Vehicle Miles (Not) Traveled: Why Fuel Economy Requirements Don t Increase Household Driving Jeremy West: MIT Mark Hoekstra: Texas A&M, NBER Jonathan Meer: Texas A&M, NBER Steven Puller: Texas A&M, NBER, E2e TE3 Conference October 30, 2015
The energy rebound effect Energy efficiency (fuel economy) reduces the cost per unit of energy service (price per vehicle mile traveled) So, as fuel economy improves, do people increase their VMT? Extent of rebound has stark implications when policy operates primarily by fuel economy mandates rather than fuel use taxes Magnitude is an empirical question: Jevons paradox (1865) and large subsequent literature quantifying energy takeback
Literature approaches to estimating the rebound effect 1. Correlate household fuel economy with VMT Highly endogenous: people select which vehicles they purchase based partly on their VMT major selection problem Even with panel data: people likely consider their anticipated VMT in deciding which new vehicle to purchase 2. Exploit plausibly (more) exogenous variation in fuel prices to identify VMT elasticity to price per mile More appropriate if estimating effect of gasoline tax 1 Gasoline prices affect price per mile but not utility per mile In contrast, fuel economy is highly correlated with other, utility-providing, vehicle characteristics 1 Caveats by Li, Linn & Muehlegger (2014) and Coglianese, Davis, Kilian, & Stock (2015) acknowledged.
Mr. Tufteland assumed he would dearly miss the 45 miles per gallon economy of his Honda [Civic hybrid], but said the trade-offs ultimately were not worth the modest fuel savings. I wish I had that extra 12 miles per gallon, but I d still rather have the Forester any day, he said. The fuel costs aren t that different, and the Subaru is so much more functional in terms of space, comfort and the ability to get me anywhere. It s really the perfect car for me.
Higher fuel economy accompanies reduced performance
Higher fuel economy accompanies reduced safety/comfort
Higher fuel economy accompanies reduced book value
Illustration of the two components of policy-induced improvement in fuel economy
The ideal experiment Randomly induce households to purchase vehicles that differ in fuel economy and in all correlated aspects Don t want exogenous change in fuel prices Want an exogenous change in vehicle choices Assess subsequent household VMT, accounting for potential rebound in new vehicles and substitution within HH fleet The truly policy-relevant parameter accounts for both standard rebound and attribute-based adjustment
Cash for Clunkers eligibility as quasi-random experiment Households with a clunker at 18 MPG were eligible; households whose clunker was 19 MPG were ineligible Had to register & insure clunker for one or more years, ruling out common forms of manipulation about cutoff National Household Travel Survey supports that barely eligible households are identical to barely ineligible households Among purchasers in our data, barely eligibles are similar to barely ineligibles in pre-treatment fleets, purchases, and driving behavior In summary, this appears to be a clean quasi-experiment
Empirical procedure: regression discontinuity designs Cash for clunkers eligibility quasi-randomly (directly) affected: Timing of purchase pull-forward Characteristics of purchase: fuel economy and correlated vehicle attributes Empirically, we need to: 1. Identify the clunker for each household vehicle most likely to be traded 2. Determine counterfactual purchase timing for each eligible household 3. Holding purchase timing equal i.e. when the pull-forward window ends determine counterfactual vehicle characteristics for eligible households purchases 4. Determine counterfactual subsequent VMT for eligible households i.e. policy-relevant rebound
Data Texas DMV registrations Names and addresses used to determine household fleets Includes purchase dates for new vehicles VIN-decoder (from DataOne Software) used to determine vehicle attributes such as fuel economy Texas emissions test records Required annually at two years since sale in 17 Texas counties (60% of population) Record odometer, from which we determine VMT Determine subsidized transactions under CfC using federal CARS data from NHTSA Link everything at vehicle level by VIN17 Texas is representative
Purchase timing: what is the pull-forward window? 1. Identify clunker for each household, the vehicle most likely to be traded-in We use household s oldest vehicle in model years Matching to actual subsidies shows this approach to be very similar to more sophisticated p-score models, while requiring less completeness of data 2. Identify counterfactual purchase timing for eligible households purchases Discontinuities in purchase frequency for all households in Texas (or just those in the 17 counties with smog check) Objective is to determine time window at which pull-forward is completed
Cumulative fraction purchasing new vehicle: CfC only
Cumulative fraction purchasing new vehicle: CfC+7 mo.
Cumulative fraction purchasing new vehicle: CfC+8 mo.
Cumulative fraction purchasing new vehicle: CfC+9 mo.
Cumulative fraction purchasing new vehicle: CfC+10 mo.
Distribution for counterfactual of subsidized purchases
New vehicle purchase characteristics Counterfactual: eligible households in these counties would have purchased a new vehicle sometime in 12 months beginning with CfC Results for vehicle characteristics are insensitive to the time window 3. Holding purchase timing equal i.e. when the pull-forward window ends determine counterfactual vehicle characteristics for eligible households purchases Fuel economy Other vehicle attributes
Fuel efficiency (miles per gallon)
All estimates are significant at p < 0.01
Book value (manufacturer suggested retail price)
Safety/comfort (curb weight)
Performance (horsepower-per-pound)
Effect on VMT Many vehicle characteristics change along with fuel economy Analogous to how CAFE standards affect new vehicle fleet 4. Determine counterfactual subsequent VMT for eligible households i.e. policy-relevant rebound Assess discontinuities in total household VMT and allocation across household vehicle fleet for multi-vehicle households
Annual total household vehicle miles traveled (VMT)
Annual total gallons of fuel consumed by household
Fraction of miles driven in newly-purchased vehicle
Conclusions Understanding the role of rebound is especially important as the U.S. increasingly uses CAFE standards as its primary transportation fuel policy Unlike gasoline prices (taxes), changes to fuel economy are accompanied by changes to other utility-providing vehicle attributes, at least in the short and medium run This study uses quasi-random variation in vehicle choices to estimate the policy-relevant rebound effect We find no evidence of increased VMT, and a reduction in fuel consumption, as an effect of increasing fuel economy along with changes to correlated vehicle attributes
Fuel economy in counties in study representative of U.S. Source: Busse, Knittel, Zettelmeyer (2012) Data