicle to complete the Federal test cycle. This model was used to predict the tractive energy required for the average vehicle of a given footprint 
and “body technology package” to complete the cycle. The body technology packages considered are defined in Table II-6, below. Using the absolute tractive energy predicted and tractive energy effectiveness values spanning possible ICE engines, fuel economy values were then estimated for different body technology packages and engine tractive energy effectiveness values.
Start Printed Page 43021
Chapter 6 of the PRIA shows the resultant CAFE levels estimated for the vehicle classes ANL simulated for this analysis, at different footprint values and by vehicle “box.” Pickups are considered 1-box, hatchbacks and minivans are 2-box, and sedans are 3-box. These estimates are compared with the MY 2021 standards finalized in 2012. The general trend of the simulated data points follows the pattern of the previous MY 2021 standards for all technology packages and tractive energy effectiveness values presented in the PRIA. The tractive energy curves are intended to validate the curve shapes against a physics-based alternative, and the analysis suggests that the curve shapes track the physical relationship between fuel economy and tractive energy for different footprint values.
Physical limitations are not the only forces manufacturers face; they must also produce vehicles that consumers will purchase. For this reason, in setting future standards, the analysis will continue to consider information from statistical analyses that do not homogenize technology applications in addition to statistical analyses which do, as well as a tractive energy analysis similar to the one presented above.
The relationship between fuel economy and footprint remains directionally discernable but quantitatively uncertain. Nevertheless, each standard must commit to only one function. Approaching the question “how is fuel economy related to footprint” from different directions and applying different approaches will provide the greatest confidence that the single function defining any given standard appropriately and reasonably reflects the relationship between fuel economy and footprint. Please provide comments on this tentative conclusion and the above discussion.
D. Characterization of Current and Anticipated Fuel-Saving Technologies
The analysis evaluates a wide array of technologies manufacturers could use to improve the fuel economy of new vehicles, in both the near future and the timeframe of this proposed rulemaking, to meet the fuel economy and CO2 standards proposed in this rulemaking. The analysis evaluated costs for these technologies, and looked at how these costs may change over time. The analysis also considered how fuel-saving technologies may be used on many types of vehicles (ranging from small cars to trucks) and how the technologies may perform in improving fuel economy and CO2 emissions in combination with other technologies. With cost and effectiveness estimates for technologies, the analysis can forecast how manufacturers may respond to potential standards and can estimate the associated costs and benefits related to technology and equipment changes. This assists the assessment of technological feasibility and is a building block for the consideration of economic practicability of potential standards.
NHTSA, EPA, and CARB issued the Draft Technical Assessment Report (Draft TAR) 
as the first step in the EPA MTE process. The Draft TAR provided an opportunity for the agencies to share with the public updated technical analysis relevant to development of future standards. For this NPRM, the analysis relies on portions of the analysis presented in the Draft TAR, along with new information that has been gathered and developed since conducting that analysis, and the significant, substantive input that was received during the public comment period.
The Draft TAR considered many technologies previously assessed in the 2012 final rule.
In some cases, manufacturers have nearly universally adopted a technology in today's new vehicle fleet (for example, electric power steering). In other cases, manufacturers occasionally use a technology in today's new vehicle fleet (like turbocharged engines). For a few technologies considered in the 2012 rulemaking, manufacturers began implementing the technologies but have since largely pivoted to other technologies due to consumer acceptance issues (for instance, in some cases drivability and performance feel issues associated with dual clutch transmissions without a torque converter) or limited commercial success. The analysis utilizes new information as manufacturers' use of technologies evolves.
Some of the emerging technologies described in the Draft TAR were not included in this analysis, but this includes some additional technologies not previously considered. As industry invents and develops new fuel-savings technologies, and as suppliers and manufacturers produce and apply the technologies, and as consumers react to the new technologies, efforts are continued to learn more about the capabilities and limitations of new technologies. While a technology is in early development, theoretical constructs, limited access to test data, and CBI is relied on to assess the technology. After manufacturers commercialize the technology and bring products to market, the technology may be studied in more detail, which generally leads to the most reliable information about the technology. In addition, once in production, the technology is represented in the fuel economy and CO2 status of the baseline fleet. The technology analysis is kept as current as possible in light of the ongoing technology development and implementation in the automotive industry.
Some technology assumptions have been updated since the MYs 2017-2025 final rule and, in many cases, since the 2016 Draft TAR. In some cases, EPA and NHTSA presented different analytical approaches in the Draft TAR; the analysis is now presented using the Start Printed Page 43022same direct manufacturing costs, retail costs, and learning rates. In addition, the effectiveness of fuel-economy technologies is now assessed based on the same assumptions, and with the same tools. Finally, manufacturers' response to stringency alternatives is forecast with the same simulation model.
Since the 2017 and later final rule, many cost assessments, including tear down studies, were funded and completed, and presented as part of the Draft TAR analysis. These studies evaluated transmissions, engines, hybrid technologies, and mass reduction.
As a result, the analysis uses updated cost estimates for many technologies, some of which have been updated since the Draft TAR. In addition to those studies, the analysis also leveraged research reports from other organizations to assess costs.
Today's analysis also updates the costs to 2016 dollars, as in many cases technology costs were estimated several years ago.
The analysis uses an updated, peer-reviewed model developed by ANL for the Department of Energy to provide a more rigorous estimate for battery costs. The new battery model provides an estimate future for battery costs for hybrids, plug-in hybrids, and electric vehicles, taking into account the different battery design characteristics and taking into account the size of the battery for different applications.
In the Draft TAR, two possible methodologies to estimate indirect costs from direct manufacturing costs, described as “indirect cost multipliers” and “retail price equivalent” were presented. Both of these methodologies attempted to relate the price of parts for fuel-saving technologies to a retail price. Today's analysis utilizes the direct manufacturing costs (DMC) and the retail price equivalent (RPE) methodology published in the Draft TAR.
Two tools to estimate effectiveness of fuel-saving technologies were used in the Draft TAR, and for today's analysis, only one tool was used (Autonomie).
Previously, EPA developed “ALPHA”, an in-house model that estimated fuel-savings for technologies, which provided a foundation for EPA's analysis. EPA's “ALPHA” results were used to calibrate a much simpler “Lumped Parameter Model” that was developed by EPA to estimate technology effectiveness for many technologies. The Lumped Parameter Model (LPM) approximated simulation modeling results instead of directly using the results and lead to less accurate estimates of technology effectiveness. Many stakeholders questioned the efficacy of the Lumped Parameter Model and ALPHA assumptions and outputs in combination,
especially as the tool was used to evaluate increasingly heterogeneous combinations of technologies in the baseline fleet.
For today's analysis, EPA and NHTSA used an updated version of the Autonomie model—an improved version of what NHTSA presented in the 2016 Draft TAR—to assess technology effectiveness of technologies and combinations of technologies. The Department of Energy's ANL developed Autonomie and the underpinning model assumptions leveraged research from the DOE's Vehicle Technologies Office and feedback from the public. Autonomie is commercially available and widely used; third parties such as suppliers, automakers, and academic researchers (who publish findings in peer reviewed academic journals) commonly use the Autonomie simulation software.
Similarly for today's analysis, only one tool is used. Previously, EPA developed “OMEGA,” a tool that looked at costs of technologies and effectiveness of technologies (as estimated by EPA's Lumped Parameter Model or ALPHA), and applied cost effective technologies to manufacturers' fleets in response to potential standards. Many stakeholders commented that the OMEGA model oversimplified fleet-wide analysis, resulting in significant shortcomings.
For instance, OMEGA assumed manufacturers would redesign all vehicles in the fleet by 2021, and then again by 2025; stakeholders purported that these assumptions did not reflect practical constraints in many manufacturers' business models.
Additionally, stakeholders commented that OMEGA did not adequately take into consideration common parts like shared engines, shared transmissions, and engineering platforms. The CAFE model does consider refresh and redesign cycles and parts sharing. The CAFE model can evaluate responses to any policy alternative on a year-by-year basis, as required by EPCA/EISA 
and as allowed by the CAA, and can also account for manufacturers' year-by-year plans that involve “carrying forward” technology from prior model years, and some other vehicles possibly applying “extra” technology in anticipation of standards in ensuing model years. For today's analysis, an updated version of the CAFE model is used—an improved version of what NHTSA presented in the 2016 Draft TAR—to assess manufacturers' response to policy alternatives. See Section II.A.1 above for further discussion of the decision to use the CAFE model for the NPRM analysis.
Each aforementioned change is discussed briefly in the remainder of this section and in much greater detail in Chapter 6 of the PRIA. A brief summary of the technologies considered in this proposal is discussed below. Please provide comments on all aspects of the analysis as discussed here and as detailed in the PRIA.Start Printed Page 43023
1. Data Sources and Processes for Developing Individual Technology Assumptions
Technology assumptions were developed that provide a foundation for conducting a fleet-wide compliance analysis. As part of this effort, the analysis estimated technology costs, projected technology effectiveness values, and identified possible limitations for some fuel-saving technologies. There is a preference to use values developed from careful review of commercialized technologies; however, in some cases for technologies that are new, and are not yet for sale in any vehicle, the analysis relied on information from other sources, including CBI and third-party research reports and publications. Many emerging technologies are still being evaluated for the analysis supporting the final rule, including those that are currently emerging.
For today's analysis, one set of cost assumptions, one set of effectiveness values (developed with one tool), and one set of assumptions about the limitations of some technologies are presented. Many sources of data were evaluated, in addition to many stakeholder comments received on the Draft TAR. Throughout the process of developing the assumptions for today's analysis, the preferred approach was to harmonize on sources and methodologies that were data-driven and reproducible in independent verification, produced using tools utilized by OEMs, suppliers, and academic institutions, and using tools that could support both CAFE and CO2 analysis. A single set of assumptions also facilitates and focuses public comment by reducing burden on stakeholders who seek to review all of the supporting documentation for this proposal.
(a) Technology Costs
The analysis estimated present and future costs for fuel-saving technologies, taking into consideration the type of vehicle, or type of engine if technology costs vary by application. Cost estimates were developed based on three main inputs. First, direct manufacturing costs (DMC), or the component costs of the physical parts and systems, were considered, with estimated costs assuming high volume production. DMCs generally do not include the indirect costs of tools, capital equipment, and financing costs, nor do they cover indirect costs like engineering, sales, and administrative support. Second, indirect costs via a scalar markup of direct manufacturing costs (the retail price equivalent, or RPE) was taken into account. Finally, costs for technologies may change over time as industry streamlines design and manufacturing processes. Potential cost improvements with learning effects (LE) were also considered. The retail cost of equipment in any future year is estimated to be equal to the product of the DMC, RPE, and LE. Considering the retail cost of equipment, instead of merely direct manufacturing costs, is important to account for the real-world price effects of a technology, as well as market realities. Absent government mandate, a manufacturer will not undertake expensive development and support costs to implement technologies without realistic prospects of consumer willingness to pay enough for such technology to allow for the manufacturer to recover its investment.
(1) Direct Manufacturing Costs
In many instances, the analysis used agency-sponsored tear-down studies of vehicles and parts to estimate the direct manufacturing costs of individual technologies. In the simplest cases, the studies produced results that confirmed third-party industry estimates, and aligned with confidential information provided by manufacturers and suppliers. In cases with a large difference between the tear-down study results and credible independent sources, study assumptions were scrutinized, and sometimes the analysis was revised or updated accordingly.
Studies were conducted on vehicles and technologies that would cover a breadth of fuel-savings technologies, but because tear-down studies can be time-intensive and expensive, the agencies did not sponsor teardown studies for every technology. For some technologies, independent tear-down studies were also utilized, in addition to other publications and confidential business information.
Due to the variety of technologies and their applications, a detailed tear-down study could not be conducted for every technology, including pre-production technologies.
Many fuel-saving technologies were considered that are pre-production, or sold in very small pilot volumes. For emerging technologies that could be applied in the rulemaking timeframe, a tear-down study cannot be conducted to assess costs because the product is not yet in the marketplace for evaluation. In these cases, third-party estimates and confidential information from suppliers and manufacturers are relied upon; however, there are some common pitfalls with relying on confidential business information to estimate costs. The agencies and the source may have had incongruent or incompatible definitions of “baseline.” 
The source may have provided direct manufacturer costs at a date many years in the future, and assumed very high production volumes, important caveats to consider for agency analysis. In addition, a source, under no contractual obligation to the agencies, may provide incomplete and/or misleading information. In other cases, intellectual property considerations and strategic business partnerships may have contributed to a manufacturer's cost information and could be difficult to account for in the model as not all manufacturer's may have access to proprietary technologies at stated costs. New information is carefully evaluated in light of these common pitfalls, especially regarding emerging technologies. The analysis used third-party, forward looking information for advanced cylinder deactivation and variable compression ratio engines, and while these cost estimates may be cursory (as is the case with many emerging technologies prior to commercialization), the agencies took care to use early information provided fairly and reasonably. While costs for fuel-saving technologies reflect the best estimates available today, technology cost estimates will likely change in the future as technologies are deployed and as production is expanded. For emerging technologies, the best information available at the time of the analysis was utilized, and cost assumptions will continue to be updated.
(2) Indirect Costs
As explained above, in addition to direct manufacturing costs, the analysis estimates and considers indirect manufacturing costs. To estimate indirect costs, direct manufacturing costs are multiplied by a factor to represent the average price for fuel-saving technologies at retail. This factor, referred to as the retail price equivalence (RPE), accounts for indirect costs like engineering, sales, and administrative support, as well as other overhead costs, business expenses, warranty costs, and return on capital Start Printed Page 43024considerations. This approach to the RPE remains unchanged from the RPE approach NHTSA presented in the Draft TAR.
The RPE was chosen for this analysis instead of indirect cost multipliers (ICM) because it provides the best estimate of indirect costs. For a more detailed discussion of the approach to indirect costs, see PRIA Chapter 9.
(3) Stranded Capital Costs
Past analyses accounted for costs associated with stranded capital when fuel economy standards caused a technology to be replaced before its costs were fully amortized. The idea behind stranded capital is that manufacturers amortize research, development, and tooling expenses over many years, especially for engines and transmissions. The traditional production life-cycles for transmissions and engines have been a decade or longer. If a manufacturer launches or updates a product with fuel-saving technology, and then later replaces that technology with an unrelated or different fuel-saving technology before the equipment and research and development investments have been fully paid off, there will be unrecouped, or stranded, capital costs. Quantifying stranded capital costs attempted to account for such lost investments. In the Draft TAR analysis, there were only a few technologies for a few manufacturers that were projected to have stranded capital costs.
As more technologies are included in this analysis, and as the CAFE model has been expanded to account for platform and engine sharing and updated with redesign and refresh cycles, accounting for stranded capital has become increasingly complex. Separately, the fact that manufacturers may be shifting their investment strategies in ways that may affect stranded capital calculations was considered. For instance, Ford and General Motors agreed to jointly develop next generation transmission technologies,
and some suppliers sell similar transmissions to multiple manufacturers. These arrangements allow manufacturers to share in capital expenditures, or amortize expenses more quickly. Manufacturers increasingly share parts on vehicles around the globe, achieving greater scale and greatly affecting tooling strategies and costs. Given these trends in the industry and their uncertain effect on capital amortization, and given the difficulty of handling this uncertainty in the CAFE model, this analysis does not account for stranded capital. However, these trends will be monitored to assess the role of stranded capital moving forward.
The analysis continues to rely on projected refresh and redesign cycles in the CAFE model to moderate the cadence for technology adoption and limit the occurrence of stranded capital and the need to account for it. Stranded capital is an important consideration to appropriately account for costs if there is too rapid of a turnover for certain technologies.
(4) Cost Learning
Manufacturers make improvements to production processes over time, often resulting in lower costs. Today's analysis estimates cost learning by considering Wright's learning theory, which states that as every time cumulative volume for a product doubles, the cost lowers by a scalar factor. The analysis accounts for learning effects with model year-based cost learning forecasts for each technology that reduce direct manufacturing costs over time. Historical use of technologies were evaluated, and industry forecasts were reviewed to estimate future volumes for the purpose of developing the model year-based technology cost learning curves. The CAFE model does not dynamically update learning curves, based on compliance pathways chosen in simulation.
As discussed above, cost inputs to the CAFE model incorporate estimates of volume-based learning. As an alternative approach, Volpe Center staff have considered modifications such that the CAFE model would calculate degrees of volume-based learning dynamically, responding to the model's application of affected technologies. While it is intuitive that the degree of cost reduction achieved through experience producing a given technology should depend on the actual accumulated experience (i.e., volume) producing that technology, staff have thus far found such dynamic implementation in the CAFE model infeasible. Insufficient data has been available regarding manufacturers' historical application of specific technology. Also, insofar as underlying direct manufacturing costs already make some assumptions about volume and scale, insufficient information is currently available to determine how to dynamically adjust these underlying costs. It should be noted that if learning responds dynamically to volume, and volume responds dynamically to learning, an internally consistent model solution would likely require iteration of the CAFE model to seek a stable solution within the model's representation multiyear planning. Thus far, these challenges suggest it would be infeasible to calculate degrees of volume-based learning in a manner that responds dynamically to modeled technology application. Nevertheless, the agencies invite comment on the issue, and seek data and methods that would provide the basis for a practicable approach to doing so.
Today's analysis also updates the way learning effects apply to costs. In the Draft TAR analysis, NHTSA applied learning curves only to the incremental direct manufacturing costs or costs over the previous technology on the tech tree. In practice, two things were observed: (1) If the incremental direct manufacturing costs were positive, technologies could not become less expensive than their predecessors on the tech tree, and (2) absolute costs over baseline technology depended on the learning curves of root technologies on the tech tree. Today's analysis applies learning effects to the incremental cost over the null technology state on the tech tree. After this step, the analysis calculates year-by-year incremental costs over preceding technologies on the tech tree to create the CAFE model inputs.
Direct manufacturing costs and learning effects for many technologies were reviewed by evaluating historical use of technologies and industry forecasts to estimate future volumes. This approach produced reasonable estimates for technologies already in production. For technologies not yet in production in MY 2016, the cumulative volume in MY 2016 is zero, because manufacturers have not yet produced the technologies. For pre-production cost estimates, the analysis often relies on confidential business information sources to predict future costs. Many sources for pre-production cost estimates include significant learning effects, often providing cost estimates assuming high volume production, and often for a timeframe late in the first production generation or early in the second generation of the technology. Rapid doubling and re-doubling of a low cumulative volume base with Wright's learning curves can provide unrealistic cost estimates. In addition, direct manufacturing cost projections can vary depending on the initial production volume assumed. Direct costs with learning were carefully examined, and adjustments were made to the starting Start Printed Page 43025point for those technologies on the learning curve to better align with the assumptions used for the initial direct cost estimate. See PRIA Chapter 9 for more detailed information on cost learning.
(b) Technology Effectiveness
(1) Technology Effectiveness Simulation Modeling
Full-vehicle simulation modeling was used to estimate the fuel economy improvements manufacturers could make to their fleet by adding new technologies, taking into account MY 2016 vehicle specifications, as well as how combinations of technologies interact. Full-vehicle simulation modeling uses computer software and physics-based models to predict how combinations of technologies perform together.
The simulation and modeling requires detailed specifications for each technology that describes its efficiency and performance-related characteristics. Those specifications generally come from design specifications, laboratory measurements, simulation or modeling, and may involve additional analysis. For example, the analysis used engine maps showing fuel use vs. engine torque vs. engine speed, and transmission maps taking into account gear efficiency for a range of loads and speeds. With physics-based technology specifications, full-vehicle simulation modeling can be used to estimate technology effectiveness for various combinations and permutations of technologies for many vehicle classes. To develop the specifications used for the simulation and modeling, laboratory test data was evaluated for production and pre-production technologies, technical publications, manufacturer and supplier CBI, and simulation modeling of specific technologies. Evaluating recently introduced production products to inform the technology effectiveness models of emerging technologies is preferred because doing so allows for a more reliable analysis of incremental improvements over previous technologies; however, some technologies were considered that are not yet in production. As technologies evolve and new applications emerge, this work will be continued and may include additional technologies and/or updated modeling for the final rule. The details of new and emerging technologies are discussed in PRIA Chapter 6.
Using full-vehicle simulation modeling has two primary advantages over using single or limited point estimates for fuel efficiency improvements of technologies. First, technology effectiveness often differs significantly depending on the type of vehicle and the other technologies that are on the vehicle, and this is shown in full-vehicle simulations. Different technologies may provide different fuel economy improvements depending on whether they are implemented alone or in tandem with other technologies. Single point estimates often oversimplify these important, complex relationships and lead to less accurate effectiveness estimates. Also, because manufacturers often implement a number of fuel-saving technologies simultaneously at vehicle redesigns, it is generally difficult to isolate the effect of individual technologies using laboratory measurement of production vehicles alone. Simulation modeling offers the opportunity to isolate the effects of individual technologies by using a single or small number of baseline configurations and incrementally adding technologies to those baseline configurations. This provides a consistent reference point for the incremental effectiveness estimates for each technology and for combinations of technologies for each vehicle type and reduces potential double counting or undercounting technology effectiveness. Note: It is most important that the incremental effectiveness of each technology and combinations be accurate and relative to a consistent baseline, because it is the incremental effectiveness that is applied to each vehicle model/configuration in the MY 2016 baseline fleet (and to each vehicle model/configuration's absolute fuel economy value) to determine the absolute fuel economy of the model/configuration with the additional technology. The absolute fuel economy values of the simulation modeling runs by themselves are used only to determine the incremental effectiveness and are never used directly to assign an absolute fuel economy value to any vehicle model/configuration for the rulemaking analysis. Therefore, commenters on technology effectiveness should be specific about the incremental effectiveness of technologies relative to other specifically defined technologies. The fuel economy of a specific vehicle or simulation modeling run in isolation may be less useful.
Second, full-vehicle simulation modeling requires explicit specifications and assumptions for each technology; therefore, these assumptions can be presented for public review and comment. For instance, transmission gear efficiencies, shift logic, and gear ratios are explicitly stated as model inputs and are available for review and comment. For today's analysis, every effort was made to make the input specifications and modeling assumptions available for review and comment. PRIA Chapter 6 and referenced documents provide more detailed information.
Technology development and application will be monitored to acquire more information for the final rule. The agencies may update the analysis for the final rule based on comments and/or new information that becomes available.
Today's analysis utilizes effectiveness estimates for technologies developed using Autonomie software,
a physics-based full-vehicle simulation tool developed and maintained by the Department of Energy's ANL. Autonomie has a long history of development and widespread application by users in industry, academia, research institutions and government.
Real-world use has contributed significantly to aspects of Autonomie important to producing realistic estimates of fuel economy and CO2 emission rates, such as estimation and consideration of performance, utility, and driveability metrics (e.g., towing capability, shift business, frequency of engine on/off transitions). This steadily increasing realism has, in turn, steadily increased confidence in the appropriateness of using Autonomie to make significant investment decisions. Notably, DOE uses Autonomie for analysis supporting budget priorities and plans for programs managed by its Vehicle Technologies Office (VTO) and to decide among competing vehicle technology R&D projects.
In the 2015 National Academies of Science (NAS) study of fuel economy improving technologies, the Committee recommended that the agencies use full-vehicle simulation to improve the analysis method of estimating technology effectiveness.
The committee acknowledged that developing and executing tens or hundreds of thousands of constantly changing vehicle packages models in Start Printed Page 43026real-time is extremely challenging. While initially this approach was not considered practical to implement, a process developed by Argonne in collaboration with NHTSA and the DOT Volpe Center has succeeded in enabling large scale simulation modeling. For more details about the Autonomie simulation model and its submodels and inputs, see PRIA Chapter 6.2.
Today's analysis modeled more than 50 fuel economy-improving technologies, and combinations thereof, on 10 vehicle types (an increase from five vehicle types in NHTSA's Draft TAR analysis). While 10 vehicle types may seem like a small number, a large portion of the production volume in the MY 2016 fleet have specifications that are very similar, especially in highly competitive segments (for instance, many mid-sized sedans, many small SUVs, and many large SUVs coalesce around similar specifications, respectively), and baseline simulations have been aligned around these modal specifications. The sequential addition of these technologies generated more than 100,000 unique technology combinations per vehicle class. The analysis included 10 technology classes, so more than one million full-vehicle simulations were run. In addition, simulation modeling was conducted to determine the appropriate amount of engine downsizing needed to maintain baseline performance across all modeled vehicle performance metrics when advanced mass reduction technology or advanced engine technology was applied, so these simulations take into account performance neutrality, given logical engine down-sizing opportunities associated with specific technologies.
Some baseline vehicle assumptions used in the simulation modeling were updated based on public comment and the assessment of the MY 2016 production fleet. The analysis included updated assumptions about curb weight, component inertia, as well as technology properties like baseline rolling resistance, aerodynamic drag coefficients, and frontal areas. Many of the assumptions are aligned with published research from the Department of Energy's Vehicle Technologies Office and other independent sources.
Additional transmission technologies and more levels of aerodynamic technologies than NHTSA presented in the Draft TAR analysis were also added for today's analysis. Having additional technologies allowed the agencies to assign baselines and estimate fuel-savings opportunities with more precision.
The 10 vehicle types (referred to as “technology classes” in the modeling documentation) are shown in Table II-7. Each vehicle type (technology class) represented a large segment of vehicles, such as medium cars, small SUVs, and medium performance SUVs.
Baseline parameters were defined with ANL for each technology class, including baseline curb weight, time required to accelerate from stop to 60 miles per hour, time required to accelerate from 50 miles per hour to 80 miles per hour, ability of the vehicle to maintain constant 65 miles per hour speed on a six percent upgrade, and (for some classes) assumptions about towing capability.
Start Printed Page 43027
From these baseline specifications, incremental combinations of fuel saving technologies were applied. As the combinations of technologies change, so too may predicted performance.
The analysis attempts to maintain performance by resizing engines at a few specific incremental technology steps. Steps from one technology to another typically associated with a major vehicle redesign, or engine redesign, were identified, and engine resizing was restricted only to these steps. The analysis allowed engine resizing when mass reduction of 10% or greater was applied to the vehicle glider mass,
and when one powertrain architecture was replaced with another architecture.
The analysis resized engines to the extent that performance was maintained for the least capable performance criteria to maintain vehicle utility for that criteria; therefore, sometimes other performance attributes may improve. For instance, the amount of engine resizing may be determined based on its high speed acceleration time if it is the least capable criteria, but that resizing may also improve the low speed acceleration time.
The analysis did not re-size the engine in response to adding technologies that have small effects on vehicle performance. For instance, if a vehicle's weight is reduced by a small amount causing the 0-60 mile per hour time to improve slightly, the analysis would not resize the engine. Manufacturers have repeatedly told the agencies that the high costs for redesign and the increased manufacturing complexity that would result from resizing engines for such small changes in the vehicle preclude doing so. The analysis should not, in fact, include engine resizing with the application of every technology or for combinations of technologies that drive small performance changes so that the analysis better reflects what is feasible for manufacturers to do.
2. CAFE model
The CAFE model is designed to simulate compliance with a given set of CAFE or CO2 standards for each manufacturer that sells vehicles in the United States. The model begins with a Start Printed Page 43028representation of the MY 2016 vehicle model offerings for each manufacturer that includes the specific engines and transmissions on each model variant, observed sales volumes, and all fuel economy improving technology that is already present on those vehicles. From there the model adds technology, in response to the standards being considered, in a way that minimizes the cost of compliance and reflects many real-world constraints faced by automobile manufacturers. The model addresses fleet year-by-year compliance, taking into consideration vehicle refresh and redesign schedules and shared platforms, engines, and transmissions among vehicles.
As a result of simulating compliance, the CAFE model provides the technology pathways that manufacturers could use to comply with regulations, including how technologies could be applied to each of their vehicle model/configurations in response to a given set of standards. The model calculates the impacts of the simulated standard: Technology costs, fuel savings (both in gallons and dollars), CO2 reductions, social costs and benefits, and safety impacts.
The current analysis reflects several changes made to the CAFE model since 2012, when NHTSA used the model to estimate the effects, costs, and benefits of final CAFE standards for light-duty vehicles produced during MYs 2017-2021 and augural standards for MYs 2022-2025. The changes are discussed in Section II.A.1, above, and PRIA Chapter 6.
3. Assumptions About Individual Technology Cost and Effectiveness Values
Cost and effectiveness values were estimated for each technology included in the analysis, with a summary list of all technologies provided in Table II-1 (List of Technologies with Data Sources for Technology Assignments) of Preamble Chapter II.B, above. In all, more than 50 technologies were considered in today's analysis, and the analysis evaluated many combinations of these technologies on many applications. Potential issues in assessing technology effectiveness and cost were identified, including:
Baseline (MY 2016) vehicle technology level assessed as too low, or too high. Compliance information was extensively reviewed and supplemented with available literature on many MY 2016 vehicle models. Manufacturers could also review the baseline technology assignments for their vehicles, and the analysis incorporates feedback received from manufacturers.
Technology costs too low or too high. Tear down cost studies, CBI, literature, and the 2015 NAS study information were referenced to estimate technology costs. In cases that one technology appeared exemplary on cost and effectiveness relative to all other technologies, information was acquired from additional sources to confirm or reject assumptions. Cost assumptions for emerging technologies are continuously being evaluated.
Technology effectiveness too high or too low in combination with other vehicle technologies. Technology effectiveness was evaluated using the Autonomie full-vehicle simulation modeling, taking into account the impact of other technologies on the vehicle and the vehicle type. Inputs and modeling for the analysis took into account laboratory test data for production and some pre-production technologies, technical publications, manufacturer and supplier CBI, and simulation modeling of specific technologies. Evaluating recently introduced production products to inform the technology effectiveness models of emerging technologies was preferred; however, some technologies that are not yet in production were considered, via CBI. Simulation modeling used carefully chosen baseline configurations to provide a consistent, reasonable reference point for the incremental effectiveness estimates.
Vehicle performance not considered or applied in an infeasible manner. Performance criteria, including low speed acceleration (0-60 mph time), high speed acceleration (50-80 mph time), towing, and gradeability (six percent grade at 65 mph) were also considered. In the simulation modeling, resizing was applied to achieve the same performance level as the baseline for the least capable performance criteria but only with significant design changes. The analysis struck a balance by employing a frequency of engine downsizing that took product complexity and economies of scale into account.
Availability of technologies for production application too soon or too late. A number of technologies were evaluated that are not yet in production. CBI was gathered on the maturity and timing of these technologies and the likely cadence at which manufacturers might adopt these technologies.
Product complexity and design cadence constraints too low or too high. Product platforms, refresh and redesign cycles, shared engines, and shared transmissions were also considered in the analysis. Product complexity and the cadence of product launches were matched to historical values for each manufacturer.
Customer acceptance under estimated or over estimated. Resale prices for hybrid vehicles, electric vehicles, and internal combustion engine vehicles were evaluated to assess consumer willingness to pay for those technologies. The analysis accounts for the differential in the cost for those technologies and the amount consumers have actually paid for those technologies. Separately, new dual-clutch transmissions and manual transmissions were applied to vehicles already equipped with these transmission architectures.
Please provide comments on all assumptions for fuel economy and CO2 technology costs, effectiveness, availability, and applicability to vehicles in the fleet.
The technology effectiveness modeling results show effectiveness of a technology often varies with the type of vehicle and the other technologies that are on the vehicle. Figure II-1 and Figure II-2 show the range of effectiveness for each technology for the range of vehicle types and technology combinations included in this NPRM analysis. The data reflect the change in effectiveness for applying each technology by itself while all other technologies are held unchanged. The data show the improvement in fuel consumption (in gallons per mile) and tailpipe CO2 over the combined 2-cycle test procedures. For many technologies, effectiveness values ranged widely; only a few technologies for which effectiveness may be reasonably represented as a fixed offset were identified.
For engine technologies, the effectiveness improvement range is relative to a comparably equipped vehicle with only variable valve timing (VVT) on the engine. For automatic transmission technologies, the effectiveness improvement range is over a 5-speed automatic transmission. For manual transmission technologies, the effectiveness improvement range is over a 5-speed manual transmission. For road load technologies like aerodynamics, rolling resistance, and mass reduction, the effectiveness improvement ranges are relative to the least advanced technology state, respectively. For hybrid and electric drive systems that wholly replace an engine and transmission, the effectiveness improvement ranges are relative to a comparably equipped vehicle with a basic engine with VVT only and a 5-speed automatic transmission. For hybrid or electrification technologies that complement other advanced engine Start Printed Page 43029and transmission technologies, the effectiveness improvement ranges are relative to a comparably equipped vehicle without the hybrid or electrification technologies (for instance, parallel strong hybrids and belt integrated starter generators retain engine technologies, such as a turbo charged engine or an Atkinson cycle engine). Many technologies have a wide range of estimated effectiveness values. Figure II-3 below shows a hierarchy of technologies discussed.
Start Printed Page 43030
Start Printed Page 43031
4. Engine Technologies
There are a number of engine technologies that manufacturers can use to improve fuel economy and CO2. Some engine technologies can be incorporated into existing engines with minor or moderate changes to the engines, but many engine technologies require an entirely new engine architecture.
In this section and for this analysis, the terms “basic engine technologies” and “advanced engine technologies” are used only to define how the CAFE model applies a specific engine technology and handles incremental costs and effectiveness improvements. “Basic engine technologies” refer to technologies that, in many cases, can be adapted to an existing engine with minor or moderate changes to the engine. “Advanced engine technologies” refer to technologies that generally require significant changes or an entirely new engine architecture. In the CAFE model, basic engine technologies may be applied in combination with other basic engine technologies; advanced engine technologies (defined by an engine map) stand alone as an exclusive engine technology. The words “basic” and “advanced” are not meant to confer any information about the level of sophistication of the technology. Also, many advanced engine technology Start Printed Page 43032definitions include some basic engine technologies, but these basic technologies are already accounted for in the costs and effectiveness values of the advance engine. The “basic engine technologies” need not be (and are not) applied in addition to the “advanced engine technologies” in the CAFE model.
Engines come in a wide variety of shapes, sizes, and configurations, and the incremental engine costs and effectiveness values often depend on engine architecture. The agencies modeled single overhead cam (SOHC), dual overhead cam (DOHC), and overhead valve (OHV) engines separately to account for differences in engine architecture. The agencies adjusted costs for some engine technologies based on the number of cylinders and number of banks in the engine, and the agencies evaluated many production engines to better understand how costs and capabilities may vary with engine configuration. Table II-8, Table II-9, Table II-10 below shows the summary of absolute costs 
for different technologies.
Start Printed Page 43033
Start Printed Page 43034
Start Printed Page 43035
Many types of production powertrains were reviewed and tested for this analysis, and engine maps were developed for each combination of Start Printed Page 43036engine technologies. For a given engine configuration, some production engines may be less efficient than the engine maps presented in the analysis, and some may be more efficient. Developing engine maps that reasonably represented most vehicles equipped with the engine technology, and that are in production today, was the preferred approach for this analysis. Additionally, some advanced engines were included in the simulation that are not yet in production. The engine maps for these engines were either based on CBI or were theoretical. The most recently released production engines are still being reviewed, and the analysis may include updated engine maps in the future or add entirely new engine maps to the analysis if either action could improve the quality of the fleet-wide analysis.
Stakeholders provided many comments on the engine maps that were presented in the Draft TAR. These comments were considered, and today's analysis utilizes several engine maps that were updated since the Draft TAR. Most notably, for turbocharged and downsized engines, the engine maps were adjusted in high torque, low speed operating conditions to address engine knock with regular octane fuel to align with the fuel octane that manufacturers recommend be used for the majority of vehicles. In the Draft TAR, NHTSA assumed high octane fuel to develop engine maps. See the discussion below and in PRIA Chapter 6.3 for more details. Please provide comment on the appropriateness of assuming the use of lower octane fuels.
(a) “Basic” Engine Technologies
The four “basic” engine technologies in today's model are Variable Valve Timing (VVT), Variable Valve Lift (VVL), Stoichiometric Gasoline Direct Injection (SGDI), and basic Cylinder Deactivation (DEAC). Over the last decade, manufacturers upgraded many engines with these engine technologies. Implementing these technologies involves changes to the cylinder head of the engine, but the engine block, crankshaft, pistons, and connecting rods require few, if any, changes. In today's analysis, manufacturers may apply the four basic engine technologies in various combinations, just as manufacturers have done recently.
(1) Variable Valve Timing (VVT)
Variable Valve Timing (VVT) is a family of valve-train designs that dynamically adjusts the timing of the intake valves, exhaust valves, or both, in relation to piston position. This family of technologies reduces pumping losses. VVT is nearly universally used in the MY 2016 fleet.
(2) Variable Valve Lift (VVL)
Variable Valve Lift (VVL) dynamically adjusts the travel of the valves to optimize airflow over a broad range of engine operating conditions. The technology increases effectiveness by reducing pumping losses and may improve efficiency by affecting in-cylinder charge (fuel and air mixture), motion, and combustion.
(3) Stoichiometric Gasoline Direct Injection (SGDI)
Stoichiometric Gasoline Direct Injection (SGDI) sprays fuel at high pressure directly into the combustion chamber, which provides cooling of the in-cylinder charge via in-cylinder fuel vaporization to improve spark knock tolerance and enable an increase in compression ratio and/or more optimal spark timing for improved efficiency. SGDI appears in about half of basic engines produced in MY 2016, and the technology is used in many advanced engines as well.
(4) Basic Cylinder Deactivation (DEAC)
Basic Cylinder Deactivation (DEAC) disables intake and exhaust valves and prevents fuel injection into some cylinders during light-load operation. The engine runs temporarily as though it were a smaller engine, which reduces pumping losses and improves efficiency. Manufacturers typically disable one-cylinder bank with basic cylinder deactivation. In the MY 2016 fleet, manufacturers used DEAC on V6, V8, V10, and V12 engines on OHV, SOHC, and DOHC engine configurations. With some engine configurations in some operating conditions, DEAC creates noise-vibration-and-harshness (NVH) challenges. NVH challenges are significant for V6 and I4 DEAC configurations. For I4 engine configurations, manufacturers can operate the DEAC function of an engine in very few operating conditions, with limited potential to save fuel. No manufacturers sold I4 DEAC engines in the MY 2016 fleet. Typically, the smaller the engine displacement, the less opportunity DEAC provides to improve fuel consumption.
Manufacturers and suppliers continue to evaluate more improved versions of cylinder deactivation, including advanced cylinder deactivation and pairing basic cylinder deactivation with turbo charged engines. No manufacturers produced such technologies in the MY 2016 fleet. Advanced cylinder deactivation and turbo technologies were modeled and considered separately in today's analysis.
(b) “Advanced” Engine Technologies
The analysis included “advanced” engine technologies that can deliver high levels of effectiveness but often require a significant engine design change or a new engine architecture. In the CAFE model, “basic” engine technologies may be considered in combination and applied before advanced engine technologies. “Advanced” engine technologies generally include one or more basic engine technologies in the simulation, without the need to layer on “basic” engine technologies on top of “advanced” engines. Once an advanced engine technology is applied, the model does not reconsider the basic engine technologies. The characterization of each advanced engine technology takes into account the prerequisite technologies.
Many of the newest advanced engine technologies improve effectiveness over their predecessors, but the engines may also include sophisticated materials or manufacturing processes that contribute to efficiency improvements. For instance, one recently introduced turbo charged engine uses sodium filled valve stems.
Another recently introduced high compression ratio engine uses a sophisticated laser cladding process to manufacture valve seats and improve airflow.
To fully consider these advancements (and their potential benefits), the incremental costs of these technologies, as well as the effectiveness improvements, must be accounted for.
(1) Turbocharged Engines
Turbo engines recover energy from hot exhaust gas and compress intake air, thereby increasing available airflow and increasing specific power level. Due to specific power improvements on turbo engines, engine displacement can be downsized. The downsizing reduces pumping losses and improves fuel economy at lower loads. For the NPRM analysis, a level of downsizing is assumed to be applied that achieves performance similar to the baseline naturally-aspirated engine. This assumes manufacturers would apply the benefits toward improved fuel economy Start Printed Page 43037and not trade off fuel economy improvements to increase overall vehicle performance. In practice, manufacturers have often also improved some vehicle performance attributes at the expense of not maximizing potential fuel economy improvements.
Manufacturers may develop engines to operate on varying levels of boost,
with higher levels of boost achieving higher engine specific power and enabling greater levels of engine downsizing and corresponding reductions in pumping losses for improved efficiency. However, engines operating at higher boost levels are generally more susceptible to engine knock,
especially at higher torques and low engine speeds. Additionally, engines with higher boost levels typically require larger induction and exhaust system components, dissipate greater amounts of heat, and with greater levels of engine downsizing have increased challenges with turbo lag.
For these reasons, three levels of turbo downsizing technologies are separately modeled in this analysis.
The analysis also modeled turbocharged engines with parallel hybrid technology. In simulations with high stringencies, many manufacturers produced turbo-hybrid electric vehicles. In the MY 2016 fleet, of the vehicles that use parallel hybrid technology, many use turbocharged engines.
Since the Draft TAR, the turbo family engine maps were updated to reflect operation on 87 AKI regular octane fuel.
In the Draft TAR, turbo engine maps were developed assuming premium fuel. For this rulemaking analyses, pathways to improving fuel economy and CO2 are analyzed, while also maintaining vehicle performance, capability, and other attributes. This includes assuming there is no change in the fuel octane required to operate the vehicle. Using 87 AKI regular octane fuel is consistent with the fuel octane that manufacturers specify for the majority of vehicles, and enables the modeling to account for important design and calibration issues associated with regular octane fuel. Using the updated criteria assures the NPRM analysis reflects real-world constraints faced by manufacturers to assure engine durability, and acceptable drivability, noise and harshness, and addresses the over-estimation of potential fuel economy improvements related to the fuel octane assumptions, which did not fully account for these constraints, in the Draft TAR. Compared with the NHTSA analysis in the Draft TAR, these engine maps adjust the fuel use at high torque and low speed operation and at high speed operation to fully account for knock limitations with regular octane fuel.
The analysis assumes engine downsizing with the addition of turbo technology. For instance, in the simulations, manufacturers may have replaced a naturally-aspirated V8 engine with a turbo V6 engine, and manufacturers may have replaced a naturally-aspirated V6 engine with a turbo I4 engine. When manufacturers reduced the number of banks or cylinders of an engine, some cost savings is projected due to fewer cylinders and fewer valves. Such cost savings is projected to help offset the additional costs of turbo charger specific hardware, making turbo downsizing a very attractive technology progression for some engines.
Level 1 Turbo Charging (TURBO1) adds a turbo charger to a DOHC engine with SGDI, VVT, and continuously VVL. The engine operates at up to 18 bar brake mean effective pressure (BMEP).
Manufacturers used Turbo1 technology in a little less than a quarter of the MY 2016 fleet with particularly high concentrations in premium vehicles.
Level 2 Turbo Charging (TURBO2) operates at up to 24 bar BMEP. The step from Turbo1 to Turbo2 is accompanied with additional displacement downsizing for reduced pumping losses. Very few manufacturers have Turbo2 technology in the MY 2016 fleet.
Turbo Charging with Cooled Exhaust Gas Recirculation (CEGR1) improves the knock resistance of Turbo2 engines by mixing cooled inert exhaust gases into the engine's air intake. That allows greater boost levels, more optimal spark timing for improved fuel economy, and performance and greater engine downsizing for lower pumping losses. CEGR1 technology is used in only a few vehicles in the MY 2016 fleet, and many of these vehicles include high-performance utility either for towing or acceleration.
(a) Turbocharged Engine Technologies Not Considered
Previous analyses considered turbo charged engines with even higher BMEP than today's Turbo2 and CEGR1 technologies, but today's analysis does not present 27 bar BMEP turbo engines. Turbo engines with very high BMEP have demonstrated limited potential to improve fuel economy due to practical limitations on engine downsizing and tradeoffs with launch performance and drivability. Based on the analysis, and based on CBI, CEGR2 turbo engine technology was not included in this NPRM analysis.
(2) High Compression Ratio Engines (Atkinson Cycle Engines)
Atkinson cycle gasoline engines use changes in valve timing (e.g., late-intake-valve-closing or LIVC) to reduce the effective compression ratio while maintaining the expansion ratio. This approach allows a reduction in top-dead-center (TDC) clearance ratio (e.g., increase in “mechanical” or “physical” compression ratio) to increase the effective expansion ratio without increasing the effective compression ratio to a point that knock-limited operation is encountered. Increasing the expansion ratio in this manner improves thermal efficiency but also lowers peak BMEP, particularly at lower engine speeds.
Often knock concerns for these engines limit applications in high load, low RPM conditions. Some manufacturers have mitigated knock concerns by lowering back pressure with long, intricate exhaust systems, but these systems must balance knock performance with emissions tradeoffs, and the increased size of the exhaust manifold can pose packaging concerns, particularly on V-engine configurations.
Only a few manufacturers produced internal combustion engine vehicles with Atkinson cycle engines in MY Start Printed Page 430382016; however, these engines are commonly paired with hybrid electric vehicle technologies due to the synergy of peak efficiency of Atkinson cycle engines and immediate torque from electric motors in strong hybrids. Atkinson cycle engines are very common on power split hybrids and are sometimes observed as part of a parallel hybrid system or plug-in hybrid system.
Atkinson cycle engines played a prominent role in EPA's January 2017 final determination, which has since been withdrawn. Today's analysis recognizes that the technology is not suitable for many vehicles due to performance, emissions and packaging issues, and/or the extensive capital and resources that would be required for manufacturers to shift from other powertrain technology pathways (such as turbocharging and downsizing) to standalone Atkinson cycle engine technology.
A number of Asian manufacturers have launched Atkinson cycle engines in smaller vehicles that do not use hybrid technologies. These production engines have been benchmarked to characterize HCR1 technology for today's analysis.
Today's analysis restricted the application of stand-alone Atkinson cycle engines in the CAFE model in some cases. The engines benchmarked for today's analysis were not suitable for MY 2016 baseline vehicle models that have 8-cylinder engines and in many cases 6-cylinder engines.
EPA conceptualized a “future” Atkinson cycle engine and published the theoretical engine map in an SAE paper. For this engine, EPA staff began with a best-in-class 2.0L Atkinson cycle engine and then increased the efficiency of the engine map further, through the theoretical application of additional technologies in combination, like cylinder deactivation, engine friction reduction, and cooled exhaust gas recirculation. This engine remains entirely speculative, as no production engine as outlined in the EPA SAE paper has ever been commercially produced or even produced as a prototype in a lab setting. Furthermore, the engine map has not been validated with hardware and bench data, even on a prototype level (as no such engine exists to test to validate the engine map).
Previously, EPA relied heavily on the HCR2 (or sometimes referred to as ATK2 in previous EPA analysis) engine as a cost effective pathway to compliance for stringent alternatives, but many engine experts questioned its technical feasibility and near term commercial practicability. Stakeholders asked for the engine to be removed from compliance simulations until the performance could be validated with engine hardware. 
While for the Draft TAR, the agencies ran full-vehicle simulations with the theoretical engine map and made these available in the CAFE model, HCR2 technology as described in EPA's SAE paper was not included in today's analysis because there has been no observable physical demonstration of the speculative technology, and many questions remain about its practicability as specified, especially in high load, low engine speed operating conditions. Simulations with EPA's HCR2 engine map produce results that approach (and sometimes exceed) diesel powertrain efficiency.
Given the prominence of this unproven technology in previous rule-makings, the CAFE model may be configured to consider the application of HCR2 technology for reference only.
As new engines emerge that achieve high thermal efficiency, questions may be raised as to whether the HCR2 engine is a simulation proxy for the new engine technology. It is important to conduct a thorough evaluation of the actual new production engines to measure the brake specific fuel consumption and to characterize the improvements attributable to friction and thermal efficiency before drawing conclusions. Using vehicle level data may misrepresent or conflate complex interactions between a high thermal efficiency engine, engine friction reduction, accessory load improvements, transmission technologies, mass reduction, aerodynamics, rolling resistance, and other vehicle technologies. For instance, some of the newest high compression ratio engines show improved thermal efficiency, in large part due to improved accessory loads or reduced parasitic losses from accessory systems.
The CAFE model allows for incremental improvement over existing HCR1 technologies with the addition of improved accessory devices (IACC), a technology that is available to be applied on many baseline MY 2016 vehicles with HCR1 engines and may be applied as part of a pathway of compliance to further improve the effectiveness of existing HCR1 engines.
(c) Emerging Gasoline Engine Technologies
Manufacturers and suppliers continue to invest in many emerging engine technologies, and some of these technologies are on the cusp of commercialization. Often, manufacturers submit information about new engine technologies that they may soon bring into production. When this happens, a collaborative effort is undertaken with suppliers and manufacturers to learn as much as possible and sometimes begin simulation modeling efforts. Bench data, or performance data for preproduction vehicles and engines, is usually closely held confidential business information. To properly characterize the technologies, it is often necessary to wait until the engine technologies are in production to study them.
(1) Advanced Cylinder Deactivation (ADEAC)
Advanced cylinder deactivation systems (or rolling or dynamic cylinder deactivation systems) allows a further degree of cylinder deactivation than DEAC. The technology allows the engine to vary the percentage of cylinders deactivated and the sequence in which cylinders are deactivated, essentially providing “displacement on demand” for low load operations, so long as the calibration avoids certain frequencies.Start Printed Page 43039
ADEAC systems may be integrated into the valvetrains with moderate modifications on OHV engines. However, while the ADEAC operating concept remains the same on DOHC engines, the valvetrain hardware configuration is very different, and application on DOHC engines is projected to be more costly per cylinder due to the valvetrain differences.
Some preproduction 8-cylinder OHV prototype vehicles were briefly evaluated for this analysis, but no production versions of the technology have been studied.
Today's analysis relied on CBI to estimate costs and effectiveness values of ADEAC. Since no engine map was available at the time of the NPRM analysis, ADEAC was estimated to improve a basic engine with VVL, VVT, SGDI, and DEAC by three percent (for 4 cylinder engines) six percent (for engines with more than 4 cylinders).
ADEAC systems will continue to be studied as production begins.
(2) Variable Compression Ratio Engines (VCR)
Engines using variable compression ratio (VCR) technology appear to be at a production-intent stage of development but also appear to be targeted primarily towards limited production, high performance and very high BMEP (27-30 bar) applications. Variable compression ratio engines work by changing the length of the piston stroke of the engine to operate at a more optimal compression ratio and improve thermal efficiency over the full range of engine operating conditions.
A number of manufacturers and suppliers provided information about VCR technologies, and several design concepts were reviewed that could achieve a similar functional outcome. In addition to design concept differences, intellectual property ownership complicates the ability of the agencies to define a VCR hardware system that could be widely adopted across the industry.
For today's analysis, VCR engines have a spot on the technology simulation tree, but VCR is not actively used in the NPRM simulation. Reasonable representations of costs and technology characterizations remain open questions for VCR engine technology and the analysis.
NHTSA is sponsoring work to develop engine maps for additional combinations of technologies. Some of these technologies being researched presently, including VCR, may be used in the analysis supporting the final rule. Please provide comment on variable compression ratio engine technology. Should VCR technology be employed in the timeframe of this proposed rulemaking? Why or why not? Do commenters believe VCR technology will see widespread adoption in the US vehicle fleet? Why or why not? What vehicle segments may it best be suited for, and which segments would it not be best suited for? Why or why not? What cost and effectiveness values should be used if VCR is modeled for analysis? Please provide supporting data. Additionally, please provide any comments on the sponsored work related to VCR, described further in PRIA Chapter 6.3.
(3) Compression Ignition Gasoline Engines (SpCCI, HCCI)
For many years, engine developers, researchers, manufacturers have explored ways to achieve the inherent efficiency of a diesel engine while maintaining the operating characteristics of a gasoline engine. A potential pathway for striking this balance is utilizing compression ignition for gasoline fueled engines, more commonly referred to as Homogeneous Charge Compression Ignition (HCCI).
Ongoing, periodic discussions with manufacturers on future fuel saving technologies and powertrain plans have, generally, included HCCI as a long-term strategy. The technology appears to always be a strong consideration as, in theory, it provides the “best of both worlds,” meaning a way to provide diesel engine efficiency with gasoline engine performance and emissions levels.
Developments in both the research and the potential production implementation of HCCI for the US market is continually assessed. In 2017, a significant, potentially production breakthrough was announced by Mazda regarding a gasoline-fueled engine employing Spark Controlled Compression Ignition (SpCCI), where HCCI is employed for a portion of its normal operation and spark ignition is used at other times.
Soon after, Mazda publicly stated they plan to introduce this engine as part of the Skyactiv family of engines in 2019.
However, HCCI was not included in the simulation and vehicle fleet modeling for past rulemakings, and is not included in this NPRM analysis, primarily because effectiveness, cost, and mass market implementation readiness data are not available.
Please comment on the potential use of HCCI technology in the timeframe covered by this rule. More specifically, should HCCI be included in the final rulemaking analysis for this proposed rulemaking? Why or why not? Please provide supporting data, including effectiveness values, costs in relation varying engine types and applications, and production timing that supports the timeframe of this rulemaking.
(d) Diesel Engines
Diesel engines have several characteristics that give superior fuel efficiency, including reduced pumping losses due to lack of (or greatly reduced) throttling, high pressure direct injection of fuel, a combustion cycle that operates at a higher compression ratio, and a very lean air/fuel mixture relative to an equivalent-performance gasoline engine. This technology requires additional enablers, such as a NOX adsorption catalyst system or a urea/ammonia selective catalytic reduction system for control of NOX emissions during lean (excess air) operation.
(e) Alternative Fuel Engines
(1) Compressed Natural Gas (CNG)
Compressed Natural Gas (CNG) engines use compressed natural gas as a fuel source. The fuel storage and supply systems for these engines differ tremendously from gasoline, diesel, and flex fuel vehicles.
(2) Flex Fuel Engines
Flex fuel engines can run on regular gasoline and fuel blended with ethanol. These vehicles may require additional equipment in the fuel system to effectively supply different blends of fuel to the engine.
(f) Lubrication and Friction Reduction
Low-friction lubricants including low viscosity and advanced low friction lubricant oils are now available (and widely used). If manufacturers choose to make use of these lubricants, they may need to make engine changes and conduct durability testing to accommodate the lubricants. The level of low friction lubricants exceeded 85% penetration in the MY 2016 fleet.
Reduction of engine friction can be achieved through low-tension piston rings, roller cam followers, improved material coatings, more optimal thermal management, piston surface treatments, and other improvements in the design of Start Printed Page 43040engine components and subsystems that improve efficient engine operation.
Manufacturers have already widely adopted both lubrication and friction reduction technologies. This analysis includes advanced engine maps that already assume application of low-friction lubricants and engine friction reduction technologies. Therefore, additional friction reduction is not considered in today's analysis.
The use and commercial development of improved lubricants and friction reduction components will continue to be monitored, including conical boring and oblong cylinders, and future analyses may be updated if new information becomes available.
5. Fuel Octane
(a) What is fuel octane level?
Gasoline octane levels are an integral part of potential engine performance. According the United States Energy Information Administration (EIA), octane ratings are measures of fuel stability. These ratings are based on the pressure at which a fuel will spontaneously combust (auto-ignite) in a testing engine.
Spontaneous combustion is an undesired condition that will lead to serious engine damage and costly repairs for consumers if not properly managed. The higher an octane number, the more stable the fuel, mitigating the potential for spontaneous combustion, also commonly known as “knock.” Modern engine control systems are sophisticated and allow manufacturers to detect when “knock” occurs during engine operation. These control systems are designed to adjust operating parameters to reduce or eliminate “knock” once detected.
In the United States, consumers are typically able to select from three distinct grades of fuel, each of which provides a different octane rating. The octane levels can vary from region to region, but on the majority, the octane levels offered are regular (the lowest octane fuel-generally 87 Anti-Knock Index (AKI) also expressed as (the average of Research Octane + Motor Octane), midgrade (the middle range octane fuel-generally 89-90 AKI), and premium (the highest octane fuel-generally 91-94 AKI).
At higher elevations, the lowest octane rating available can drop to 85 AKI.
Currently, throughout the United States, pump fuel is a blend of 90% gasoline and 10% ethanol. It is standard practice for refiners to manufacture gasoline and ship it, usually via pipelines, to bulk fuel terminals across the country. In many cases, refiners supply lower octane fuels than the minimum 87-octane required by law to these terminals. The terminals then perform blending operations to bring the fuel octane level up to the minimum required by law, and higher. In some cases, typically to lowest fuel grade, the “base fuel” is blended with ethanol, which has a typical octane rating of approximately 113. For example, in 2013, the State of Nebraska Ethanol Board defined requirements for refiners to 84-octane gas for blending to achieve 87-octane prior to final dispensing to consumers.
(b) Fuel Octane Level and Engine Performance
A typical, overarching goal of optimal spark-ignited engine design and operation is to maximize the greatest amount of energy from the fuel available, without manifesting detrimental impacts to the engine over its expected operating conditions. Design factors, such as compression ratio, intake and exhaust value control specifications, combustion chamber and piston characteristics, among others, are all impacted by octane (stability) of the fuel consumers are anticipated to use.
Vehicle manufacturers typically develop their engines and engine control system calibrations based on the fuel available to consumers. In many cases, manufacturers may recommend a fuel grade for best performance and to prevent potential damage. In some cases, manufacturers may require a specific fuel grade for both best performance and/or to prevent potential engine damage.
Consumers, though, may or may not choose to follow the recommendation or requirement for a specific fuel grade. Additionally, regional fuel availability could also limit consumer choice, or, in the case of higher elevation regions, present an opportunity for consumers to use a fuel grade that is below the minimum recommended. As such, vehicle manufacturers employ strategies for scenarios where a lower than recommended, or required, fuel grade is used, mitigating engine damage over the life of a vehicle.
When knock (also referred to as detonation) is encountered during engine operation, at the most basic level, non-turbo charged engines can reduce or eliminate knock by adjusting the timing of the spark that ignites the fuel, as well as the amounts of fuel injected at each intake stroke (“fueling”). In turbo-charged applications, boost levels are typically reduced along with spark timing and fueling adjustments. Past rulemakings have also discussed other techniques that may be employed to allow higher compression ratios, more optimal spark timing to be used without knock, such as the addition of cooled exhaust gas recirculation (EGR). Regardless of the type of spark-ignition engine or technology employed, reducing or preventing knock results in the loss of potential power output, creating a “knock-limited” constraint on performance and efficiency.
Despite limits imposed by available fuel grades, manufacturers continue to make progress in extracting more power and efficiency from spark-ignited engines. Production engines are safely operating with regular 87 AKI fuel with compression ratios and boost levels once viewed as only possible with premium fuel. According to the Department of Energy, the average gasoline octane level has remained fundamentally flat starting in the early 1980's and decreased slightly starting in the early 2000s. During this time, however, the average compression ratio for the U.S. fleet has increased from 8.4 to 10.52, a more than 20% increase, yielding the statement that, “There is some concern that in the future, auto manufacturers will reach the limit of technological increases in compression ratios without further increases in the octane of the fuel.” 
As such, manufacturers are still limited by the available fuel grades to consumers and the need to safeguard the durability of their products for all of the available fuels; thus, the potential Start Printed Page 43041improvement in the design of spark-ignition engines continues to be overshadowed by the fuel grades available to consumers.
(c) Potential of Higher Octane Fuels
Automakers and advocacy groups have expressed support for increases to fuel octane levels for the U.S. market and are actively participating in Department of Energy research programs on the potential of higher octane fuel usage. Some positions for potential future octane levels include advocacy for today's premium grade becoming the base grade of fuel available, which could enable low cost design changes that would improve fuel economy and CO2. Challenges associated with this approach include the increased fuel cost to consumers who drive vehicles designed for current regular octane grade fuel that would not benefit from the use of the higher cost higher octane fuel. The net costs for a shift to higher octane fuel would persist well into the future. Net benefits for the transition would not be achieved until current regular octane fuel is not available in the North American market, causing manufacturers to redesign all engines to operate the higher octane fuel, and then after those vehicles have been in production a sufficient number of model years to largely replace the current on-road vehicle fleet. The transition to net positive benefits could take many years.
In anticipation of this proposed rulemaking, organizations such as the High Octane Low Carbon Alliance (HOLC) and the Fuel Freedom Foundation (FFA), have shared their positions on the potential for making higher octane fuels available for the U.S. market. Other stakeholders also commented to past NHTSA rulemakings and/or the Draft TAR regarding the potential for increasing octane levels for the U.S. market.
In the meetings with HOLC and the FFA, the groups advocated for the potential benefits high octane fuels could provide via the blending of non-petroleum feedstocks to increase octane levels available at the pump. The groups' positions on benefits took both a technical approach by suggesting an octane level of 100 is desired for the marketplace, as well as, the benefits from potential increased national energy security by reduced dependencies on foreign petroleum.
(d) Fuel Octane—Request for Comments
Please comment on the potential benefits, or dis-benefits, of considering the impacts of increased fuel octane levels available to consumers for purposes of the model. More specifically, please comment on how increasing fuel octane levels would play a role in product offerings and engine technologies. Are there potential improvements to fuel economy and CO2 reductions from higher octane fuels? Why or why not? What is an ideal octane level for mass-market consumption balanced against cost and potential benefits? What are the negatives associated with increasing the available octane levels and, potentially, eliminating today's lower octane fuel blends? Please provide supporting data for your position(s).
6. Transmission Technologies
Transmissions transmit torque from the engine to the wheels. Transmissions may improve fuel efficiency primarily through two mechanisms: (1) Transmissions with more gears allow the engine to operate more regularly at the most efficient speed-load points, and (2) transmissions may have improvements in friction (gears, bearings, seals, and so on), or improvements in shift efficiency that help the transmission transfer torque more efficiently, lowering parasitic losses. These mechanisms are very different, so full-vehicle simulation is helpful to understand how a transmission may work with complementary equipment to improve fuel economy.
Today's analysis significantly increased the number of transmissions modeled in full-vehicle simulations, attempting to more closely align the Department of Energy full-vehicle simulations with existing vehicles. Previously, EPA included just five transmissions 
by vehicle class in their analysis, and often EPA represented upgrades among manual, automatic, continuously variable, and dual clutch transmissions with the same effectiveness 
and cost values 
within a vehicle class. Today's analysis simulated nearly 20 transmissions, with explicit assumptions about gear ratios, gear efficiencies, gear spans, shift logic, and transmission architecture. This analysis improves transparency by making clear the assumptions underlying the transmissions in the full-vehicle simulations and by increasing the number of transmissions simulated since the Draft TAR. Methods will be continuously evaluated to improve transmission models in full-vehicle simulations. For the box plots of effectiveness values, as shown in the PRIA Chapter 6, all automatic transmissions are relative to a 5-speed automatic, and all manual transmissions are relative to a 5-speed manual. Table II-11 below shows the absolute costs of transmission used for this analysis including learning and retail price equivalent.
Start Printed Page 43042
(a) Automatic Transmissions
Five-, six-, seven-, eight-, nine- and ten-speed automatic transmissions are optimized by changing the gear ratios to enable the engine to operate in a more efficient operating range over a broader range of vehicle operating conditions. While a six speed transmission application was most prevalent for the MYs 2012-2016 final rule, eight and higher speed transmissions were more prevalent in the MY 2016 fleet.
“L2” and “L3” transmissions designate improved gear efficiency and reduced parasitic losses. Few transmissions in the MY 2016 fleet have achieved “L2” efficiency, and the highest level of transmission efficiencies modeled are assumed to be available in MY 2022.
(1) Continuously Variable Transmissions
Continuously variable transmission (CVT) commonly uses V-shaped pulleys connected by a metal belt rather than gears to provide ratios for operation. Unlike manual and automatic transmissions with fixed transmission ratios, continuously variable transmissions can provide fully variable and an infinite number of transmission Start Printed Page 43043ratios that enable the engine to operate in a more efficient operating range over a broader range of vehicle operating conditions. In this NPRM, two levels of CVTs are considered for future vehicles. The second level CVT would have a wider transmission ratio, increased torque capacity, improvements in oil pump efficiency, lubrication improvements, and friction reduction. While CVTs work well with light loads, the technology as modeled is not suitable for larger vehicles such as trucks and large SUVs.
(2) Dual Clutch Transmissions
Dual clutch or automated shift manual transmissions (DCT) are similar to manual transmissions except for the vehicle controls shifting and launch functions. A dual-clutch automated shift manual transmission uses separate clutches for even-numbered and odd-numbered gears, so the next expected gear is pre-selected, which allows for faster and smoother shifting. The 2012-2016 final rule limited DCT applications to a maximum of 6-speeds. Both 6-speed and 8-speed DCT transmissions are considered in today's proposal.
Dual clutch transmissions are very effective transmission technologies, and previous rule-making projected rapid, and wide adoption into the fleet. However, early DCT product launches in the U.S. market experienced shift harshness and poor launch performance, resulting in customer satisfaction issues—some so extreme as to prompt vehicle buyback campaigns.
Most manufacturers are not using DCTs in the U.S. market due to the customer satisfaction issues. Manufacturers used DCTs in about three percent of the MY 2016 fleet. Today's analysis limits the application of improved DCTs to vehicles that already use DCTs. Many of these vehicles are imported performance products.
(b) Manual Transmissions
Manual 6- and 7-speed transmissions offer an additional gear ratio, sometimes with a higher overdrive gear ratio, over a 5-speed manual transmission. Similar to automatic transmissions, more gears often means the engine may operate in the efficient zone more frequently.
7. Vehicle Technologies
As discussed earlier in Section II.D.1.b)(1), several technologies were considered for this analysis, and Table II-12, Table II-13, and Table II-14 below shows the full list of vehicle technologies analyzed and the associated absolute cost.
Start Printed Page 43044
Start Printed Page 43045
Start Printed Page 43046
(a) Reduced Rolling Resistance
Lower-rolling-resistance tires have characteristics that reduce frictional losses associated with the energy dissipated mainly in the deformation of the tires under load, thereby improving fuel economy and reducing CO2 emissions. New for this proposal, and also marking an advance over low rolling resistance tires considered during the heavy duty greenhouse gas rulemaking,
is a second level of lower rolling resistance tires that reduce frictional losses even further. The first level of low rolling resistance tires will have 10% rolling resistance reduction while the second level would have 20% rolling resistance reduction. In this NPRM, baseline vehicle reference rolling resistance values were determined based on the MY 2016 vehicles rather than the MY 2008 vehicles used in the 2012 final rule. Rolling resistance values were assigned based on CBI shared by manufacturers.
In some cases, low rolling resistance tires can affect traction, which may be untenable for some high performance vehicles. For cars and SUVs with more than 405 horsepower, the analysis restricted the application of the highest levels of rolling resistance. For cars and SUVs with more than 500 horsepower, the analysis restricted the application of any additional rolling resistance technology.
(b) Reduced Aerodynamic Drag Coefficient
Aerodynamic drag reduction can be achieved via two approaches, either by reducing the drag coefficients or reducing vehicle frontal area. To reduce the drag coefficient, skirts, air dams, underbody covers, and more aerodynamic side view mirrors can be applied. In the MY 2017-2025 final rule and the 2016 Draft TAR, the analysis included two levels of aerodynamic technologies. The second level included active grille shutters, rear visors, and larger under body panels. This NPRM expanded the aerodynamic drag improvements from two levels to four to provide more discrete levels. The NPRM levels are 5%, 10%, 15%, and 20% Start Printed Page 43047improvement relative to baseline reference vehicles. The agencies relied on the wind tunnel testing performed by National Research Council (NRC), Canada, Transport Canada (TC), and Environment and Climate Change, Canada (ECCC) to quantify the aerodynamic drag impacts of various OEM aerodynamic technologies and to explore the improvement potential of these technologies by expanding the capability and/or improving the design of MY 2016 state-of-the-art aerodynamic treatments. The agencies estimated the level of aerodynamic drag in each vehicle model in the MY 2016 baseline fleet and gathered CBI on aerodynamic drag coefficients, so each vehicle has an appropriate initial value for further improvements.
Notably, today's analysis assumes aerodynamic drag reduction can only come from reduction in the aerodynamic drag coefficient and not from reduction of frontal area.
For some bodystyles, the agencies have no evidence that manufacturers may be able to achieve 15% or 20% aerodynamic drag coefficient reduction relative to baseline for some bodystyles (for instance, with pickup trucks) due to form drag limitions. Previously, EPA analysis assumed some vehicles from all bodystyles could (and would) reduce aerodynamic forces by 20%, which in some cases led to future pickup trucks having aerodynamic drag coefficients better than some of today's typical cars, if frontal area were held constant. While ANL created full-vehicle simulations for trucks with 20% drag reduction, those simulations were not used in the CAFE modeling. That level of drag reduction is likely not technologically feasible with today's technology, and the analysis accordingly restricted the application of advanced levels of aerodynamics in some instances, such as in this case, due to bodystyle form drag limitations. Separate from form drag limitations, some high performance vehicles already use advanced aerodynamics technologies to generate down force, and sometimes these applications must trade-off between aerodynamic drag coefficient reduction and down force. Today's analysis does not apply 15% or 20% aerodynamic drag coefficient reduction to cars and SUVs with more than 405 horsepower.
(c) Mass Reduction
Mass Reduction can be achieved in many ways, such as material substitution, design optimization, part consolidation, improving manufacturing process, etc. The analysis utilizes mass reduction levels of 5, 10, 15, and 20% relative to a reference glider vehicle for each vehicle subsegment. For HEV, PHEV, and BEV vehicles, net mass reduction was considered, including the mass reduction applied to the glider and the added mass of electrification components. An extensive discussion of mass reduction technologies as well as the cost of mass reduction is located in Chapter 6.3 of the PRIA. The analysis included an estimated level of mass reduction technology in each vehicle model in the MY 2016 baseline fleet so that each vehicle model has an appropriate initial value for further improvements.
(d) Low Drag Brakes (LDB)
Low-drag brakes reduce the sliding friction of disc brake pads on rotors when the brakes are not engaged because the brake pads are pulled away from the rotors.
(e) Secondary Axle Disconnect (SAX)
Front or secondary axle disconnect for all-wheel drive systems provides a torque distribution disconnect between front and rear axles when torque is not required for the non-driving axle. This results in the reduction of associated parasitic energy losses.
8. Electrification Technologies
For this NPRM, the analysis of electrification technologies relies primarily on research published by the Department of Energy, ANL.
ANL's assumptions regarding all hybrid systems, including belt-integrated starter generators, strong parallel and series hybrids, plug-in hybrids,
and battery electric vehicles, and most projected technology costs were adopted for this analysis. In addition, the most recent ANL BatPaC model is used to estimate battery costs. Table II-15 and Table II-16 below show the absolute costs of all electrification technologies estimated for this NPRM analysis relative to a basic internal combustion engine vehicle with a 5-speed automatic transmission.
Start Printed Page 43048
Start Printed Page 43049
(a) Hybrid Technologies
(1) 12-Volt Stop-Start
12-volt Stop-Start, sometimes referred to as idle-stop or 12-volt micro hybrid, is the most basic hybrid system that facilitates idle-stop capability. These systems typically incorporate an enhanced performance battery and other features such as electric transmission pump and cooling pump to maintain vehicle systems during idle-stop.
(2) Higher Voltage Stop-Start/Belt Integrated Starter Generator
Higher Voltage Stop-Start/Belt Integrated Starter Generator (BISG), sometimes referred to as a mild hybrid system, provides idle-stop capability and uses a higher voltage battery with increased energy capacity over typical automotive batteries. The higher system voltage allows the use of a smaller, more powerful electric motor. This system replaces a standard alternator with an enhanced power, higher voltage, higher efficiency starter-alternator, that is belt driven and that can recover braking energy while the vehicle slows down (regenerative braking). Today's analysis assumes 48V systems on cars and small SUVs and high voltage systems for large SUVs and trucks. Future analysis may reference the application and operation of 48V systems on large SUVs and trucks, if applicable.
(3) Integrated Motor Assist (IMA)/Crank Integrated Starter Generator
Integrated Motor Assist (IMA)/Crank integrated starter generator (CISG) provides idle-stop capability and uses a high voltage battery with increased energy capacity over typical automotive batteries. The higher system voltage allows the use of a smaller, more powerful electric motor and reduces the weight of the wiring harness. This system replaces a standard alternator with an enhanced power, higher voltage, higher efficiency starter alternator that is crankshaft-mounted and can recover braking energy while the vehicle slows down (regenerative braking).
(4) P2 Hybrid
P2 Hybrid (SHEVP2) is a newly emerging hybrid technology that uses a transmission-integrated electric motor placed between the engine and a gearbox or CVT, much like the IMA system described above except with a wet or dry separation clutch that is used to decouple the motor/transmission from the engine. In addition, a P2 hybrid would typically be equipped with a larger electric machine. Disengaging the clutch allows all-electric operation and more efficient brake-energy recovery. Engaging the clutch allows efficient coupling of the engine and electric motor and, when combined with a DCT transmission, reduces gear-train losses relative to power-split or 2-mode hybrid systems. Battery costs are now considered separately from other HEV hardware.
P2 Hybrid systems typically rely on the internal combustion engine to deliver high, sustained power levels. While many vehicles may use HCR1 engines as part of a hybrid powertrain, HCR1 engines may not be suitable for all vehicles, especially high performance vehicles, or vehicles designed to carry or tow large loads. Many manufacturers may prefer turbo engines (with high specific power output) for P2 Hybrid systems.
(5) Power-Split Hybrid
Power-split Hybrid (SHEVPS) is a hybrid electric drive system that replaces the traditional transmission with a single planetary gearset and a motor/generator. This motor/generator uses the engine to either charge the battery or supply additional power to the drive motor. A second, more powerful motor/generator is permanently connected to the vehicle's final drive and always turns with the wheels. The planetary gear splits engine power between the first motor/generator and the drive motor to either charge the battery or supply power to the wheels. The power-split hybrid technology is included in this analysis as an enabling technology supporting this proposal, (the agencies evaluate the P2 hybrid technology discussed above where power-split hybrids might otherwise have been appropriate). As stated above, battery costs are now considered separately from other HEV hardware. Power-split hybrid technology as modeled in this analysis is not suitable for large vehicles that must handle high loads.
The ANL Autonomie simulations assumed all power-split hybrids use a high compression ratio engine. Therefore, all vehicles equipped with SHEVPS technology in the CAFE model inputs and simulations are assumed to have high compression ratio engines.
(6) Plug-in Hybrid Electric
Plug-in hybrid electric vehicles (PHEV) are hybrid electric vehicles with the means to charge their battery packs from an outside source of electricity (usually the electric grid). These vehicles have larger battery packs with more energy storage and a greater capability to be discharged than other hybrid electric vehicles. They also use a control system that allows the battery pack to be substantially depleted under electric-only or blended mechanical/electric operation and batteries that can be cycled in charge sustaining operation at a lower state of charge than is typical of other hybrid electric vehicles. These vehicles are sometimes referred to as Range Extended Electric Vehicles (REEV). In this NPRM analysis, PHEVs with two all-electric ranges—both a 30 mile and a 50 mile all-electric range—have been included as potential technologies. Again, battery costs are now considered separately from other PHEV hardware.
The ANL Autonomie simulations assumed all PHEVs use a high compression ratio engine. Therefore, all vehicles equipped with PHEV technology in the CAFE model inputs and simulations are assumed to have high compression ratio engines. In practice, many PHEVs recently introduced in the marketplace use turbo-charged engines in the PHEV system, and this is particularly true for PHEVs produced by European manufacturers and for other PHEV performance vehicle applications.
Please provide comment on the modeling of PHEV systems. Should turbo PHEVs be considered instead, or in addition to high compression ratio PHEVs? Why or why not? What vehicle segments may turbo PHEVs best be suited for, and which segments would it not be best suited for? What vehicle segments may high compression ratio PHEVs best be suited for, and which segments would it not be best suited for? Similarly, the analysis currently considers PHEVs with 30-mile and 50-mile all-electric range, and should other ranges be considered? For instance, a 20-mile all-electic range may decrease the battery pack size, and hence the battery pack cost relative to a 30-mile all-electric range system, while still providing electric-vehicle functionality in many applications. Do commenters believe PHEV technology will see widespread adoption in the US vehicle fleet? Why or why not? Please provide supporting data.
(b) Full Electrification and Fuel Cell Vehicles
(1) Battery Electric
Electric vehicles (EV) are equipped with all-electric drive and with systems powered by energy-optimized batteries charged primarily from grid electricity. EVs with range of 200 miles have been included as a potential technology in this NPRM. Battery costs are now considered separately from other EV hardware.Start Printed Page 43050
(2) Fuel Cell Electric
Fuel cell electric vehicles (FCEVs) utilize a full electric drive platform but consume electricity generated by an onboard fuel cell and hydrogen fuel. Fuel cells are electrochemical devices that directly convert reactants (hydrogen and oxygen via air) into electricity, with the potential of achieving more than twice the efficiency of conventional internal combustion engines. High pressure gaseous hydrogen storage tanks are used by most automakers for FCEVs. The high pressure tanks are similar to those used for compressed gas storage in more than 10 million CNG vehicles worldwide, except that they are designed to operate at a higher pressure (350 bar or 700 bar vs. 250 bar for CNG). FCEVs are currently produced in limited numbers and are available in limited geographic areas.
(c) Electric Vehicle Infrastructure
BEVs and PHEVs may be charged at home or elsewhere. Home chargers may access electricity from a regular wall outlet, or they may require special equipment to be installed at the home. Commercial chargers may sometimes be found at businesses or other travel locations. These chargers often may supply power to the vehicle at a faster rate of charge but often require significant capital investment to install.
Time to charge, and availability and convenience of charging are significant factors for plug-in vehicle operators. For many consumers, accessible charging stations present inconveniences that may deter the adoption of battery electric and plug-in hybrid vehicles.
More detail about charging and charging infrastructure, including a discussion of potential electric vehicle impacts on the electric grid, is available in the PRIA, Chapter 6. For today's analysis, costs for installing chargers and charge convenience is not taken into account in the CAFE model. Also, today's analysis assumes HEVs, PHEVs, and BEVs have the same survival rates and mileage accumulation schedules as vehicles with conventional powertrains, and that HEVs, PHEVs, and BEVs never receive replacement batteries before being scrapped. The agencies invite comment on these assumptions and on data and practicable methods to implement any alternatives.
9. Accessory Technologies
Two accessory technologies, electric power steering (EPS) and improved accessories (IACC) (accessory technologies categorized for the 2012 rule) were considered in this analysis, and are described below.
Table II-17 and Table II-18 below shows the estimated absolute costs including learning effects and retail price equivalent utilized in today's analysis.
(a) Electric Power Steering (EPS)
Electric power steering (EPS)/Electrohydraulic power steering (EHPS) is an electrically-assisted steering system that has advantages over traditional hydraulic power steering because it replaces a continuously operated hydraulic pump, thereby reducing parasitic losses from the accessory drive. Manufacturers have informed the agencies that full EPS systems are being developed for all types of light-duty vehicles, including large trucks. However, this analysis applies the EHPS technology to large trucks and the EPS technology to all other light-duty vehicles.Start Printed Page 43051
(b) Improved Accessories (IACC)
Improved accessories (IACC) may include high efficiency alternators, electrically driven (i.e., on-demand) water pumps, variable geometry oil pumps, cooling fans, a mild regeneration strategy, and high efficiency alternators. It excludes other electrical accessories such as electric oil pumps and electrically driven air conditioner compressors. In the MY 2017-2025 final rule, two levels of IACC were offered as a technology path (a low improvement level and a high improvement level). Since much of the market has incorporated some of these technologies in the MY 2016 fleet, the analysis assumes all vehicles have incorporated what was previously the low level, so only the high level remains as an option for some vehicles.
10. Other Technologies Considered but Not Included in This Aanalysis
Manufacturers, suppliers, and researchers continue to create a diverse set of fuel economy technologies. Many high potential technologies that are still in the early stages of the development and commercialization process are still being evaluated for any final analysis. Due to uncertainties in the cost and capabilities of emerging technologies, some new and pre-production technologies are not yet a part of the CAFE model simulation. Evaluating and benchmarking promising fuel economy technologies continues to be a priority as commercial development matures.
(a) Engine Technologies
- Variable compression ratio (VCR)—varies the compression ratio and swept volume by changing the piston stroke on all cylinders. Manufacturers accomplish this by changing the effective length of the piston connecting rod, with some prototypes having a range of 8:1 to 14:1 compression ratio. In turbocharged form, early publications suggest VCR may be possible to deliver up to 35% improved efficiency over the existing equivalent-output naturally-aspirated engine.
- Opposed-piston engine—sometimes known as opposed-piston opposed-cylinder (OPOC), operates with two pistons per cylinder working in opposite reciprocal motion and running on a two-stroke combustion cycle. It has no cylinder head or valvetrain but requires a turbocharger and supercharger for engine breathing. The efficiency may be significantly higher than MY 2016 turbocharged gasoline engines with competitive costs. This engine architecture could run on many fuels, including gasoline and diesel. Packaging constraints, emissions compliance, and performance across a wide range of operating conditions remain as open considerations. No production vehicles have been publicly announced, and multiple manufacturers continue to evaluate the technology.
- Dual-fuel—engine concepts such as reactivity controlled compression ignition (RCCI) combine multiple fuels (e.g. gasoline and diesel) in cylinder to improve brake thermal efficiency while reducing NOX and particulate emissions. This technology is still in the research phase.
- Smart accessory technologies—can improve fuel efficiency through smarter controls of existing systems given imminent or expected controls inputs in real world driving conditions. For instance, a vehicle could adjust the use of accessory systems to conserve energy and fuel as a vehicle approaches a red light. Vehicle connectivity and sensors can further refine the operation for more benefit and smoother operation.
- High Compression Miller Cycle Engine with Variable Geometry Turbocharger or Electric Supercharger—Atkinson cycle gasoline engines with sophisticated forced induction system that requires advanced controls. The benefits of these technologies provide better control of EGR rates and boost which is achieved with electronically controlled turbocharger or supercharger. The electric version of this technology which incorporates 48V is called E-boost.
(b) Electrified Vehicle Powertrain
- Advanced battery chemistries—many emerging battery technologies promise to eventually improve the cost, safety, charging time, and durability in comparison to the MY 2016 automotive lithium-ion batteries. For instance, many view solid state batteries as a promising medium-term automotive technology. Solid state batteries replace the battery's liquid electrolyte with a solid electrolyte to potentially improve safety, power and energy density, durability, and cost. Some variations use ceramic, polymer, or sulfide-based solid electrolytes. Multiple automakers and suppliers are exploring the technology and possible commercialization that may occur in the early 2020s.
- Supercapacitors/Ultracapacitors—An electrical energy storage device with higher power density but lower energy density than batteries. Advanced capacitors may reduce battery degradation associated with charge and discharge cycles, with some tradeoffs to cost and engineering complexity. Supercapacitors/Ultracapacitors are currently not used in parallel or as a standalone traction motor energy storage device.
○ Lower-cost magnets for Brushless Direct Current (BLDC) motors—BLDC motor technology, common in hybrid and battery electric vehicles, uses rare earth magnets. By substituting and eliminating rare earths from the magnets, motor cost can be significantly reduced. This technology is announced, but not yet in production. The capability and material configuration of these systems remains a closely guarded trade secret.
Start Printed Page 43052
○ Integrated multi-phase integrated electric vehicle drivetrains. Research has been conducted on 6-phase and 9-phase integrated systems to potentially reduce cost and improve power density. Manufacturers may improve system power density through integration of the motor, inverter, control, and gearing. These systems are in the research phase.
(c) Transmission Technologies
- Beltless CVT—Most MY 2016, commercially available CVTs rely on belt technology. A new architecture of CVT replaces belts or pulleys with a continuously variable variator, which is a special type of planetary set with balls and rings instead of gears. The technology promises to improve efficiency, handle higher torques, and change modes more quickly. This technology may be commercially available as early as 2020.
- Multi-speed electric motor transmission—MY 2016 battery electric vehicle transmissions are single-speed. Multiple gears can allow for more torque multiplication at lower speeds or a downsized electric machine, increased efficiency, and higher top speed. Two-speed transmission designs are available but not currently in production.
(d) Energy-Harvesting Technology
- Vehicle waste heat recovery systems—Internal combustion engines convert the majority of the fuel's energy to heat. Thermoelectric generators and heat pipes can convert this heat to electricity.
Thermoelectric generators, often made of semiconductors, have been tested by automakers but have traditionally not been implemented due to low efficiency and high cost.
These systems are not yet in production.
- Suspension energy recovery—Multiple electromechanical and electrohydraulic suspension technologies exist that can convert motion from uneven roads into electricity. These technologies are limited to luxury vehicles with limited production volumes. This technology is not produced in 2016 but planned for production as early as 2018.
11. Air Conditioning Efficiency and Off-Cycle Technologies
(a) Air Conditioning Efficiency Technologies
Air conditioning (A/C) is a virtually standard automotive accessory, with more than 95% of new cars and light trucks sold in the United States equipped with mobile air conditioning (MAC) systems. Most of the additional air conditioning related load on an engine is due to the compressor, which pumps the refrigerant around the system loop. The less the compressor operates or the more efficiently it operates, the less load the compressor places on the engine, and the better fuel consumption will be. This high penetration means A/C systems can significantly impact energy consumed by the light duty vehicle fleet.
Vehicle manufacturers can generate credits for improved A/C systems under EPA's GHG program and receive a fuel consumption improvement value (FCIV) equal to the value of the benefit not captured on the 2-cycle test under NHTSA's CAFE program.
Table II-19 provides a “menu” of qualifying A/C technologies, with the magnitude of each improvement value or credit estimated based on the expected reduction in CO2 emissions from the technology.
NHTSA converts the improvement in grams per mile to a FCIV for each vehicle for purposes of measuring CAFE compliance. As part of a manufacturer's compliance data, manufacturers will provide information about which off-cycle technologies are present on which vehicles (see Section X for further discussion of reporting off-cycle technology information).
The 2012 final rule for MYs 2017 and later outlined two test procedures to determine credit or FCIV eligibility for A/C efficiency menu credits, the idle test, and the AC17 test. The idle test, performed while the vehicle is at idle, determined the additional CO2 generated at idle when the A/C system is operated.
The AC17 test is a four-part performance test that combines the existing SC03 driving cycle, the fuel economy highway test cycle, and a pre-conditioning cycle, and solar soak period.
Manufacturers could use the idle test or AC17 test to determine improvement values for MYs 2014-2016, while for MYs 2017 and later, the AC17 test is the exclusive test that manufacturers can use to demonstrate eligibility for menu A/C improvement values.
In MYs 2020 and later, manufacturers will use the AC17 test to demonstrate eligibility for A/C credits and to partially quantify the amount of the credit earned. AC17 test results equal to or greater than the menu value will allow manufacturers to claim the full menu value for the credit. A test result less than the menu value will limit the amount of credit to that demonstrated on the AC17 test. In addition, for MYs 2017 and beyond, A/C fuel consumption improvement values will be available for CAFE calculations, whereas efficiency credits were previously only available for GHG compliance. The agencies proposed these changes in the 2012 final rule for MYs 2017 and later largely as a result of new data collected, as well as the extensive technical comments submitted on the proposal.
The pre-defined technology menu and associated car and light truck credit value is shown in Table II-19 below. The regulations include a definition of each technology that must be met to be eligible for the menu credit.
Manufacturers are not required to submit any other emissions data or information beyond meeting the definition and useful life requirements 
to use the pre-defined Start Printed Page 43053credit value. Manufacturers' use of menu-based credits for A/C efficiency is subject to a regulatory cap: 5.7 g/mi for cars and trucks through MY 2016 and separate caps of 5.0 g/mi for cars and 7.2g/mi for trucks for later MYs.
In the 2012 final rule for MYs 2017 and later, the agencies estimated that manufacturers would employ significant advanced A/C technologies throughout their fleets to improve fuel economy, and this was reflected in the stringency of the standards.
Many manufacturers have since incorporated A/C technology throughout their fleets, and the utilization of advanced A/C technologies has become a significant contributor to industry compliance plans. As summarized in the EPA Manufacturer Performance Report for the 2016 model year,
15 auto manufacturers included A/C efficiency credits as part of their compliance demonstration in the 2016 MY. These amounted to more than 12 million Mg of fuel consumption improvement values of the total net fuel consumption improvement values reported. This is equivalent to approximately four grams per mile across the 2016 fleet. Accordingly, a significant amount of new information about A/C technology and the efficacy of test procedures has become available since the 2012 final rule.
The sections below provide a brief history of the AC17 test procedure for evaluating A/C efficiency improving technology and discuss stakeholder comments on the AC17 test procedure approach and discuss A/C efficiency technology valuation through the off-cycle program.
Start Printed Page 43054
(1) Evaluation of the AC17 Test Procedure Since the Draft TAR
In developing the AC17 test procedure, the agencies sought to develop a test procedure that could more reliably generate an appropriate fuel consumption improvement value based on an “A” to “B” comparison, that is, a comparison of substantially similar vehicles in which one has the technology and the other does not.
The agencies believe that the AC17 test procedure is more capable of detecting the effect of more efficient A/C components and controls strategies during a transient drive cycle rather than during just idle (as measured in the old idle test procedure). As described above and in the 2012 final rule,
the AC17 test is a four-part performance test that combines the existing SC03 driving Start Printed Page 43055cycle, the fuel economy highway cycle, as well as a pre-conditioning cycle, and a solar soak period.
The agencies received several comments on the Draft TAR evaluation of the AC17 test procedure. FCA commented generally that A/C efficiency technologies “are not showing their full effect on this AC17 test as most technologies provide benefit at different temperatures and humidity conditions in comparison to a standard test conditions. All of these technologies are effective at different levels at different conditions. So there is not one size fits all in this very complex testing approach. Selecting one test that captures benefits of all of these conditions has not been possible.” 
The agencies acknowledge that any single test procedure is unlikely to equally capture the real-world effect of every potential technology in every potential use case. Both the agencies and stakeholders understood this difficulty when developing the AC17 test procedure. While no test is perfect, the AC17 test procedure represents an industry best effort at identifying a test that would greatly improve upon the idle test by capturing a greater range of operating conditions. General industry evaluation of the AC17 test procedure is in agreement that the test achieves this objective.
FCA also noted that “[i]t is a major problem to find a baseline vehicle that is identical to the new vehicle but without the new A/C technology. This alone makes the test unworkable.” 
The agencies disagree this makes the test unworkable. The regulation describes the baseline vehicle as a “similar” vehicle, selected with good engineering judgment (such that the test comparison is not unduly affected by other differences). Also, OEMs expressed confidence in using A-to-B testing to qualify for fuel consumption improvement values for software-based A/C efficiency technologies. While hardware technologies may pose a greater challenge in locating a sufficiently similar “A” baseline vehicle, the engineering analysis provision under 40 CFR 86.1868-12(g)(2) provides an alternative to locating and performing an AC17 test on such a vehicle. Further, as the USCAR program in general and the GM Denso SAS compressor application specifically have shown, the test is able to resolve small differences in CO2 effectiveness (1.3 grams in the latter case) when carefully conducted.
Commenters on the Draft TAR also expressed a desire for improvements in the process by which manufacturers without an “A” vehicle (for the A-to-B comparison) could apply under the engineering analysis provision, such as development of standardized engineering analysis and bench testing procedures that could support such applications. For example, Toyota requested that “EPA consider an optional method for validation via an engineering analysis, as is currently being developed by industry.” 
Similarly, the Alliance commented that, “[t]he future success of the MAC credit program in generating emissions reductions will depend to a large extent on the manner in which it is administered by EPA, especially with respect to making the AC17 A-to-B provisions function smoothly, without becoming a prohibitive obstacle to fully achieving the MAC indirect credits.” 
As described in the Draft TAR, in 2016, USCAR members initiated a Cooperative Research Program (CRP) through the Society of Automotive Engineers (SAE) to develop bench testing standards for the four hardware technologies in the fuel consumption improvement value menu (blower motor control, internal heat exchanger, improved evaporators and condensers, and oil separator). The intent of the program is to streamline the process of conducting bench testing and engineering analysis in support of an application for A/C credits under 40 CFR part 86.1868-12(g)(2), by creating uniform standards for bench testing and for establishing the expected GHG effect of the technology in a vehicle application.
An update to the list of SAE standards under development originally presented in the Draft TAR is listed in Table II-20. Since completion of the Draft TAR, work has continued on these standards, which appear to be nearing completion. The agencies seek comment with the latest completion of these SAE standards.
(2) A/C Efficiency Technology Valuation Through the Off-Cycle Program
The A/C technology menu, discussed at length above, includes several A/C efficiency-improving technologies that were well defined and had been quantified for effectiveness at the time of the 2012 final rule for MYs 2017 and beyond. Manufacturers claimed the vast majority of A/C efficiency credits to date by utilizing technologies on the menu; however, the agencies recognize that manufacturers will develop additional technologies that are not currently listed on the menu. These additional A/C efficiency-improving technologies are eligible for fuel consumption improvement values on a case-by-case basis under the off-cycle program. Approval under the off-cycle program also requires “A-to-B” comparison testing under the AC17 test, that is, testing substantially similar vehicles in which one has the technology and the other does not.
To date, the agencies have received one type off-cycle application for an A/C efficiency technology. In December 2014, General Motors submitted an off-cycle application for the Denso SAS A/Start Printed Page 43056C compressor with variable crankcase suction valve technology, requesting an off-cycle GHG credit of 1.1 grams CO2 per mile. In December 2017, BMW of North America, Ford Motor Company, Hyundai Motor Company, and Toyota petitioned and received approval to receive the off-cycle improvement value for the same A/C efficiency technology. EPA, in consultation with NHTSA, evaluated the applications and found methodologies described therein were sound and appropriate.
Accordingly, the agencies approved the fuel economy improvement value applications.
The agencies received additional stakeholder comments on the off-cycle approval process as an alternate route to receiving A/C technology credit values. The Alliance requested that EPA “simplify and standardize the procedures for claiming off-cycle credits for the new MAC technologies that have been developed since the creation of the MAC indirect credit menu.” 
Other commenters noted the importance of continuing to incentivize further innovation in A/C efficiency technologies as new technologies emerge that are not listed on the menu or when manufacturers begin to reach regulatory caps. The commenters suggested that EPA should consider adding new A/C efficiency technologies to the menu and/or update the fuel consumption improvement values for technology already listed on the menu, particularly in cases where manufacturers can show through an off-cycle application that the technology actually deserves more credit than that listed on the menu. For example, Toyota commented that “the incentive values for A/C efficiency should be updated along with including new technologies being deployed.” 
The agencies note that some of these comments are directed towards the off-cycle technology approval process generally, which is described in more detail in Section X of this preamble. Regarding the A/C technology menu specifically, the agencies do anticipate that new A/C technologies not currently on the menu will emerge over the time frame of the MY 2021-2026 standards. This proposal requests comment on adding one additional A/C technology to the menu—the A/C compressor with variable crankcase suction valve technology, discussed below (and also one off-cycle technology, discussed below). The agencies also request comment on whether to change any fuel economy improvement values currently assigned to technologies on the menu.
Next, as mentioned above, the menu-based improvement values for A/C efficiency established in the 2012 final rule for MYs 2017 and by end are subject to a regulatory cap. The rule set a cap of 5.7 g/mi for cars and trucks through MY 2016 and separate caps of 5.0 g/mi for cars and 7.2g/mi for trucks for later MYs.
Several commenters asked EPA to reconsider the applicability of the cap to non-menu A/C efficiency technologies claimed through the off-cycle process and questioned the applicability of this cap on several different grounds. These comments appear to be in response to a Draft TAR passage that stated: “Applications for A/C efficiency credits made under the off-cycle credit program rather than the A/C credit program will continue to be subject to the A/C efficiency credit cap” (Draft TAR, p. 5-210). The agencies considered these comments and present clarification below. As additional context, the 2012 TSD states:
“. . . air conditioner efficiency is an off-cycle technology. It is thus appropriate [. . .] to employ the standard off-cycle credit approval process [to pursue a larger credit than the menu value]. Utilization of bench tests in combination with dynamometer tests and simulations [. . .] would be an appropriate alternate method of demonstrating and quantifying technology credits (up to the maximum level of credits allowed for A/C efficiency) [emphasis added]. A manufacturer can choose this method even for technologies that are not currently included in the menu.” 
This suggests the concept of placing a limit on total A/C fuel consumption improvement values, even when some are granted under the off-cycle program, is not entirely new and that EPA considered the menu cap as being appropriate at the time.
A/C regulatory caps specified under 40 CFR 86.1868-12(b)(2) apply to A/C efficiency menu-based improvement values and are not part of the off-cycle regulation (40 CFR 86.1869-12). However, it should be noted that off-cycle applications submitted via the public process pathway are decided individually on merits through a process involving public notice and opportunity for comment. In deciding whether to approve or deny a request, the agencies may take into account any factors deemed relevant, including such issues as the realization of claimed fuel consumption improvement value in real-world use. Such considerations could include synergies or interactions among applied technologies, which could potentially be addressed by application of some form of cap or other applicable limit, if warranted. Therefore, applying for A/C efficiency fuel consumption improvement values through the off-cycle provisions in 40 CFR 86.1869-12 should not be seen as a route to unlimited A/C fuel consumption improvement values. The agencies discuss air conditioning efficiency improvement values further in Section X of this NPRM.
(b) Off-Cycle Technologies
“Off-cycle” emission reductions and fuel consumption improvements can be achieved by employing off-cycle technologies resulting in real-world benefits but where that benefit is not adequately captured on the test procedures used to demonstrate compliance with fuel economy emission standards. EPA initially included off-cycle technology credits in the MY 2012-2016 rule and revised the program in the MY 2017-2025 rule.
NHTSA adopted equivalent off-cycle fuel consumption improvement values for MYs 2017 and later in the MY 2017-2025 rule.
Manufacturers can demonstrate the value of off-cycle technologies in three ways: First, they may select fuel economy improvement values and CO2 credit values from a pre-defined “menu” for off-cycle technologies that meet certain regulatory specifications. As part of a manufacturer's compliance data, manufacturers will provide information about which off-cycle technologies are present on which vehicles.
The pre-defined list of technologies and associated off-cycle light-duty vehicle fuel economy improvement values and GHG credits is shown in Table II-21 and Table II-22 below.
A Start Printed Page 43057definition of each technology equipment must meet to be eligible for the menu credit is included at 40 CFR 86.1869-12(b)(4). Manufacturers are not required to submit any other emissions data or information beyond meeting the definition and useful life requirements to use the pre-defined credit value. Credits based on the pre-defined list are subject to an annual manufacturer fleet-wide cap of 10 g/mile.
Manufacturers can also perform their own 5-cycle testing and submit test results to the agencies with a request explaining the off-cycle technology. The additional three test cycles have different operating conditions including high speeds, rapid accelerations, high temperature with A/C operation and cold temperature, enabling improvements to be measured for technologies that do not impact operation on the 2-cycle tests. Credits determined according to this methodology do not undergo public review.
The third pathway allows manufacturers to seek EPA approval to use an alternative methodology for determining the value of an off-cycle technology. This option is only available if the benefit of the technology cannot be adequately demonstrated using the 5-cycle methodology. Manufacturers may also use this option to demonstrate reductions that exceed Start Printed Page 43058those available via use of the predetermined menu list. The manufacturer must also demonstrate that the off-cycle technology is effective for the full useful life of the vehicle. Unless the manufacturer demonstrates that the technology is not subject to in-use deterioration, the manufacturer must account for the deterioration in their analysis.
Manufacturers must develop a methodology for demonstrating the benefit of the off-cycle technology, and EPA makes the methodology available for public comment prior to an EPA determination, in consultation with NHTSA, on whether to allow the use of the methodology to measure improvements. The data needed for this demonstration may be extensive.
Several manufacturers have requested and been granted use of alternative test methodologies for measuring improvements. In 2013, Mercedes requested off-cycle credits for the following off-cycle technologies in use or planned for implementation in the 2012-2016 model years: Stop-start systems, high-efficiency lighting, infrared glass glazing, and active seat ventilation. EPA approved methodologies for Mercedes to determine these off-cycle credits in September 2014.
Subsequently, FCA, Ford, and GM requested off-cycle credits using this same methodology. FCA and Ford submitted applications for off-cycle credits from high efficiency exterior lighting, solar reflective glass/glazing, solar reflective paint, and active seat ventilation. Ford's application also demonstrated off-cycle benefits from active aerodynamic improvements (grille shutters), active transmission warm-up, active engine warm-up technologies, and engine idle stop-start. GM's application described real-world benefits of an air conditioning compressor with variable crankcase suction valve technology. EPA approved the credits for FCA, Ford, and GM in September 2015.
Note, however, that although EPA granted the use of alternative methodologies to determine credit values, manufacturers have yet to report credits to EPA based on those alternative methodologies.
As discussed below, all three methods have been used by manufacturers to generate off-cycle improvement values and credits.
(1) Use of Off-Cycle Technologies to Date
Manufacturers used a wide array of off-cycle technologies in MY 2016 to generate off-cycle GHG credits using the pre-defined menu. Table II-23 below shows the percent of each manufacturer's production volume using each menu technology reported to EPA for MY 2016 by manufacturer. Table II-24 shows the g/mile benefit each manufacturer reported across its fleet from each off-cycle technology. Like Table II-23, Table II-24 provides the mix of technologies used in MY 2016 by manufacturer and the extent to which each technology benefits each manufacturer's fleet. Fuel consumption improvement values for off-cycle technologies were not available in the CAFE program until MY 2017; therefore, only GHG off-cycle credits have been generated by manufacturers thus far.
Start Printed Page 43059
Start Printed Page 43060
In 2016, manufacturers generated the vast majority of credits using the pre-defined menu.
Although MY 2014 was the first year that manufacturers could generate credits using pre-defined menu values, manufacturers have acted quickly to generate substantial off-cycle improvements. FCA and Jaguar Land Rover generated the most off-cycle credits on a fleet-wide basis, reporting credits equivalent to approximately 6 g/mile and 5 g/mile, respectively. Several other manufacturers report fleet-wide credits in the range of approximately 1 to 4 g/mile. In MY 2016, the fleet total across manufacturers equaled approximately 2.5 g/mile. The agencies Start Printed Page 43061expect that as manufacturers continue expanding their use of off-cycle technologies, the fleet-wide effects will continue to grow with some manufacturers potentially approaching the 10 g/mile fleet-wide cap.
E. Development of Economic Assumptions and Information Used as Inputs to the Analysis
1. Purpose of Developing Economic Assumptions for Use in Modeling Analysis
(a) Overall Framework of Costs and Benefits
It is important to report the benefits and costs of this proposed action in a format that conveys useful information about how those impacts are generated and that also distinguishes the impacts of those economic consequences for private businesses and households from the effects on the remainder of the U.S. economy. A reporting format will accomplish the first objective to the extent that it clarifies the benefits and costs of the proposed action's impacts on car and light truck producers, illustrates how these are transmitted to buyers of new vehicles, shows the action's collateral economic effects on owners of used cars and light trucks, and identifies how these impacts create costs and benefits for the remainder of the U.S. economy. It will achieve the second objective by showing clearly how the economy-wide or “social” benefits and costs of the proposed action are composed of its direct effects on vehicle producers, buyers, and users, plus the indirect or “external” benefits and costs it creates for the general public.
Table II-25 through Table II-28 present the economic benefits and costs of the proposed action to reduce CAFE and CO2 emissions standards for model years 2021-26 at three percent and seven percent discount rates in a format that is intended to meet these objectives. Note: They include costs which are transfers between different economic actors—these will appear as both a cost and a benefit in equal amounts (to separate affected parties). Societal cost and benefit values shown elsewhere in this document do not show costs which are transfers for the sake of simplicity but report the same net societal costs and benefits. As it indicates, the proposed action first reduces costs to manufacturers for adding technology necessary to enable new cars and light trucks to comply with fuel economy and emission regulations (line 1). It may also reduce fine payments by manufacturers who would have failed to comply with the more demanding baseline standards. Manufacturers are assumed to transfer these cost savings on to buyers by charging lower prices (line 5); although this reduces their revenues (line 3), on balance, the reduction in compliance costs and lower sales revenue leaves them financially unaffected (line 4).
Start Printed Page 43062
Start Printed Page 43063
Start Printed Page 43064
Start Printed Page 43065
Start Printed Page 43066
As the tables show, most impacts of the proposed action will fall on the businesses and individuals who design, manufacture, and sell (at retail and wholesale) cars and light trucks, the consumers who purchase, drive, and subsequently sell or trade-in new models (and ultimately bear the cost of fuel economy technology), and owners of used cars and light trucks produced during model years prior to those covered by this action. Compared to the baseline standards, if the preferred alternative is finalized, buyers of new cars and light trucks will benefit from Start Printed Page 43067their lower purchase prices and financing costs (line 5). They will also avoid the increased risks of being injured in crashes that would have resulted from manufacturers' efforts to reduce the weight of new models to comply with the baseline standards, which represents another benefit from reducing stringency vis-à-vis the baseline (line 6).
At the same time, new cars and light trucks will offer lower fuel economy with more lenient standards in place, and this imposes various costs on their buyers and users. Drivers will experience higher costs as a consequence of new vehicles' increased fuel consumption (line 7), and from the added inconvenience of more frequent refueling stops required by their reduced driving range (line 8). They will also forego some mobility benefits as they use newly-purchased cars and light trucks less in response to their higher fueling costs, although this loss will be almost fully offset by the fuel and other costs they save by driving less (line 9). On balance, consumers of new cars and light trucks produced during the model years subject to this proposed action will experience significant economic benefits (line 10).
By lowering prices for new cars and light trucks, this proposed action will cause some owners of used vehicles to retire them from service earlier than they would otherwise have done, and replace them with new models. In effect, it will transfer some driving that would have been done in used cars and light trucks under the baseline scenario to newer and safer models, thus reducing costs for injuries (both fatal and less severe) and property damages sustained in motor vehicle crashes. This improvement in safety results from the fact that cars and light trucks have become progressively more protective in crashes over time (and also slightly less prone to certain types of crashes, such as rollovers). Thus, shifting some travel from older to newer models reduces injuries and damages sustained by drivers and passengers because they are traveling in inherently safer vehicles and not because it changes the risk profiles of drivers themselves. This reduction in injury risks and other damage costs produces benefits to owners and drivers of older cars and light trucks. This also results in benefits in terms of improved fuel economy and significant reductions of emissions from newer vehicles (line 11).
Table II-27 through Table II-28 also show that the changes in fuel consumption and vehicle use resulting from this proposed action will in turn generate both benefits and costs to the remainder of the U.S. economy. These impacts are “external,” in the sense that they are by-products of decisions by private firms and individuals that alter vehicle use and fuel consumption but are experienced broadly throughout the U.S. economy rather than by the firms and individuals who indirectly cause them. Increased refining and consumption of petroleum-based fuel will increase emissions of carbon dioxide and other greenhouse gases that theoretically contribute to climate change, and some of the resulting (albeit uncertain) increase in economic damages from future changes in the global climate will be borne throughout the U.S. economy (line 13). Similarly, added fuel production and use will increase emissions of more localized air pollutants (or their chemical precursors), and the resulting increase in the U.S. population's exposure to harmful levels of these pollutants will lead to somewhat higher costs from its adverse effects on health (line 14). On the other hand, it is expected that the proposed standards, by reducing new vehicle prices relative to the baseline, will accelerate fleet turnover to cleaner, safer, more efficient vehicles (as compared to used vehicles that might otherwise continue to be driven or purchased).
As discussed in PRIA Section 9.8, increased consumption and imports of crude petroleum for refining higher volumes of gasoline and diesel will also impose some external costs throughout the U.S. economy, in the form of potential losses in production and costs for businesses and households to adjust rapidly to sudden changes in energy prices (line 15 of the table), although these costs should be tempered by increasing U.S. oil production.
Reductions in driving by buyers of new cars and light trucks in response to their higher operating costs will also reduce the external costs associated with their contributions to traffic delays and noise levels in urban areas, and these additional benefits will be experienced throughout much of the U.S. economy (line 17). Finally, some of the higher fuel costs to buyers of new cars and light trucks will consist of increased fuel taxes; this increase in revenue will enable Federal and State government agencies to provide higher levels of road capacity or maintenance, producing benefits for all road and transit users (line 18).
On balance, Table II-27 through Table II-28 show that the U.S. economy as a whole will experience large net economic benefits from the proposed action (line 22). While the proposal to establish less stringent CAFE and GHG emission standards will produce net external economic costs, as the increase in environmental and energy security externalities outweighs external benefits from reduced driving and higher fuel tax revenue (line 19), the table also shows that combined benefits to vehicle manufacturers, buyers, and users of cars and light trucks, and the general public (line 20), including the value of the lives saved and injuries avoided, will greatly outweigh the combined economic costs they experience as a consequence of this proposed action (line 21).
The finding that this action to reduce the stringency of previously-established CAFE and GHG standards will create significant net economic benefits—when it was initially claimed that establishing those standards would also generate large economic benefits to vehicle buyers and others throughout the economy—is notable. This contrast with the earlier finding is explained by the availability of updated information on the costs and effectiveness of technologies that will remain available to improve fuel economy in model years 2021 and beyond, the fleet-wide consequences for vehicle use, fuel consumption, and safety from requiring higher fuel economy (that is, considering these consequences for used cars and light trucks as well as new ones), and new estimates of some external costs of fuel in petroleum use.
2. Macroeconomic Assumptions That Affect the Benefit Cost Analysis
Unlike previous CAFE and GHG rulemaking analyses, the economic context in which the alternatives are simulated is more explicit. While both this analysis and previous analyses contained fuel price projections from the Annual Energy Outlook, which has embedded assumptions about future macroeconomic conditions, this analysis requires explicit assumptions about future GDP growth, labor force participation, and interest rates in order to evaluate the alternatives.
Start Printed Page 43068
The analysis simulates compliance through MY 2032 explicitly and must consider the full useful lives of those vehicles, approximately 40 years, in order to estimate their lifetime mileage accumulation and fuel consumption. This means that any macroeconomic forecast influencing those factors must cover a similar span of years. Due to the long time horizon, a source that regularly produces such lengthy forecasts of these factors was selected: Start Printed Page 43069the 2017 OASDI Trustees Report from the U.S. Social Security Administration. While Table-II-29 only displays assumptions through CY 2050, the remaining years merely continue the trends present in the table.
The analysis once again uses fuel price projections from the 2017 Annual Energy Outlook.
The projections by fuel calendar year and fuel type are presented in Table-II-30, in real 2016 dollars. Fuel prices in this analysis affect not only the value of each gallon of fuel consumed but relative valuation of fuel-saving technologies demanded by the market as a result of their associated fuel savings.
Start Printed Page 43070
3. New Vehicle Sales and Employment Assumptions
In all previous CAFE and GHG rulemaking analyses, static fleet forecasts that were based on a combination of manufacturer compliance data, public data sources, and proprietary forecasts were used. When simulating compliance with regulatory alternatives, the analysis projected identical sales across the alternatives, for each manufacturer down to the make/model level where the exact same number of each model variant was simulated to be sold in a given model year under both the least stringent alternative (typically the Start Printed Page 43071baseline) and the most stringent alternative considered. To the extent that an alternative matched the assumptions made in the production of the proprietary forecast, using a static fleet based upon those assumptions may have been warranted. However, it seems intuitive that any sufficiently large span of regulatory alternatives would contain alternatives for which that static forecast was unrepresentative. A number of commenters have encouraged consideration of the potential impact of CAFE/GHG standards on new vehicle prices and sales, and the changes to compliance strategies that those shifts could necessitate.
In particular, the continued growth of the utility vehicle segment creates compliance challenges within some manufacturers' fleets as sales volumes shift from one region of the footprint curve to another.
Any model of sales response must satisfy two requirements: It must be appropriate for use in the CAFE model, and it must be econometrically reasonable. The first of these requirements implies that any variable used in the estimation of the econometric model, must also be available as a forecast throughout the duration of the years covered by the simulations (this analysis explicitly simulates compliance through MY 2032). Some values the model calculates endogenously, making them available in future years for sales estimation, but others must be known in advance of the simulation. As the CAFE model simulates compliance, it accumulates technology costs across the industry and over time. By starting with the last known transaction price and adding the accumulated technology cost to that value, the model is able to represent the average selling price in each future model year assuming that manufacturers are able to pass all of their compliance costs on to buyers of new vehicles. Other variables used in the estimation must enter the model as inputs prior to the start of the compliance simulation.
(a) How do car and light truck buyers value improved fuel economy?
How potential buyers value improvements in the fuel economy of new cars and light trucks is an important issue in assessing the benefits and costs of government regulation. If buyers fully value the savings in fuel costs that result from higher fuel economy, manufacturers will presumably supply any improvements that buyers demand, and vehicle prices will fully reflect future fuel cost savings consumers would realize from owning—and potentially re-selling—more fuel-efficient models. In this case, more stringent fuel economy standards will impose net costs on vehicle owners and can only result in social benefits by correcting externalities, since consumers would already fully incorporate private savings into their purchase decisions. If instead consumers systematically undervalue the cost savings generated by improvements in fuel economy when choosing among competing models, more stringent fuel economy standards will also lead manufacturers to adopt improvements in fuel economy that buyers might not choose despite the cost savings they offer.
The potential for car buyers to forego improvements in fuel economy that offer savings exceeding their initial costs is one example of what is often termed the “energy-efficiency gap.” This appearance of such a gap, between the level of energy efficiency that would minimize consumers' overall expenses and what they actually purchase, is typically based on engineering calculations that compare the initial cost for providing higher energy efficiency to the discounted present value of the resulting savings in future energy costs.
There has long been an active debate about why such a gap might arise and whether it actually exists. Economic theory predicts that individuals will purchase more energy-efficient products only if the savings in future energy costs they offer promise to offset their higher initial costs. However, the additional cost of a more energy-efficient product includes more than just the cost of the technology necessary to improve its efficiency; it also includes the opportunity cost of any other desirable features that consumers give up when they choose the more efficient alternative. In the context of vehicles, whether the expected fuel savings outweigh the opportunity cost of purchasing a model offering higher fuel economy will depend on how much its buyer expects to drive, his or her expectations about future fuel prices, the discount rate he or she uses to value future expenses, the expected effect on resale value, and whether more efficient models offer equivalent attributes such as performance, carrying capacity, reliability, quality, or other characteristics.
Published literature has offered little consensus about consumers' willingness-to-pay for greater fuel economy, and whether it implies over-, under- or full-valuation of the expected fuel savings from purchasing a model with higher fuel economy. Most studies have relied on car buyers' purchasing behavior to estimate their willingness-to-pay for future fuel savings; a typical approach has been to use “discrete choice” models that relate individual buyers' choices among competing vehicles to their purchase prices, fuel economy, and other attributes (such as performance, carrying capacity, and reliability), and to infer buyers' valuation of higher fuel economy from the relative importance of purchase prices and fuel economy.
Empirical estimates using this approach span a wide range, extending from substantial undervaluation of fuel savings to significant overvaluation, thus making it difficult to draw solid conclusions about the influence of fuel economy on vehicle buyers' choices (see Helfand & Wolverton, 2011; Green (2010) for detailed reviews of these cross-sectional studies). Because a vehicle's price is often correlated with its other attributes (both measured and unobserved), analysts have often used instrumental variables or other approaches to address endogeneity and other resulting concerns (e.g., Barry, et al. 1995).
Despite these efforts, more recent research has criticized these cross-sectional studies; some have questioned the effectiveness of the instruments they use (Allcott & Greenstone, 2012), while others have observed that coefficients estimated using non-linear statistical methods can be sensitive to the optimization algorithm and starting values (Knittel & Metaxoglou, 2014). Collinearity (i.e., high correlations) among vehicle attributes—most notably among fuel economy, performance or power, and vehicle size—and between vehicles' measured and unobserved features also raises questions about the reliability and interpretation of coefficients that may conflate the value of fuel economy with other attributes (Sallee, et al., 2016; Busse, et al., 2013; Allcott & Wozny, 2014; Allcott & Greenstone, 2012; Helfand & Wolverton, 2011).
In an effort to overcome shortcomings of past analyses, three recently published studies rely on panel data from sales of individual vehicle models to improve their reliability in identifying the association between vehicles' prices and their fuel economy (Sallee, et al. 2016; Allcott & Wozny, 2014; Busse, et al., 2013). Although they differ in certain details, each of these Start Printed Page 43072analyses relates changes over time in individual models' selling prices to fluctuations in fuel prices, differences in their fuel economy, and increases in their age and accumulated use, which affects their expected remaining life, and thus their market value. Because a vehicle's future fuel costs are a function of both its fuel economy and expected gasoline prices, changes in fuel prices have different effects on the market values of vehicles with different fuel economy; comparing these effects over time and among vehicle models reveals the fraction of changes in fuel costs that is reflected in changes in their selling prices (Allcott & Wozny, 2014). Using very large samples of sales enables these studies to define vehicle models at an extremely disaggregated level, which enables their authors to isolate differences in their fuel economy from the many other attributes, including those that are difficult to observe or measure, that affect their sale prices.
These studies point to a somewhat narrower range of estimates than suggested by previous cross-sectional studies; more importantly, they consistently suggest that buyers value a large proportion—and perhaps even all—of the future savings that models with higher fuel economy offer.
Because they rely on estimates of fuel costs over vehicles' expected remaining lifetimes, these studies' estimates of how buyers value fuel economy are sensitive to the strategies they use to isolate differences among individual models' fuel economy, as well as to their assumptions about buyers' discount rates and gasoline price expectations, among others. Since Anderson et al. (2013) find evidence that consumers expect future gasoline prices to resemble current prices, we use this assumption to compare the findings of the three studies and examine how their findings vary with the discount rates buyers apply to future fuel savings.
As Table 1 indicates, Allcott & Wozny (2014) find that consumers incorporate 55% of future fuel costs into vehicle purchase decisions at a six percent discount rate, when their expectations for future gasoline prices are assumed to reflect prevailing prices at the time of their purchases. With the same expectation about future fuel prices, the authors report that consumers would fully value fuel costs only if they apply discount rates of 24% or higher. However, these authors' estimates are closer to full valuation when using gasoline price forecasts that mirror oil futures markets because the petroleum market expected prices to fall during this period (this outlook reduces the discounted value of a vehicle's expected remaining lifetime fuel costs). With this expectation, Allcott & Wozny (2014) find that buyers value 76% of future cost savings (discounted at six percent) from choosing a model that offers higher fuel economy, and that a discount rate of 15% would imply that they fully value future cost savings. Sallee et al. (2016) begin with the perspective that buyers fully internalize future fuel costs into vehicles' purchase prices and cannot reliably reject that hypothesis; their base specification suggests that changes in vehicle prices incorporate slightly more than 100% of changes in future fuel costs. For discount rates of five to six percent, the Busse et al. (2013) results imply that vehicle prices reflect 60 to 100% of future fuel costs. As Table II-31 suggests, higher private discount rates move all of the estimates closer to full valuation or to over-valuation, while lower discount rates imply less complete valuation in all three studies.
Start Printed Page 43073
The studies also explore the sensitivity of the results to other parameters that could influence their results. Busse et al. (2013) and Allcott & Wozny (2014) find that relying on data that suggest lower annual vehicle use or survival probabilities, which imply that vehicles will not last as long, moves their estimates closer to full valuation, an unsurprising result because both reduce the changes in expected future fuel costs caused by fuel price fluctuations. Allcott & Wozny's (2014) base results rely on an instrumental variables estimator that groups miles-per-gallon (MPG) into two quantiles to mitigate potential attenuation bias due to measurement error in fuel economy, but they find that greater disaggregation of the MPG groups implies greater undervaluation (for example, it reduces the 55% estimated reported in Table 1 to 49%). Busse et al. (2013) allow gasoline prices to vary across local markets in their main specification; using national average gasoline prices, an approach more directly comparable to the other studies, results in estimates that are closer to or above full valuation. Sallee et al. (2016) find modest undervaluation by vehicle fleet operators or manufacturers making large-scale purchases, compared to retail dealer sales (i.e., 70 to 86%).
Since they rely predominantly on changes in vehicles' prices between repeat sales, most of the valuation estimates reported in these studies apply most directly to buyers of used vehicles. Only Busse et al. (2013) examine new vehicle sales; they find that consumers value between 75 to 133% of future fuel costs for new vehicles, a higher range than they estimate for used vehicles. Allcott & Wozny (2014) examine how their estimates vary by vehicle age and find that fluctuations in purchase prices of younger vehicles imply that buyers whose fuel price expectations mirror the petroleum futures market value a higher fraction of future fuel costs: 93% for one- to three-year-old vehicles, compared to their estimate of 76% for all used vehicles assuming the same price expectation.
Accounting for differences in their data and estimation procedures, the three studies described here suggest that car buyers who use discount rates of five to six percent value at least half—and perhaps all—of the savings in future fuel costs they expect from choosing models that offer higher fuel economy. Perhaps more important in assessing the case for regulating fuel economy, one study suggests that buyers of new cars and light trucks value three-quarters or more of the savings in future fuel costs they anticipate from purchasing higher-mpg models, although this result is based on more limited information.
In contrast, previous regulatory analyses of fuel economy standards implicitly assumed that buyers undervalue even more of the benefits they would experience from purchasing models with higher fuel economy so that without increases in fuel economy standards little improvement would occur, and the entire value of fuel savings from raising CAFE standards represented private benefits to car and light truck buyers themselves. For instance, in the EPA analysis of the 2017-2025 model year greenhouse gas emission standards, fuel savings alone added up to $475 billion (at three percent discount rate) over the lifetime of the vehicles, far outweighing the compliance costs: $150 billion). The assertion that buyers were unwilling to take voluntary advantage of this opportunity implies that collectively, they must have valued less than a third ($150 billion/$475 billion = 32%) of the fuel savings that would have resulted from those standards.
The evidence Start Printed Page 43074reviewed here makes that perspective extremely difficult to justify and would call into question any analysis that claims to show large private net benefits for vehicle buyers.
What analysts assume about consumers' vehicle purchasing behavior, particularly about potential buyers' perspectives on the value of increased fuel economy, clearly matters a great deal in the context of benefit-cost analysis for fuel economy regulation. In light of recent evidence on this question, a more nuanced approach than assuming that buyers drastically undervalue benefits from higher fuel economy, and that as a consequence, these benefits are unlikely to be realized without stringent fuel economy standards, seems warranted. One possible approach would be to use a baseline scenario where fuel economy levels of new cars and light trucks reflected full (or nearly so) valuation of fuel savings by potential buyers in order to reveal whether setting fuel economy standards above market-determined levels could produce net social benefits. Another might be to assume that, unlike in the agencies' previous analyses, where buyers were assumed to greatly undervalue higher fuel economy under the baseline but to value it fully under the proposed standards, buyers value improved fuel economy identically under both the baseline scenario and with stricter CAFE standards in place. The agencies ask for comment on these and any alternative approaches they should consider for valuing fuel savings, new peer-reviewed evidence on vehicle buyers' behavior that casts light on how they value improved fuel economy, the appropriate private discount rate to apply to future fuel savings, and thus the degree to which private fuel savings should be considered as private benefits of increasing fuel economy standards.
(b) Sales Data and Relevant Macroeconomic Factors
Developing a procedure to predict the effects of changes in prices and attributes of new vehicles is complicated by the fact that their sales are highly pro-cyclical—that is, they are very sensitive to changes in macroeconomic conditions—and also statistically “noisy,” because they reflect the transient effects of other factors such as consumers' confidence in the future, which can be difficult to observe and measure accurately. At the same time, their average sales price tends to move in parallel with changes in economic growth; that is, average new vehicle prices tend to be higher when the total number of new vehicles sold is increasing and lower when the total number of new sales decreases (typically during periods of low economic growth or recessions). Finally, counts of the total number of new cars and light trucks that are sold do not capture shifts in demand among vehicle size classes or body styles (“market segments”); nor do they measure changes in the durability, safety, fuel economy, carrying capacity, comfort, or other aspects of vehicles' quality.
The historical series of new light-duty vehicle sales exhibits cyclic behavior over time that is most responsive to larger cycles in the macro economy—but has not increased over time in the same way the population, for example, has. While U.S. population has grown over 35 percent since 1980, the registered vehicle population has grown at an even faster pace—nearly doubling between 1980 and 2015.
But annual vehicle sales did not grow at a similar pace -even accounting for the cyclical nature of the industry. Total new light-duty sales prior to the 2008 recession climbed as high as 16 million, though similarly high sales years occurred in the 1980's and 1990's as well. In fact, when considering a 10-year moving average to smooth out the effect of cycles, most 10-year averages between 1992 and 2015 are within a few percent of the 10-year average in 1992. And although average transaction prices for new vehicles have been rising steadily since the recession ended, prices are not yet at historical highs when adjusted for inflation. The period of highest inflation-adjusted transaction prices occurred from 1996-2006, when the average transaction price for a new light-duty vehicle was consistently higher than the price in 2015.
In an attempt to overcome these analytical challenges, various approaches were experimented with to predict the response of new vehicle sales to the changes in prices, fuel economy, and other features. These included treating new vehicle demand as a product of changes in total demand for vehicle ownership and demand necessary to replace used vehicles that are retired, analyzing total expenditures to purchase new cars and light trucks in conjunction with the total number sold, and other approaches. However, none of these methods offered a significant improvement over estimating the total number of vehicles sold directly from its historical relationship to directly measurable factors such as their average sales price, macroeconomic variables such as GDP or Personal Disposable Income, U.S. labor force participation, and regularly published surveys of consumer sentiment or confidence.
Quarterly, rather than annual data on total sales of new cars and light trucks, their average selling price, and macroeconomic variables was used to develop an econometric model of sales, in order to increase the number of observations and more accurately capture the causal effects of individual explanatory variables. Applying conventional data diagnostics for time-series economic data revealed that most variables were non-stationary (i.e., they reflected strong underlying time trends) and displayed unit roots, and statistical tests revealed co-integration between the total vehicle sales—the model's dependent variable—and most candidate explanatory variables.
(c) Current Estimation of Sales Impacts
To address the complications of the time series data, the analysis estimated an autoregressive distributed-lag (ARDL) model that employs a combination of lagged values of its dependent variable—in this case, last year's and the prior year's vehicle sales—and the change in average vehicle price, quarterly changes in the U.S. GDP growth rate, as well as current and lagged values of quarterly estimates of U.S. labor force participation. The number of lagged values of each explanatory variable to include was determined empirically (using the Bayesian information criterion), by examining the effects of including different combinations of their lagged values on how well the model “explained” historical variation in car and light truck sales.
The results of this approach were encouraging: The model's predictions fit the historical data on sales well, each of its explanatory variables displayed the expected effect on sales, and analysis of its unexplained residual terms revealed little evidence of autocorrelation or other indications of statistical problems. The model coefficients suggest that positive GDP growth rates and increases in labor force participation are both indicators of increases in new vehicle sales, while positive changes in average new vehicle price reduce new sales. However, the magnitude of the Start Printed Page 43075coefficient on change in average price is not as determinative of total sales as the other variables.
Based on the model, a $1,000 increase in the average new vehicle price causes approximately 170,000 lost units in the first year, followed by a reduction of another 600,000 units over the next ten years as the initial sales decrease propagates over time through the lagged variables and their coefficients. The price elasticity of new car and light truck sales implied by alternative estimates of the model's coefficients ranged from −0.2 to −0.3—meaning that changes in their prices have moderate effects on total sales—which contrasts with estimates of higher sensitivity to prices implied by some models.
The analysis was unable to incorporate any measure of new car and light truck fuel economy in the model that added to its ability to explain historical variation in sales, even after experimenting with alternative measures of such as the unweighted and sales-weighted averages fuel economy of models sold in each quarter, the level of fuel economy they were required to achieve, and the change in their fuel economy from previous periods.
Despite the evidence in the literature, summarized above, that consumers value most, if not all, of the fuel economy improvements when purchasing new vehicles, the model described here operates at too high a level of aggregation to capture these preferences. By modeling the total number of new vehicles sold in a given year, it is necessary to quantify important measures, like sales price or fuel economy, by averages. Our model operates at a high level of aggregation, where the average fuel economy represents an average across many vehicle types, usage profiles, and fuel economy levels. In this context, the average fuel economy was not a meaningful value with respect to its influence on the total number of new vehicles sold. A number of recent studies have indeed shown that consumers value fuel savings (almost) fully. Those studies are frequently based on large datasets that are able to control for all other vehicle attributes through a variety of econometric techniques. They represent micro-level decisions, where a buyer is (at least theoretically) choosing between a more or less efficient version of a pickup truck (for example) that is otherwise identical. In an aggregate sense, the average is not comparable to the decision an individual consumer faces.
Estimating the sales response at the level of total new vehicle sales likely fails to address valid concerns about changes to the quality or attributes of new vehicles sold—both over time and in response to price increases resulting from CAFE standards. However, attempts to address such concerns would require significant additional data, new statistical approaches, and structural changes to the CAFE model over several years. It is also the case that using absolute changes in the average price may be more limited than another characterization of price that relies on distributions of household income over time or percentage change in the new vehicle price. The former would require forecasting a deeply uncertain quantity many years into the future, and the latter only become relevant once the simulation moves beyond the magnitude of observed price changes in the historical series. Future versions of this model may use a different characterization of cost that accounts for some of these factors if their inclusion improves the model estimation and corresponding forecast projections are available.
The changes in selling prices, fuel economy, and other features of cars and light trucks produced during future model years that result from manufacturers' responses to lower CAFE and GHG emission standards are likely to affect both sales of individual models and the total number of new vehicles sold. Because the values of changes in fuel economy and other features to potential buyers are not completely understood; however, the magnitude, and possibly even the direction, of their effect on sales of new vehicles is difficult to anticipate. On balance, it is reasonable to assume that the changes in prices, fuel economy, and other attributes expected to result from their proposed action to amend and establish fuel economy and GHG emission standards are likely to increase total sales of new cars and light trucks during future model years. Please provide comment on the relationship between price increases, fuel economy, and new vehicle sales, as well as methods to appropriately account for these relationships.
(d) Projecting New Vehicle Sales and Comparisons to Other Forecasts
The purpose of the sales response model is to allow the CAFE model to simulate new vehicle sales in a given future model year, accounting for the impact of a regulatory alternative's stringency on new vehicle prices (in a macro-economic context that is identical across alternatives). In order to accomplish this, it is important that the model of sales response be dynamically stable, meaning that it responds to shocks not by “exploding,” increasing or decreasing in a way that is unbounded, but rather returns to a stable path, allowing the shock to dissipate. The CAFE model uses the sales model described above to dynamically project future sales; after the first year of the simulation, lagged values of new vehicle sales are those that were produced by the model itself rather than observed. The sales response model constructed here uses two lagged dependent variables and simple econometric conditions determine if the model is dynamically stable. The coefficients of the one-year lag and the two-year lag, β1 and β2, respectively must satisfy three conditions. Their sum must be less than one, β2 − β1 <1, and the absolute value of β2 must be less than one. The coefficients of this model satisfy all three conditions.
Using the Augural CAFE standards as the baseline, it is possible to produce a series of future total sales as shown in Table-II-32. For comparison, the table includes the calculated total light-duty sales of a proprietary forecast purchased to support the 2016 Draft TAR analysis, the total new light-duty sales in EIA's 2017 Annual Energy Outlook, and a (short) forecast published in the Center for Automotive Research's Q4 2017 Automotive Outlook. All of the forecasts in Table-II-32 assume the Augural Standards are in place through MY 2025, though assumptions about the costs required to comply with them likely differ. As the table shows, despite differences among them, the dynamically produced sales projection from the CAFE model is not qualitatively different from the others.
Start Printed Page 43076
While this forecast projects a relatively high, but flat, level of new vehicle sales into the future, it is worth noting that it continues another trend observed in the historical data. The time series of annual new vehicle sales is volatile from year to year, but multi-year averages are less so being sufficient to wash out the variation associated with them peaks and valleys of the series. Despite the fact that the moving average annual new vehicle sales has been growing over the last four decades, it has not kept pace with U.S. population growth. Data from the Federal Reserve Bank of St. Louis shows that the per-capita sales of new vehicles peaked in 1986 and has declined more than 25% from this peak to today's level.
While the sales projection in Table-II-32 would represent a historically high average of new vehicle sales over the analysis period, it would not be sufficient to reverse the trend of declining per-capita sales of new vehicles during the analysis period, though it would continue the trend at a slower rate.
In addition to the statistical model that estimates the response of total new vehicle sales to changes in the average new vehicle price, the CAFE model incorporates a dynamic fleet share model that modifies the light truck (and, symmetrically, passenger car) share of the new vehicle market. A version of this model first appeared in the 2012 final rule, when this fleet share component was introduced to ensure greater internal consistency within inputs in the uncertainty analysis. For today's analysis, this dynamic fleet share is enabled throughout the analysis of alternatives.
The dynamic fleet share model is a series of difference equations that determine the relative share of light trucks and passenger cars based on the average fuel economy of each, the fuel price, and average vehicle attributes like horsepower and vehicle mass (the latter of which explicitly evolves as a result of the compliance simulation). While this model was taken from EIA's National Energy Modeling System (NEMS), it is applied at a different level. Rather than apply the shares based on the regulatory class distinction, the CAFE model applies the shares to body-style. This is done to account for the large-scale shift in recent years to crossover utility vehicles that have model variants in both the passenger car and light truck regulatory fleets. The agencies have always modified their static forecasts of new vehicle sales to reflect the PC/LT split present in the Annual Energy Outlook; this integration continues that approach in a way that ensures greater internal consistency when simulating multiple regulatory alternatives (and conducting sensitivity analysis on any of the factors that influence fleet share).
(e) Vehicle Choice Models as an Alternative Method To Estimate New Vehicle Sales
Another potential option to estimate future new vehicle sales would be to use a full consumer choice model. The agencies simulate compliance with CAFE and CO2 standards for each manufacturer using a disaggregated representation of its regulated vehicle fleets. This means that each manufacturer may have hundreds of vehicle model variants (e.g., the Honda Civic with the 6-cylinder engine, and the Honda Civic with the 4-cylinder engine would each be treated as different, in some ways, during the compliance simulation).
While the analysis accounts for a wide variety of attributes across these vehicles, only a few of them change during the compliance simulation. However, all of those attributes are relevant in the context of consumer choice models.
Aside from the computational intensity of simulating new vehicle sales at the level of individual models—for all manufacturers, under each regulatory alternative, over the next decade or more—it would be necessary to include additional relationships Start Printed Page 43077about how consumers trade off among vehicle attributes, which types of consumers prefer which types of attributes (and how much), and how manufacturers might strategically price these modified vehicles. This requires a strategic pricing model, which each manufacturer has and would likely be unwilling to share. Some of this strategic pricing behavior occurs on small time-scale through the use of dealer incentives, rebates on specific models, and creative financing offers. When simulating compliance at the annual scale, it is effectively impossible to account for these types of strategic decisions.
It is also true consumers have heterogeneous preferences that change over time and determine willingness-to-pay for a variety of vehicle attributes. These preferences change in response to marketing, distribution, pricing, and product strategies that manufacturers may change over time. With enough data, a consumer choice model could stratify new vehicle buyers into types and attempt to measure the strength of each type's preference for fuel economy, acceleration, safety rating, perceived quality and reliability, interior volume, or comfort. However, other factors also influence customers' purchase decision, and some of these can be challenging to model. Consumer proximity to dealerships, quality of service and customer experience at dealerships, availability and terms of financing, and basic product awareness may significantly factor into sales success.
Manufacturers' marketing choices may significantly and unpredictably affect sales. Ad campaigns may increase awareness in the market, and campaigns may reposition consumers' perception of the brands and products. For example, in 2011 the Volkswagen Passat featured an ad with a child in a Darth Vader costume (and showcased remote start technology on the Passat). In MY 2012, Kia established the Kia Soul with party rocking, hip-hop hamster commercials showcasing push-button ignition, a roomy interior, and design features in the brake lights. Both commercials raised awareness and highlighted basic product features. Each commercial also impressed demographic groups with pop culture references, product placement, and co-branding. While the marketing budget of individual manufacturers may help a consumer choice model estimate market share for a given brand, estimating the impact of a given campaign on new sales is more challenging as consumers make purchasing decisions based upon their own needs and desires.
Modelers must understand how consumers and commercial buyers select vehicles in order to effectively develop and implement a consumer choice model in a compliance simulation. Consumers purchase vehicles for a variety of reasons such as family need, need for more space, new technology, changes to income and affordability of a new vehicle, improved fuel economy, operating costs of current vehicles, and others. Once committed to buying a vehicle, consumers use different processes to narrow down their shopping list. Consumer choice decision attributes include factors both related and not related to the vehicle design. The vehicle's utility for those attributes is researched across many different information sources as listed in the table below.
An objective, attribute-based consumer choice model could lead to projected swings in manufacturer market shares and individual model volumes. The current approach simulates compliance for each manufacturer assuming that it produces the same set of vehicles that it produced in the initial year of the simulation (MY 2016 in today's analysis). If a consumer choice model were to drive projected sales of a given vehicle model below some threshold, as consumers have done in the real market, the simulation currently has no way to generate a new vehicle model to take its place. As demand changes across specific market segments and models, manufacturers adapt by supplying new vehicle nameplates and models (e.g., the proliferation of crossover utility vehicles in recent years). Absent that flexibility in the compliance simulation, even the more accurate consumer choice model may produce unrealistic projections of future sales volumes at the model, segment, or manufacturer level.
Comment is sought on the development and use of potential consumer choice model in compliance simulations. Comment is also sought on the appropriate breadth, depth, and complexity of considerations in a consumer choice model.
(f) Industry Employment Baseline (Including Multiplier Effect) and Data Description
In the first two joint CAFE/CO2 rulemakings, the agencies considered an analysis of industry employment impacts in some form in setting both CAFE and emissions standards; NHTSA conducted an industry employment analysis in part to determine whether the standards the agency set were economically practicable, that is, whether the standards were “within the financial capability of the industry, but not so stringent as to” lead to “adverse economic consequences, such as a significant loss of jobs or unreasonable elimination of consumer choice.” 
EPA similarly conducted an industry employment analysis under the broad authority granted to the agency under the Clean Air Act.
Both agencies recognized the uncertainties inherent in estimating industry employment impacts; in fact, both agencies dedicated a substantial amount of discussion to uncertainty in industry employment analyses in the 2012 final rule for MYs 2017 and beyond.
Notwithstanding these uncertainties, CAFE and CO2 standards do impact industry labor hours, and providing the best analysis practicable better informs stakeholders Start Printed Page 43078and the public about the standards' impact than would omitting any estimates of potential labor impacts.
Today many of the effects that were previously qualitatively identified, but not considered, are quantified. For instance, in the PRIA for the 2017-2025 rule EPA identified “demand effects,” “cost effects,” and “factor shift effects” as important considerations for industry labor, but the analysis did not attempt to quantify either the demand effect or the factor shift effect.
Today's industry labor analysis quantifies direct labor changes that were qualitatively discussed previously.
Previous analyses and new methodologies to consider direct labor effects on the automotive sector in the United States were improved upon and developed. Potential changes that were evaluated include (1) dealership labor related to new light duty vehicle unit sales; (2) changes in assembly labor for vehicles, for engines and for transmissions related to new vehicle unit sales; and (3) changes in industry labor related to additional fuel savings technologies, accounting for new vehicle unit sales. All automotive labor effects were estimated and reported at a national level,
in job-years, assuming 2,000 hours of labor per job-year.
The analysis estimated labor effects from the forecasted CAFE model technology costs and from review of automotive labor for the MY 2016 fleet. For each vehicle in the CAFE model analysis, the locations for vehicle assembly, engine assembly, and transmission assembly and estimated labor in MY 2016 were recorded. The percent U.S. content for each vehicle was also recorded. Not all parts are made in the United States, so the analysis also took into account the percent U.S. content for each vehicle as manufacturers add fuel-savings technologies. As manufacturers added fuel-economy technologies in the CAFE model simulations, the analysis assumed percent U.S. content would remain constant in the future, and that the U.S. labor added would be proportional to U.S. content. From this foundation, the analysis forecasted automotive labor effects as the CAFE model added fuel economy technology and adjusted future sales for each vehicle.
The analysis also accounts for sales projections in response to the different regulatory alternatives; the labor analysis considers changes in new vehicle prices and new vehicle sales (for further discussion of the sales model, see Section 2.E). As vehicle prices rise, the analysis expected consumers to purchase fewer vehicles than they would have at lower prices. As manufacturers sell fewer vehicles, the manufacturers may need less labor to produce the vehicles and less labor to sell the vehicles. However, as manufacturers add equipment to each new vehicle, the manufacturers will require human resources to develop, sell, and produce additional fuel-saving technologies. The analysis also accounts for the potential that new standards could shift the relative shares of passenger cars and light trucks in the overall fleet (see Section 2.E); insofar as different vehicles involved different amounts of labor, this shifting impacts the quantity of estimated labor. The CAFE model automotive labor analysis takes into account reduction in vehicle sales, shifts in the mix of passenger cars and light trucks, and addition of fuel-savings technologies.
For today's analysis, it was assumed that some observations about the production of MY 2016 vehicles would carry forward, unchanged into the future. For instance, assembly plants would remain the same as MY 2016 for all products now, and in the future. The analysis assumed percent U.S. content would remain constant, even as manufacturers updated vehicles and introduced new fuel-saving technologies. It was assumed that assembly labor hours per unit would remain at estimated MY 2016 levels for vehicles, engines, and transmissions, and the factor between direct assembly labor and parts production jobs would remain the same. When considering shifts from one technology to another, the analysis assumed revenue per employee at suppliers and original equipment manufacturers would remain in line with MY 2016 levels, even as manufacturers added fuel-saving technologies and realized cost reductions from learning.
The analysis focused on automotive labor because adjacent employment factors and consumer spending factors for other goods and services are uncertain and difficult to predict. The analysis did not consider how direct labor changes may affect the macro economy and possibly change employment in adjacent industries. For instance, the analysis did not consider possible labor changes in vehicle maintenance and repair, nor did it consider changes in labor at retail gas stations. The analysis did not consider possible labor changes due to raw material production, such as production of aluminum, steel, copper and lithium, nor did the agencies consider possible labor impacts due to changes in production of oil and gas, ethanol, and electricity. The analysis did not analyze effects of how consumers could spend money saved due to improved fuel economy, nor did the analysis assess the effects of how consumers would pay for more expensive fuel savings technologies at the time of purchase; either could affect consumption of other goods and services, and hence affect labor in other industries. The effects of increased usage of car-sharing, ride-sharing, and automated vehicles were not analyzed. The analysis did not estimate how changes in labor from any industry could affect gross domestic product and possibly affect other industries as a result.
Finally, no assumptions were made about full-employment or not full-employment and the availability of human resources to fill positions. When the economy is at full employment, a fuel economy regulation is unlikely to have much impact on net overall U.S. employment; instead, labor would primarily be shifted from one sector to another. These shifts in employment impose an opportunity cost on society, approximated by the wages of the employees, as regulation diverts workers from other activities in the economy. In this situation, any effects on net employment are likely to be transitory as workers change jobs (e.g., some workers may need to be retrained or require time to search for new jobs, while shortages in some sectors or regions could bid up wages to attract workers). On the other hand, if a regulation comes into effect during a period of high unemployment, a change in labor demand due to regulation may affect net overall U.S. employment because the labor market is not in equilibrium. Schmalansee and Stavins point out that net positive employment effects are possible in the near term when the economy is at less than full employment due to the potential hiring of idle labor resources by the regulated sector to meet new requirements (e.g., to install new equipment) and new economic activity in sectors related to the regulated sector longer run, the net effect on employment is more difficult to predict and will depend on the way in which the related industries respond to the regulatory requirements. For that reason, this analysis does not include multiplier effects but instead focuses on Start Printed Page 43079labor impacts in the most directly affected industries. Those sectors are likely to face the most concentrated labor impacts.
Comment is sought on these assumptions and approaches in the labor analysis.
4. Estimating Labor for Fuel Economy Technologies, Vehicle Components, Final Assembly, and Retailers
The following sections discuss the approaches to estimating factors related to dealership labor, final assembly labor and parts production, and fuel economy technology labor.
(a) Dealership Labor
The analysis evaluated dealership labor related to new light-duty vehicle sales, and estimated the labor hours per new vehicle sold at dealerships, including labor from sales, finance, insurance, and management. The effect of new car sales on the maintenance, repair, and parts department labor is expected to be limited, as this need is based on the vehicle miles traveled of the total fleet. To estimate the labor hours at dealerships per new vehicle sold, the National Automobile Dealers Association 2016 Annual Report, which provides franchise dealer employment by department and function, was referenced.
The analysis estimated that slightly less than 20% of dealership employees' work relates to new car sales (versus approximately 80% in service, parts, and used car sales), and that on average dealership employees working on new vehicle sales labor for 27.8 hours per new vehicle sold.
(b) Final Assembly Labor and Parts Production
How the quantity of assembly labor and parts production labor for MY 2016 vehicles would increase or decrease in the future as new vehicle unit sales increased or decreased was estimated.
Specific assembly locations for final vehicle assembly, engine assembly, and transmission assembly for each MY 2016 vehicle were identified. In some cases, manufacturers assembled products in more than one location, and the analysis identified such products and considered parallel production in the labor analysis.
The analysis estimated industry average direct assembly labor per vehicle (30 hours), per engine (four hours), and per transmission (five hours) based on a sample of U.S. assembly plant employment and production statistics and other publicly available information. The analysis recognizes that some plants may use less labor than the analysis estimates to produce the vehicle, the engine, or the transmission, and other plants may have used more labor. The analysis used the assembly locations and industry averages for labor per unit to estimate U.S. assembly labor hours for each vehicle. U.S. assembly labor hours per vehicle ranged from as high as 39 hours if the manufacturer assembled the vehicle, engine, and transmission at U.S. plants, to as low as zero hours if the manufacturer imported the vehicle, engine, and transmission.
The analysis also considered labor for part production in addition to labor for final assembly. Motor vehicle and equipment manufacturing labor statistics from the U.S. Census Bureau, the Bureau of Labor Statistics,
and other publicly available sources were surveyed. Based on these sources, the analysis noted that the historical average ratio of vehicle assembly manufacturing employment to employment for total motor vehicle and equipment manufacturing for new vehicles remained roughly constant over the period from 2001 through 2013, at a ratio of 5.26. Observations from 2001-2013 spanned many years, many combinations of technologies and technology trends, and many economic conditions, yet the ratio remained about the same. Accordingly, the analysis scaled up estimated U.S. assembly labor hours by a factor of 5.26 to consider U.S. parts production labor in addition to assembly labor for each vehicle.
The industry estimates for vehicle assembly labor and parts production labor for each vehicle scaled up or down as unit sales scaled up or down over time in the CAFE model.
(c) Fuel Economy Technology Labor
As manufacturers spend additional dollars on fuel-saving technologies, parts suppliers and manufacturers require human resources to bring those technologies to market. Manufacturers may add, shift, or replace employees in ways that are difficult for the agencies to predict in response to adding fuel-savings technologies; however, it is expected that the revenue per labor hour at original equipment manufacturers (OEMs) and suppliers will remain about the same as in MY 2016 even as industry includes additional fuel-saving technology.
To estimate the average revenue per labor hour at OEMs and suppliers, the analysis looked at financial reports from publicly traded automotive businesses.
Based on recent figures, it was estimated that OEMs would add one labor year per $633,066 revenue 
and that suppliers would add one labor year per $247,648 in revenue.
These global estimates are applied to all revenues, and U.S. content is applied as a later adjustment. In today's analysis, it was assumed these ratios would remain constant for all technologies rather than that the increased labor costs would be shifted toward foreign countries. Comment is sought on the realism of this assumption.
(d) Labor Calculations
The analysis estimated the total labor as the sum of three components: Dealership hours, final assembly and parts production, and labor for fuel-economy technologies (at OEM's and suppliers). The CAFE model calculated additional labor hours for each vehicle, based on current vehicle manufacturing locations and simulation outputs for additional technologies, and sales changes. The analysis applied some constants to all vehicles,
but other constants were vehicle specific,
or year specific for a vehicle.
While a multiplier effect of all U.S. automotive related jobs on non-auto related U.S. jobs was not considered for today's analysis, the analysis did program a “global multiplier” that can be used to scale up or scale down the total labor hours. This multiplier exists in the parameters file, and for today's analysis the analysis set the value at 1.00.
5. Additional Costs and Benefits Incurred by New Vehicle Buyers
Some costs of purchasing and owning a new or used vehicle scale with the Start Printed Page 43080value of the vehicle. Where fuel economy standards increase the transaction price of vehicles, they will affect both the absolute amount paid in sales tax and the average amount of financing required to purchase the vehicle. Further, where they increase the MSRP, they increase the appraised value upon which both value-related registration fees and a portion of insurance premiums are based. The analysis assumes that the transaction price is a set share of the MSRP, which allows calculation of these factors as shares of MSRP. Below the assumptions made about how each of these additional costs of vehicle purchase and ownership scale with the MSRP and how the analysis arrived at these assumptions are discussed.
(a) Sales Taxes
The analysis took auto sales taxes by state 
and weighted them by population by state to determine a national weighted-average sales tax of 5.46%. The analysis sought to weight sales taxes by new vehicle sales by state; however, such data were unavailable. It is recognized that for this purpose, new vehicle sales by state is a superior weighting mechanism to Census population; in effort to approximate new vehicle sales by state, a study of the change in new vehicle registrations (using R.L. Polk data) by state across recent years was conducted, resulting in a corresponding set of weights. Use of the weights derived from the study of vehicle registration data resulted in a national weighted-average sales tax rate almost identical to that resulting from the use of Census population estimates as weights, just slightly above 5.5%. The analysis opted to utilize Census population rather than the registration-based proxy of new vehicle sales as the basis for computing this weighted average, as the end results were negligibly different and the analytical approach involving new vehicle registrations had not been as thoroughly reviewed. Note: Sales taxes and registration fees are transfer payments between consumers and the Federal government and are therefore not considered a cost in the societal perspective. However, these costs are considered as additional costs in the private consumer perspective.
(b) Financing Costs
The analysis assumes 85% of automobiles are financed based on Experian's quarter 4, 2016 “State of the Automotive Finance Market,” which notes that 85.2% of 2016 new vehicles were financed, as were 85.9% of 2015 new vehicle purchases.
The analysis used data from Wards Automotive and JD Power on the average transaction price of new vehicle purchases, average financed new auto beginning principal, and the average incentive as a percent of MSRP to compute the ratio of the average financed new auto principal to the average new vehicle MSRP for calendar years 2011-2016. Table-II-34 shows that the average financed auto principal is between 82 and 84% of the average new vehicle MSRP. Using the assumption that 85% of new vehicle purchases involve some financing, the average share of the MSRP financed for all vehicles purchased, including non-financed transactions, rather than only those that are financed, was computed. Table-II-34 shows that this share ranges between 70 and 72%. From this, the analysis assumed that on an aggregate level, including all new vehicle purchases, 70% of the value of all vehicles' MSRP is financed. It is likely that the share financed is correlated with the MSRP of the new vehicle purchased, but for simplification purposes, it is assumed that 70% of all vehicle costs are financed, regardless of the MSRP of the vehicle. In measurements of the impacts on the average consumer, this assumption will not affect the outcome of our calculation, though this assumption will matter for any discussions about how many, or which, consumers bear the brunt of the additional cost of owning more expensive new vehicles. For sake of simplicity, the model also assumes that increasing the cost of new vehicles will not change the share of new vehicle MSRP that is financed; the relatively constant share from 2011-2016 when the average MSRP of a vehicle increased 10% supports this assumption. It is recognized that this is not indicative of average individual consumer transactions but provides a useful tool to analyze the aggregate marketplace.
From Wards Auto data, the average 48- and 60-month new auto interest rates were 4.25% in 2016, and the average finance term length for new autos was 68 months. It is recognized that longer financing terms generally include higher interest rates. The share financed, interest rate, and finance term length are added as inputs in the Start Printed Page 43081parameters file so that they are easier to update in the future. Using these inputs the model computes the stream of financing payments paid for the average financed purchases as the following:
Note: The above assumes the interest is distributed evenly over the period, when in reality more of the interest is paid during the beginning of the term. However, the incremental amount calculated as attributable to the standard will represent the difference in the annual payments at the time that they are paid, assuming that a consumer does not repay early. This will represent the expected change in the stream of financing payments at the time of financing.
The above stream does not equate to the average amount paid to finance the purchase of a new vehicle. In order to compute this amount, the share of financed transactions at each interest rate and term combination would have to be known. Without having projections of the full distribution of the auto finance market into the future, the above methodology reasonably accounts for the increased amount of financing costs due to the purchase of a more expensive vehicle, on an average basis taking into account non-financed transactions. Financing payments are also assumed to be an intertemporal transfer of wealth for a consumer; for this reason, it is not included in the societal cost and benefit analysis. However, because it is an additional cost paid by the consumer, it is calculated as a part of the private consumer welfare analysis.
It is recognized that increased finance terms, combined with rising interest rates, lead to a longer period of time before a consumer will have positive equity in the vehicle to trade in toward the purchase of a newer vehicle. This has impacts in terms of consumers either trading vehicles with negative equity (thereby increasing the amount financed and potentially subjecting the consumer to higher interest rates and/or rendering the consumer unable to obtaining financing) or delaying the replacement of the vehicle until they achieve suitably positive equity to allow for a trade. Comment is sought on the effect these developments will have on the new vehicle market, both in general, and in light of increased stringency of fuel economy and GHG emission standards. Comment is also sought on whether and how the model should account for consumer decisions to purchase a used vehicle instead of a new vehicle based upon increased new vehicle prices in response to increased CAFE standard stringency.
(c) Insurance Costs
More expensive vehicles will require more expensive collision and comprehensive (e.g., fire and theft) car insurance. Actuarially fair insurance premiums for these components of value-based insurance will be the amount an insurance company will pay out in the case of an incident type weighted by the risk of that type of incident occurring. It is expected that the same driver in the same vehicle type will have the same risk of occurrence for the entirety of a vehicle's life so that the share of the value of a vehicle paid out should be constant over the life of a vehicle. However, the value of vehicles will decline at some depreciation rate so that the absolute amount paid in value-related insurance will decline as the vehicle depreciates. This is represented in the model as the following stream of expected collision and comprehensive insurance payments:
To utilize the above framework, estimates of the share of MSRP paid on collision and comprehensive insurance and of annual vehicle depreciations are needed to implement the above equation. Wards has data on the average annual amount paid by model year for new light trucks and passenger cars on collision, comprehensive and damage and liability insurance for model years 1992-2003; for model years 2004-2016, they only offer the total amount paid for insurance premiums. The share of total insurance premiums paid for collision and comprehensive coverage was computed for 1979-2003. For cars the share ranges from 49 to 55%, with the share tending to be largest towards the end of the series. For trucks the share ranges from 43 to 61%, again, with the share increasing towards the end of the series. It is assumed that for model years 2004-2016, 60% of insurance premiums for trucks, and 55% for cars, is paid for collision and comprehensive. Using these shares the absolute amount paid for collision and comprehensive coverage for cars and trucks is computed. Then each regulatory class in the fleet is weighted by share to estimate the overall average amount paid for collision and comprehensive insurance by model year as shown in Table-II-35. The average share of the initial MSRP paid in collision and comprehensive insurance by model year is then computed. The average share paid for model years 2010-2016 is 1.83% of the initial MSRP. This is used as the share of the value of a new vehicle paid for collision and comprehensive in the future.
Start Printed Page 43082
2017 data from Fitch Black Book was used as a source for vehicle depreciation rates; two- to six-year-old vehicles in 2016 had an average annual depreciation rate of 17.3%.
It is assumed that future depreciation rates will be like recent depreciation, and the analysis used the same assumed depreciation. Table-II-36 shows the cumulative share of the initial MSRP of a vehicle assumed to be paid in collision and comprehensive insurance in five-year age increments under this depreciation assumption, conditional on a vehicle surviving to that age—that is, the expected insurance payments at the time of purchase will be weighted by the probability of surviving to that age. If a vehicle lives to 10 years, 9.9% of the initial MSRP is expected to be paid in collision and comprehensive payments; by 20 years 11.9% of the initial MSRP; finally, if a vehicle lives to age 40, 12.4% of the initial MSRP. As can be seen, the majority of collision and comprehensive payments are paid by the time the vehicle is 10 years old.
The increase in insurance premiums resulting from an increase in the average value of a vehicle is a result of an increase in the expected amount insurance companies will have to pay out in the case of damage occurring to the driver's vehicle. In this way, it is a cost to the private consumer, attributable to the CAFE standard that caused the price increase.
(d) Consumer Acceptance of Specific Technologies
In previous rulemaking analyses, NHTSA imposed an economic cost of lost welfare to buyers of advanced electric vehicles. NHTSA chose to model a 75-mile EV for early adopters, who we assume would not be concerned with the lower range, and a 150-mile EV for the broader market. The initial five percent of EV sales were assumed to go to early adopters, with the remainder being 150-mile EVs. The broader market was assumed to have some lower utility for the 150-mile EV, due to the lower driving range between refueling events relative to a conventional vehicle. Thus, an additional social cost of about $3,500 per vehicle was assigned to the EV150 to capture the lost utility to consumers.
Additionally, NHTSA imposed a “relative value loss” of 1.94% of the vehicle's MSRP to reflect the economic value of the difference between the useful life of a conventional ICE and the 150-mile EV when it reaches a 55% battery capacity (as a result of battery deteroriation).
In subsequent analyses (the 2016 Draft TAR analysis and today's analysis), NHTSA removed the low-range EVs from its technology set due to both weak consumer demand for low-range EVs in the marketplace and subsequent technology advances that make 200-mile EVs a more practical option for new EVs produced in future model years. The exclusion of low-range EVs in the technology set reduced the need to account for consumer welfare losses Start Printed Page 43083attributable to reduced driving range. While the sensitivity analysis explores some potential for continuing consumer value loss, even in the improved electrified powertrain vehicles, the central analysis assumes that no value loss exists for electrified powertrains. However, ongoing low sales volumes and a growing body of literature suggest that consumer welfare losses may still exist if manufacturers are forced to produce electric vehicles in place of vehicles with internal combustion engines (forcing sacrifices to cargo capacity or driving range) in order to comply with standards. This topic will receive ongoing investigation and revision before the publication of the final rule. Please provide comments and any relevant data that would help to inform the estimation of implementation of any value loss related to sacrificed attributes in electric vehicles.
One reason it was necessary to account for welfare losses from reduced driving range in this way is that, in previous rulemakings, the agencies implicitly assumed that every vehicle in the forecast would be produced and purchased and that manufacturers would pass on the entire incremental cost of fuel-saving technologies to new car (and truck) buyers. However, many stakeholders commented that consumers are not willing to pay the full incremental costs for hybrids, plug-in hybrids, and battery electric vehicles.
For this analysis, consumer willingness to pay for HEVs, PHEVs, BEVs relative to comparable ICE vehicles was investigated. The analysis compared the estimated price premium the electrified vehicles command in the used car market and estimated the willingness to pay premium for new vehicles with electrification technologies at age zero relative to their internal combustion engine counterparts. For the analysis, the willingness to pay was compared with the expected incremental cost to produce electrification technologies. Manufacturers also contributed confidential business information about the costs, revenues, and profitability of their electrified vehicle lines. The CBI provided a valuable check on the empirical work described below. As a result of this examination, we no longer assume manufacturers can pass on the entire incremental cost of hybrid, plug-in hybrid, and battery electric vehicles to buyers of those vehicles. The difference between the buyer's willingness-to-pay for those technologies, and the cost to produce them, must be recovered from buyers of other vehicles in a manufacturer's product portfolio or sacrificed from its profits, or sacrificed from dealership profits, or supplemented with State or Federal incentives (or, some combination of the four).
Using data from the used vehicle market, statistical models were fit to estimate consumer willingness to pay for new vehicles with varying levels of electrification relative to comparable internal combustion engine vehicles was evaluated in four steps. The analysis (1) gathered used car fair market value for select vehicles; (2) developed regression models to estimate the portion of vehicle depreciation rate attributable to the vehicle nameplate and the portion attributable to the vehicle's technology content at each age (using fixed effects for nameplates and specific electrification technologies); (3) estimated the value of vehicles at age zero (i.e., when the vehicles were new); and (4) compared new vehicle values for comparable vehicles across different electrification levels (i.e., internal combustion, HEV, PHEV, and BEV) to estimate willingness-to-pay for the electric technology relative to an ICE.
The dataset used for estimation consisted of vehicle attribute data from Edmunds and transaction data from Kelley Blue Book published online in June and July of 2017 for select vehicles of interest. 
The dataset was constructed to contain pairs of vehicles that were nearly the same, except for type of powertrain (internal combustion versus some amount of electrification). For instance, the dataset contained used vehicle prices for the Honda Accord and Honda Accord Hybrid, Toyota Camry and Toyota Camry Hybrid, Ford Fusion and Ford Fusion Hybrid, Kia Soul and Kia Soul EV, and so on for several model years. In some cases, the manufacturer produced no identically equivalent internal combustion engine vehicle, so a similar internal combustion vehicle produced by the same manufacturer was used as the point of comparison. For example, the Nissan Leaf was paired with the Nissan Versa, as well as the Toyota Prius and Toyota Corolla. Only vehicles available for private sale, and in good vehicle condition were included in the analysis.
The dataset contains fewer observations for PHEVs and BEVs because manufacturers have produced fewer examples of vehicles with these technologies, compared to HEV and ICE vehicles. In all of these cases, trim level and options packages were matched between ICE and electric powertrains to minimize the degree of non-powertrain difference between vehicle pairs. The resale price data spanned many model years, but most observations in the dataset represent MY 2013 through MY 2016.
The regression models used to estimate the transaction price (or “Value”) as a function of age, control for the type of powertrain (ICE, HEV, PHEV, and BEV) and nameplate to account for their impact on the value of the vehicle as it ages.
The regression takes the following form, with ICE, HEV, PHEV, and BEV binary variables (0, or 1), and age defined as 2017 minus the model year was used:
1n(Value = ,β1 (ICE * Age) + β2 (HEV * Age) + β3 (PHEV * Age) + β4 (BEV * Age) + β5 (HEV) + β6 (PHEV) + β7 (BEV) + FENameplate
For each observation in the dataset, the “Value” at age zero is determined by setting the age variable to zero and solving.
Start Printed Page 43084
The estimated willingness-to-pay for electrified powertrain packages over an internal combustion engine in an otherwise similar vehicle is computed as the difference between their estimated initial values, using the functions above. These pair-wise differences are averaged to estimate a price premium for new vehicles with HEV, PHEV, and BEV technologies. This analysis suggests that consumers are willing to pay more for new electrified vehicles than their new internal engine combustion counterparts, but only a little more, and not necessarily enough to cover the relatively large projected incremental cost to produce these vehicles. Specifically, the analysis estimated consumers are willing to pay between $2,000 and $3,000 more for the electrified powertrains considered here than their internal combustion engine counterparts.
Table-II-37 illustrates the variation in willingness-to-pay by electrification level (although the statistical model did not distinguish between PHEV30 and PHEV50 due to the small number of available operations for plug-in hybrids). As the table demonstrates, the difference between the median and mean predicted price premium for PHEVs is significant. The limited number of PHEV observations were not uniformly distributed among the nameplates present, and some of the luxury vehicles in the set retained value in a way that skewed the average. The CBI acquired from manufacturers was more consistent with the mean than median value (except for the PHEVs).
Additionally, the Kelley Blue Book data suggest that the used electrified vehicles were often worth less than their used internal combustion engine counterpart vehicles after a few years of use.
As Table-II-38 illustrates, the value of the price premium shrinks as the vehicles age and depreciate. Using the statistical model, we estimate that strong hybrids hold less than $100 of the initial price premium by age eight (on average). While the battery electric vehicles appear to be worth less than their ICE counterparts by age eight, there is limited data about this emerging segment of the new vehicle market. These independently-produced results using publicly available data were in line with manufacturers' reported confidential business information.
The “technology cost burden” numbers used in today's analysis represent the amount of a given technology's incremental cost that manufacturers are unable to pass along to the buyer of a given vehicle at the time of purchase. The burden is defined as the difference between estimated willingness-to-pay, itself a combination of the estimated values and confidential business information received from manufacturers any tax credits that can be passed through in the price, and the cost of the technology. In general, the incremental willingness-to-pay falls well short of the costs currently projected for HEVs, PHEVs, and BEVs; for example, BEV technology can add roughly $18,000 in equipment costs to the vehicle after standard retail price equivalent markups (with a large portion of those costs being batteries), but the estimated willingness-to-pay is only about $3,000. While tax credits offset some, if not most of that difference for PHEVs and BEVs, there is some residual amount that buyers of new electrified vehicles are currently unwilling to cover, and that must either come from forgone profits or be passed Start Printed Page 43085along to buyers of other vehicles in a manufacturer's portfolio.
Manufacturers may be able to recover some or all of these costs by charging higher prices for their other models, in which case it will represent a welfare loss to buyers of other vehicles (even if not to buyers of HEVs, PHEVs, or BEVs themselves). To the extent that they are unable to do so and must absorb part or all of these costs, their profits will decline, and in effect this cost will be borne by their investors. In practice, the analysis estimates benefits and costs to car and light truck manufacturers and buyers under the assumption that each manufacturer recovers all technology costs and civil penalties it incurs from buyers via higher average prices for the models it produces and sells, although sufficient information to support specific assumptions about price increases for individual models is not present. In effect, this means that any part of a manufacturer's costs to convert specific models to electric drive technologies that it cannot recover by charging higher prices to their buyers will be borne collectively by buyers of the other models they produce. Each of those buyers is in effect assumed to pay a slight premium (or “markup”) over the manufacturer's cost to produce the models they purchase (including the cost of any technology used to improve its fuel economy), this premium on average is modeled to recover the full cost of technology applied to all vehicles to improve the fuel economy of the fleet. So, even though electrified vehicles are modeled as if their buyers are unwilling to pay the full cost of the technology associated with their fuel economy improvement, the price borne by the average new vehicle buyer represents the average incremental technology cost for all applied technology, the sum of all technology costs divided by the number of units sold, across all classes, for each manufacturer.
The willingness-to-pay analysis described above relies on used vehicle data that is widely available to the public. Market tracking services update used vehicle price estimates regularly as fuel prices and other market conditions change, making the data easy to update in the future as market conditions change. The used vehicle data also account for consumer willingness-to-pay absent State and Federal rebates at the time of sale, which are reflected in both the initial purchase price of the vehicle and its later value in the used vehicle market. As such, the analysis would continue to be relevant even if incentive programs for vehicle electrification change or phase out in the future. By considering a variety of nameplates and body styles produced by several manufacturers, this analysis produces average willingness-to-pay estimates that can be applied to the whole industry. By evaluating matched pairs of vehicles from the same manufacturer, the analysis accounts for many additional factors that may be tied to the brand, rather than the technology, and influence the fair market price of vehicles. In particular, the data inherently include customer valuations for fuel-savings and vehicle maintenance schedules, as well as other factors like noise-vibration-and-harshness, interior space,
and fueling convenience in the context of the vehicles considered.
There are some limitations to this approach. There are currently few observations of PHEV and BEV technologies in the data, and most of the observations for BEVs are sedans and small cars, the values for which are extrapolated to other market segments. Additionally, the used vehicle data supporting these estimates inherently includes both older and newer generations of technology, so the historical regression may be slow to react to rapid changes in the new vehicle marketplace. As new vehicle nameplates emerge, and existing nameplates improve their implementation of electrification technologies, this model will require re-estimation to determine how these new entrants impact the estimated industry average willingness-to-pay.
Additionally, the willingness-to-pay analysis does not consider electric vehicles with no direct ICE counterpart. For example, today's evaluation does not consider Tesla because the Tesla brand has no ICE equivalent, and because the free-market prices for used Tesla vehicles have been difficult (if not impossible) to obtain, primarily due to factory guaranteed resale values (which is a program that still affects the used market for many Tesla vehicles). Still, Tesla vehicles have a large share of the BEV market by both unit sales and dollar sales, it may be possible to include Tesla data in a future update to this analysis. Similarly, the analysis did not include ICE vehicles with no similar HEV, PHEV, or BEV nameplate or counterpart, so the analysis presented here looks at a small portion of all transactions and is more likely to include fuel efficient models where market demand for hybrid (or higher) versions may exist. One possible alternative is to rely on new vehicle transaction prices to estimate consumer willingness-to-pay for new vehicles with certain attributes. However, new vehicle transaction data is highly proprietary and difficult to obtain in a form that may be disclosed to the public.
While estimating willingness-to-pay for electrification technologies from depreciation and MSRP data is appealing, many manufacturers handle MSRP and pricing strategies differently, with some preferring to deviate only a little from sticker price and others preferring to offer high discounts. There is evidence of large differences between MSRP and effective market prices to consumers for many vehicles, especially BEVs.
Please provide comments on methods and data used to evaluate consumer willingness-to-pay for electrification technologies.
(e) Refueling Surplus
Direct estimates of the value of extended vehicle range are not available in the literature, so the reduction in the required annual number of refueling cycles due to improved fuel economy was calculated and the economic value of the resulting benefits assessed. Chief among these benefits is the time that owners save by spending less time both in search of fueling stations and in the act of pumping and paying for fuel.
The economic value of refueling time savings was calculated by applying DOT-recommended valuations for travel time savings to estimates of how much time is saved.
The value of travel time depends on average hourly valuations of personal and business time, which are functions of total hourly compensation costs to employers. The total hourly compensation cost to employers, inclusive of benefits, in 2010$ is $29.68.
Table-II-39 below demonstrates the approach to estimating the value of travel time ($/hour) for both urban and rural (intercity) driving. This approach relies on the use of DOT-recommended weights that assign a lesser valuation to personal travel time than to business travel time, as well as Start Printed Page 43086weights that adjust for the distribution between personal and business travel.
The estimates of the hourly value of urban and rural travel time ($15.67 and $21.93, respectively) shown in Table-II-39 above must be adjusted to account for the nationwide ratio of urban to rural driving. By applying this adjustment (as shown in Table-II-40 below), an overall estimate of the hourly value of travel time—independent of urban or rural status—may be produced.
The calculations above assume only one adult occupant per vehicle. To fully estimate the average value of vehicle travel time, the presence of additional adult passengers during refueling trips must be accounted for. The analysis applies such an adjustment as shown in Table-II-40; this adjustment is performed separately for passenger cars and for light trucks, yielding occupancy-adjusted valuations of vehicle travel time during refueling trips for each fleet.
Children (persons under age 16) are excluded from average vehicle occupancy counts, as it is assumed that the opportunity cost of children's time is zero.
Start Printed Page 43087
The analysis estimated the amount of refueling time saved using (preliminary) survey data gathered as part of our 2010-2011 National Automotive Sampling System's Tire Pressure Monitoring System (TPMS) study.
The study was conducted at fueling stations nationwide, and researchers made observations regarding a variety of characteristics of thousands of individual fueling station visits from August 2010 through April 2011.
Among these characteristics of fueling station visits is the total amount of time spent pumping and paying for fuel. From a separate sample (also part of the TPMS study), researchers conducted interviews at the pump to gauge the distances that drivers travel in transit to and from fueling stations, how long that transit takes, and how many gallons of fuel are being purchased.
This analysis of refueling benefits considers only those refueling trips which interview respondents indicated the primary reason was due to a low reading on the gas gauge.
This restriction was imposed so as to exclude drivers who refuel on a fixed (e.g., weekly) schedule and may be unlikely to alter refueling patterns as a result of increased driving range. The relevant TPMS survey data on average refueling trip characteristics are presented below in Table-II-41.
As an illustration of how the value of extended refueling range was estimated, assume a small light truck model has an average fuel tank size of approximately 20 gallons and a baseline actual on-road fuel economy of 24 mpg (its assumed level in the absence of a higher CAFE standard for the given model year). TPMS survey data indicate that drivers who indicated the primary reason for their refueling trips was a low reading on the gas gauge typically refuel when their tanks are 35% full (i.e. as shown in Table-II-41, with 7.0 gallons in reserve, and the consumer purchases 13 gallons). By this measure, a typical driver would have an effective driving range of 312 miles (= 13.0 gallons × 24 Start Printed Page 43088mpg) before he or she is likely to refuel. Increasing this model's actual on-road fuel economy from 24 to 25 mpg would therefore extend its effective driving range to 325 miles (= 13.0 gallons × 25 mpg). Assuming that the truck is driven 12,000 miles/year,
this one mpg improvement in actual on-road fuel economy reduces the expected number of refueling trips per year from 38.5 (= 12,000 miles per year/312 miles per refueling) to 36.9 (= 12,000 miles per year/325 miles per refueling), or by 1.6 refuelings per year. If a typical fueling cycle for a light truck requires a total of 6.83 minutes, then the annual value of time saved due to that one mpg improvement would amount to $3.97 (= (6.83/60) × $21.81 × 1.6).
In the central analysis, this calculation was repeated for each future calendar year that light-duty vehicles of each model year affected by the standards considered in this rule would remain in service. The resulting cumulative lifetime valuations of time savings account for both the reduction over time in the number of vehicles of a given model year that remain in service and the reduction in the number of miles (VMT) driven by those that stay in service. The analysis also adjusts the value of time savings that will occur in future years both to account for expected annual growth in real wages 
and to apply a discount rate to determine the net present value of time saved.
A further adjustment is made to account for evidence from the interview-based portion of the TPMS study which suggests that 40% of refueling trips are for reasons other than a low reading on the gas gauge. It is therefore assumed that only 60% of the theoretical refueling time savings will be realized, as it was assumed that owners who refuel on a fixed schedule will continue to do. Based on peer reviewer comments to NHTSA's initial implementation of refueling time savings (subsequent to the CAFE NPRM issued in 2011), the analysis of refueling time savings was updated for the final rule to reflect peer reviewer suggestions.
Beyond updating time values to current dollars, that analysis has been used, unchanged, in today's analysis as well.
Because a reduction in the expected number of annual refueling trips leads to a decrease in miles driven to and from fueling stations, the value of consumers' fuel savings associated with this decrease can also be calculated. As shown in Table-II-41, the typical incremental round-trip mileage per refueling cycle is 1.08 miles for light trucks and 0.97 miles for passenger cars. Going back to the earlier example of a light truck model, a decrease of 1.6 in the number of refuelings per year leads to a reduction of 1.73 miles driven per year (= 1.6 refuelings × 1.08 miles driven per refueling). Again, if this model's actual on-road fuel economy was 24 mpg, the reduction in miles driven yields an annual savings of approximately 0.07 gallons of fuel (= 1.73 miles/24 mpg), which at $3.25/gallon 
results in a savings of $0.23 per year to the owner.
This example is illustrative only of the approach used to quantify this benefit. In practice, the societal value of this benefit excludes fuel taxes (as they are transfer payments) from the calculation and is modeled using fuel price forecasts specific to each year the given fleet will remain in service.
The annual savings to each consumer shown in the above example may seem like a small amount, but the reader should recognize that the valuation of the cumulative lifetime benefit of this savings to owners is determined separately for passenger car and light truck fleets and then aggregated to show the net benefit across all light-duty vehicles, which is much more significant at the macro level. Calculations of benefits realized in future years are adjusted for expected real growth in the price of gasoline, for the decline in the number of vehicles of a given model year that remain in service as they age, for the decrease in the number of miles (VMT) driven by those that stay in service, and for the percentage of refueling trips that occur for reasons other than a low reading on the gas gauge; a discount rate is also applied in the valuation of future benefits. Using this direct estimation approach to quantify the value of this benefit by model year was considered; however, it was concluded that the value of this benefit is implicitly captured in the separate measure of overall valuation of fuel savings. Therefore, direct estimates of this benefit are not added to net benefits calculations. It is noted that there are other benefits resulting from the reduction in miles driven to and from fueling stations, such as a reduction in greenhouse gas emissions—CO2 in particular—which, as per the case of fuel savings discussed in the preceding paragraph, are implicitly accounted for elsewhere.
Special mention must be made with regard to the value of refueling time savings benefits to owners of electric and plug-in electric (both referred to here as EV) vehicles. EV owners who routinely drive daily distances that do not require recharging on-the-go may eliminate the need for trips to fueling or charging stations. It is likely that early adopters of EVs will factor this benefit into their purchasing decisions and maintain driving patterns that require once-daily at-home recharging (a process which generally takes five to eleven hours for a full charge) 
for those EV owners who have purchased and installed a Level Two charging station to a high-voltage outlet at their home or parking place. However, EV owners who regularly or periodically need to drive distances further than the fully-charged EV range may need to recharge at fixed locations. A distributed network of charging stations (e.g., in parking lots, at parking meters) may allow some EV owners to recharge their vehicles while at work or while shopping, yet the lengthy charging cycles of current charging technology may pose a cost to owners due to the value of time spent waiting for EVs to charge and potential EV shoppers who do not have access to charging at home (e.g., because they live in an apartment without a vehicle charging station, only Start Printed Page 43089have street parking, or have garages with insufficient voltage). Moreover, EV owners who primarily recharge their vehicles at home will still experience some level of inconvenience due to their vehicle being either unavailable for unplanned use or to its range being limited during this time should they interrupt the charging process. Therefore, at present EVs hold potential in offering significant time savings but only to owners with driving patterns optimally suited for EV characteristics. If fast-charging technologies emerge and a widespread network of fast-charging stations is established, it is expected that a larger segment of EV vehicle owners will fully realize the potential refueling time savings benefits that EVs offer. This is an area of significant uncertainty.
6. Vehicle Use and Survival
To properly account for the average value of consumer and societal costs and benefits associated with vehicle usage under various CAFE and GHG alternatives, it is necessary to estimate the portion of these costs and benefits that will occur at each age (or calendar year) for each model year cohort. Doing so requires some estimate of how many miles the average vehicle of a given type 
is expected to drive at each age and what share of the initial model year cohort is expected to remain at each age. The first estimates are referred to as the vehicle miles travelled (VMT) schedules and the second as the survival rate schedules. In this section the data sources and general methodologies used to develop these two essential inputs are briefly discussed. More complete discussions of the development of both the VMT schedules and the survival rate schedules are present in the PRIA Chapter 8.
(a) Updates to Vehicle Miles Traveled Schedules Since 2012 FR
The MY 2017-2021 FRM built estimates of average lifetime mileage accumulation by body style and age using the 2009 National Household Travel Survey (NHTS), which surveys odometer readings of the vehicles present from the approximately 113,000 households sampled. Approximately 210,000 vehicles were in the sample of self-reported odometer readings collected between April 2008 and April 2009. This represents a sample size of less than one percent of the more than 250 million light-duty vehicles registered in 2008 and 2009. The NHTS sample is now 10 years old and taken during the Great Recession. The 2017 NHTS was not available at the time of this rulemaking. Because of the age of the last available NHTS and the unusual economic conditions under which it was collected, NHTSA built the new schedule using a similar method from a proprietary dataset collected in the fall of 2015. This new data source has the advantages of both being newer, a larger sample, and collected by a third party.
(1) Data Sources and Estimation (Polk Odometer Data)
To develop new mileage accumulation schedules for vehicles regulated under the CAFE program (classes 1-3), NHTSA purchased a data set of vehicle odometer readings from IHS/Polk (Polk). Polk collects odometer readings from registered vehicles when they encounter maintenance facilities, state inspection programs, or interactions with dealerships and OEMs—these readings are more likely to be precise than the self-reported odometer readings collected in the NHTS. The average odometer readings in the data set NHTSA purchased are based on more than 74 million unique odometer readings across 16 model years (2000-2015) and vehicle classes present in the data purchase (all registered vehicles less than 14,000 lbs. GVW). This sample represents approximately 28% of the light-duty vehicles registered in 2015, and thus has the benefit of not only being a newer, but also, a larger, sample.
Comparably to the NHTS, the Polk data provide a measure of the cumulative lifetime vehicle miles traveled (VMT) for vehicles, at the time of measurement, aggregated by the following parameters: Make, model, model year, fuel type, drive type, door count, and ownership type (commercial or personal). Within each of these subcategories they provide the average odometer reading, the number of odometer readings in the sample from which Polk calculated the averages, and the total number of that subcategory of vehicles in operation.
In estimating the VMT models, each data point was weighted (make/model classification) by the share of each make/model in the total population of the corresponding vehicle body style. This weighting ensures that the predicted odometer readings, by body style and model year, represent each vehicle classification among observed vehicles (i.e., the vehicles for which Polk has odometer readings), based on each vehicles' representation in the registered vehicle population of its body style. Implicit in this weighting scheme is the assumption that the samples used to calculate each average odometer reading by make, model, and model year are representative of the total population of vehicles of that type. Several indicators suggest that this is a reasonable assumption.
First, the majority of vehicle make/models is well-represented in the sample. For more than 85% of make/model combinations, the average odometer readings are collected for 20% or more of the total population. Most make/model observations have sufficient sample sizes, relative to their representation in the vehicle population, to produce meaningful average odometer totals at that level. Second, we considered whether the representativeness of the odometer sample varies by vehicle age because VMT schedules in the CAFE model are specific to each age. It is possible that, for some of those models, an insufficient number of odometer readings is recorded to create an average that is likely to be representative of all of those models in operation for a given year. For all model years other than 2015, approximately 95% or more of vehicles types are represented by at least five percent of their population. For this reason, observations from all model years, other than 2015, were included in the estimation of the new VMT schedules.
Because model years are sold in in the Fall of the previous calendar year, throughout the same calendar year, and even into the following calendar year—not all registered vehicles of a make/model/model year will have been registered for at least a year (or more) until age three. The result is that some MY 2014 vehicles may have been driven for longer than one year, and some less, at the time the odometer was observed. In order to consider this in the definition of age, an age of a vehicle is assigned to be the difference between the average reading date of a make/model and the average first registration date of that make/model. The result is that the continuous age variable reflects the amount of time that a car has been registered at the time of odometer reading and presumably the time span that the car has accumulated the miles.
After creating the “age” variable, the analysis fits the make/model lifetime VMT data points to a weighted quartic polynomial regression of the age of the vehicle (stratified by vehicle body styles). The predicted values of the quartic regressions are used to calculate the marginal annual VMT by age for each body style by calculating differences in estimated lifetime mileage accumulation by age. However, the Polk data acquired by NHTSA only contains Start Printed Page 43090observations for vehicles newer than 16 years of age. In order to estimate the schedule for vehicles older than the age 15 vehicles in the Polk data, information about that portion of the schedule from the VMT schedules used in both the 2017-2021 Final Light Duty Rule and 2019-2025 Medium-Duty NPRM was combined. The light-duty schedules were derived from the survey data contained in the 2009 National Household Travel Survey (NHTS).
From the old schedules, the annual VMT is expected to be decreasing for all ages. Towards the end of the sample, the predictions for annual VMT increase. In order to force the expected monotonicity, a triangular smoothing algorithm is performed until the schedule is monotonic. This performs a weighted average which weights the observations close to the observation more than those farther from it. The result is a monotonic function, that predicts similar lifetime VMT for the sample span as the original function. Because the analysis does not have data beyond 15 years of age, it is not able to correctly capture that part of the annual VMT curve using only the new dataset. For this reason, trends in the old data to extrapolate the new schedule for ages beyond the sample range are used.
To use the VMT information from the newer data source for ages outside of the sample, final in-sample age (15 years) are used as a seed and then applied to the proportional trend from the old schedules to extrapolate the new schedules out to age 40. To do this, the annual percentage difference in VMT of the old schedule for ages 15-40 is calculated. The same annual percentage difference in VMT is applied to the new schedule to extend beyond the final in-sample value. This assumes that the overall proportional trend in the outer years is correctly modeled in the old VMT schedule and imposes this same trend for the outer years of the new schedule. The extrapolated schedules are the final input for the VMT schedules in the CAFE model. PRIA Chapter 8 contains a lengthier discussion of both the data source and the methodology used to create the new schedules.
(2) Using New Schedules in the CAFE Model/Analysis
While the Polk registration data set contains odometer readings for individual vehicles, the CAFE model tabulates “mileage accumulation” schedules, which relate average annual miles driven to vehicle age, based on vehicles' body style. For the purposes of VMT accounting, the CAFE model classifies vehicles in the analysis fleet as being one of the following: Passenger car, SUV, pickup truck, passenger van, or medium-duty pickup/van.
In order to use the Polk data to develop VMT schedules for each of these vehicle classes in the CAFE model, a mapping between the classification of each model in the Polk data and the classes in the CAFE model was first constructed. This mapping enabled separate tabulations of average annual miles driven at each age for each of the vehicle classes included in the CAFE model.
The only revision made to the mappings used to construct the new VMT schedules was to merge the SUV and passenger van body styles into a single class. These body styles were merged because there were very few examples of vans—only 38 models were in use during 2014, where every other body style had at least three times as many models. Further, as shown in the PRIA Chapter 8, there was not a significant difference between the 2009 NHTS van and SUV mileage schedules, nor was there a significant difference between the schedules built with the two body styles merged or kept separate using the 2015 Polk data. Merging these body styles does not change the workings of the CAFE model in any way, and the merged schedule is simply entered as an input for both vans and SUVs.
Although there is a single VMT by age schedule used as an input for each body style, the assumptions about the rebound effect require that this schedule be scaled for future analysis years to reflect changes in the cost of travel from the time the Polk sample was originally collected. These changes result from both changes in fuel prices between the time the sample was collected and any future analysis year and differences in fuel economy between the vehicles included in the sample used to build the mileage schedules and the future-year vehicles analyzed within the CAFE Model simulation.
As discussed in Section 0, recent literature supports a 20% “rebound effect” for light-duty vehicle use, which represents an elasticity of annual use with respect to fuel cost per mile of −0.2. Because fuel cost per mile is calculated as fuel price per gallon divided by fuel economy (in miles per gallon), this same elasticity applies to changes in fuel cost per mile that result from variation in fuel prices or differences in fuel economy. It suggests that a five percent reduction in the cost per mile of travel for vehicles of a certain body style will result in a one percent increase in the average number of miles they are driven annually.
The average cost per mile (CPM) of a vehicle of a given age and vehicle style in CY 2016 (the first analysis year of the simulation) was used as the reference point to calculate the rebound effect within the CAFE model. However, this does not perfectly align with the time of the collection of the Polk dataset. The Polk data were collected in 2015 (so that 2014 fuel prices were the last to influence sampled vehicles' odometer readings), and represents the average odometer reading at a single point in time for age (model year) included in the cross-section. We use the difference in the average odometer reading for each vintage during 2014 to calculate the number of miles vehicles are driven at each age (see PRIA Chapter 8 for specific details on the analysis). For example, we interpret the difference in the average odometer reading between the five- and six-year-old vehicles of a given body style as the average number of miles they are driven during the year when they were five years old. However, vehicles produced during different model years do not have the same average fuel economy, so it is important to consider the average fuel economy of each vintage (or model year) used to measure mileage accumulation at a given age when scaling VMT for the rebound calculation.
The first step in doing so is to adjust for any change in average annual use that would have been caused by differences in fuel prices between CYs 2014 and 2016. This is done by scaling the original schedules of annual VMT by age tabulated from the Polk sample using the following equation:
Start Printed Page 43091
Here, the average fuel economy for vehicles of a given body style and age refers to a different MY in 2016 than it did in 2014; for example, a MY 2014 vehicle had reached age two vehicle during CY 2016, whereas a 2012 model year vehicle was age two during CY 2014.
To estimate the average annual use of vehicles of a specified body type and age during future calendar years under a specific regulatory alternative, the CAFE model adjusts the resulting estimates of vehicle use by age for that body type during CY 2016 to reflect (1) the projected change in fuel prices from 2016 to each future calendar year; and (2) the difference between the average fuel economy for vehicles of that body type and age during a future calendar year and the average fuel economy for vehicles of that same body type and age during 2016. These two factors combine to determine the average fuel cost per mile for vehicles of that body type and age during each future calendar year and the average fuel cost per mile for vehicles of that same body type and age during 2016.
The elasticity of annual vehicle use with respect to fuel cost per mile is applied to the difference between these two values because vehicle use is assumed to respond identically to differences in fuel cost per mile that result from changes in fuel prices or from differences in fuel economy. The model then repeats this calculation for each calendar year during the lifetimes of vehicles of other body types, and subsequently repeats this entire set of calculations for each regulatory alternative under consideration. The resulting differences in average annual use of vehicles of each body type at each age interact with the number estimated to remain in use at that age to determine total annual VMT by vehicles of each body type.
This adjustment is defined by the equation below:
This equation uses the observed cost per mile of a vehicle of each age and style in CY 2016 as the reference point for all future calendar years. That is, the reference fuel price is fixed at 2016 levels, and the reference fuel economy of vehicles of each age is fixed to the average fuel economy of the vintage that had reached that age in 2016. For example, the reference CPM for a one-year-old SUV is always the CPM of the average MY 2015 SUV in CY 2016, and the CPM for a two-year-old SUV is always the CPM of the average MYv2014 SUV in CY 2016.
This referencing ensures that the model's estimates of annual mileage accumulation for future calendar years reflect differences in the CPM of vehicles of each given type and age relative to CPM resulting from the average fuel economy of vehicles of that type and age and observed fuel prices during the year when the mileage accumulation schedules were originally measured. This is consistent with a definition of the rebound effect as the elasticity of annual vehicle use with respect to changes in the fuel cost per mile of travel, regardless of the source of changes in fuel cost per mile. Alternative forms of referencing are possible, but none can guarantee that projected future vehicle use will respond to both projected changes in fuel prices and differences in individual models' fuel economy among regulatory alternatives.
The mileage estimates described above are a crucial input in the CAFE model's calculation of fuel consumption and savings, energy security benefits, consumer surplus from cheaper travel, recovered refueling time, tailpipe emissions, and changes in crashes, fatalities, noise and congestion.
(3) Comparison to other VMT projections (2012 FR, AEO average lifetime miles, totals?)
Across all body styles and ages, the previous VMT schedules estimate higher average annual VMT than the updated schedules. Table-II—42 compares the lifetime VMT under the 2009 NHTS and the 2015 Polk dataset. The 40-year lifetime VMT gives the Start Printed Page 43092expected lifetime VMT of a vehicle conditional on surviving to age 40. The new schedules predict between 24 and 31% fewer miles for a 40-year old vehicle depending on the body style. The new schedules predict that the average 40-year old vehicle will drive between approximately 260k and 280k miles depending on the body style versus between approximately 350k and 380k for the previous schedules.
The static survival-weighted lifetime VMT represents the expected number of miles the average vehicle of each body style will drive, weighting by the likelihood it survives to each age using the previous static scrappage schedules. The dynamic survival-weighted lifetime VMT represents the expected number of miles driven by each body style, weighting by the dynamic survival schedules under baseline assumptions.274 There is a similar proportional reduction in expected lifetime VMT under both survival assumptions, with the dynamic scrappage model predicting lifetime mileage accumulation within 10,000 miles of the previous static model under both VMT schedules. The expected lifetime mileage accumulation reduces between 13 and 15% under the current VMT schedules when compared to the previous schedules—a smaller proportional reduction than the unweighted lifetime assumptions. Using the updated schedules, the expected lifetime mileage accumulation is between approximately 150k and 170k miles depending on the body style, rather than the approximately 180k to 210k miles under the previous schedules. For more detail on when the mileage and survival rates occur, chapter 8 of the PRIA gives the full VMT schedules by age. The section below gives further estimates of how lifetime VMT estimates vary under different assumptions within the dynamic scrappage model.
We have several reasons for preferring the new VMT schedules over the prior iterations. Before discussing these reasons, it is important to note that NHTSA uses the same general methodology in developing both schedules. We consider data on average odometer readings by age and body style collected once during a given window of time; we then estimate a weighted polynomial function between vehicle age and lifetime accumulation for a given vehicle style. As with the previous schedules, we use the inter-annual differences as the estimate of annual miles traveled for a given age.
The primary advantage of the current schedules is the data source. The previous schedules are based on data that is outdated and self-reported, while the observations from Polk are between five and seven years newer than those in the NHTS and represent valid odometer readings (rather than self-reported information). Further, the 2009 NHTS represents approximately one percent of the sample of vehicles registered in 2008/2009, while the 2015 Polk dataset represents approximately 30% of all registered light-duty vehicles; it is a much larger dataset, and less likely to oversample certain vehicles. Additionally, while the NHTS may be a representative sample of households, it is less likely to be a representative sample of vehicles. However, by properly accounting for vehicle population weights in the new averages and models, we corrected for this issue in the derivation of the new schedules.
Importantly, this methodology treats the cross-section of ages in a single calendar year as a panel of the same model year vehicle, when in reality each age represents a single model year, and not a true panel. We have some concern that where the most heavily driven vehicles drop out of the sample that the lifetime odometer readings will be lower than they would be if the scrapped vehicles had been left in the dataset without additional mileage accumulation. This would bias our estimates of inter-annual mileage accumulation downward and may result in an undervaluation of costs and benefits associated with additional travel for vehicles of older ages. For the next VMT schedule iteration, NHTSA intends to use panel data to test the magnitude of any attrition effect that may exist. While this caveat is important, all previous iterations were also built from a single calendar year cross-section and contain the same inherent bias.
(b) How does CAFE affect vehicle retirement rates?
Lightly used vehicles are a close substitute for new vehicles; thus, there is relationship between the two markets. As the price for new vehicles increases, there is an upward shift in the demand for used vehicles. As a result of the upward shift in the demand curve, the equilibrium price and quantity of used vehicles both increase; the value of used vehicles increases as a result. The decision to scrap or maintain a used vehicle is closely linked with the value of the vehicle; when the value is lesser than the cost to maintain the vehicle, it will be scrapped. In general, as a result of new vehicle price increases, the scrappage rate, or the proportion of vehicles remaining on the road unregistered in a given year, of used vehicles will decline. Because older vehicles are on average less efficient and less safe, this will have important implications for the evaluations of costs Start Printed Page 43093and benefits of fuel economy standards, which increase the cost of new vehicles and reduce the average cost per mile of fuel costs.
Fuel economy standards result in the application of more fuel saving technologies for at least some models, which result in a higher cost for manufacturers to produce otherwise identical vehicles. This increase in production cost amounts to an upward shift in the supply curve for new vehicles. This increases the equilibrium price and reduces the quantity of vehicles demanded. While the cost of new vehicles increases under increased fuel economy standards, the fuel cost per mile of travel declines. Consumers will place some value on the fuel savings associated with the additional technology, to the extent that they value reduced operating expenses against the increased price of a new vehicle, increased financing costs (and impediments to obtaining financing), and increased insurance costs.
There is a trade-off between fuel economy and other attributes that consumers value such as: Vehicle performance, interior volume, etc. Where the additional value of fuel savings associated with a technology is greater than any loss of value from trade-offs with other attributes, the demand for new vehicles will also shift upwards. Where the additional evaluation of fuel savings is lesser than any loss of value from changes to other attributes, the demand will shift downwards. Thus, the direction of the demand shift is unknown. However, if we assume that manufacturers pass all costs associated with a model off to the consumer of that vehicle, then the per vehicle profit remains constant. If we also assume that manufacturers are good predictors of the valuation and elasticity of certain vehicle attributes, then we can assume that even if there is some positive demand shift, it is not enough to increase demand above the original equilibrium levels, or manufacturers would apply those technologies even in the absence of regulation.
As noted above, the increase in the price of new vehicles will result in increased demand for used vehicles as substitutes, extending the expected age and lifetime vehicle miles travelled of less efficient, and generally, less safe vehicles. The additional usage of older vehicles will result in fewer gallons saved and more total on-road fatalities under more stringent CAFE alternatives. For more on the topic of safety, the relative safety of specific model year vehicles is discussed in Section 0 of the preamble and PRIA Chapter 11. Both the erosion of fuel savings and the increase in incremental fatalities will decrease the societal net benefits of increasing new vehicle fuel economy standards.
Our previous estimates of vehicle scrappage did not include a dynamic response to new vehicle price, but recent literature has continued to illustrate that this an omission which could rival the rebound effect in magnitude (Jacobsen & van Bentham, 2015). For this reason, we worked to develop an econometric survival model which captures the effect of increasing the price of new vehicles on the survival rate of used vehicles discussed in the following sections and in more detail in the PRIA Chapter 8. We discuss the literature on vehicle scrappage rate and discuss in the succeeding section. A brief explanation of why we develop our own models and the data sources and econometric estimations we use to do so, follows. We conclude the discussion of the updates to vehicle survival estimates with a summary of the results, a description of how we use them in the CAFE model, and finally, how the updated schedules compare with the previous static scrappage schedules.
(1) What does the literature say about the relationship?
(a) How Fuel Economy Standards Impact Vehicle Scrappage
The effects of differentiated regulation 
in the context of fuel economy (particularly, emission standards only affecting new vehicles) was discussed in detail in Gruenspecht (1981) and (1982), and has since been coined the “Gruenspecht effect.” Gruenspecht recognized that because fuel economy standards affect only new vehicles, any increase in price (net of the portion of reduced fuel savings valued by consumers) will increase the expected life of used vehicles and reduce the number of new vehicles entering the fleet. In this way, increased fuel economy standards slow the turnover of the fleet and the entrance of any regulated attributes tied only to new vehicles. Although Gruenspecht acknowledges that a structural model which allows new vehicle prices to affect used vehicle scrappage only through their effect on used vehicle prices would be preferable, the data available on used vehicle prices was (and still is) limited. Instead he tested his hypothesis in his 1981 dissertation using new vehicle price and other determinants of used car prices as a reduced form to approximate used car scrappage in response to increasing fuel economy standards.
Greenspan & Cohen (1996) offer additional foundations from which to think about vehicle stock and scrappage. Their work identifies two types of scrappage: Engineering scrappage and cyclical scrappage. Engineering scrappage represents the physical wear on vehicles, which results in their being scrapped. Cyclical scrappage represents the effects of macroeconomic conditions on the relative value of new and used vehicles; under economic growth the demand for new vehicles increases and the value of used vehicles declines, resulting in increased scrappage. In addition to allowing new vehicle prices to affect cyclical vehicle scrappage à la the Gruenspecht effect, Greenspan and Cohen also note that engineering scrappage seems to increase where EPA emission standards also increase; as more costs goes towards compliance technologies, it becomes more expensive to maintain and repair more complicated parts, and scrappage increases. In this way, Greenspan and Cohen identify two ways that fuel economy standards could affect vehicle scrappage: (1) Through increasing new vehicle prices, thereby increasing used vehicle prices, and finally, reducing on-road vehicle scrappage, and (2) by shifting resources towards fuel-saving technologies—potentially reducing the durability of new vehicles by making them more complex.
(b) Aggregate vs. Atomic Data Source in the Literature
One important distinction between the literatures on vehicles scrappage is between those that use atomic vehicle data, data following specific individual vehicles, and those that use some level of aggregated data, data that counts the total number of vehicles of a given type. The decision to scrap a vehicle is an atomic one—that is, made on an individual vehicle basis. The decision relates to the cost of maintaining a vehicle, and the value of the vehicle both on the used car market, and as scrap metal. Generally, a used car owner will decide to scrap a vehicle where the value of the vehicle is less than the value of the vehicle as scrap metal plus the cost to maintain or repair the vehicle. In other words, the owner gets more value from scrapping the vehicle than continuing to drive it or from selling it.
Recent work is able to model scrappage as an atomic decision due to the availability of a large database of used vehicle transactions. Following works by other authors including: Start Printed Page 43094Busse, Knittel, & Zettelmeyer (2013); Sallee, West, & Fan (2010); Alcott & Wozny (2013); and Li, Timmins, & von Haefen (2009)—Jacobsen & van Benthem (2015) considers the impact of changes in gasoline prices on used vehicle values and scrappage rates. In turn, they consider the impact of an increase in used vehicle values on the scrappage rate of those vehicles. They find that increases in gasoline price result in a reduction in the scrappage rate of the most fuel efficient vehicles and an increase in the scrappage rate of the least fuel efficient vehicles. This has important implications for the validity of the average fuel economy values linked to model years and assumed to be constant over the life of that model year fleet within this study. Future iterations of this study could further investigate the relationship between fuel economy, vehicle usage, and scrappage, as noted in other places in this discussion.
While the decision to scrap a vehicle is made atomically, the data available to NHTSA on scrappage rates and variables that influence these scrappage rates are aggregate measures. This influences the best available methods to measure the impacts of new vehicle prices on existing vehicle scrappage. The result is that this study models aggregate trends in vehicle scrappage and not the atomic decisions that make up these trends. Many other works within the literature use the same data source and general scrappage construct, such as: Walker (1968); Park (1977), Greene & Chen (1981); Gruenspecht (1981); Gruenspecht (1982); Feeney & Cardebring (1988); Greenspan & Cohen (1996); Jacobsen & van Bentham (2015); and Bento, Roth, & Zhuo (2016) all use the same aggregate vehicle registration data as the source to compute vehicle scrappage.
Walker (1968) and Bento, Roth, & Zhuo (2016) use aggregate data to directly compute the elasticity of scrappage from measures of used vehicle prices. Walker (1968) uses the ratio of used vehicle Consumer Price Index (CPI) to repair and maintenance CPI. Bento, Roth, & Zhuo (2016) use used vehicle prices directly. While the direct measurement of the elasticity of scrappage is preferable in a theoretical sense, the CAFE model does not predict future values of used vehicles, only future prices of new vehicles. For this reason, any model compatible with the current CAFE model must estimate a reduced form similar to Park (1977); Gruenspecht (1981); Greenspan & Cohen (1996), who use some form of new vehicle prices or the ratio of new vehicle prices to maintenance and repair prices to impute some measure of the effect of new vehicle prices on vehicle scrappage.
(c) Historical Trends in Vehicle Durability
Waker (1968); Park (1977); Feeney & Cardebring (1988); Hamilton & Macauley (1999); and Bento, Ruth, & Zhuo (2016) all note that vehicles change in durability over time. Walker (1968) simply notes a significant distinction in expected vehicle lifetimes pre- and post-World War I. Park (1977) discusses a `durability factor' set by the producer for each year so that different vintages and makes will have varying expected lifecycles. Feeney & Cardebring (1988) show that durability of vehicles appears to have generally increased over time both in the U.S. and Swedish fleets using registration data from each country. They also note that the changes in median lifetime between the Swedish and U.S. fleet track well, with a 1.5 year lag in the U.S. fleet. This lag is likely due to variation in how the data is collected—the Swedish vehicle registry requires a title to unregister a vehicle, and therefore gets immediate responses, where the U.S. vehicle registry requires re-registration, which creates a lag in reporting.
Hamilton & Macauley (1999) argue for a clear distinction between embodied versus disembodied impacts on vehicle longevity. They define embodied impacts as inherent durability similar to Park's producer supplied `durability factor' and Greenspan's `engineering scrappage' and disembodied effects those which are environmental, not unlike Greenspan and Cohen's `cyclical scrappage.' They use calendar year and vintage dummy variables to isolate the effects—concluding that the environmental factors are greater than any pre-defined `durability factor.' Some of their results could be due to some inflexibility of assuming model year coefficients are constant over the life of a vehicle, and there may be some correlation between the observed life of the later model years of their sample and the `stagflation' 
of the 1970's. Bento, Ruth, & Zhuo (2016) find that the average vehicle lifetime has increased 27% from 1969 to 2014 by sub-setting their data into three model year cohorts. To implement these findings in the scrappage model incorporated into the CAFE model, this study takes pains to estimate the effect of durability changes in such a way that the historical durability trend can be projected into the future; for this reason, a continuous `durability' factor as a function of model year vintage is included.
(d) Models of the Gruenspecht Effect Used in Other Policy Analyses
This is not the first estimation of the `Gruenspecht Effect' for policy considerations. In their Technical Support Document (TSD) for the 2004 proposal to reduce greenhouse gas emissions from motor vehicles, California Air Resources Board (CARB) outlines how they utilized the CARBITS vehicle transaction choice model in an attempt to capture the effect of increasing new vehicle prices on vehicle replacement rates. They consider data from the National Personal Transportation Survey (NPTS) as a source of revealed preferences and a University of California (UC) study as a source of stated preferences for the purchase and sale of household fleets under different prices and attributes (including fuel economy) of new vehicles.
The transaction choice model represents the addition and deletion of a vehicle from a household fleet within a short period of time as a “replacement” of a vehicle, rather than as two separate actions. Their final data set consists of 790 vehicle replacements, 292 additions, and 213 deletions; they do not include the deletions, but assume any vehicle over 19 years old that is sold is scrapped. This allows them to capture a slowing of vehicle replacement under higher new vehicle prices, but because their model does not include deletions, does not explicitly model vehicle scrappage, but assumes all vehicles aged 20 and older are scrapped rather than resold. They calibrate the model so that the overall fleet size is benchmarked to Emissions FACtors (EMFAC) fleet predictions for the starting year; the simulation then produces estimates that match the EMFAC predictions without further calibration.
The CARB study captures the effect on new vehicle prices on the fleet replacement rates and offers some precedence for including some estimate of the Gruenspecht Effect. One important thing to note is that because vehicles that exited the fleet without replacement were excluded, the effect of new vehicle prices on scrappage rates where the scrapped vehicle is not replaced is not captured. Because new and used vehicles are substitutes, it is expected that used vehicle prices will increase with new vehicle prices. Because higher used vehicle prices will lower the number of vehicles whose cost of maintenance is higher than their value, it is expected that not only will Start Printed Page 43095replacements of used vehicles slow, but also, that some vehicles that would have been scrapped without replacement under lower new vehicle prices will now remain on the road because their value will have increased. Aggregate measures of the Gruenspecht effect will include changes to scrappage rates both from slower replacement rates, and slower non-replacement scrappage rates.
(2) Description of Data Sources
NHTSA purchases proprietary data on the registered vehicle population from IHS/Polk for safety analyses. IHS/Polk has annual snapshots of registered vehicle counts beginning in calendar year (CY) 1975 and continuing until calendar year 2015. The data includes the following regulatory classes as defined by NHTSA: Passenger cars, light trucks (classes 1 and 2a), and medium and heavy-duty trucks (classes 2b and 3). Polk separates these vehicles into another classification scheme: Cars and trucks. Under their schema, pickups, vans, and SUVs are treated as trucks, and all other body styles are included as cars. In order to build scrappage models to support the model year (MY) 2021-2026 light duty vehicle (LDV) standards, it was important to separate these vehicle types in a way compatible with the existing CAFE model.
There were two compatible choices to aggregate scrappage rates: (1) By regulatory class or (2) by body style. Because for NHTSA's purposes vans/SUVs are sometimes classified as passenger cars and sometimes as light trucks, and there was no quick way to reclassify some SUVs as passenger cars within the Polk dataset, NHTSA chose to aggregate survival schedules by body style. This approach is also preferable because NHTSA uses body style specific lifetime VMT schedules. Vehicles experience increased wear with use; many maintenance and repair events are closely tied to the number of miles on a vehicle. The current version of the CAFE model considers separate lifetime VMT schedules for cars, vans/SUVs, pickups and classes 2b and 3 vehicles. These vehicles are assumed to serve different purposes and, as a result, are modelled to have different average lifetime VMT patterns. These different uses likely also result in different lifetime scrappage patterns.
Once stratified into body style level buckets, the data can be aggregated into population counts by vintage and age. These counts represent the population of vehicles of a given body style and vintage in a given calendar year. The difference between the counts of a given vintage and vehicle type from one calendar year to the next is assumed to represent the number of vehicles of that vintage and type scrapped in a given year. There were a couple other important data considerations for the calculations of the historical scrappage rates not discussed here but discussed in detail in the PRIA Chapter 8.
For historical data on vehicle transaction prices, the models use data from the National Automobile Dealers Association (NADA), which records the average transaction price of all light-duty vehicles. These transaction prices represent the prices consumers paid for new vehicles but do not include any value of vehicles that may have been traded in to dealers. Importantly, these transaction prices were not available by vehicle body styles; thus, the models will miss any unique trends that may have occurred for a particular vehicle body style. This may be particularly relevant for pickup trucks, which observed considerable average price increases as luxury and high option pickups entered the market. Future models will further consider incorporating price series that consider the price trends for cars, SUVs and vans, and pickups separately.
The models use the NADA price series rather than the Bureau of Labor Statistics (BLS) New Vehicle Consumer Price Index (CPI), used by Park (1977) and Greenspan & Cohen (1997), because the BLS New Vehicle CPI makes quality adjustments to the new vehicle prices. BLS assumes that additions of safety and fuel economy equipment are a quality adjustment to a vehicle model, which changes the good and should not be represented as an increase in its price. While this is good for some purposes, it presumes consumers fully value technologies that improve fuel economy. Because it is the purpose to this study to measure whether this is true, it is important that vehicle prices adjusted to fully value fuel economy improving technologies, which would obscure the ability to measure the preference for more fuel efficient and expensive new vehicles, are not used. As further justification for using the NADA price series over the BLS New Vehicle CPI, Park (1977) cites a discontinuity found in the amount of quality adjustments made to the series so that more adjustments are made over time. This could further limit the ability for the BLS New Vehicle CPI to predict changes in vehicle scrappage.
Vehicle scrappage rates are also influenced by fuel economy and fuel prices. Historical data on the fuel economy by vehicle style from model years 1979-2016 was obtained from the 2016 EPA Motor Trends Report.
The van/SUV fuel economy values represent a sales-weighted harmonic average of the individual body styles. Fuel prices were obtained from Department of Energy (DOE) historical values, and future fuel prices within the CAFE model use the Annual Energy Outlook (AEO) future oil price projections.
From these values the average cost per 100 miles of travel for the cohort of new vehicles in a given calendar year and the average cost per 100 miles of travel for each used model year cohort in that same calendar year are computed.
It is expected that as the new vehicle fleet becomes more efficient (holding all other attributes constant) that it will be more desirable, and the demand for used vehicles should decrease (increasing their scrappage). As a given model year cohort becomes more expensive to operate due to increases in fuel prices, it is expected the scrappage of that model year will increase. It is perhaps worth noting that more efficient model year vintages will be less susceptible to changes in fuel prices, as Start Printed Page 43096absolute changes in their cost per mile will be smaller. The functional forms of the cost per mile measures are further discussed in the model specification subsection 3 below.
Aggregate measures that cyclically affect the value of used vehicles include macroeconomic factors like the real interest rate, the GDP growth rate, unemployment rates, cost of maintenance and repairs, and the value of a vehicle as scrap metal or as parts. Here only the GDP growth rate is discussed, as this is the only measure included in the final model. Extended reasoning as to why other variables are not included in the final model in the PRIA Chapter 8 is offered, but the discussion was omitted here for brevity in describing only the final model. Generally economic growth will result in a higher demand for new vehicles—cars in aggregate are normal goods—and a reduction in the value of used vehicles. The result should be an increase in the scrappage rate of existing vehicles so that we expect the GDP growth rate to be an important predictor of vehicle scrappage rates.
NHTSA sourced the GDP growth rate from the 2017 OASDI Trustees Report.
The Trustees Report offers credible projections beyond 2032. Because the purpose of building this scrappage model is to project vehicle survival rates under different fuel economy alternatives and the current fuel economy projections go as far forward as calendar year 2032, using a data set that encompasses projections at least through 2032 is an essential characteristic of any source used for this analysis.
(3) Summary of Model Estimation
The most predictive element of vehicle scrappage is what Greenspan and Cohen deem `engineering scrappage.' This source of scrappage is largely determined by the age of a vehicle and the durability of a specific model year vintage. Vehicle scrappage typically follows a roughly logistic function with age—that is, instantaneous scrappage increases to some peak, and then declines, with age as noted in Walker (1968); Park (1977); Greene & Chen (1981); Gruenspecht (1981); Feeney & Cardebring (1988); Greenspan & Cohen (1996); Hamilton & Macauley (1999); and Bento, Roth, & Zhuo (2016). Thus, this analysis also uses a logistic function to capture this trend of vehicle scrappage with age but allows non-linear terms to capture any skew to the logistic relationship. Specific details about the final and considered forms of engineering scrappage by body styles is presented in the PRIA Chapter 8.
The final and considered independent variables intended to capture cyclical elements of vehicle scrappage and the considered forms of each are discussed in PRIA Chapter 8; here only inclusion of the GDP growth rate is discussed. The GDP growth rate is not a single-period effect; both the current and previous GDP growth rates will affect vehicle scrappage rates. A single year increase will affect scrappage differently than a multi-period trend. For this reason, an optimal number of lagged terms are included: The within-period GDP growth rate, the previous period GDP growth rate, and the growth rate from two prior years for the car model, while for vans/SUVs, and pickups, the current and previous period GDP growth rate are sufficient.
Similarly, the considered model allows that one-period changes in new vehicle prices will affect the used vehicle market differently than a consistent trend in new vehicle prices. The optimal number of lags is three so that the price trend from the current year and the three prior years influences the demand for and scrappage of used vehicles. Note: The average lease length is three years 
so that the price of an average vehicle coming off lease is estimated to affect the scrappage rate of used vehicles—this is a major source of the newest used vehicles that enter the used car fleet. Further, because increases in new vehicle prices due to increased stringency of CAFE standards is the primary mechanism through which CAFE standards influence vehicle scrappage and the CAFE Model assumes that usage, efficiency, and safety vary with the age of the vehicle, particular attention is paid to the form of this effect. It is important to know the likelihood of scrappage by the age of the vehicle to correctly account for the additional costs of additional fatalities and increased fuel consumption from deferred scrappage. Thus, the influence of increasing new vehicle prices is allowed to influence the demand for used vehicles (and reduce their scrappage) differently for different ages of vehicles in the scrappage model. We discuss both how we determined the correct form and number of lags for each body style in PRIA Chapter 8.
The final cyclical factor affecting vehicle scrappage in the preferred model is the cost per 100 miles of travel both of new vehicles and of the vehicle which is the subject of the decision to scrap or not to scrap. The new vehicle cost per 100 miles is defined as the ratio of the average fuel price faced by new vehicles in a given calendar year and the average new vehicle fuel economy for 100 miles in the same calendar year, and varies only with calendar year:
The cost per 100 miles of the potentially scrapped vehicle is described as the ratio of the average fuel price faced by that model year vintage in a given calendar year and the average fuel economy for 100 miles of travel for that model year when it was new, and varies both with calendar year and model year:
Start Printed Page 43097
The average per-gallon fuel price faced by a model year vintage in a given calendar year is the annual average fuel price of all fuel types present in that model year fleet for the given calendar year, weighted by the share of each fuel type in that model year fleet. Or the following, where FT represents the set of fuel types present in a given model year vintage:
For these variables, the best fit model includes the cost per mile of both the new and the used vehicle for the current and prior year. This is congruent with research that suggests consumers respond to current fuel prices and fuel price changes. The selection process of this form for the cost per mile and the implications is discussed in PRIA Chapter 8.
There are a couple other controlling factors considered in our final model. The 2009 Car Allowance Rebate System (CARS) is not outlined here but is outlined in PRIA Chapter 8. This program aimed to accelerate the retirement of less fuel efficient vehicles and replace them with more fuel efficient vehicles. Further discussion of how this is controlled for is located in PRIA Chapter 8. Finally, evidence of autocorrelation was found, and including three lagged values of the dependent variable addresses the concern. Treatment of autocorrelation is discussed in PRIA Chapter 8.
One additional issue encountered in the estimations of scrappage rates is that the models predict too many vehicles remain on the road in the later years. This issue occurs because the data beyond age 15 are progressively more sparsely populated; vehicles over 15 years were not captured in the Polk data until 1994, when each successive collection year added an additional age of vehicles until 2005 when all ages began to be collected. This means that for vehicles over the age of 25 there are only 10 years of data. In order to correct for this issue the fact that the final fleet share converges to roughly the same share for most model years for a given vehicle type is used. The predicted versus historical relationships seem to deviate beginning around age 20; thus, for scrappage rates for vehicles beyond age 20 an exponential decay function which guarantees that by age 40 the final fleet share reaches the convergence level observed in the historical data is applied. The application of the decay function and mathematical definition is further defended in PRIA Chapter 8.
A sensitivity case is also developed to isolate the magnitude of the Greunspecht effect. The impacts on costs and benefits are presented in section VII.H.1 of this document. In order to isolate the effect, the price of new vehicles is held constant at CY 2016 levels. The specific methodology used to do so is described in detail in PRIA Chapter 8, as is the leakage implied by comparing the reference and no Gruenspecht effect sensitivity cases. It is important to note here that the leakage calculated ranges between 12 and 18% across regulatory alternatives. This is in line with Jacobsen & van Bentham (2015) estimates which put leakage for their central case between 13 and 16%. Their high gasoline price case is more in line this analysis' central case—with fuel prices of $3/gallon—and predicts leakage of 21%. This further validates the scrappage model effects against examples in the literature.
The models used for this analysis are able to capture the relationship for vehicle scrappage as it varies with age and how this relationship changes with increases to new vehicle price, the cost per mile of travel of new and used vehicles, and how the rate varies cyclically with the GDP growth rate. It also controls for the CARS program and checks the influence of a change in Polk's data collection procedures. The goodness of fit measures and the plausibility of the predictions of the model are discussed at some length in PRIA Chapter 8. In the next section, the impacts of updating the static scrappage models to the dynamic models on average vehicle age and usage, by body styles, and across different regulatory assumptions are discussed.
(c) What is the estimated effect on vehicle retirement and how do results compare to previously estimated fleets and VMT?
The expected lifetime of a car estimated using the static scrappage schedule from the 2012 final rule, both in years and miles, is between the expected lifetime of the dynamic scrappage model in the absence of CAFE standards and under the baseline standards. Estimated by the dynamic scrappage model, the average vehicle is expected to live 15.1 years under the influence of only market demand for new technology, and 15.6 years under the baseline scenario, a four percent increase. However, given the distribution of the mileage accumulation schedule by age, this amounts only to a two percent increase in the expected lifetime mileage accumulation of an individual vehicle. This range is consistent with DOT expectations in terms of direction and magnitude.
The use of a static retirement schedule, while deemed a reasonable approach in the past, is a limited representation of scrappage behavior. It fails to account for increasing vehicle durability—occurring for the last several decades—and the resulting increase in average vehicle age in the on-road fleet, which has nearly doubled since 1980.
Thus, turning off the dynamic scrappage model described above would not impose a perspective on the analysis that is neutral with respect to observed scrappage behavior but would instead represent a strong assumption that asserts important trends in the historical record will abruptly cease or change direction.
As discussed above, the dynamic scrappage model implemented to support this proposal affects total fleet size through several mechanisms. Although the model accounts for the influence of changes to average new vehicle price and U.S. GDP growth, the most influential mechanism, by far, is the observed trend of increasing vehicle durability over successive model years. This phenomenon is prominently discussed in the academic literature related to vehicle retirement, where there is no disagreement about its existence or direction.
In fact, when the CAFE model is exercised in a way that keeps average new vehicle prices at (approximately) MY 2016 levels, the on-road fleet grows from an initial level of 228 million in 2016 to 340 million in 2050, an increase of 49% over the 35-year period from 2016 to 2050.
The historical data show the size of the registered vehicle population (i.e., the on-road fleet) growing by about 60% in the 35 years between 1980 and Start Printed Page 430982015.
In the 35 years between 2016 and 2050, our simulation shows the on-road fleet growing from about 230 million vehicles to about 345 million vehicles when the market adopts only the amount of fuel economy, which it naturally demands. The simulated growth over this period is about 50% from today's level, rather than the 60% observed in the historical data over the last 35 years. Under the baseline regulatory scenario, the growth over the next 35 years is simulated to be about 54%—still short of the observed growth over a comparable period of time. In fact, the simulated annual growth rate in the size of the on-road fleet in this analysis, about 1.3%, is lower than the long-term average annual growth rate of about two percent dating back to the 1970s.
Additionally, there are inherent precision limitations in measuring something as vast and complex as the registered vehicle population. For decades, the two authoritative sources for the size of the on-road fleet have been R.L. Polk (now IHS/Polk) and FHWA. For two decades these two sources differed by more than 10% each year, only lately converging to within a few percent of each other. These discrepancies over the correct interpretation of the data by each source have consistently represented differences of more than 10 million vehicles.
The total number of new vehicles projected to enter the fleet is slightly higher than the historical trend (though the impact of the great recession makes it hard to say by how much). More generally, the projections used in the analysis cover long periods of time without exhibiting the kinds of fluctuation that are present in the historical record. For example, the forecast of GDP growth in our analysis posits a world in which the United States sees uninterrupted positive annual growth in real GDP for four decades. The longest such period in the historical record is 17 years and still included several years of low (but positive) growth during that interval.
Over such a long period of time, in the absence of deep insight into the future of the U.S. auto industry, it is sensible to assume that the trends observed over the course of decades are likely to persist. Analyzing fuel economy standards requires an understanding of the mechanisms that influence new vehicle sales, the size of the on-road fleet, and vehicle miles traveled. It is upon these mechanisms that the policy acts: Increasing/decreasing new vehicle prices changes the rate at which new vehicles are sold, changing the attributes and prices of these vehicles influences the rates at which all used vehicles are retired, the overall size of the on-road fleet determines the total amount of VMT, which in turn affects total fuel consumption, fatalities, and other externalities. The fact that DOT's bottom-up approach produces results in line with historical trends is both expected and intended.
This is not to say that all details of this new approach will be immediately intuitive for reviewers accustomed to results that do not include a dynamic sales model or dynamic scrappage model, much less results that combine the two. For example, some reviewers may observe that today's analysis shows that, compared to the baseline standards, the proposed standards produce a somewhat smaller on-road fleet (i.e., fewer vehicles in service) despite somewhat increased sales of new vehicles (consistent with reduced new vehicle prices) and decreased prices for used vehicles. While it might be natural to assume that reduced prices of new vehicles and increased sales should lead to a larger on-road fleet, in our modelling, the increased sales are more than offset by the somewhat accelerated scrappage that accompanies the estimated decrease in new vehicle prices. This outcome represents an on-road fleet that is both smaller and a little younger on average (relative to the baseline) and “turns over” more quickly.
To further test the validity of the scrappage model, a dynamic forecast was constructed for calendar years 2005 through 2015 to see how well it predicts the fleet size for this period. The last true population the scrappage model “sees” is the 2005 registered vehicle population. It then takes in known production volumes for the new model year vehicles and dynamically estimates instantaneous scrappage rates for all registered vehicles at each age for CYs 2006-2015, based only on the observed exogenous values that inform the model (GDP growth rate, observed new vehicle prices, and cost per mile of operation), fleet attributes of the vehicles (body style, age, cost per mile of operation), and estimated scrappage rates at earlier ages. Within this exercise, the scrappage model relies on its own estimated values as the previous scrappage rates at earlier ages, forcing any estimation errors to propagate through to future years. This exercise is discussed further in PRIA Chapter VII. While the years of the recession represent a significant shock to the size of the fleet, briefly reversing many years of annual growth, the model recovers quickly and produces results within one percent of the actual fleet size, as it did prior to the recession.
In order to compare the magnitudes of the sales and scrappage effects across different fuel economy standards considered it is important to define comparable measures. The sales effect in a single calendar year is simply the difference in new vehicle sales across alternatives. However, the scrappage effect in a single calendar year is not simply the change in fleet size across regulatory alternatives. The scrappage model predicts the probability that a vehicle will be scrapped in the next year conditional on surviving to that age; the absolute probability that a vehicle survives to a given age is conditional on the scrappage effect for all previous analysis years. In other words, if successive calendar years observe lower average new vehicle prices, the effect of increased scrappage on fleet size will accumulate with each successive calendar year—because fewer vehicles survived to previous ages, the same probability of scrappage would result in a smaller fleet size for the following year as well, though fewer vehicles will have been scrapped than in the previous year.
To isolate the number of vehicles not scrapped in a single calendar year because of the change in standards, the first step is to calculate the number of vehicles scrapped in every calendar year for both the proposed standards and the baseline; this is calculated by the inter-annual change in the size of the used vehicle fleet (vehicles ages 1-39) for each alternative. The difference in this measure across regulatory alternatives represents the change in vehicle scrappage because of a change in the standards. The resulting scrappage effect for a single calendar year can be compared to the difference across regulatory alternatives in new vehicle sales for the same calendar year as a comparison of the relative magnitudes of the two effects. In most years, under the proposed standards relative to the baseline standards, the analysis shows that for each additional new vehicles sold, two to four used vehicles are removed from the fleet. Over the time period of the analysis these predicted differences in the numbers of vehicles accumulate, resulting in a maximum of Start Printed Page 43099seven million fewer vehicles by CY 2033 for the proposed CAFE standards relative to the augural standards, and nine million fewer vehicles by CY 2035 for the proposed GHG standards relative to the current GHG standards. Tables 11-29 and 11-30 in the PRIA show the difference in the fleet size by calendar year for the proposed standards relative to the augural standards for the CAFE and GHG programs, respectively.
To understand why the sales and scrappage effects do not perfectly offset each other to produce a constant fleet size across regulatory alternatives it is important to remember that the decision to buy a new vehicle and the decision to scrap a used vehicle are often not made by the same household as a joint decision. The average length of initial ownership for new vehicles is approximately 6.5 years (and increasing over time). Cumulative scrappage up to age seven is typically less than 10%of the initial fleet. This suggests that most vehicles belong to more than one household over the course of their lifetimes. The household that is deciding whether or not to purchase a new vehicle is rarely the same household deciding whether or not to scrap a vehicle. So a vehicle not scrapped in a given year is seldom the direct substitute for a new vehicle purchased by that household. Considering this, it is not expected that for every additional vehicle scrapped, there is also an additional new one sold, under the proposed standards relative to the baseline standards.
Further, while sales and scrappage decisions are both influenced by changes in new vehicle prices, the mechanism through which these decisions change are different for the two effects. A decrease in average new vehicle prices will directly increase the demand for new vehicles along the same demand curve. This decrease in new vehicle prices will cause a substitution towards new vehicles and away from used vehicles, shifting the entire demand curve for used vehicles downwards. This will decrease both the equilibrium prices of used vehicles, as shown in Figure 8-16 of the PRIA. Since the decision to scrap a vehicle in a given year is closely related to the difference between the vehicle's value and the cost to maintain it, if the value of a vehicle is lower than the cost to maintain it, the current owner will not choose to maintain the vehicle for their own use or for resale in the used car market, and the vehicle will be scrapped. That is, a current owner will only supply a vehicle to the used car market if the price of the vehicle is greater than the cost of supplying it. Lowering the equilibrium price of used vehicles will lower the increase the number of scrapped vehicles, lowering the supply of used vehicles, and decreasing the equilibrium quantity. The change in new vehicle sales is related to demand of new vehicles at a given price, but the change in used vehicle scrappage is related to the shift in the demand curve for used vehicles, and the resulting change in the quantity current owners will supply; these effects are likely not exactly offsetting.
Our models indicate that the ratio of the magnitude of the scrappage effect to the sales effect is greater than one so that the fleet grows under more stringent scenarios. However, it is important to remember that not all vehicles are driven equally; used vehicles are estimated to deliver considerably less annual travel than new vehicles. Further, used vehicles only have a portion of their original life left so that it will take more than one used vehicle to replace the full lifetime of a new vehicle, at least in the long-run. The result of the lower annual VMT and shorter remaining lifetimes of used vehicles, is that although the fleet is 1.5% bigger in CY 2050 for the augural baseline than it is for the proposed standards, the total non-rebound VMT for CY 2050 is 0.4% larger in the augural baseline than in the proposed standards. This small increase in VMT is consistent with a larger fleet size; if more used vehicles are supplied, there likely is some small resulting increase in VMT.
Our models face some limitations, and work will continue toward developing methods for estimating vehicle sales, scrappage, and mileage accumulation. For example, our scrappage model assumes that the average VMT for a vehicle of a particular vintage is fixed—that is, aside from rebound effects, vehicles of a particular vintage drive the same amount annually, regardless of changes to the average expected lifetimes. The agencies seek comment on ways to further integrate the survival and mileage accumulation schedules. Also, our analysis uses sales and scrappage models that do not dynamically interact (though they are based on similar sets of underlying factors); while both models are informed by new vehicle prices, the model of vehicle sales does not respond to the size and age profile of the on-road fleet, and the model of vehicle scrappage rates does not respond to the quantity of new vehicles sold. As one potential option for development, the potential for an integrated model of sales and scrappage, or for a dynamic connection between the two models will be considered. Comment is sought on both the sales and scrappage models, on potential alternatives, and on data and methods that may enable practicable integration of any alternative models into the CAFE model.
7. Accounting for the Rebound Effect Caused by Higher Fuel Economy
(a) What is the rebound effect and how is it measured?
Amending and establishing fuel economy and GHG standards at a lesser stringency than the augural standards for future model years will lead to comparatively lower fuel economy for new cars and light trucks, thus increasing the amount of fuel they consume in traveling each mile than they would under the augural standard. The resulting increase in their per-mile fuel and total driving costs will lead to a reduction in the number of miles they are driven each year over their lifetimes, and example of the rebound effect that is usually associated with energy efficiency improvements working in reverse. The fuel economy rebound effect—a specific example of the energy efficiency rebound effect for the case of motor vehicles—refers to the well-documented tendency of vehicles' use to increase when their fuel economy is improved and the cost of driving each mile declines as a result.
(b) What does the literature say about the magnitude of this effect?
Table-II-43 summarizes estimates of the fuel economy rebound effect for light-duty vehicles from studies conducted through 2008, when the agencies originally surveyed research on this subject.
After summarizing all of the estimates reported in published and other publicly-available research available at that time, it distinguishes among estimates based on the type of data used to develop them. As the table reports, estimates of the rebound effect ranged from 6% to as high as 75%, and the range spanned by published estimates was nearly as wide (7-75%).
Start Printed Page 43100
Most studies reported more than one empirical estimate, and the authors of published studies typically identified the single estimate in which they were most confident; these preferred estimates spanned only a slightly narrower range (9-75%).
Despite their wide range, these estimates displayed a strong central tendency, as Table-II-43 also shows. The average values of all estimates, those that were published, and authors' preferred estimates from published studies were 22-23%, and the median estimates in each category were close to these values, indicating nearly symmetric distributions. The estimates in each category also clustered fairly tightly around their respective average values, as shown by their standard deviations in the table's last column. When classified by the type of data they relied on, U.S. aggregate time-series data produced slightly smaller values (averaging 18%) than did panel-type data for individual states (23%) or household survey data (25%). In each category, the median estimate was again quite close to the average reported value, and comparing the standard deviations of estimates based on each type of data again suggests a fairly tight scatter around their respective means.
Of these studies, a then recently-published analysis by Small & Van Dender (2007), which reported that the rebound effect appeared to be declining over time in response to increasing income of drivers, was singled out. These authors theorized that rising income increased the opportunity cost of drivers' time, leading them to be less responsive over time to reductions in the fuel cost of driving each mile. Small and Van Dender reported that while the rebound effect averaged 22% over the entire time period they analyzed (1967-2001), its value had declined by half—or to 11%—during the last five years they studied (1997-2001). Relying primarily on forecasts of its continued decline over time, the analysis reduced the 20% rebound effect that NHTSA used to analyze the effects of CAFE standards for light trucks produced during model years 2005-07 and 2008-11 to 10% for their analysis of CAFE and GHG standards for MY 2012-16 passenger cars and light trucks.
Table-II-44 summarizes estimates of the rebound effect reported in research that has become available since the agencies' original survey, which extended through 2008, and the following discussion briefly summarizes the approaches used by these more recent studies. Bento et al. (2009) combined demographic characteristics of more than 20,000 U.S. households, the manufacturer and model of each vehicle they owned, and their annual usage of each vehicle from the 2001 National Household Travel Survey with detailed data on fuel economy and other attributes for each vehicle model obtained from commercial publications. The authors aggregated vehicle models into 350 categories representing combinations of manufacturer, vehicle type, and age, and use the resulting data to estimate the parameters of a complex model of households' joint choices of the number and types of vehicles to own, and their annual use of each vehicle.
Start Printed Page 43101
Bento et al. estimate the effect of vehicles' operating costs per mile, including fuel costs, which depend in part on each vehicle's fuel economy, as well as maintenance and insurance expenses, on households' annual use of each vehicle they own. Combining the authors' estimates of the elasticity of vehicle use with respect to per-mile operating costs with the reported fraction of total operating costs accounted for by fuel (slightly less than one-half) yields estimates of the rebound effect. The resulting values vary by household composition, vehicle size and type, and vehicle age, ranging from 21 to 38%, with a composite estimate of 34% for all households, vehicle models, and ages. The smallest values apply to new luxury cars, while the largest estimates are for light trucks and households with children, but the implied rebound effects differ little by vehicle age.
Barla et al. (2009) analyzed the responses of car and light truck ownership, vehicle travel, and average fuel efficiency to variation in fuel prices and aggregate economic activity (measured by gross product) using panel-type data for the 10 Canadian provinces over the period from 1990 through 2004. The authors estimated a system of equations for these three variables using statistical procedures appropriate for models where the variables of interest are simultaneously determined (that is, where each variable is one of the factors explaining variation in the others). This procedure enabled them to control for the potential “reverse influence” of households' demand for vehicle travel on their choices of how many vehicles to own and their fuel efficiency levels when estimating the effect of variation in fuel efficiency on vehicle use.
Their analysis found that provincial-level aggregate economic activity had moderately strong effects on car and light truck ownership and use but that fuel prices had only modest effects on driving and the average fuel efficiency of the light-duty vehicle fleet. Each of these effects became considerably stronger over the long term than in the year when changes in economic activity and fuel prices initially occurred, with three to five years typically required for behavioral adjustments to stabilize. After controlling for the joint relationship among vehicle ownership, driving demand, and the fuel efficiency of cars and light trucks, Barla et al. estimated elasticities of average vehicle use with respect to fuel efficiency that corresponded to a rebound effect of eight percent in the short run, rising to nearly 20% within five years. A notable feature of their analysis was that variation in average fuel efficiency among the individual Canadian provinces and over the time period they studied was adequate to identify its effect on vehicle use, without the need to combine it with variation in fuel prices in order to identify its effect.
Wadud et al. (2009) combine data on U.S. households' demographic characteristics and expenditures on gasoline over the period 1984-2003 from the Consumer Expenditure Survey with data on gasoline prices and an estimate of the average fuel economy of Start Printed Page 43102vehicles owned by individual households (constructed from a variety of sources). They employ these data to explore variation in the sensitivity of individual households' gasoline consumption to differences in income, gasoline prices, the number of vehicles owned by each household, and their average fuel economy. Using an estimation procedure intended to account for correlation among unmeasured characteristics of households and among estimation errors for successive years, the authors explore variation in the response of fuel consumption to fuel economy and other variables among households in different income categories and between those residing in urban and rural areas.
Dividing U.S. households into five equally-sized income categories, Wadud et al. estimate rebound effects ranging from 1-25%, with the smallest estimates (8% and 1%) for the two lowest income categories, and significantly larger estimates for the middle (18%) and two highest income groups (18 and 25%). In a separate analysis, the authors estimate rebound effects of seven percent for households of all income levels residing in U.S. urban areas and 21% for rural households.
West & Pickrell (2011) analyzed data on more than 100,000 households and 300,000 vehicles from the 2009 Nationwide Household Transportation Survey to explore how households owning multiple vehicles chose which of them to use and how much to drive each one on the day the household was surveyed. Their study focused on how the type and fuel economy of each vehicle a household owned, as well as its demographic characteristics and location, influenced household members' decisions about whether and how much to drive each vehicle. They also investigated whether fuel economy and fuel prices exerted similar influences on vehicle use, and whether households owning more than one vehicle tended to substitute use of one for another—or vary their use of all of them similarly—in response to fluctuations in fuel prices and differences in their vehicles' fuel economy.
Their estimates of the fuel economy rebound effect ranged from as low as nine percent to as high as 34%, with their lowest estimates typically applying to single-vehicle households and their highest values to households owning three or more vehicles. They generally found that differences in fuel prices faced by households who were surveyed on different dates or who lived in different regions of the U.S. explained more of the observed variation in daily vehicle use than did differences in vehicles' fuel economy. West and Pickrell also found that while the rebound effect for households' use of passenger cars appeared to be quite large—ranging from 17% to nearly twice that value—it was difficult to detect a consistent rebound effect for SUVs.
Anjovic & Haas (2012) examined variation in vehicle use and fuel efficiency among six European nations over an extended period (1970-2006), using an elaborate model and estimation procedure intended to account for the existence of common underlying trends among the variables analyzed and thus avoid identifying spurious or misleading relationships among them. The six nations included in their analysis were Austria, Germany, Denmark, France, Italy, and Sweden; the authors also conducted similar analyses for the six nations combined. The authors focused on the effects of average income levels, fuel prices, and the fuel efficiency of each nation's fleet of cars on the total distance they were driven each year and their total fuel energy consumption. They also tested whether the responses of energy consumption to rising and falling fuel prices appeared to be symmetric in the different nations.
Anjovic and Haas report a long-run aggregate rebound effect of 44% for the six nations their study included, with corresponding values for individual nations ranging from a low of 19% (for Austria) to as high as 56% (Italy). These estimates are based on the estimated response of vehicle use to variation in average fuel cost per kilometer driven in each of the six nations and for their combined total. Other information reported in their study, however, suggests lower rebound effects; their estimates of the response of total fuel energy consumption to fuel efficiency appear to imply an aggregate rebound effect of 24% for the six nations, with values ranging from as low as 0-3% (for Austria and Denmark) to as high as 70% (Sweden), although the latter is very uncertain. These results suggest that vehicle use in European nations may be somewhat less sensitive to variation in driving costs caused by changes in fuel efficiency than to changes in driving costs arising from variation in fuel prices, but they find no evidence of asymmetric responses of total fuel consumption to rising and falling prices. Using data on household characteristics and vehicle use from the 2009 Nationwide Household Transportation Survey (NHTS), Su (2012) analyzes the effects of locational and demographic factors on household vehicle use and investigates how the magnitude of the rebound effect varies with vehicles' annual use. Using variation in the fuel economy and per-mile cost of and detailed controls for the demographic, economic, and locational characteristics of the households that owned them (e.g., road and population density) and each vehicle's main driver (as identified by survey respondents), the author employs specialized regression methods to capture the variation in the rebound effect across 10 different categories of vehicle use.
Su estimated the overall rebound effect for all vehicles in the sample averaged 13%, and that its magnitude varied from 11-19% among the 10 different categories of annual vehicle use. The smallest rebound effects were estimated for vehicles at the two extremes of the distribution of annual use—those driven comparatively little, and those used most intensively—while the largest estimated effects applied to vehicles that were driven slightly more than average. Controlling for the possibility that high-mileage drivers respond to the increased importance of fuel costs by choosing vehicles that offer higher fuel economy narrowed the range of Su's estimated rebound effects slightly (to 11-17%), but did not alter the finding that they are smallest for lightly- and heavily-driven vehicles and largest for those with slightly above average use.
Linn (2013) also uses the 2009 NHTS to develop a linear regression approach to estimate the relationship between the VMT of vehicles belonging to each household and a variety of different factors: Fuel costs, vehicle characteristics other than fuel economy (e.g., horsepower, the overall “quality” of the vehicle), and household characteristics (e.g., age, income). Linn reports a fuel economy rebound effect with respect to VMT of between 20-40%.
One interesting result of the study is that when the fuel efficiency of all vehicles increases, which would be the long-run effect of rising fuel efficiency standards, two factors have opposing effects on the VMT of a particular vehicle. First, VMT increases when that vehicle's fuel efficiency increases. But the increase in the fuel efficiency of the household's other vehicles causes the vehicle's own VMT to decrease. Because the effect of a vehicle's own fuel efficiency is larger than the other vehicles' fuel efficiency, VMT increases if the fuel efficiency of all vehicles increases proportionately. Linn also finds that VMT responds much more strongly to vehicle fuel economy than to gasoline prices, which is at variance with the Hymel et al. and Greene results discussed above.Start Printed Page 43103
Like Su and Linn, Liu et al. (2014) employed the 2009 NHTS to develop an elaborate model of an individual household's choices about how many vehicles to own, what types and ages of vehicles to purchase, and how much combined driving to do using all of them. Their analysis used a complex mathematical formulation and statistical methods to represent and measure the interdependence among households' choices of the number, types, and ages of vehicles to purchase, as well as how intensively to use them.
Liu et al. employed their model to simulate variation in households' total vehicle use to changes in their income levels, neighborhood characteristics, and the per-mile fuel cost of driving averaged over all vehicles each household owns. The complexity of the relationships among the number of vehicles owned, their specific types and ages, fuel economy levels, and use incorporated in their model required them to measure these effects by introducing variation in income, neighborhood attributes, and fuel costs, and observing the response of households' annual driving. Their results imply a rebound effect of approximately 40% in response to significant (25-50%) variation in fuel costs, with almost exactly symmetrical responses to increases and declines.
A study of the rebound effect by Frondel et al. (2012) used data from travel diaries recorded by more than 2,000 German households from 1997 through 2009 to estimate alternative measures of the rebound effect, and to explore variation in their magnitude among households. Each household participating in the survey recorded its automobile travel and fuel purchases over a period of one to three years and supplied information on its composition and the personal characteristics of each of its members. The authors converted households' travel and fuel consumption to a monthly basis, and used specialized estimation procedures (quantile and random-effects panel regression) to analyze monthly variation in their travel and fuel use in relation to differences in fuel prices, the fuel efficiency of each vehicle a household owned, and the fuel cost per mile of driving each vehicle.
Frondel et al. estimate four separate measures of the rebound effect, three of which capture the response of vehicle use to variation in fuel efficiency, fuel price, and fuel cost per mile traveled, and a fourth capturing the response of fuel consumption to changes in fuel price. Their first three estimates range from 42% to 57%, while their fourth estimate corresponds to a rebound effect of 90%. Although their analysis finds no significant variation of the rebound effect with household income, vehicle ownership, or urban versus rural location, it concludes that the rebound effect is substantially larger for households that drive less (90%) than for those who use their vehicles most intensively (56%).
Gillingham (2014) analyzed variation in the use of approximately five million new vehicles sold in California from 2001 to 2003 during the first several years after their purchase, focusing particularly on how their use responded to geographic and temporal variation in fuel prices. His sample consisted primarily of personal or household vehicles (87%) but also included some that were purchased by businesses, rental car companies, and government agencies. Using county-level data, he analyzed the effect of differences in the monthly average fuel price paid by their drivers on variation in their monthly use and explored how that effect varied with drivers' demographic characteristics and household incomes.
Gillingham's analysis did not include a measure of vehicles' fuel economy or fuel cost per mile driven, so he could not measure the rebound effect directly, but his estimates of the effect of fuel prices on vehicle use correspond to a rebound effect of 22-23% (depending on whether he controlled for the potential effect of gasoline demand on its retail price). His estimation procedure and results imply that vehicle use requires nearly two years to adjust fully to changes in fuel prices. He found little variation in the sensitivity of vehicle use to fuel prices among car buyers with different demographic characteristics, although his results suggested that it increases with their income levels.
Weber & Farsi (2014) analyzed variation in the use of more than 70,000 individual cars owned by Swiss households who were included in a 2010 survey of travel behavior. Their analysis focuses on the simultaneous relationships among households' choices of the fuel efficiency and size (weight) of the vehicles they own, and how much they drive each one, although they recognize that fuel efficiency cannot be chosen independently of vehicle weight. The authors employ a model specification and statistical estimation procedures that account for the likelihood that households intending to drive more will purchase more fuel-efficient cars but may also choose more spacious and comfortable—and thus heavier—models, which affects their fuel efficiency indirectly, since heavier vehicles are generally less fuel-efficient. The survey data they rely on includes both owners' estimates of their annual use of each car and the distance it was actually driven on a specific day; because they are not closely correlated, the authors employ them as alternative measures of vehicle use to estimate the rebound effect, but this restricts their sample to the roughly 8,100 cars for which both measures are available. Weber and Farsi's estimates of the rebound effect are extremely large: 75% using estimated annual driving and 81% when they measure vehicle use by actual daily driving. Excluding vehicle size (weight) and limiting the choices that households are assumed to consider simultaneously to just vehicles' fuel efficiency and how much to drive approximately reverses these estimates, but both are still very large. Using a simpler procedure that does not account for the potential effect of driving demand on households' choices among vehicle models of different size and fuel efficiency produces much smaller values for the rebound effect: 37% using annual driving and 19% using daily travel. The authors interpret these latter estimates as likely to be too low because actual on-road fuel efficiency has not improved as rapidly as suggested by the manufacturer-reported measure they employ. This introduces an error in their measure that may be related to a vehicle's age, and their more complex estimation procedure may reduce its effect on their estimates. Nevertheless, even their lower estimates exceed those from many other studies of the rebound effect, as Table 8-2 shows.
Hymel, Small, & Van Dender (2010)—and more recently, Hymel & Small (2015)—extended the simultaneous equations analysis of time-series and state-level variation in vehicle use originally reported in Small & Van Dender (2007) and to test the effect of including more recent data. As in the original 2007 study, both subsequent extensions found that the fuel economy rebound effect had declined over time in response to increasing personal income and urbanization but had risen during periods when fuel prices increased. Because they rely on the response of vehicle use to fuel cost per mile to estimate the rebound effect, however, none of these three studies is able to detect whether its apparent decline in response to rising income levels over time truly reflects its effect on drivers' responses to changing fuel economy—the rebound effect itself—or simply captures the effect of rising income on their sensitivity to fuel Start Printed Page 43104
These updated studies each revised Small and Van Dender's original estimate of an 11% rebound effect for 1997-2011 upward when they included more recent experience: To 13% for the period 2001-04, and subsequently to 18% for 2000-2009.
In their 2015 update, Hymel and Small hypothesized that the recent increase in the rebound effect could be traced to a combination of expanded media coverage of changing fuel prices, increased price volatility, and an asymmetric response by drivers to variation in fuel costs. The authors estimated that about half of the apparent increase in the rebound effect for recent years could be attributed to greater volatility in fuel prices and more media coverage of sudden price changes. Their results also suggest that households curtail their vehicle use within the first year following an increase in fuel prices and driving costs, while the increase in driving that occurs in response to declining fuel prices—and by implication, to improvements in fuel economy—occurs more slowly.
West et al. (2015) attempted to infer the fuel economy rebound effect using data from Texas households who replaced their vehicles with more fuel-efficient models under the 2009 “Cash for Clunkers” program, which offered sizeable financial incentives to do so. Under the program, households that retired older vehicles with fuel economy levels of 18 miles per gallon (MPG) or less were eligible for cash incentives ranging from $3,500-4,000, while those retiring vehicles with higher fuel economy were ineligible for such rebates. The authors examined the fuel economy, other features, and subsequent use of new vehicles households in Texas purchased to replace older models that narrowly qualified for the program's financial incentives because their fuel economy was only slightly below the 18 MPG threshold. They then compared these to the fuel economy, features, and use of new vehicles that demographically comparable households bought to replace older models, but whose slightly higher fuel economy—19 MPG or above—made them barely ineligible for the program.
The authors reported that the higher fuel economy of new models that eligible households purchased in response to the generous financial incentives offered under the “Cash for Clunkers” did not prompt their buyers to use them more than the older, low-MPG vehicles they replaced. They attributed this apparent absence of a fuel economy rebound effect—which they described as an “attribute-adjusted” measure of its magnitude—to the fact that eligible households chose to buy less expensive, smaller, and lower-performing models to replace those they retired. Because these replacements offered lower-quality transportation service, their buyers did not drive them more than the vehicles they replaced.
The applicability of this result to the proposal's analysis is doubtful because previous regulatory analyses assumed that manufacturers could achieve required improvements in fuel economy without compromising the performance, carrying and towing capacity, comfort, or safety of cars and light trucks from recent model years.
While this may be technically true, doing so would come at a combined greater cost. If this argument is correct, then amending future standards at a reduced stringency from their previously-adopted levels would lead to less driving attributable to rebound, and should therefore not lead to artificial constraints in new vehicles' other features that offset the reduction in their use stemming from lower fuel economy.
Most recently, De Borger et al. (2017) analyze the response of vehicle use to changes in fuel economy among a sample of nearly 350,000 Danish households owning the same model vehicle, of which almost one-third replaced it with a different model sometime during the period from 2001 to 2011. By comparing the changes in households' driving from the early years of this period to its later years among those who replaced their vehicles during the intervening period to the changes in driving among households who kept their original vehicles, the authors attempted to isolate the effect of changes in fuel economy on vehicle use from those of other factors. They measured the rebound effect as the change in households' vehicle use in response to differences in the fuel economy between vehicles they had owned previously and the new models they purchased to replace them, over and above any change in vehicle use among households who did not buy new cars (and thus saw no change in fuel economy).
These authors' data enabled them to control for the effects of changes over time in household characteristics and vehicle features other than fuel economy that were likely to have contributed to observed changes in vehicle use. They also employed complex statistical methods to account for the fact that some households replacing their vehicles may have done so in anticipation of changes in their driving demands (rather than the reverse), as well as for the possibility that some households who replaced their cars may have done so because their driving behavior was more sensitive to fuel prices than other households. Their estimates ranged from 8-10%, varying only minimally among alternative model specifications and statistical estimation procedures or in response to whether their sample was restricted to households that replaced their vehicles or also included households that kept their original vehicles throughout the period.
Finally, De Borger et al. found no evidence that the rebound effect is smaller among lower-income households than among their higher-income counterparts.
(c) What value have the agencies assumed in this rule?
On the basis of all of the evidence summarized here, a fuel economy rebound effect of 20% has been chosen to analyze the effects of the proposed action. This is a departure from the 10% value used in regulatory analyses for MYs 2012-2016 and previous analyses for MYs 2017-2025 CAFE and GHG standards and represents a return to the value employed in the analyses for MYs 2005-2011 CAFE standards. There are several reasons the estimate of the fuel economy rebound effect for this analysis has been increased.
First, the 10% value is inconsistent with nearly all research on the magnitude of the rebound effect, as Table-II-43 and Table-II-44 indicate. Instead, it is based almost exclusively Start Printed Page 43105on the finding of the 2007 study by Small and Van Dender that the rebound effect had been declining over time in response to drivers' rising incomes and on extending that decline through future years using an assumption of steady income growth. As indicated above, however, subsequent extensions of Small and Van Dender's original research have produced larger estimates of the rebound effect for recent years: While their original study estimated the rebound effect at 11% for 1997-2001, the 2010 update by Hymel, Small, and Van Dender reported a value of 13% for 2004, and Hymel and Small's 2015 update estimated the rebound effect at 18% for 2003-09. Further, the issues with state-level measures of vehicle use, fuel consumption, and fuel economy identified previously raise some doubt about the reliability of these studies' estimates of the rebound effect.
At the same time, the continued increases in income that were anticipated to produce a continued decline in the rebound effect have not materialized. The income measure (real personal income per Capita) used in these analyses has grown only approximately one percent annually over the past two decades and is projected to grow at approximately 1.5% for the next 30 years, in contrast to the two to three percent annual growth assumed by the agencies when developing earlier forecasts of the future rebound effect. Further, another recent study by DeBorger et al. (2016) analyzed the separate effects of variation in household income on the sensitivity of their vehicle use to fuel prices and the fuel economy of vehicles they own. These authors' results indicate that the decline in the fuel economy rebound effect with income reported in Small & Van Dender (2007) and subsequent research results entirely from a reduction in drivers' sensitivity to fuel prices as their incomes rise rather than from any effect of rising income on the sensitivity of vehicle use to fuel economy itself. This latter measure, which DeBorger et al. find has not changed significantly as incomes have risen over time, is the correct measure of the fuel economy rebound effect, so their analysis calls into question its assumed sensitivity to income.
Some studies of households' use of individual vehicles also find that the fuel economy rebound effect increases with the number of vehicles they own. Because vehicle ownership is strongly associated with household income, this common finding suggests that the overall value of the rebound effect is unlikely to decline with rising incomes as the agencies had previously assumed. In addition, buyers of new cars and light trucks belong disproportionately to higher-income households that already own multiple vehicles, which further suggests that the higher values of the rebound effect estimated by many studies for such households are more relevant for analyzing use of newly-purchased cars and light trucks.
Finally, research on the rebound effect conducted since the agencies' original 2008 review of evidence almost universally reports estimates in the 10-40% (and larger) range, as Table-II-43 shows. Thus, the 20% rebound effect used in this analysis more accurately represents the findings from both the studies considered in 2008 review and the more recent analyses.
(1) What are the implications of the rebound effect for VMT?
The assumed rebound effect not only influences the use of new vehicles in today's analysis but also affects the response of the initial registered vehicle population to changes in fuel price throughout their remaining useful lives. The fuel prices used in this analysis are lower than the projections used to inform the 2012 Final Rule but generally increase from today's level over time. As they do so, the rebound effect acts as a price elasticity of demand for travel—as the cost-per-mile of travel increases, owners of all vehicles in the registered population respond by driving less. In particular, they drive 20% less than the difference between the cost-per-mile of travel when they were observed in calendar year 2016 and the relevant cost-per-mile at any future age. For the new vehicles subject to this proposal (and explicitly simulated by the CAFE model), fuel economies increase relative to MY 2016 levels, and generally improve enough to offset the effect of rising fuel prices—at least during the years covered by the proposal. For those vehicles, the difference between the initial cost-per-mile of travel and future travel costs is negative. As the vehicles become less expensive to operate, they are driven more (20% more than the difference between initial and present travel costs, precisely). Of course, each of the regulatory alternatives considered in the analysis would result in lower fuel economy levels for vehicles produced in model year 2020 and later than if the baseline standards remained in effect, so total VMT is lower under these alternatives than under the baseline.
(2) What is the mobility benefit that accrues to vehicle owners?
The increase in travel associated with the rebound effect produces additional benefits to vehicle owners, which reflect the value to drivers and other vehicle occupants of the added (or more desirable) social and economic opportunities that become accessible with additional travel. As evidenced by the fact that they elect to make more frequent or longer trips when the cost of driving declines, the benefits from this added travel exceed drivers' added outlays for the fuel it consumes (measured at the improved level of fuel economy resulting from stricter CAFE standards). The amount by which the benefits from this increased driving travel exceed its increased fuel costs measures the net benefits they receive from the additional travel, usually are referred to as increased consumer surplus.
NHTSA's analysis estimates the economic value of the decreased consumer surplus provided by reduced driving using the conventional approximation, which is one half of the product of the increase in vehicle operating costs per vehicle-mile and the resulting decrease in the annual number of miles driven. Because it depends on the extent of the change in fuel economy, the value of economic impacts from decreased vehicle use changes by model year and varies among alternative CAFE standards.
(d) Societal Externalities Associated With CAFE Alternatives
(1) Energy Security Externalities
Higher U.S. fuel consumption will produce a corresponding increase in the nation's demand for crude petroleum, which is traded actively in a worldwide market. The U.S. accounts for a large enough share of global oil consumption that the resulting boost in global demand will raise its worldwide price. The increase in global petroleum prices that results from higher U.S. demand causes a transfer of revenue to oil producers worldwide from not only buyers of new cars and light trucks, but also other consumers of petroleum products in the U.S. and throughout the world, all of whom pay the higher price that results.
Although these effects will be tempered by growing U.S. oil production, uncertainty in the long-term import-export balance makes it difficult to precisely project how these effects might change in response to that increased production. Growing U.S. petroleum consumption will also increase potential costs to all U.S. petroleum users from possible interruptions in the global supply of petroleum or rapid increases in global oil prices, not all of which are borne by Start Printed Page 43106the households or businesses who increase their petroleum consumption (that is, they are partly “external” to petroleum users). If U.S. demand for imported petroleum increases, it is also possible that increased military spending to secure larger oil supplies from unstable regions of the globe will be necessary.
These three effects are often referred to collectively as “energy security externalities” resulting from U.S. petroleum consumption, and increases in their magnitude are sometimes cited as potential social costs of increased U.S. demand for oil. To the extent that they represent real economic costs that would rise incrementally with increases in U.S. petroleum consumption of the magnitude likely to result from less stringent CAFE and GHG standards, these effects represent potential additional costs of this proposed action. Chapter 7 of the Regulatory Impact Analysis for this proposed action defines each of these energy security externalities in detail, assesses whether its magnitude is likely to change as a consequence of this action, and identifies whether that change represents a real economic cost or benefit of this action.
(2) Environmental Externalities
The change in criteria pollutant emissions that result from changes in vehicle usage and fuel consumption is estimated as part of this analysis. Criteria air pollutants include carbon monoxide (CO), hydrocarbon compounds (usually referred to as “volatile organic compounds,” or VOC), nitrogen oxides (NOX), fine particulate matter (PM2.5), and sulfur oxides (SOX). These pollutants are emitted during vehicle storage and use, as well as throughout the fuel production and distribution system. While increases in domestic fuel refining, storage, and distribution that result from higher fuel consumption will increase emissions of these pollutants, reduced vehicle use associated with the fuel economy rebound effect will decrease their emissions. The net effect of less stringent CAFE standards on total emissions of each criteria pollutant depends on the relative magnitudes of increases in its emissions during fuel refining and distribution, and decreases in its emissions resulting from additional vehicle use. Because the relationship between emissions in fuel refining and vehicle use is different for each criteria pollutant, the net effect of increased fuel consumption from the proposed standards on total emissions of each pollutant is likely to differ.
The social damage costs associated with changes in the emissions of criteria pollutants and CO2 was calculated, attributing benefits and costs to the regulatory alternatives considered based on the sign of the change in each pollutant. In previous rulemakings, the agencies have considered the social cost of CO2 emissions from a global perspective, accumulating social costs for CO2 emissions based on adverse outcomes attributable to climate change in any country. In this analysis, however, the costs of CO2 emissions and resulting climate damages from both domestic and global perspectives were considered. Chapter 9 of the Regulatory Impact Analysis provides a detailed discussion of how the agencies estimate changes in emissions of criteria air pollutants and CO2 and reports the values the agencies use to estimate benefits or costs associated with those changes in emissions.
(3) Traffic Externalities (Congestion, Noise)
Increased vehicle use associated with the rebound effect also contributes to increased traffic congestion and highway noise. To estimate the economic costs associated with these consequences of added driving, the estimates of per-mile congestion and noise costs caused by increased use of automobiles and light trucks developed previously by the Federal Highway Administration (FHWA) were applied. These values are intended to measure the increased costs resulting from added congestion and the delays it causes to other drivers and passengers and noise levels contributed by automobiles and light trucks. NHTSA previously employed these estimates in its analysis accompanying the MY 2011 final CAFE rule as well as in its analysis of the effects of higher CAFE standards for MY 2012-16 and MY 2017-2021. After reviewing the procedures used by FHWA to develop them and considering other available estimates of these values and recognizing that no commenters have addressed these costs directly in their comments on previous rules, the values continue to be appropriate for use in this proposal. For this analysis, FHWA's estimates of per-mile costs are multiplied by the annual increases in automobile and light truck use from the rebound effect to yield the estimated increases in total congestion and noise externality costs during each year over the lifetimes of the cars and light trucks in the on-road fleet. Due to the fact that this proposal represents a decrease in stringency, the fuel economy rebound effect results in fewer miles driven under the action alternatives relative to the baseline, which generates savings in congestion and road noise relative to the baseline.
F. Impact of CAFE Standards on Vehicle Safety
In past CAFE rulemakings, NHTSA has examined the effect of CAFE standards on vehicle mass and the subsequent effect mass changes will have on vehicle safety. While setting standards based on vehicle footprint helps reduce potential safety impacts associated with CAFE standards as compared to setting standards based on some other vehicle attribute, footprint-based standards cannot entirely eliminate those impacts. Although prior analyses noted that there could also be impacts because of other factors besides mass changes, those impacts were not estimated quantitatively.
In this current analysis, the safety analysis has been expanded to include a broader and more comprehensive measure of safety impacts, as discussed below. A number of factors can influence motor vehicle fatalities directly by influencing vehicle design or indirectly by influencing consumer behavior. These factors include:
(1) Changes, which affect the crashworthiness of vehicles impact other vehicles or roadside objects, in vehicle mass made to reduce fuel consumption. NHTSA's statistical analysis of historical crash data to understand effects of vehicle mass and size on safety indicates reducing mass in light trucks generally improves safety, while reducing mass in passenger cars generally reduces safety. NHTSA's crash simulation modeling of vehicle design concepts for reducing mass revealed similar trends.
(2) The delay in the pace of consumer acquisition of newer safer vehicles that results from higher vehicle prices associated with technologies needed to meet higher CAFE standards. Because of a combination of safety regulations and voluntary safety improvements, passenger vehicles have become safer over time. Compared to prior decades, fatality rates have declined significantly Start Printed Page 43107because of technological safety improvements as well as behavioral shifts such as increased seat belt use. The results of this analysis project that vehicle prices will be nearly $1,900 higher under the augural CAFE standards compared to the preferred alternative that would hold stringency at MY 2020 levels in MYs 2021-2026. This will induce some consumers to delay or forgo the purchase of newer safer vehicles and slow the transition of the on-road fleet to one with the improved safety available in newer vehicles. This same factor can also shift the mix of passenger cars and light trucks.
(3) Increased driving because of better fuel economy. The “rebound effect” predicts consumers will drive more when the cost of driving declines. More stringent CAFE standards reduce vehicle operating costs, and in response, some consumers may choose to drive more. Driving more increases exposure to risks associated with on-road transportation, and this added exposure translates into higher fatalities.
Although all three factors influence predicted fatality levels that may occur, only two of them, the changes in vehicle mass and the changes in the acquisition of safer vehicles—are actually imposed on consumers by CAFE standards. The safety of vehicles has improved over time and is expected to continue improving in the future commensurate with the pace of safety technology innovation and implementation and motor vehicle safety regulation. Safety improvements will likely continue regardless of changes to CAFE standards. However, its pace may be modified if manufacturers choose to delay or forgo investments in safety technology because of the demand CAFE standards impose on research, development, and manufacturing budgets. Increased driving associated with rebound is a consumer choice. Improved CAFE will reduce driving costs, but nothing in the higher CAFE standards compels consumers to drive additional miles. If consumers choose to do so, they are making a decision that the utility of more driving exceeds the marginal operating costs as well as the added crash risk it entails. Thus, while the predicted fatality impacts with all three factors embedded into the model are measured, the fatalities associated with consumer choice decisions are accounted for separately from those resulting from technologies implemented in response to CAFE regulations or economic limitations resulting from CAFE regulation. Only those safety impacts associated with mass reduction and those resulting from higher vehicle prices are directly attributed to CAFE standards.
This is reflected monetarily by valuing extra rebound miles at the full value of their added driving cost plus the added safety risk consumers experience, which completely offsets the societal impact of any added fatalities from this voluntary consumer choice.
The safety component of CAFE analysis has evolved over time. In the 2012 final rule, the analysis accounted for the change in projected fatalities attributable to mass reduction of new vehicles. The model assumed that manufacturers would choose mass reduction as a compliance method across vehicle classes such that the net effect of mass reduction on fatalities was zero. However, in the 2016 draft Technical Assessment Report, DOT made two consequential changes to the analysis of fatalities associated with the CAFE standards. In particular, first, the modelling assumed that mass reduction technology was available to all vehicles, regardless of net safety impact, and second, it accounted for the incremental safety costs associated with additional miles traveled due to the rebound effect. The current analysis extends the analysis to report incremental fatality impacts associated with additional miles traveled due to the rebound effect, and identifies the increase in fatalities associated with additional driving separately from changes in fatalities attributable other sources.
The current analysis adds another element: The effect that higher new vehicle prices have on new vehicle sales and on used vehicle scrappage, which influences total expected fatalities because older vehicle vintages are associated with higher rates of involvement in fatal crashes than newer vehicles. Finally, a dynamic fleet share model also predicts the effects of changes in the standards on the share of light trucks and passenger cars in future model year light-duty vehicle fleets. Vehicles of different body styles have different rates of involvement in fatal crashes, so that changing the share of each in the projected future fleet has safety impacts; the implied safety effects are captured in the current modelling. The agencies seek comment on changes to the safety analysis made in this proposal, they seek particular comment on the following changes:
(1) The sales scrappage models as independent models: Two separate models capture the effects of new vehicle prices on new vehicle demand and used vehicle retirement rates—the sales model and the scrappage model, respectively. We seek public comment on the methods used for each of these models, in particular we seek comment on:
- The assumptions and variables included in the independent models
- The techniques and data used to estimate the independent models
- The structure and implementation of the independent models
(2) Integration of the sales and scrappage models: The new sales and scrappage models use many of the same predictors, but are not directly integrated. We seek public comment on, and data supporting whether integrating the two models is appropriate.
(3) Integration of the scrappage rates and mileage accumulation: The current model assumes that annual mileage accumulation and scrappage rates are independent of one another. We seek public comment on the appropriateness of this assumption, and data that would support developing an interaction between scrappage rates and mileage accumulation, or testing whether such an interaction is important to include.
(4) Increased risk of older vehicles: The observed increase in crash and injury risk associated with older vehicles is likely due to a combination of vehicle factors and driver factors. For example, older vehicles are less crashworthy because in general they're equipped with fewer or less modern safety features, and drivers of older cars are on average younger and may be less skilled drivers or less risk-averse than drivers of new vehicles. We fit a model which includes both an age and vintage affect, but assume that the age effect is entirely a result of changes in average driver demographics, and not impacted by changes in CAFE or GHG standards. We seek comment on this approach for attributing increased older vehicle risk. Is the analysis likely to overestimate or underestimate the safety benefits under the proposed alternative?
(5) Changes in the mix of light trucks and passenger cars: The dynamic fleet share model predicts changes in the future share of light truck and passenger car vehicles. Changes in the mix of vehicles may result in Start Printed Page 43108increased or decreased fatalities. Does the dynamic fleet share model reasonably capture consumers' decisions about how they substitute between different types and sizes of vehicles depending on changes in fuel economy, relative and absolute prices, and other vehicle attributes? We seek comment on whether our safety analysis provides a reasonable estimate of the effects of changes in fleet mix on future fatalities.
1. Impact of Weight Reduction on Safety
The primary goals of CAFE and CO2 standards are reducing fuel consumption and CO2 emissions from the on-road light-duty vehicle fleet; in addition to these intended effects, the potential of the standards to affect vehicle safety is also considered.
As a safety agency, NHTSA has long considered the potential for adverse safety consequences when establishing CAFE standards, and under the CAA, EPA considers factors related to public health and human welfare, including safety, in regulating emissions of air pollutants from mobile sources.
Safety trade-offs associated with fuel economy increases have occurred in the past, particularly before NHTSA CAFE standards were attribute-based; past safety trade-offs may have occurred because manufacturers chose at the time, in response to CAFE standards, to build smaller and lighter vehicles. Although the agency now uses attribute-based standards, in part to protect against excessive vehicle downsizing, the agency must be mindful of the possibility of related safety trade-offs in the future. In cases where fuel economy improvements were achieved through reductions in vehicle size and mass, the smaller, lighter vehicles did not fare as well in crashes as larger, heavier vehicles, on average.
Historically, as shown in FARS data analyzed by NHTSA, the safest cars generally have been heavy and large, while cars with the highest fatal-crash rates have been light and small. The question, then, is whether past is necessarily a prologue when it comes to potential changes in vehicle size (both footprint and “overhang”) and mass in response to the more stringent future CAFE and GHG standards.
Manufacturers stated they will reduce vehicle mass as one of the cost-effective means of increasing fuel economy and reducing CO2 to meet standards, and this approach is incorporated this expectation into the modeling analysis supporting the standards. Because the analysis discerns a historical relationship between vehicle mass, size, and safety, it is reasonable to assume these relationships will continue in the future.
(a) Historical Analyses of Vehicle Mass and Safety
Researchers have been using statistical analysis to examine the relationship of vehicle mass and safety in historical crash data for many years and continue to refine their techniques. In the MY 2012-2016 final rule, the agencies stated we would conduct further study and research into the interaction of mass, size, and safety to assist future rulemakings and start to work collaboratively by developing an interagency working group between NHTSA, EPA, DOE, and CARB to evaluate all aspects of mass, size, and safety. The team would seek to coordinate government-supported studies and independent research to the greatest extent possible to ensure the work is complementary to previous and ongoing research and to guide further research in this area.
The agencies also identified three specific areas to direct research in preparation for future CAFE/CO2 rulemaking regarding statistical analysis of historical data. First, NHTSA would contract with an independent institution to review statistical methods NHTSA and DRI used to analyze historical data related to mass, size, and safety, and to provide recommendations on whether existing or other methods should be used for future statistical analysis of historical data. This study would include a consideration of potential near multicollinearity in the historical data and how best to address it in a regression analysis. The 2010 NHTSA report (hereinafter 2010 Kahane report) was also peer reviewed by two other experts in the safety field—Farmer (Insurance Institute for Highway Safety) and Lie (Swedish Transport Administration).
Second, NHTSA and EPA, in consultation with DOE, would update the MY 1991-1999 database where safety analyses in the NPRM and final rule are based with newer vehicle data and create a common database that could be made publicly available to address concerns that differences in data were leading to different results in statistical analyses by different researchers.
And third, to assess if the design of recent model year vehicles incorporating various mass reduction methods affect relationships among vehicle mass, size, and safety, the agencies sought to identify vehicles using material substitution and smart design and to assess if there is sufficient crash data involving those vehicles for statistical analysis. If sufficient data exists, statistical analysis would be conducted to compare the relationship among mass, size, and safety of these smart design vehicles to vehicles of similar size and mass with more traditional designs.
By the time of the MY 2017-2025 final rule, significant progress was made on these tasks: The independent review of recent and updated statistical analyses of the relationship between vehicle mass, size, and crash fatality rates had been completed. NHTSA contracted with the University of Michigan Transportation Research Institute (UMTRI) to conduct this review, and the UMTRI team led by Green evaluated more than 20 papers, including studies done by NHTSA's Kahane, Wenzel of the U.S. Department of Energy's Lawrence Berkeley National Laboratory, Dynamic Research, Inc., and others. UMTRI's basic findings are discussed in Chapter 11 of the PRIA accompanying this NPRM.
Some commenters in recent CAFE rulemakings, including some vehicle manufacturers, suggested designs and materials of more recent model year vehicles may have weakened the historical statistical relationships between mass, size, and safety. It was agreed that the statistical analysis would be improved by using an updated database reflecting more recent safety technologies, vehicle designs and materials, and reflecting changes in the vehicle fleet. An updated database was created and employed for assessing safety effects for that final rule. The agencies also believed, as UMTRI found, different statistical analyses may have produced different results because they used slightly different datasets for their analyses.
To try to mitigate this issue and to support the current rulemaking, NHTSA created a common, updated database for statistical analysis consisting of crash data of model years 2000-2007 vehicles in calendar years 2002-2008, as Start Printed Page 43109compared to the database used in prior NHTSA analyses, which was based on model years 1991-1999 vehicles in calendar years 1995-2000. The new database was the most up-to-date possible, given the processing lead time for crash data and the need for enough crash cases to permit statistically meaningful analyses. NHTSA made the preliminary version of the new database, which was the basis for NHTSA's 2011 preliminary report (hereinafter 2011 Kahane report), available to the public in May 2011, and an updated version in April 2012 (used in NHTSA's 2012 final report, hereinafter 2012 Kahane report),
enabling other researchers to analyze the same data and hopefully minimize discrepancies in results because of inconsistencies across databases.
Since the publication of the MYs 2017-2025 final rule, NHTSA has sponsored, and is sponsoring, new studies and research to inform the current CAFE and CO2 rulemaking. In addition, the National Academy of Sciences published a new report in this area.
Throughout the rulemaking process, NHTSA's goal is to publish as much of our research as possible. In establishing standards, all available data, studies, and information objectively without regard to whether they were sponsored by the agencies, will be considered.
Undertaking these tasks has helped come closer to resolving ongoing debates in statistical analysis research of historical crash data. It is intended that these conclusions will be applied going forward in future rulemakings, and it is believed the research will assist the public discussion of the issues. Specific historical analyses (in addition to NHTSA's own analysis) on vehicle mass and safety used to support this rulemaking include:
- The 2011 and 2013 NHTSA Workshops on Vehicle Mass, Size, and Safety;
- the University of Michigan Transportation Research Institute (UMTRI) independent review of a set of statistical relationships between vehicle curb weight, footprint variables (track width, wheelbase), and fatality rates from vehicle crashes;
- the 2012 Lawrence Berkeley National Laboratory (LBNL) Phase 1 and Phase 2 reports on the sensitivity of NHTSA's baseline results and casualty risk per VMT;
- the 2012 DRI reports on, among other things, the effects of mass reduction on crash frequency and fatality risk per crash;
- LBNL's subsequent review of DRI's study;
- the 2015 National Academy of Sciences Report; and
- the 2017 NBER working paper analyzing the relationships among traffic fatalities, CAFE standards, and distributions of MY 1989-2005 light-duty vehicle curb weights.
A detailed discussion of each analysis is discussed in Chapter 11 of the PRIA accompanying this proposed rule.
(b) Recent NHTSA Analysis Supporting CAFE Rulemaking
As mentioned previously, NHTSA and EPA's 2012 joint final rule for MYs 2017 and beyond set “footprint-based” standards, with footprint being defined as roughly equal to the wheelbase multiplied by the average of the front and rear track widths. Basing standards on vehicle footprint ideally helps to discourage vehicle manufacturers from downsizing their vehicles; the agencies set higher (more stringent) mile per gallon (mpg) targets for smaller-footprint vehicles but would not similarly discourage mass reduction that maintains footprint while potentially improving fuel economy. Several technologies, such as substitution of light, high-strength materials for conventional materials during vehicle redesigns, have the potential to reduce weight and conserve fuel while maintaining a vehicle's footprint and maintaining or possibly improving the vehicle's structural strength and handling.
In considering what technologies are available for improving fuel economy, including mass reduction, an important corollary issue for NHTSA to consider is the potential effect those technologies may have on safety. NHTSA has thus far specifically considered the likely effect of mass reduction that maintains footprint on fatal crashes. The relationship between a vehicle's mass, size, and fatality risk is complex, and it varies in different types of crashes. As mentioned above, NHTSA, along with others, has been examining this relationship for more than a decade.
The safety chapter of NHTSA's April 2012 final regulatory impact analysis (FRIA) of CAFE standards for MY 2017-2021 passenger cars and light trucks included a statistical analysis of relationships between fatality risk, mass, and footprint in MY 2000-2007 passenger cars and LTVs (light trucks and vans), based on calendar year (CY) 2002-2008 crash and vehicle-registration data; 
this analysis was also detailed in the 2012 Kahane report.
The principal findings and conclusions of the 2012 Kahane report were mass reduction in the lighter cars, even while holding footprint constant, would significantly increase fatality risk, whereas mass reduction in the heavier LTVs would reduce societal fatality risk by reducing the fatality risk of occupants of lighter vehicles colliding with those heavier LTVs. NHTSA concluded, as a result, any reasonable combination of mass reductions that held footprint constant in MY 2017-2021 vehicles—concentrated, at least to some extent, in the heavier LTVs and limited in the lighter cars—would likely be approximately safety-neutral; it would not significantly increase fatalities and might well decrease them.
NHTSA released a preliminary report (2016 Puckett and Kindelberger report) on the relationship between fatality risk, mass, and footprint in June 2016 in advance of the Draft TAR. The preliminary report covered the same scope as the 2012 Kahane report, offering a detailed description of the databases, modeling approach, and analytical results on relationships among vehicle size, mass, and fatalities that informed the Draft TAR. Results in the Draft TAR and the 2016 Puckett and Kindelberger report are consistent with results in the 2012 Kahane report; chiefly, societal effects of mass reduction are small, and mass reduction concentrated in larger vehicles is likely to have a beneficial effect on fatalities, while mass reduction concentrated in smaller vehicles is likely to have a detrimental effect on fatalities.
For the 2016 Puckett and Kindelberger report and Draft TAR, NHTSA, working closely with EPA and the DOE, performed an updated statistical analysis of relationships between fatality rates, mass and footprint, updating the crash and exposure databases to the latest available model years. The agencies analyzed updated databases that included MY 2003-2010 vehicles in CY 2005-2011 crashes. For this proposed Start Printed Page 43110rule, databases are the most up-to-date possible (MY 2004-2011 vehicles in CY 2006-2012), given the processing time for crash data and the need for enough crash cases to permit statistically meaningful analyses. As in previous analyses, NHTSA has made the new databases available to the public on its website, enabling other researchers to analyze the same data and hopefully minimizing discrepancies in results that would have been because of inconsistencies across databases.
(c) Updated Analysis for This Rulemaking
The basic analytical method used to analyze the impacts of weight reduction on safety in this proposed rule is the same as in NHTSA's 2012 Kahane report, 2016 Puckett and Kindelberger report, and the Draft TAR: The agency analyzed cross sections of the societal fatality rate per billion vehicle miles of travel (VMT) by mass and footprint, while controlling for driver age, gender, and other factors, in separate logistic regressions by vehicle class and crash type. “Societal” fatality rates include fatalities to occupants of all the vehicles involved in the collisions, plus any pedestrians.
The temporal range of the data is now MY 2004-2011 vehicles in CY 2006-2012, updated from previous databases of MY 2000-2007 vehicles in CY 2002-2008 (2012 Kahane Report) and MY 2003-2010 vehicles in CY 2005-2011 (2016 Puckett and Kindelberger report and Draft TAR). NHTSA purchased a file of odometer readings by make, model, and model year from Polk that helped inform the agency's improved VMT estimates. As in the 2012 Kahane report, 2016 Puckett and Kindelberger report, and the Draft TAR, the vehicles are grouped into three classes: Passenger cars (including both two-door and four-door cars); CUVs and minivans; and truck-based LTVs.
There are nine types of crashes specified in the analysis. Single-vehicle crashes include first-event rollovers, collisions with fixed objects, and collisions with pedestrians, bicycles and motorcycles. Two-vehicle crashes include collisions with: heavy-duty vehicles; car, CUV, or minivan < 3,187 pounds (the median curb weight of other, non-case, cars, CUVs and minivans in fatal crashes in the database); car, CUV, or minivan ≥ 3,187 pounds; truck-based LTV < 4,360 pounds (the median curb weight of other truck-based LTVs in fatal crashes in the database); and truck-based LTV ≥ 4,360 pounds. An additional crash type includes all other fatal crash types (e.g., collisions involving more than two vehicles, animals, or trains). Splitting the “other” vehicles into a lighter and a heavier group permits more accurate analyses of the mass effect in collisions of two light vehicles. Grouping partner-vehicle CUVs and minivans with cars rather than LTVs is more appropriate because their front-end profile and rigidity more closely resembles a car than a typical truck-based LTV.
The curb weight of passenger cars is formulated, as in the 2012 Kahane report, 2016 Puckett and Kindelberger report, and Draft TAR, as a two-piece linear variable to estimate one effect of mass reduction in the lighter cars and another effect in the heavier cars. The boundary between “lighter” and “heavier” cars is 3,201 pounds (which is the median mass of MY 2004-2011 cars in fatal crashes in CY 2006-2012, up from 3,106 for MY 2000-2007 cars in CY 2002-2008 in the 2012 NHTSA safety database, and up from 3,197 for MY 2003-2010 cars in CY 2005-2011 in the 2016 NHTSA safety database).
Likewise, for truck-based LTVs, curb weight is a two-piece linear variable with the boundary at 5,014 pounds (again, the MY 2004-2011 median, higher than the median of 4,594 for MY 2000-2007 LTVs in CY 2002-2008 and the median of 4,947 for MY 2003-2010 LTVs in CY 2005-2011). Curb weight is formulated as a simple linear variable for CUVs and minivans. Historically, CUVs and minivans have accounted for a relatively small share of new-vehicle sales over the range of the data, resulting in less crash data available than for cars or truck-based LTVs.
For a given vehicle class and weight range (if applicable), regression coefficients for mass (while holding footprint constant) in the nine types of crashes are averaged, weighted by the number of baseline fatalities that would have occurred for the subgroup MY 2008-2011 vehicles in CY 2008-2012 if these vehicles had all been equipped with electronic stability control (ESC). The adjustment for ESC, a feature of the analysis added in 2012, takes into account results will be used to analyze effects of mass reduction in future vehicles, which will all be ESC-equipped, as required by NHTSA's regulations.
Techniques developed in the 2011 (preliminary) and 2012 (final) Kahane reports have been retained to test statistical significance and to estimate 95 percent confidence bounds (sampling error) for mass effects and to estimate the combined annual effect of removing 100 pounds of mass from every vehicle (or of removing different amounts of mass from the various classes of vehicles), while holding footprint constant.
NHTSA considered the near multicollinearity of mass and footprint to be a major issue in the 2010 Kahane report 
and voiced concern about inaccurately estimated regression coefficients.
High correlations between mass and footprint and variance inflation factors (VIF) have not changed from MY 1991-1999 to MY 2004-2011; large vehicles continued to be, on the average, heavier than small vehicles to the same extent as in the previous decade.
Nevertheless, multicollinearity appears to have become less of a problem in the 2012 Kahane, 2016 Puckett and Kindelberger/Draft TAR, and current NHTSA analyses. Ultimately, only three of the 27 core models of fatality risk by vehicle type in the current analysis indicate the potential presence of effects of multicollinearity, with estimated effects of mass and footprint reduction greater than two percent per 100-pound mass reduction and one-square-foot footprint reduction, respectively; these three models include passenger cars and CUVs in first-event rollovers, and CUVs in collisions with LTVs greater than 4,360 pounds. This result is consistent with the 2016 Puckett and Kindelberger report, which also found only three cases out of 27 models with estimated effects of mass and footprint reduction greater than two percent per 100-pound mass reduction and one-square-foot footprint reduction.
Table II-45 presents the estimated percent increase in U.S. societal fatality risk per 10 billion VMT for each 100-Start Printed Page 43111pound reduction in vehicle mass, while holding footprint constant, for each of the five vehicle classes:
None of the estimated effects have 95-percent confidence bounds that exclude zero, and thus are not statistically significant at the 95-percent confidence level. Two estimated effects are statistically significant at the 85-percent level. Societal fatality risk is estimated to: (1) Increase by 1.2 percent if mass is reduced by 100 pounds in the lighter cars; and (2) decrease by 0.61 percent if mass is reduced by 100 pounds in the heavier truck-based LTVs. The estimated increases in societal fatality risk for mass reduction in the heavier cars and the lighter truck-based LTVs, and the estimated decrease in societal fatality risk for mass reduction in CUVs and minivans are not significant, even at the 85-percent confidence level.
Confidence bounds estimate only the sampling error internal to the data used in the specific analysis that generated the point estimate. Point estimates are also sensitive to the modification of components of the analysis, as discussed at the end of this section. However, this degree of uncertainty is methodological in nature rather than statistical.
It is useful to compare the new results in Table II-45 to results in the 2012 Kahane report (MY 2000-2007 vehicles in CY 2002-2008) and the 2016 Puckett and Kindelberger report and Draft TAR (MY 2003-2010 vehicles in CY 2005-2011), presented in Table II-46 below:
New results are directionally the same as in 2012; in the 2016 analysis, the estimate for lighter LTVs was of opposite sign (but small magnitude). Consistent with the 2012 Kahane and 2016 Puckett and Kindelberger reports, mass reductions in lighter cars are estimated to lead to increases in fatalities, and mass reductions in heavier LTVs are estimated to lead to decreases in fatalities. However, NHTSA does not consider this conclusion to be definitive because of the relatively wide confidence bounds of the estimates. The estimated mass effects are similar among analyses for both classes of passenger cars; for all reports, the estimate for lighter passenger cars is statistically significant at the 85-percent confidence level, while the estimate for heavier passenger cars is insignificant.
The estimated mass effect for heavier truck-based LTVs is stronger in this analysis and in the 2016 Puckett and Kindelberger report than in the 2012 Kahane report; both estimates are statistically significant at the 85-percent confidence level, unlike the corresponding insignificant estimate in the 2012 Kahane report. The estimated mass effect for lighter truck-based LTVs is insignificant and positive in this analysis and the 2012 Kahane report, while the corresponding estimate in the 2016 Puckett and Kindelberger report was insignificant and negative.
Vehicle mass continued an historical upward trend across the MYs in the newest databases. The average (VMT-weighted) masses of passenger cars and CUVs both increased by approximately three percent from MYs 2004 to 2011 (3,184 pounds to 3,289 pounds for passenger cars, and 3,821 pounds to 3,924 pounds for CUVs). Over the same period, the average mass of minivans increased by six percent (from 4,204 pounds to 4,462 pounds), and the average mass of LTVs increased by 10% (from 4,819 pounds to 5,311 pounds). Start Printed Page 43112Historical reasons for mass increases within vehicle classes include: Manufacturers discontinuing lighter models; manufacturers re-designing models to be heavier and larger; and shifting consumer preferences with respect to cabin size and overall vehicle size.
The principal difference between heavier vehicles, especially truck-based LTVs, and lighter vehicles, especially passenger cars, is mass reduction has a different effect in collisions with another car or LTV. When two vehicles of unequal mass collide, the change in velocity (delta V) is greater in the lighter vehicle. Through conservation of momentum, the degree to which the delta V in the lighter vehicle is greater than in the heavier vehicle is proportional to the ratio of mass in the heavier vehicle to mass in the lighter vehicle:
Because fatality risk is a positive function of delta V, the fatality risk in the lighter vehicle in two-vehicle collisions is also higher. Removing some mass from the heavy vehicle reduces delta V in the lighter vehicle, where fatality risk is higher, resulting in a large benefit, offset by a small penalty because delta V increases in the heavy vehicle where fatality risk is low—adding up to a net societal benefit. Removing some mass from the lighter vehicle results in a large penalty offset by a small benefit—adding up to net harm.
These considerations drive the overall result: Mass reduction is associated with an increase in fatality risk in lighter cars, a decrease in fatality risk in heavier LTVs, CUVs, and minivans, and has smaller effects in the intermediate groups. Mass reduction may also be harmful in a crash with a movable object such as a small tree, which may break if hit by a high mass vehicle resulting in a lower delta V than may occur if hit by a lower mass vehicle which does not break the tree and therefore has a higher delta V. However, in some types of crashes not involving collisions between cars and LTVs, especially first-event rollovers and impacts with fixed objects, mass reduction may not be harmful and may be beneficial. To the extent lighter vehicles may respond more quickly to braking and steering, or may be more stable because their center of gravity is lower, they may more successfully avoid crashes or reduce the severity of crashes.
Farmer, Green, and Lie, who reviewed the 2010 Kahane report, again peer-reviewed the 2011 Kahane report.
In preparing his 2012 report (along with the 2016 Puckett and Kindelberger report and Draft TAR), Kahane also took into account Wenzel's 
assessment of the preliminary report and its peer reviews, DRI's analyses published early in 2012, and public comments such as the International Council on Clean Transportation's comments submitted on NHTSA and EPA's 2010 notice of joint rulemaking.
These comments prompted supplementary analyses, especially sensitivity tests, discussed at the end of this section.
The regression results are best suited to predict the effect of a small change in mass, leaving all other factors, including footprint, the same. With each additional change from the current environment (e.g., the scale of mass change, presence and prevalence of safety features, demographic characteristics), the model may become less accurate. It is recognized that the light-duty vehicle fleet in the MY 2021-2026 timeframe will be different from the MY 20042011 fleet analyzed here.
Nevertheless, one consideration provides some basis for confidence in applying regression results to estimate effects of relatively large mass reductions or mass reductions over longer periods. This is NHTSA's sixth evaluation of effects of mass reduction and/or downsizing,
comprising Start Printed Page 43113databases ranging from MYs 1985 to 2011.
Results of the six studies are not identical, but they have been consistent to a point. During this time period, many makes and models have increased substantially in mass, sometimes as much as 30-40%.
If the statistical analysis has, over the past years, been able to accommodate mass increases of this magnitude, perhaps it will also succeed in modeling effects of mass reductions of approximately 10-20%, should they occur in the future.
(d) Calculation of MY 2021-2026 Safety Impact
Neither CAFE standards nor this analysis mandate mass reduction, or mandate mass reduction occur in any specific manner. However, mass reduction is one of the technology applications available to manufacturers, and thus a degree of mass reduction is allowed within the CAFE model to: (1) Determine capabilities of manufacturers; and (2) to predict cost and fuel consumption effects of improved CAFE standards.
The agency utilized the relationships between weight and safety from the new NHTSA analysis, expressed as percentage increases in fatalities per 100-pound weight reduction, and examined the weight impacts assumed in this CAFE analysis. The effects of mass reduction on safety were estimated relative to estimated baseline levels of safety across vehicle classes and model years. To identify baseline levels of safety, the agency examined effects of identifiable safety trends over lifetimes of vehicles produced in each model year. The projected effectiveness of existing and forthcoming safety technologies and expected on-road fleet penetration of safety technologies were incorporated into observed trends in fatality rates to estimate baseline fatality rates in future years across vehicle classes and model years.
The agency assumed safety trends will result in a reduction in the target population of fatalities from which the vehicle mass impacts are derived. Table II-47 through Table II-52 show results of NHTSA's vehicle mass-size-safety analysis over the cumulative lifetime of MY 1977-2029 vehicles, for both the CAFE and GHG programs, based on the MY 2016 baseline fleet, accounting for the projected safety baselines. The reported fatality impacts are undiscounted, but the monetized safety impacts are discounted at three-percent and seven-percent discount rates. The reported fatality impacts are estimated increases or decreases in fatalities over the lifetime of the model year fleet. A positive number means that fatalities are projected to increase; a negative number (in parentheses) means that fatalities are projected to decrease.
Results are driven extensively by the degree to which mass is reduced in relatively light passenger cars and in relatively heavy vehicles because their coefficients in the logistic regression analysis have the most significant values. We assume any impact on fatalities will occur over the lifetime of the vehicle, and the chance of a fatality occurring in any particular year is directly related to the weighted vehicle miles traveled in that year.
Start Printed Page 43114
Start Printed Page 43115
Start Printed Page 43116
Start Printed Page 43117
Start Printed Page 43118
Start Printed Page 43119
For all light-duty vehicles, mass changes are estimated to lead to a decrease in fatalities over the cumulative lifetime of MY 1977-2029 vehicles in all alternatives evaluated. The effects of mass changes on fatalities Start Printed Page 43120range from a combined decrease (relative to the augural standards, the baseline) of 12 fatalities for Alternative #7 to a combined decrease of 173 fatalities for Alternative #4. The difference in results by alternative depends upon how much weight reduction is used in that alternative and the types and sizes of vehicles to which the weight reduction applies. The decreases in fatalities are driven by impacts within passenger cars (decreases of between 17 and 281 fatalities) and are offset by impacts within light trucks (increases of between 6 and 120 fatalities).
Additionally, social effects of increasing fatalities can be monetized using NHTSA's estimated comprehensive cost per life of $9,900,000 in 2016 dollars. This consists of a value of a statistical life of $9.6 million in 2015 dollars plus external economic costs associated with fatalities such as medical care, insurance administration costs and legal costs, updated for inflation to 2016 dollars.
Typically, NHTSA would also estimate the effect on injuries and add that to social costs of fatalities, but in this case NHTSA does not have a model estimating the effect of vehicle mass on injuries. Blincoe et al. estimates that fatalities account for 39.5% of total comprehensive costs due to injury.
If vehicle mass impacts non-fatal injuries proportionally to its impact on fatalities, then total costs would be approximately 2.53 (1/0.395) times the value of fatalities alone or around $25.07 million per fatality. NHTSA has selected this value as representative of the relationship between fatality costs and injury costs because this approach is internally consistent among NHTSA studies.
Changes in vehicle mass are estimated to decrease social safety costs over the lifetime of the nine model years by between $176 million (for Alternative #7) and $2.7 billion (for Alternative #4) relative to the augural standards at a three-percent discount rate and by between $97 million and $1.6 billion at a seven-percent discount rate. The estimated decreases in social safety costs are driven by estimated decreases in costs associated with passenger cars, ranging from $264 million (for Alternative #7) to $4.4 billion (for Alternative #1) relative to the Augural standards at a three-percent discount rate and by between $146 million and $2.5 billion at a seven-percent discount rate. The estimated decreases in costs associated with passenger cars are offset by estimated increases in costs associated with light trucks, ranging from $88 million (for Alternative #7) to $2.0 billion (for Alternative #1) relative to the Augural standards at a three-percent discount rate and by between $49 million and $1.3 billion at a seven-percent discount rate.
Table II-53 through Table II-55 presents average annual estimated safety effects of vehicle mass changes, for CYs 2035-2045:
Start Printed Page 43121
Start Printed Page 43122
Start Printed Page 43123
Start Printed Page 43124
Start Printed Page 43125
Start Printed Page 43126
For all light-duty vehicles, mass changes are estimated to lead to an average annual decrease in fatalities in all alternatives evaluated for CYs 2035-2045. The effects of mass changes on fatalities range from a combined Start Printed Page 43127decrease (relative to the Augural standards) of 1 fatality per year for Alternative #7 to a combined increase of 22 fatalities per year for Alternative #1. The difference in the results by alternative depends upon how much weight reduction is used in that alternative and the types and sizes of vehicles to which the weight reduction applies. The decreases in fatalities are generally driven by impacts within passenger cars (decreases of between 1 and 33 fatalities per year relative to the Augural standards) and are generally offset by impacts within light trucks (increases of between 1 and 12 fatalities per year).
Changes in vehicle mass are estimated to decrease average annual social safety costs in CY 2035-2045 by between $2 million (for Alternative #7) and $271 million (for Alternative #1) relative to the Augural standards at a three-percent discount rate and by between $1 million and $111 million at a seven-percent discount rate. The estimated decreases in social safety costs are generally driven by estimated decreases in costs associated with passenger cars, decreasing between $13 million (for Alternative #7) and $424 million (for Alternative #1) relative to the Augural standards at a three-percent discount rate and decreasing between $5 million and $175 million at a seven-percent discount rate. The estimated decreases in costs associated with passenger cars are generally offset by estimated increases in costs associated with light trucks, decreasing between $11 million (for Alternative #7) and $153 million (for Alternative #1) relative to the Augural standards at a three-percent discount rate and decreasing between $5 million and $64 million at a seven-percent discount rate.
To help illuminate effects at the model year level, Table II-59 presents the lifetime fatality impacts associated with vehicle mass changes for passenger cars, light trucks, and all light-duty vehicles by model year under Alternative #1, relative to the Augural standards for the CAFE Program. Table II-59 presents an analogous table for the GHG Program.
Start Printed Page 43128
Under Alternative #1, passenger car fatalities associated with mass changes are estimated to decrease generally from MY 2017 (decrease of three fatalities) through MY 2029 (decrease of 36 fatalities), peaking in MY 2025 (37 fatalities). Corresponding estimates of light truck fatalities associated with mass changes are generally positive, ranging from a decrease of one fatality in MYs 2017 and 2018 to an increase of 14 fatalities in MYs 2026 through 2029. Altogether, light-duty vehicle fatality reductions associated with mass changes under Alternative #1 are Start Printed Page 43129estimated to be concentrated among MY 2023 through MY 2029 vehicles (146 out of 165, or 91% of net fatalities mitigated).
Table II-61 and Table II-62 present estimates of monetized lifetime social safety costs associated with mass changes by model year at three-percent and seven-percent discount rates, respectively for the CAFE Program. Table II-63 and Table II-64 show comparable tables from the perspective of the GHG Program.
Start Printed Page 43130
Start Printed Page 43131
Lifetime social safety costs are estimated to decrease generally by model year, with decreases associated with passenger cars generally offset partially by increases associated with light trucks. At a three-percent discount Start Printed Page 43132rate, decreases in lifetime social safety costs related to passenger cars are estimated to range from $13 million for existing (MY 1977 through MY 2016) cars, to $230 million for MY 2025 cars. The corresponding estimates at a seven-percent discount rate range from $7 million to $136 million. At a three-percent discount rate, impacts on lifetime social safety costs related to light trucks are estimated to range from a decrease of $5 million for MY 2017 light trucks to an increase of $96 million for MY 2022 light trucks. The corresponding estimates at a seven-percent discount rate range from $3 million to $65 million.
Consistent with the analysis of fatality impacts by model year in Table II-61, decreases in lifetime social safety costs associated with mass changes are generally concentrated in MY 2023 through MY 2029 light-duty vehicles under Alternative #1. At a three-percent discount rate, 93% of the reduction in total lifetime costs ($872 million out of $937 million) is attributed to MY 2023 through MY 2029 light-duty vehicles; at a seven-percent discount rate, 97% of the reduction in total lifetime costs ($486 million out of $501 million) is attributed to MY 2023 through MY 2029 light-duty vehicles.
(e) Sensitivity Analyses
Table II-65 shows the principal findings and includes sampling-error confidence bounds for the five parameters used in the CAFE model. The confidence bounds represent the statistical uncertainty that is a consequence of having less than a census of data. NHTSA's 2011, 2012, and 2016 reports acknowledged another source of uncertainty: The baseline statistical model can be varied by choosing different control variables or redefining the vehicle classes or crash types, which for example, could produce different point estimates.
Beginning with the 2012 Kahane report, NHTSA has provided results of 11 plausible alternative models that serve as sensitivity tests of the baseline model. Each alternative model was tested or proposed by: Farmer (IIHS) or Green (UMTRI) in their peer reviews; Van Auken (DRI) in his public comments; or Wenzel in his parallel research for DOE. The 2012 Kahane and 2016 Puckett and Kindelberger reports provide further discussion of the models and the rationales behind them.
Alternative models use NHTSA's databases and regression-analysis approach but differ from the baseline model in one or more explanatory variables, assumptions, or data restrictions. NHTSA applied the 11 techniques to the latest databases to generate alternative CAFE model coefficients. The range of estimates produced by the sensitivity tests offers insight to the uncertainty inherent in the formulation of the models, subject to the caveat these 11 tests are, of course, not an exhaustive list of conceivable alternatives.
The baseline and alternative results follow, ordered from the lowest to the highest estimated increase in societal risk per 100-pound reduction for cars weighing less than 3,201 pounds:
Start Printed Page 43133
The sensitivity tests illustrate both the fragility and the robustness of baseline estimates. On the one hand, the variation among NHTSA's coefficients is quite large relative to the baseline estimate: In the preceding example of cars < 3,201 pounds, the estimated coefficients range from almost zero to almost double the baseline estimate. This result underscores the key relationship that the societal effect of mass reduction is small and, as Wenzel has said, it “is overwhelmed by other known vehicle, driver, and crash factors.” 
In other words, varying how to model some of these other vehicle, driver, and crash factors, which is exactly what sensitivity tests do, can appreciably change the estimate of the societal effect of mass reduction.
On the other hand, variations are not particularly large in absolute terms. The ranges of alternative estimates are generally in line with the sampling-error confidence bounds for the baseline estimates. Generally, in alternative models as in the baseline models, mass reduction tends to be relatively more harmful in the lighter vehicles and more beneficial in the heavier vehicles, just as they are in the central analysis. In all models, the point estimate of NHTSA's coefficient is positive for the lightest vehicle class, cars < 3,201 pounds. In nine out of 11 models, the point estimate is negative for CUVs and minivans, and in eight out of 11 models the point estimate is negative for LTVs ≥ 5,014 pounds.
(f) Fleet Simulation Model
NHTSA has traditionally used real world crash data as the basis for projecting the future safety implications for regulatory changes. However, because lightweight vehicle designs are introducing fundamental changes to the structure of the vehicle, there is some concern that historical safety trends may not apply. To address this concern, NHTSA developed an approach to utilize lightweight vehicle designs to evaluate safety in a subset of real-world representative crashes. The methodology focused on frontal crashes because of the availability of existing vehicle and occupant restraint models. Representative crashes were simulated between baseline and lightweight vehicles against a range of vehicles and roadside objects using two different size belted driver occupants (adult male and small female) only. No passenger(s) or unbelted driver occupants were considered in this fleet simulation. The occupant injury risk from each simulation was calculated and summed to obtain combined occupant injury risk. The combined occupant injury risk was weighted according to the frequency of real world occurrences to develop overall societal risk for baseline and light-weighted vehicles. Note: The generic restraint system developed and used in the baseline occupant simulations was also used in the light-weighted vehicle occupant simulations as the purpose of this fleet simulation was to understand changes in societal injury risks because of mass reduction for different classes of vehicles in frontal crashes. No modifications to the restraint systems were made for light-weighted vehicle occupant simulations. Any modifications to restraint systems to improve occupant injury risks or societal injury risks in the light-weighted vehicle would have conflated results without identifying effects of mass reduction only. The following sections provide an overview of the fleet simulation study:
NHTSA contracted with George Washington University to develop a fleet simulation model 
to study the impact and relationship of light-weighted vehicle design with injuries and fatalities. In this study, there were eight vehicles as follows:
- 2001 model year Ford Taurus finite element model baseline and two simple design variants included a 25% lighter vehicle while maintaining the same vehicle front end stiffness and 25% overall stiffer vehicle while maintaining the same overall vehicle mass.
- 2011 model year Honda Accord finite element baseline vehicle and its 20% light-weight vehicle designed by Electricore. (This mass reduction study was sponsored by NHTSA).
- 2009/2010 model year Toyota Venza finite element baseline vehicle and two design variants included a 20% light-weight vehicle model (2010 Venza) (Low option mass reduction vehicle funded by EPA and International Council on Clean Transportation (ICCT)) and a 35% light-weight vehicle (2009 Venza) (High option mass reduction vehicle funded by California Air Resources Board).
Light weight vehicles were designed to have similar vehicle crash pulses as baseline vehicles. More than 440 vehicle crash simulations were conducted for the range of crash speeds and crash configurations to generate crash pulse and intrusion data points. The crash pulse data and intrusion data points will be used as inputs in the occupant simulation models.
For vehicle to vehicle impact simulations, four finite element models were chosen to represent the fleet. The partner vehicle models were selected to represent a range of vehicle types and weights. It was assumed vehicle models would reflect the crash response for all vehicles of the same type, e.g. mid-size car. Only the safety or injury risk for the driver in the target vehicle and in the partner vehicle were evaluated in this study.
As noted, vehicle simulations generated vehicle deformations and acceleration responses utilized to drive occupant restraint simulations and predict the risk of injury to the head, neck, chest, and lower extremities. In all, more than 1,520 occupant restraint simulations were conducted to evaluate the risk of injury for mid-size male and small female drivers.
The computed societal injury risk (SIR) for a target vehicle v in frontal crashes is an aggregate of individual serious crash injury risks weighted by real-world frequency of occurrence (v) of a frontal crash incident. A crash incident corresponds to a crash with different partners (Npartner) at a given impact speed (Pspeed), for a given driver occupant size (Loccsize), in the target or partner vehicle (T/P), in a given crash configuration (Mconfig), and in a single- or two-vehicle crash (Kevent). C IR (v) represents the combined injury risk (by body region) in a single crash incident. (v) designates the weighting factor, i.e., percent of occurrence, derived from National Automotive Sampling System Crashworthiness Data System (NASS CDS) for the crash incident. A driver age group of 16 to 50 Start Printed Page 43134years old was chosen to provide a population with a similar, i.e., more consistent, injury tolerance.
The fleet simulation was performed using the best available engineering models, with base vehicle restraint and airbag settings, to estimate societal risks of future lightweight vehicles. The range of the predicted risks for the baseline vehicles is from 1.25% to 1.56%, with an average of 1.39%, for the NASS frontal crashes that were simulated. The change in driver injury risk between the baseline and light-weighted vehicles will provide insight into the estimate of modification needed in the restraint and airbag systems of lightweight vehicles. If the difference extends beyond the expected baseline vehicle restraint and airbag capability, then adjustments to the structural designs would be needed. Results from the fleet simulation study show the trend of increased societal injury risk for light-weighted vehicle designs, as compared to their baselines, occurs for both single vehicle and two-vehicle crashes. Results are listed in Table II-66.
In general, the societal injury risk in the frontal crash simulation associated with the small size driver is elevated when compared to that of the mid-size driver. However, both occupant sizes had reasonable injury risk in the simulated impact configurations representative of the regulatory and consumer information testing. NHTSA examined three methods for combining injuries with different body regions. One observation was the baseline mid-size CUV model was more sensitive to leg injuries.
This study only looked at lightweight designs for a midsize sedan and a mid-size CUV and did not examine safety implications for heavier vehicles. The study was also limited to only frontal crash configurations and considered just mid-size CUVs whereas the statistical regression model considered all CUVs and all crash modes.
The change in the safety risk from the MY 2010 fleet simulation study was directionally consistent with results for passenger cars from NHTSA 2012 regression analysis study,
which covered data for MY 2000-MY 2007. The NHTSA 2012 regression analysis study was updated in 2016 to reflect newer MY 2003 to MY 2010. Comparing the fleet simulation societal risk to the 2016 update of the NHTSA 2012 regression analysis and the updated analysis used in this NPRM, the risk assessment from the fleet simulation is similarly directionally consistent with the passenger car risk assessment from the regression analysis. As noted, fleet simulations were performed only in frontal crash mode and did not consider other crash modes including rollover crashes.
This fleet simulation study does not provide information that can be used to modify coefficients derived for the NPRM regression analysis because of the restricted types of crashes 
and vehicle designs. As explained earlier, the fleet simulation study assumed restraint equipment to be as in the baseline model, in which restraints/airbags are not redesigned to be optimal with light-weighting.
2. Impact of Vehicle Scrappage and Sales Response on Fatalities
Previous versions of the CAFE model, and the accompanying regulatory analyses relying on it, did not carry a representation of the full on-road vehicle population, only those vehicles from model years regulated under proposed (or final) standards. The omission of an on-road fleet implicitly assumed the population of vehicles registered at the time a set of CAFE standards is promulgated is not affected by those standards. However, there are several mechanisms by which CAFE standards can affect the existing vehicle Start Printed Page 43135population. The most significant of these is deferred retirement of older vehicles. CAFE standards force manufacturers to apply fuel saving technologies to offered vehicles and then pass along the cost of those technologies (to the extent possible) to buyers of new vehicles. These price increases affect the length of loan terms and the desired length of ownership for new vehicle buyers and can discourage some buyers on the margin from buying a new vehicle in a given year. To the extent new vehicle purchases offset pending vehicle retirements, delaying new purchases in favor of continuing to use an aging vehicle affects the overall safety of the on-road fleet even if the vehicle whose retirement was delayed was not directly subject to a binding CAFE standard in the model year during its production.
The sales response in the CAFE model acts to modify new vehicle sales in two ways:
1. Changes in new vehicle prices either increase or decrease total sales (passenger cars and light trucks combined) each year in the context of forecasted macroeconomic conditions.
2. Changes in new vehicle attributes and fuel prices influence the share of new vehicles sold that are light trucks, and therefore also passenger cars.
These two responses change the total number of new vehicles sold in each model year across regulatory alternatives and the relative proportion of new vehicles that are passenger cars and light trucks. This response has two effects on safety. The first response slows the rate at which new vehicles, and their associated safety improvements, enter the on-road population. The second response influences the mix of vehicles on the road—with more stringent CAFE standards leading to a higher share of light trucks sold in the new vehicle market, assuming all else is equal. Light trucks have higher rates of fatal crashes when interacting with passenger cars and, as earlier sections discussed, different directional responses to mass reduction technology based on the existing mass and body style of the vehicle.
The sales response and scrappage response influence safety outcomes through the same basic mechanism, fleet turnover. In the case of the scrappage response, delaying fleet turnover keeps drivers in older vehicles likely to be less safe than newer model year vehicles that could replace them. Similarly, delaying the sale of new vehicles can force households to keep older vehicles in use longer, reallocate VMT within their household fleet, and generally meet travel demand through the use of older, less safe vehicles. As an illustration, if we simplify by ignoring that the share of new vehicles that are passenger cars changes with the stringency of the alternatives, simply changing the number of new vehicles between scenarios affects the mileage accumulation of the fleet and therefore all fleet level effects. Reducing the number of new vehicles sold, relative to a baseline forecasted value, reduces the size of the registered vehicle fleet that is able to service the underlying demand for travel.
Consider a simple example where we show sales effects operating on a micro-scale for a single household whose choices of whether to purchase a new vehicle is affected by vehicle price. A household starts with three vehicles, aged three, five, and eight years old. In a scenario with no CAFE standards and therefore no related changes in vehicle sales prices, the household buys a new car and scraps the eight-year old car; the other two cars in the fleet each get a year older. In a scenario where CAFE standards become more stringent causing vehicle sales prices to increase, this household chooses to delay buying a new car and each of their three existing cars gets a year older. In both cases, all three vehicles (including the new car in the first scenario, and the year-year-old car in the second scenario) have to serve the family's travel demand.
The scrappage effect is visible in the household's vehicle fleet as it moves from the first scenario to the second scenario with changes in CAFE standards. In the second scenario, the nine-year-old car remains in the household's fleet to service demand for travel, when it would otherwise have been retired. While the scrappage effect can be symmetrical to the sales effect, it need not be. The “new car” in the scenario without CAFE standards could be a new vehicle from the current model year or a used car that is of a newer vintage than the 8-year-old vehicle it replaces. The latter instance is an effect of scrappage decisions that do not directly affect new vehicle sales. Eventually, new vehicles transition to the used car market, but that on average take several years, and the shift is slow. At the household level, the scrappage decision occurs in a single year, each year, for every vehicle in the fleet. To the extent CAFE standards affect new vehicle prices and fuel economies, relative to vehicles already owned, scrappage could accelerate or decelerate depending upon the direction (and magnitude) of the changes.
3. Safety Model
The analysis supporting the CAFE rule for MYs 2017 and beyond did not account for differences in exposure or inherent safety risk as vehicles aged throughout their useful lives. However, the relationship between vehicle age and fatality risk is an important one. In a 2013 Research Note,
NHTSA's National Center for Statistics and Analysis concluded a driver of a vehicle that is four to seven years old is 10% more likely to be killed in a crash than the driver of a vehicle zero to three years old, accounting for the other factors related to the crash. This trend continued for older vehicles more generally, with a driver of a vehicle 18 years or older being 71% more likely to be killed in a crash than a driver in a new vehicle. While there are more registered vehicles that are zero to three years old than there are 20 years or older (nearly three times as many) because most of the vehicles in earlier vintages are retired sooner, the average age of vehicles in the United States is 11.6 years old and has risen significantly in the past decade.
This relationship reflects a general trend visible in the Fatality Analysis Reporting System (FARS) when looking at a series of calendar years: Newer vintages are safer than older vintages, over time, at each age. This is likely because of advancements in safety technology, like side-impact airbags, electronic stability control, and (more recently) sophisticated crash avoidance systems starting to work their way into the vehicle population. In fact, the 2013 Research Note indicated that the percentage of occupants fatally injured in fatal crashes increased with vehicle age: From 27% for vehicles three or fewer years old, to 41% for vehicles 12-14 years old, to 50% for vehicles 18 or more years old.
With an integrated fleet model now part of the analytical framework for CAFE analysis, any effects on fleet turnover (either from delayed vehicle retirement or deferred sales of new vehicles) will affect the distribution of both ages and model years present in the on-road fleet. Because each of these vintages carries with it inherent rates of fatal crashes, and newer vintages are generally safer than older ones, changing that distribution will change Start Printed Page 43136the total number of on-road fatalities under each regulatory alternative.
To estimate the empirical relationship between vehicle age, model year vintage, and fatalities, DOT conducted a statistical analysis linking data from the FARS database, a time series of Polk registration data to represent the on-road vehicle population, and assumed per-vehicle mileage accumulation rates (the derivation of which is discussed in detail in PRIA Chapter 11). These data were used to construct per-mile fatality rates that varied by vehicle vintage, accounting for the influence of vehicle age. However, unlike the NCSA study referenced above, any attempt to account for this relationship in the CAFE analysis faces two challenges. The first challenge is the CAFE model lacks the internal structure to account for other factors related to observed fatal crashes—for example, vehicle speed, seat belt use, drug use, or age of involved drivers or passengers. Vehicle interactions are simply not modeled at this level; the safety analysis in the CAFE model is statistical, using aggregate values to represent the totality of fleet interactions over time. The second challenge is perhaps the more significant of the two: The CAFE analysis is inherently forward-looking. To implement a statistical model analogous to the one developed by NCSA, the CAFE model would require forecasts of all factors considered in the NCSA model—about vehicle speeds in crashes, driver behavior, driver and passenger ages, vehicle vintages, and so on. In particular, the model would require distributions (joint distributions, in most cases) of these factors over a period of time spanning decades. Any such forecasts would be highly uncertain and would be likely to assume a continuation of current conditions.
Instead of trying to replicate the NCSA work at a similar level of detail, DOT conducted a simpler statistical analysis to separate the safety impact of the two factors the CAFE model explicitly accounts for: The distribution of vehicle ages in the fleet and the number of miles driven by those vehicles at each age. To accomplish this, DOT used data from the FARS database at a lower level of resolution; rather than looking at each crash and the specific factors that contributed to its occurrence, staff looked at the total number of fatal crashes involving light-duty vehicles over time with a focus on the influence of vehicle age and vehicle vintage. When considering the number of fatalities relative to the number of registered vehicles for a given model year (without regard to the passenger car/light-truck distinction, which has evolved over time and can create inconsistent comparisons), a somewhat noisy pattern develops. Using data from calendar year 1996 through 2015, some consistent stories develop. The points in Figure II-4 represent the number of fatalities per registered vehicle with darker circles associated with increasingly current calendar years.
As shown in Figure II-4, fatalities per registered vehicle have generally declined over time across all vehicle ages (the darker points representing newer vintages being closer to the x-axis) and, across most recent calendar years, fatality rates (per registered vehicle) start out at a low point, rise through age 15 or so, then decline through age 30 (at which point little of the initial model year cohort is still registered). While this pattern is evident in the registration data, it is magnified by imposing a mileage accumulation schedule on the registered population and examining fatalities per billion miles of VMT.
The mileage accumulation schedule used in this analysis was developed using odometer readings of vehicles aged 0-15 years in calendar year 2015. Start Printed Page 43137The years spanned by the FARS database cover all model years from calendar year 1996 through 2015. Given that there is a significant number of years between the older vehicles in the 1996 CY data and the most recent model years in the odometer data the informed the mileage accumulation schedules, staff applied an elasticity of −0.20 to the change in the average cost per mile of vehicles over their lives. While the older vehicles had lower fuel economies, which would be associated with higher per-mile driving costs, they also (mostly) faced lower fuel prices. This adjustment increased the mileage accumulation for older vehicles, but not by large amounts. Because the CAFE model uses the mileage accumulation schedule and applies it to all vehicles in the fleet, it is necessary to use the same schedule to estimate per-mile fatality rates in the statistical analysis—even if the schedule is based on vehicles that look different than the oldest vehicles in the FARS dataset.
When the per-vehicle fatality rates are converted into per-mile fatality rates, the pattern observed in the registration comparison becomes clearer. As Figure II-5 shows, the trend present in the fatality data on a per-registration basis is even clearer on a per-mile basis: Newer vintages are safer than older vintages, at each age, over time.
The shape of the curve in Figure II-5 suggests a polynomial relationship between fatality rate and vehicle age, so DOT's statistical model is based on that structure.
The final model is a weighted quartic polynomial regression (by number of registered vehicles) on vehicle age with fixed effects for the model years present in the dataset: 
The coefficient estimates and model summary are in Table II-67.
Start Printed Page 43138
Start Printed Page 43139
This function is now embedded in the CAFE model, so the combination of VMT per vehicle and the distribution of ages and model years present in the on-road fleet determine the number of fatalities in a given calendar year. The model reproduces the observed fatalities of a given model year, at each age, reasonably well with more recent model years (to which the VMT schedule is a better match) estimated with smaller errors.
While the final specification was not the only one considered, the fact this model was intended to live inside the CAFE model to dynamically estimate fatalities for a dynamically changing on-road vehicle population was a constraining factor.
(a) Predicting Future Safety Trends
The base model predicts a net increase in fatalities due primarily to slower adoption of safer vehicles and added driving because of less costly vehicle operating costs. In earlier calendar years, the improvement in safety of the on-road fleet produces a net reduction in fatalities, but from the mid-2020s forward, the baseline model predicts no further increase in safety, and the added risk from more VMT and older vehicles produces a net increase in fatalities. This model thus reflects a conservative limitation; it implicitly assumes the trend toward increasingly safe vehicles that has been apparent for the past 3 decades will flatten in mid-2020s. The agency does not assert this is the most likely case. In fact, the development of advanced crash avoidance technologies in recent years indicates some level of safety improvement is almost certain to occur. The difficulty is for most of these technologies, their effectiveness against fatalities and the pace of their adoption are highly uncertain. Moreover, autonomous vehicles offer the possibility of significantly reducing or eventually even eliminating the effect of human error in crash causation, a contributing factor in roughly 94% of all crashes. This conservative assumption may cause the NPRM to understate the beneficial effect of proposed standards on improving (reducing) the number of fatalities.
Advanced technologies that are currently deployed or in development include:
Forward Collision Warning (FCW) systems are intended to passively assist the driver in avoiding or mitigating the impact of rear-end collisions (i.e., a vehicle striking the rear portion of a vehicle traveling in the same direction directly in front of it). FCW uses forward-looking vehicle detection capability, such as RADAR, LIDAR (laser), camera, etc., to detect other vehicles ahead and use the information from these sensors to warn the driver and to prevent crashes. FCW systems provide an audible, visual, or haptic warning, or any combination thereof, to alert the driver of an FCW-equipped vehicle of a potential collision with another vehicle or vehicles in the anticipated forward pathway of the vehicle.
Crash Imminent Braking (CIB) systems are intended to actively assist the driver by mitigating the impact of rear-end collisions. These safety systems have forward-looking vehicle detection capability provided by sensing technologies such as RADAR, LIDAR, video camera, etc. CIB systems mitigate crash severity by automatically applying the vehicle's brakes shortly before the expected impact (i.e., without requiring the driver to apply force to the brake pedal).
Dynamic Brake Support (DBS) is a technology that actively increases the amount of braking provided to the driver during a rear-end crash avoidance maneuver. If the driver has applied force to the brake pedal, DBS uses forward-looking sensor data provided by technologies such as RADAR, LIDAR, video cameras, etc. to assess the potential for a rear-end crash. Should DBS ascertain a crash is likely (i.e., the sensor data indicate the driver has not applied enough braking to avoid the crash), DBS automatically intervenes. Although the manner in which DBS has been implemented differs among vehicle manufacturers, the objective of the interventions is largely the same: To supplement the driver's commanded brake input by increasing the output of the foundation brake system. In some situations, the increased braking provided by DBS may allow the driver to avoid a crash. In other cases, DBS interventions mitigate crash severity.
Pedestrian AEB (PAEB) systems provide automatic braking for vehicles when pedestrians are in the forward path of travel and the driver has taken insufficient action to avoid an imminent crash. Like CIB, PAEB safety systems use information from forward-looking sensors to automatically apply or supplement the brakes in certain driving Start Printed Page 43140situations in which the system determines a pedestrian is in imminent danger of being hit by the vehicle. Many PAEB systems use the same sensors and technologies used by CIB and DBS.
Rear Automatic Braking feature means installed vehicle equipment that has the ability to sense the presence of objects behind a reversing vehicle, alert the driver of the presence of the object(s) via auditory and visual alerts, and automatically engage the available braking system(s) to stop the vehicle.
Semi-automatic Headlamp Beam Switching device provides either automatic or manual control of headlamp beam switching at the option of the driver. When the control is automatic, headlamps switch from the upper beam to the lower beam when illuminated by headlamps on an approaching vehicle and switch back to the upper beam when the road ahead is dark. When the control is manual, the driver may obtain either beam manually regardless of the conditions ahead of the vehicle.
Rear Turn Signal Lamp Color Turn signal lamps are the signaling element of a turn signal system, which indicates the intention to turn or change direction by giving a flashing light on the side toward which the turn will be made. FMVSS No. 108 permits a rear turn signal lamp color of amber or red.
Lane Departure Warning (LDW) system is a driver assistance system that monitors lane markings on the road and alerts the driver when their vehicle is about to drift beyond a delineated edge line of their current travel lane.
Blind Spot Detection (BSD) systems uses digital camera imaging technology or radar sensor technology to detect one or more vehicles in either of the adjacent lanes that may not be apparent to the driver. The system warns the driver of an approaching vehicle's presence to help facilitate safe lane changes.
These technologies are either under development or are currently being offered, typically in luxury vehicles, as either optional or standard equipment.
To estimate baseline fatality rates in future years, NHTSA examined predicted results from a previous NCSA study 
that measured the effect of known safety regulations on fatality rates. This study relied on statistical evaluations of the effectiveness of motor vehicle safety technologies based on real world performance in the on-road vehicle fleet to determine the effectiveness of each safety technology. These effectiveness rates were applied to existing fatality target populations and adjusted for current technology penetration in the on-road fleet, taking into account the retirement of existing vehicles and the pace of future penetration required to meet statutory compliance requirements, as well as adjustments for overlapping target populations. Based on these factors, as well as assumptions regarding future VMT, the study predicted future fatality levels and rates. Because the safety impact in the CAFE model independently predicts future VMT, we removed the VMT growth rate from the NCSA study and developed a prediction of vehicle fatality trends based only on the penetration pace of new safety technologies into the on-road fleet. These data were then normalized into relative safety factors with CY 2015 as the baseline (to match the baseline fatality year used in this CAFE analysis). These factors were then converted into equivalent fatality rates/100 million VMT by anchoring them to the 2015 fatality rate/100 million VMT published by NHTSA. Figure II-6 below illustrates the modelling output and projected fatality trend from the analysis of the NCSA study, prior to adjustment to fatality rates/100 million VMT.
Start Printed Page 43141
This model was based on inputs representing the impact of technology improvement through CY 2020. Projecting this trend beyond 2020 can be justified based on the continued transformation of the on-road fleet to 100% inclusion of the known safety technologies. Based on projections in the NCSA study, significant further technology penetration can be expected in the on-road fleet for side impact improvements (FMVSSS 214), electronic stability control (FMVSS 126), upper interior head impact protection (FMVSS 301), tire pressure monitoring systems (FMVSS 138), ejection mitigation (FMVSS 226), and heavy truck stopping distance improvements (FMVSS 121). These technologies were estimated to be installed in only 40-70% of the on-road fleet as of CY 2020, implying further safety improvement well beyond the 2020 calendar year.
The NCSA study focused on projections to reflect known technology adaptation requirements, but it was conducted prior to the 2008 recession, which disrupted the economy and changed travel patterns throughout the country. Thus, while the relative trends it predicts seem reasonable, they cannot account for the real-world disruption and recovery that occurred in the 2008-2015 timeframe. In addition, the NCSA study did not attempt to adjust for safety impacts that may have resulted from changes in the vehicle sales mix (vehicle types and sizes creating different interactions in crashes), in commuting patterns, or in shopping or socializing habits associated with internet access and use. To address this, NHTSA also examined the actual change in the fatality rate as measured by fatality counts and VMT estimates. Figure II-7 below illustrates the actual fatality rates measured from 2000 through 2016 and the modeled fatality rate trend based on these historical data.
The effect of the recession and subsequent recovery can be seen in chaotic shift in the fatality rate trend starting in 2008. The generally gradual decline that had been occurring over the previous decade was interrupted by a slowdown in the rate of change followed by subsequent upward and downward shifts. More recently, the rate has begun to increase. These shifts reflect some combination of factors not captured in the NCSA analysis mentioned above. The significance of this is that although there was a steady increase in the penetration of safety technologies into the on-road fleet between 2008 and 2015, other unknown factors offset their positive influence and eventually reversed the trend in vehicle safety rates. Because of the upward shift over the 2014-2015 period, this model, which does not reflect technology trend savings after 2015, will predict an upward shift of fatality rates after 2020.
Predicting future safety trends has significant uncertainty. Although further safety improvements are expected because of advanced safety technologies such as automatic braking and eventually, fully automated vehicles, the pace of development and extent of consumer acceptance of these improvements is uncertain. Thus, two imperfect models exist for predicting future safety trends. The NCSA model reflects the expected trend from required technologies and indicates continued improvement well beyond the 2020 timeframe, which is when the historical fatality rate based model breaks down. By contrast, the historical fatality rate model reflects shifts in safety not captured by the NCSA model, but gives arguably implausible results after 2020. It essentially represents a scenario in which economic, market, or behavioral factors minimize or offset much of the potential impact of future safety technology.
For the NPRM, the analysis examines a scenario projecting safety improvements beyond 2015 using a simple average of the NCSA and historical fatality rate models, accepting each as an illustration of different and conflicting possible future scenarios. As Start Printed Page 43142both models eventually curve up because of their quadratic form, each models' results are flattened at the point where they begin to trend upward. This occurs in 2045 for the NCSA model and in 2021 for the historical model. The results are shown in Figure II-8 below. The results indicate roughly a 19% reduction in fatality rates between 2015 and 2050. This is a slower pace than what has historically occurred over the past several decades, but the biggest influence on historical rates was significant improvement in safety belt use, which was below 10% in 1960 and had risen to roughly 70% by 2000, and is now more than 90%. Because belt use is now above 90%, further such improvements are unlikely unless they come from new technologies.
A difficulty with these trend models is they are based on calendar year predictions, which are derived from the full on-road vehicle fleet rather than the model year fleet, which is the basis for calculations in the CAFE model. As such they are useful primarily as indicators that vehicle safety has steadily improved over the past several decades, and given the advanced safety technologies under current development, we would expect some continuation of improvement in MY vehicle safety over the near and mid-term future. To account for this, NHTSA approximated a model year safety trend continuing through about 2035 (Figure II-9). For this trend the agency used actual data from FARS to calculate the change in fatality rates through 2007. The recession, which struck our economy in 2008, distorted normal behavioral patterns and affected both VMT and the mix of drivers and type of driving to an extent we do not believe the recession era gives an accurate picture of the safety trends inherent in the vehicles themselves. Therefore, beginning in 2008, NHTSA approximated a trend for safety improvement through about MY 2035 to reflect the continued effect of improved safety technologies such as advanced automatic braking, which manufacturers have announced will be in all new vehicles by MY 2022. The agency recognize this is only an estimate, and actual MY trends could be above or below this line. NHTSA examined alternate trends in a sensitivity analysis and request comments on the best way to address future safety trends.
NHTSA also notes although we project vehicles will continue to become safer going forward to about 2035, we do not have corresponding cost information for technologies enabling this improvement. In a standard elasticity model, sales impacts are a function of the percent change in vehicle price. Hypothetically, increasing the base price for added safety technologies would decrease the impact of higher prices due to impacts of CAFE standards on vehicle sales. The percentage change in baseline price would decrease, which would mean a lower elasticity effect, which would mean a lower impact on sales. NHTSA will consider possible ways to address this issue before the final rule, and we request comments on the need and/or practicability for such an adjustment, as well as any data and other relevant information that could support such an analysis of these costs, as well as the future pace of technological adoption within the vehicle fleet.
Start Printed Page 43143
(b) Adjusting for Behavioral Impacts
The influence of delayed purchases of new vehicles is estimated to have the most significant effect on safety imposed by CAFE standards. Because of a combination of safety regulations and voluntary safety improvements, passenger vehicles have become safer over time. Compared to prior decades, fatality rates have declined significantly because of technological improvements, as well as behavioral shifts, such as increased seat belt use. As these safer vehicles replace older less safe vehicles in the fleet, the on-road fleet is replaced with vehicles reflecting the improved fatality rates of newer, safer vehicles. However, fatality rates associated with different model year vehicles are influenced by the vehicle itself and by driver behavior. Over time, used vehicles are purchased by drivers in different demographic circumstances who also tend to have different behavioral characteristics. Drivers of older vehicles, on average, tend to have lower belt use rates, are more likely to drive inebriated, and are more likely to drive over the speed limit. Additionally, older vehicles are more likely to be driven on rural roadways, which typically have higher speeds and produce more serious crashes. These relationships are illustrated graphically in Chapter 11 of the PRIA accompanying this proposed rule.
The behavior being modelled and ascribed to CAFE involves decisions by drivers who are contemplating buying a new vehicle, and the purchase of a newer vehicle will not in itself cause those drivers to suddenly stop wearing seat belts, speed, drive under the influence, or shift driving to different land use areas. The goal of this analysis is to measure the effect of different vehicle designs that change by model year. The modelling process for estimating safety essentially involves substituting fatality rates of older MY vehicles for improved rates that would have been experienced with a newer vehicle. Therefore, it is important to control for behavioral aspects associated with vehicle age so only vehicle design differences are reflected in the estimate of safety impacts. To address this, the CAFE safety model was run to control for vehicle age. That is, it does not reflect a decision to replace an older model year vehicle that is, for example, 10 years old with a new vehicle. Rather, it reflects the difference in the average fatality rate of each model year across its entire lifespan. This will account for most of the difference because of vehicle age, but it may still reflect a bias caused by the upward trend in societal seat belt use over time. Because of this secular trend, each subsequent model year's useful life will occur under increasingly higher average seat belt use rates. This could cause some level of behavioral safety improvement to be ascribed to the model year instead of the driver cohort. However, it is difficult to separate this effect from the belt use impacts of changing driver cohorts as vehicles age.
Glassbrenner (2012) analyzed the effect of improved safety in newer vehicles for model years 2001 through 2008. She developed several statistical regression models that specifically controlled for most behavioral factors to isolate model year vehicle characteristics. However, her study did not specifically report the change in MY fatality rates—rather, she reported total fatalities that could have been saved in a baseline year (2008) had all vehicles in the on-road fleet had the same safety features as the MY 2001 through MY 2008 vehicles. This study potentially provides a basis for comparison with results of the CAFE safety estimates. To make this comparison, the CY 2008 passenger car and light truck fatalities total from FARS were modified by subtracting the values found in Figure II-9 of her study. This gives a stream of comparable hypothetical CY 2008 fatality totals under progressively less safe model year designs. Results indicated that had the 2008 on-road fleet been equipped with MY 2008 safety equipment and vehicle characteristics, total fatalities would have been reduced by 25% compared to vehicles that were actually on the road in 2008. Similar results were calculated for each model years' vehicle characteristics back to 2001.
For comparison, predicted MY fatality rates were derived from the CAFE safety model and applied to the CY 2008 VMT calculated by that model. This gives an estimate of CY 2008 fatalities under each model years' fatality rate, which, when compared to the predicted CY fatality total, gives a trendline Start Printed Page 43144comparable to the Glassbrenner trendline illustrating the change in MY fatality rates. Both models are sensitive to the initial 2008 baseline fatality total, and because the predicted CAFE total is somewhat lower than the actual total, the agency ran a third trendline to examine the influence of this difference. Results are shown in Figure II-10.
Using the corrected fatality count, but retaining the predicted VMT changes the initial 2018 CY fatality rate to 12.62 (instead of 12.15) and produces the result shown in Figure II-10. The CAFE model trendline shifts up, which narrows the difference in early years but expands it in later years. However, VMT and fatalities are linked in the CAFE model, so the actual level of the MY safety predicted by the CAFE curve has uncertainty. Perhaps the most meaningful result from this comparison is the difference in slopes; the CAFE model predicts more rapid change through 2006, but in the last few years change decreases. This might reflect the trend in societal belt use, which rose steadily through 2005 and levelled off. Later model years' fatality rates would benefit from this trend while earlier model years would suffer. This seems consistent with our using lifetime MY fatality rates to reflect MY change rather than first year MY fatality rates (although even first year rates would reflect this bias, but not as much).
To provide another perspective on safety impacts, NHTSA accessed data from a comprehensive study of the effects of safety technologies on motor vehicle fatalities. Kahane (2015) 
examined all safety effects of vehicle safety technologies from 1960 through 2012 and found these technologies saved more than 600,000 lives during that time span. Kahane is currently working under contract for NHTSA to update this study through 2016. At NHTSA's request, Kahane accessed his database to provide a measure of relative MY vehicle design safety by controlling for seat belt use. The result was a MY safety index illustrating the progress in vehicle safety by model year which isolates vehicle design from the primary behavioral impact—seat belt usage. We normalized Kahane's index to MY 1975 and did the same to the “fixed effects” we are currently using from our safety model to compare the trends in MY safety from the two methods. Results are shown in Figure II-11.
Start Printed Page 43145
From Figure II-11 both approaches show similar long-term downward trends, but this model shows a steeper slope than Kahane's model. The two models involve completely different approaches, so some difference is to be expected. However, it is also possible this reflects different methods used to isolate vehicle design safety from behavioral impacts. As discussed previously, NHTSA addressed this issue by removing vehicle age impacts from its model, whereas Kahane's model does it by controlling for belt use. As noted previously, aside from the age impact on belt use associated with the different demographics driving older vehicles, there is a secular trend toward more belt use reflecting the increase in societal awareness of belt use importance over time. This trend is illustrated in Figure II-12 below.
NHTSA's current approach removes the age trend in belt use, but it's not clear whether it accounts for the full impacts of the secular trend as well. If not, some portion of the gap between the two trendlines could reflect behavioral impacts rather than vehicle design.
These models (NHTSA, Glassbrenner, and Kahane) involve differing approaches and assumptions contributing to uncertainty, and given this, their differences are not surprising. It is encouraging they show similar directional trends, reinforcing the basic concept we are measuring. NHTSA recognizes predicting future fatality impacts, as well as sales impacts that cause them, is a difficult and imprecise task. NHTSA will continue to investigate this issue, and we seek comment on these estimates as well as alternate methods for predicting the safety effects associated with delayed new vehicle purchases.
Start Printed Page 43146
4. Impact of Rebound Effect on Fatalities
Based on historical data, it is possible to calculate a baseline fatality rate for vehicles of any model year vintage. By simply taking the total number of vehicles involved in fatal accidents over all ages for a model year and dividing by the cumulative VMT over the useful life of every vehicle produced in that model year, one arrives at a baseline hazard rate denominated in fatalities per billion miles. The fatalities associated with vehicles produced in that model year are then proportional to the cumulative lifetime VMT, where total fatalities equal the product of the baseline hazard rate and VMT. A more comprehensive discussion of the rebound effect and the basis for calculating its impact on mileage and risk is in Chapter 8 of the PRIA accompanying this proposed rule.
5. Adjustment for Non-Fatal Crashes
Fatalities estimated to be caused by various alternative CAFE standards are valued as a societal cost within the CAFE models' cost/benefit accounting. Their value is based on the comprehensive value of a fatality derived from data in Blincoe et al. (2015), adjusted to 2016 economics and updated to reflect the official DOT guidance on the value of a statistical life in 2016. This gives a societal value of $9.9 million for each fatality. The CAFE safety model estimates effects on traffic fatalities but does not address corresponding effects on non-fatal injuries and property damage that would result from the same factors influencing fatalities. To address this, we developed an adjustment factor that would account for these crashes.
Development of this factor is based on the assumption nonfatal crashes will be affected by CAFE standards in proportion to their nationwide incidence and severity. That is, NHTSA assumes the same injury profile, the relative number of cases of each injury severity level, that occur nationwide, will be increased or decreased because of CAFE. The agency recognizes this may not be the case, but the agency does not have data to support individual estimates across injury severities. There are reasons why this may not be true. For example, because older model year vehicles are generally less safe than newer vehicles, fatalities may make up a larger portion of the total injury picture than they do for newer vehicles. This would imply lower ratios across the non-fatal injury and PDO profile and would imply our adjustment may overstate total societal impacts. NHTSA requests comments on this assumption and alternative methods to estimate injury impacts.
The adjustment factor is derived from Tables 1-8 and I-3 in Blincoe et al. (2015). Incidence in Table I-3 reflects the Abbreviated Injury Scale (AIS), which ranks nonfatal injury severity based on an ascending 5 level scale with the most severe injuries ranked as level 5. More information on the basis for these classifications is available from the Association for the Advancement of Automotive Medicine at https://www.aaam.org/abbreviated-injury-scale-ais/.
Table 1-3 in Blincoe lists injured persons with their highest (maximum) injury determining the AIS level (MAIS). This scale is represented in terms of MAIS level, or maximum abbreviated injury scale. MAIS0 refers to uninjured occupants in injury vehicles, MAIS1 are generally considered minor injuries, MAIS2 moderate injuries, MAIS3 serious injuries, MAIS4 severe injuries, and MAIS5 critical injuries. PDO refers to property damage only crashes, and counts for PDOs refer to vehicles in which no one was injured. From Table II-68, ratios of injury incidence/fatality are derived for each injury severity level as follows:
Start Printed Page 43147
For each fatality that occurs nationwide in traffic crashes, there are 561 vehicles involved in PDOs, 139 uninjured occupants in injury vehicles, 105 minor injuries, 10 moderate injuries, 3 serious injuries, and fractional numbers of the most serious categories which include severe and critical nonfatal injuries. For each fatality ascribed to CAFE it is assumed there will be nonfatal crashes in these same ratios.
Property damage costs associated with delayed new vehicle purchases must be treated differently because crashes that subsequently occur damage older used vehicles instead of newer vehicles. Used vehicles are worth less and will cost less to repair, if they are repaired at all. The consumer's property damage loss is thus reduced by longer retention of these vehicles. To estimate this loss, average new and used vehicle prices were compared. New vehicle transaction prices were estimated from a study published by Kelley Blue Book.
Based on these data, the average new vehicle transaction price in January 2017 was $34,968. Used vehicle transaction prices were obtained from Edmonds Used Vehicle Market Report published in February of 2017.
Edmonds data indicate the average used vehicle transaction price was $19,189 in 2016. There is a minor timing discrepancy in these data because the new vehicle data represent January 2017, and the used vehicle price is for the average over 2016. NHTSA was unable to locate exact matching data at this time, but the agency believes the difference will be minor.
Based on these data, new vehicles are on average worth 82% more than used vehicles. To estimate the effect of higher property damage costs for newer vehicles on crashes, the per unit property damage costs from Table I-9 in Blincoe et al. (2015) were multiplied by this factor. Results are illustrated in Table II-69.
The total property damage cost reduction was then calculated as a function of the number of fatalities reduced or increased by CAFE as follows:
S = total property damage savings from retaining used vehicles longer
F = change in fatalities estimated for CAFE due to retaining used vehicles
r = ratio of nonfatal injuries or PDO vehicles to fatalities (F)
p = value of property damage prevented by retaining older vehicleStart Printed Page 43148
n = the 8 injury severity categories
The number of fatalities ascribed to CAFE because of older vehicle retention was multiplied by the unit cost per fatality from Table I-9 in Blincoe et al. (2015) to determine the societal impact accounted for by these fatalities.
From Table I-8 in Blincoe et al. (2015), NHTSA subtracted property damage costs from all injury severity levels and recalculated the total comprehensive value of societal losses from crashes. The agency then divided the portion of these crashes because of fatalities by the resulting total to estimate the portion of crashes excluding property damage that are accounted for by fatalities. Results indicate fatalities accounted for approximately 40% of all societal costs exclusive of property damage. NHTSA then divided the total cost of the added fatalities by 0.4 to estimate the total cost of all crashes prevented exclusive of the savings in property damage. After subtracting the total savings in property damage from this value, we divided the fatality cost by it to estimate that overall, fatalities account for 43% of the total costs that would result from older vehicle retention.
For the fatalities that occur because of mass effects or to the rebound effect, the calculation was more direct, a simple application of the ratio of the portion of costs produced by fatalities. In this case, there is no need to adjust for property damage because all impacts were derived from the mix of vehicles in the on-road fleet. Again, from Table I-8 in Blincoe et al (2015), we derive this ratio based on all cost factors including property damage to be .36. These calculations are summarized as follows:
SV = Value of societal Impacts of all crashes
F = change in fatalities estimated for CAFE due to retaining used vehicles
v = Comprehensive societal value of preventing 1 fatality
x = Percent of total societal loss from crashes excluding property damage accounted for by fatalities
S = total property damage savings from retaining used vehicles longer
M = change in fatalities due to changes in vehicle mass to meet CAFE standards
c = Percent of total societal loss from all cost factors in all crashes accounted for by fatalities
For purposes of application in the CAFE model, these two factors were combined based on the relative contribution to total fatalities of different factors. As noted, although a safety impact from the rebound effect is calculated, these impacts are considered to be freely chosen rather than imposed by CAFE and imply personal benefits at least equal to the sum of their added costs and safety consequences. The impacts of this nonfatal crash adjustment affect costs and benefits equally. When considering safety impacts actually imposed by CAFE standards, only those from mass changes and vehicle purchase delays are considered. NHTSA has two different factors depending on which metric is considered. The agency created these factors by weighting components by the relative contribution to changes in fatalities associated with each component. This process and results are shown in Table II-70. Note: For the NPRM, NHTSA applied the average weighted factor to all fatalities. This will tend to slightly overstate costs because of sales and scrappage and understate costs associated with mass and rebound. The agency will consider ways to adjust this minor discrepancy for the final rule.
Table II-71, Table II-72, Table II-73, and Table II-74 summarize the safety effects of CAFE standards across the various alternatives under the 3% and 7% discount rates. As noted in Section II.F.5, societal impacts are valued using a $9.9 million value per statistical life (VSL). Fatalities in these tables are undiscounted; only the monetized societal impact is discounted.
Start Printed Page 43149
Start Printed Page 43150
Start Printed Page 43151
Start Printed Page 43152
Start Printed Page 43153
Table II-75 through Table II-78 summarize the safety effects of GHG standards across the various alternatives under the 3% and 7% discount rates. As noted in Section II.F.5, societal impacts are valued using a $9.9 million value per statistical life (VSL). Fatalities in these tables are undiscounted; only the monetized societal impact is discounted.
Start Printed Page 43154
Start Printed Page 43155
Start Printed Page 43156
Start Printed Page 43157
Start Printed Page 43158
While NHTSA notes the value of rebound effect fatalities, as well as total fatalities from all causes, the agency does not add rebound effects to the other CAFE-related impacts because rebound-related fatalities and injuries result from risk that is freely chosen and offset by societal valuations that at a minimum exceed the aggregate value of safety consequences plus added vehicle operating and maintenance costs.
These costs implicitly involve a cost and a benefit that are offsetting. The relevant safety impacts attributable to CAFE are highlighted in bold in the above tables.
G. How the Model Analyzes Different Potential CAFE and CO 2 Standards
1. Specification of No-Action and Other Regulatory Alternatives
(a) Mathematical Functions Defining Passenger Car and Light Trucks Standards for Each Model Year During 2016-2032
In the U.S. market, the stringency of CAFE and CO2 standards can influence the design of new vehicles offered for sale by requiring manufacturers to produce increasingly fuel efficient vehicles in order to meet program Start Printed Page 43159requirements. This is also true in the CAFE model simulation, where the standards can be defined with a great deal of flexibility to examine the impact of different program specifications on the auto industry. Standards are defined for each model year and can represent different slopes that relate fuel economy to footprint, different regions of flat slopes, and different rates of increase for each of three regulatory classes covered by the CAFE program (domestic passenger cars, imported passenger cars, and light trucks).
The CAFE model takes, as inputs, the coefficients of the mathematical functions described in Sections III and IV. It uses these coefficients and the function to which they belong to define the target for each vehicle in the fleet, then computes the standard using the harmonic average of the targets for each manufacturer and fleet. The model also allows the user to define the extent and duration of various compliance flexibilities (e.g., limits on the amount of credit that a manufacturer may claim related to air conditioning efficiency improvements or off-cycle fuel economy adjustments) as well as limits on the number of years that CAFE credits may be carried forward or the amount that may be transferred between a manufacturer's fleets.
(b) Off-Cycle and A/C Efficiency Adjustments Anticipated for Each Model Year
Another aspect of credit accounting is partially implemented in the CAFE model at this point—those related to the application of off-cycle and A/C efficiency adjustments, which manufacturers earn by taking actions such as special window glazing or using reflective paints that provide fuel economy improvements in real-world operation but do not produce measurable improvements in fuel consumption on the 2-cycle test.
NHTSA's inclusion of off-cycle and A/C efficiency adjustments began in MY 2017, while EPA has collected several years' worth of submissions from manufacturers about off-cycle and A/C efficiency technology deployment. Currently, the level of deployment can vary considerably by manufacturer with several claiming extensive Fuel Consumption Improvement Values (FCIV) for off-cycle and A/C efficiency technologies and others almost none. The analysis of alternatives presented here does not attempt to project how future off-cycle and A/C efficiency technology use will evolve or speculate about the potential proliferation of FCIV proposals submitted to the agencies. Rather, this analysis uses the off-cycle credits submitted by each manufacturer for MY 2017 compliance and carries these forward to future years with a few exceptions. Several of the technologies described in Section II.D are associated with A/C efficiency and off-cycle FCIVs. In particular, stop-start systems, integrated starter generators, and full hybrids are assumed to generate off-cycle adjustments when applied to vehicles to improve their fuel economy. Similarly, higher levels of aerodynamic improvements are assumed to include active grille shutters on the vehicle, which also qualify for off-cycle FCIVs.
The analysis assumes that any off-cycle FCIVs that are associated with actions outside of the technologies discussed in Section II.D (either chosen from the pre-approved “pick list,” or granted in response to individual manufacturer petitions) remain at the levels claimed by manufacturers in MY 2017. Any additional A/C efficiency and off-cycle adjustments that accrue as the result of explicit technology application are calculated dynamically in each model year for each alternative. The off-cycle FCIVs for each manufacturer and fleet, denominated in grams CO2 per mile,
are provided in Table II-79.
Start Printed Page 43160
The model currently accounts for any off-cycle adjustments associated with technologies that are included in the set of fuel-saving technologies explicitly simulated as part of this proposal (for example, start-stop systems that reduce fuel consumption during idle or active grille shutters that improve aerodynamic drag at highway speeds) and accumulates these adjustments up to the 10 g/mi cap. As a practical matter, most of the adjustments for which manufacturers are claiming off-cycle FCIV exist outside of the technology tree, so the cap is rarely reached during compliance simulation. If those FCIVs become a more important compliance mechanism, it may be necessary to model their application explicitly. However, doing so will require data on which vehicle models already possess these improvements as well as the cost and expected value of applying them to other models in the future. Comment is sought on both the data requirements and strategic decisions associated with manufacturers' use of A/C efficiency and off-cycle technologies to improve CAFE and CO2 compliance.
(c) Civil Penalty Rate and OEMs' Anticipated Willingness To Treat Civil Penalties as a Program Flexibility
Throughout the history of the CAFE program, some manufacturers have consistently achieved fuel economy levels below their standard. As in previous versions of the CAFE model, the current version allows the user to specify inputs identifying such manufacturers and to consider their compliance decisions as if they are willing to pay civil penalties for non-compliance with the CAFE program. The assumed civil penalty rate in the current analysis is $5.50 per 1/10 of a mile per gallon, per vehicle sold.
It is worth noting that treating a manufacturer as if they are willing to pay civil penalties does not necessarily mean that it is expected to pay penalties in reality. It merely implies that the manufacturer will only apply fuel economy technology up to a point, and then stop, regardless of whether or not its corporate average fuel economy is above its standard. In practice, we expect that many of these manufacturers will continue to be active in the credit market, using trades with other manufacturers to transfer credits into specific fleets that are challenged in any given year, rather than paying penalties to resolve CAFE deficits. The CAFE model calculates the amount of penalties paid by each manufacturer, but it does not simulate trades between manufacturers. In practice, some (possibly most) of the total estimated penalties may be a transfer from one OEM to another.
While the Energy Policy and Conservation Act (EPCA), as amended in 2007 by the Energy Independence and Security Act, prescribes these specific civil penalty provisions for CAFE standards, the Clean Air Act (CAA) does not contain similar provisions. Rather, the CAA's provisions regarding noncompliance constitute a de facto prohibition against selling vehicles failing to comply with emissions standards. Therefore, inputs regarding civil penalties—including inputs regarding manufacturers' potential willingness to treat civil penalty payment as an economic choice—apply only to simulation of CAFE standards.
(d) Treatment of Credit Provisions for “Standard Setting” and “Unconstrained” Analyses
NHTSA may not consider the application of CAFE credits toward compliance with new standards when establishing the standards themselves.
As such, this analysis considers 2020 to be the last model year in which carried-forward or transferred credits can be applied for the CAFE program. Beginning in model year 2021, Start Printed Page 43161today's “standard setting” analysis is conducted assuming each fleet must comply with the CAFE standard separately in every model year.
The “unconstrained” perspective acknowledges that these flexibilities exist as part of the program and, while not considered in NHTSA's decision of the preferred alternative, are important to consider when attempting to estimate the real impact of any alternative. Under the “unconstrained” perspective, credits may be earned, transferred, and applied to deficits in the CAFE program throughout the full range of model years in the analysis. The Draft Environmental Impact Analysis (DEIS) accompanying today's NPRM presents results of “unconstrained” modeling. Also, because the CAA provides no direction regarding consideration of any CO2 credit provisions, today's analysis includes simulation of carried-forward and transferred CO2 credits in all model years.
(e) Treatment of AFVs for “Standard Setting” and “Unconstrained” Analyses
NHTSA is also prohibited from considering the possibility that a manufacturer might produce alternatively fueled vehicles as a compliance mechanism,
taking advantage of credit provisions related to AFVs that significantly increase their fuel economy for CAFE compliance purposes. Under the “standard setting” perspective, these technologies (pure battery electric vehicles and fuel cell vehicles 
) are not available in the compliance simulation to improve fuel economy. Under the “unconstrained” perspective, such as is documented in the DEIS, the CAFE model considers these technologies in the context of all other available technologies and may apply them if they represent cost-effective compliance pathways. However, under both perspectives, the analysis continues to include dedicated AFVs that already exist in the MY 2016 fleet (and their projected future volumes) in CAFE calculations. Also, because the CAA provides no direction regarding consideration of alternative fuels, today's analysis includes simulation of the potential that some manufacturers might introduce new AFVs in response to CO2 standards. To fully represent the compliance benefit from such a response, NHTSA modified the CAFE model to include the specific provisions related to AFVs under the CO2 standards. In particular, the CAFE model now carries a full representation of the production multipliers related to electric vehicles, fuel cell vehicles, plug-in hybrids, and CNG vehicles, all of which vary by year through MY 2021.
2. Simulation of Manufacturers' [and Buyers'] Potential Responses to Each Alternative
The CAFE model provides a way of estimating how manufacturers could attempt to comply with a given CAFE standard by adding technology to fleets that the agencies anticipate they will produce in future model years. This exercise constitutes a simulation of manufacturers' decisions regarding compliance with CAFE or CO2 standards.
This compliance simulation begins with the following inputs: (a) The analysis fleet of vehicles from model year 2016 discussed above in Section II.B, (b) fuel economy improving technology estimates discussed above in Section II.D, (c) economic inputs discussed above in Section II.E, and (d) inputs defining baseline and potential new CAFE standards. For each manufacturer, the model applies technologies in both a logical sequence and a cost-minimizing strategy in order to identify a set of technologies the manufacturer could apply in response to new CAFE or CO2 standards. The model applies technologies to each of the projected individual vehicles in a manufacturer's fleet, considering the combined effect of regulatory and market incentives while attempting to account for manufacturers' production constraints. Depending on how the model is exercised, it will apply technology until one of the following occurs:
(1) The manufacturer's fleet achieves compliance 
with the applicable standard and continuing to add technology in the current model year would be attractive neither in terms of stand-alone (i.e., absent regulatory need) cost-effectiveness nor in terms of facilitating compliance in future model years;
(2) The manufacturer “exhausts” available technologies; 
(3) For manufacturers assumed to be willing to pay civil penalties (in the CAFE program), the manufacturer reaches the point at which doing so would be more cost-effective (from the manufacturer's perspective) than adding further technology.
The model accounts explicitly for each model year, applying technologies when vehicles are scheduled to be redesigned or freshened and carrying forward technologies between model years once they are applied (until, if applicable, they are superseded by other technologies). The model then uses these simulated manufacturer fleets to generate both a representation of the U.S. auto industry and to modify a representation of the entire light-duty registered vehicle population. From these fleets, the model estimates changes in physical quantities (gallons of fuel, pollutant emissions, traffic fatalities, etc.) and calculates the relative costs and benefits of regulatory alternatives under consideration.
The CAFE model accounts explicitly for each model year, in turn, because manufacturers actually “carry forward” most technologies between model years, tending to concentrate the application of new technology to vehicle redesigns or mid-cycle “freshenings,” and design cycles vary widely among manufacturers and specific products. Comments by manufacturers and model peer reviewers strongly support explicit year-by-year simulation. Year-by-year accounting also enables accounting for credit banking (i.e., carry-forward), as discussed above, and at least four environmental organizations recently submitted comments urging the agencies to consider such credits, citing NHTSA's 2016 results showing impacts of carried-forward credits.
Moreover, EPCA/EISA requires that NHTSA make a year-by-year determination of the appropriate level of stringency and then set the standard at that level, while ensuring ratable increases in average fuel economy through MY 2020. The multi-year planning capability, (optional) simulation of “market-driven overcompliance,” and EPCA credit mechanisms (again, for purposes of modeling the CAFE program) increase the model's ability to simulate manufacturers' real-world behavior, accounting for the fact that Start Printed Page 43162manufacturers will seek out compliance paths for several model years at a time, while accommodating the year-by-year requirement. This same multi-year planning structure is used to simulate responses to standards defined in grams CO2/mile, and utilizing the set of specific credit provisions defined under EPA's program.
(a) Representation of Manufacturers' Production Constraints
After the light-duty rulemaking analysis accompanying the 2012 final rule that finalized NHTSA's standards through MY 2021, NHTSA began work on changes to the CAFE model with the intention of better reflecting constraints of product planning and cadence for which previous analyses did not account.
(b) Product Cadence
Past comments on the CAFE model have stressed the importance of product cadence—i.e., the development and periodic redesign and freshening of vehicles—in terms of involving technical, financial, and other practical constraints on applying new technologies, and DOT has steadily made changes to both the CAFE model and its inputs with a view toward accounting for these considerations. For example, early versions of the model added explicit “carrying forward” of applied technologies between model years, subsequent versions applied assumptions that most technologies will be applied when vehicles are freshened or redesigned, and more recent versions applied assumptions that manufacturers would sometimes apply technology earlier than “necessary” in order to facilitate compliance with standards in ensuing model years. Thus, for example, if a manufacturer is expected to redesign many of its products in model years 2018 and 2023, and the standard's stringency increases significantly in model year 2021, the CAFE model will estimate the potential that the manufacturer will add more technology than necessary for compliance in MY 2018, in order to carry those product changes forward through the next redesign and contribute to compliance with the MY 2021 standard. This explicit simulation of multiyear planning plays an important role in determining year-by-year analytical results.
As in previous iterations of CAFE rulemaking analysis, the simulation of compliance actions that manufacturers might take is constrained by the pace at which new technologies can be applied in the new vehicle market. Operating at the Make/Model level (e.g., Toyota Camry) allows the CAFE model to explicitly account for the fact that individual vehicle models undergo significant redesigns relatively infrequently. Many popular vehicle models are only redesigned every six years or so, with some larger/legacy platforms (the old Ford Econoline Vans, for example) stretching more than a decade between significant redesigns. Engines, which are often shared among many different models and platforms for a single manufacturer, can last even longer—eight to ten years in most cases.
While these characterizations of product cadence are important to any evaluation of the impacts of CAFE or CO2 standards, they are not known with certainty—even by the manufacturers themselves over time horizons as long as those covered by this analysis. However, lack of certainty about redesign schedules is not license to ignore them. Indeed, when manufacturers meet with the agencies to discuss manufacturers' plans vis-à-vis CAFE and CO2 requirements, manufacturers typically present specific and detailed year-by-year information that explicitly accounts for anticipated redesigns. Such year-by-year analysis is also essential to manufacturers' plans to make use of provisions (for CAFE, statutory and specific) allowing credits to be carried forward to future model years, carried back from future model years, transferred between regulated fleets, and traded with other manufacturers. Manufacturers are never certain about future plans, but they spend considerable effort developing, continually adjusting, and implementing them.
For every model that appears in the MY 2016 analysis fleet, the model years have been estimated in which future redesigns (and less significant “freshenings,” which offer manufacturers the opportunity to make less significant changes to models) will occur. These appear in the market data file for each model variant. Mid-cycle freshenings provide additional opportunities to add some technologies in years where smaller shares of a manufacturer's portfolio is scheduled to be redesigned. In addition, the analysis accounts for multiyear planning—that is, the potential that manufacturers may apply “extra” technology in an early model year with many planned redesigns in order to carry technology forward to facilitate compliance in a later model year with fewer planned redesigns. Further, the analysis accounts for the potential that manufacturers could earn CAFE and/or CO2 credits in some model years and use those credits in later model years, thereby providing anot