The price elasticity of the demand for gasoline has been extensively studied over the last 40 years, and for good reason. It is critical for determining gasoline tax rates and evaluating alternative policies that target the negative externalities associated with automobile use (pollution, road congestion, etc.). Continuing pressure to address climate change has prompted a variety of policy proposals around the world that use a carbon price or fee to reduce the demand for gasoline and other fossil fuels. Gasoline prices have also become increasingly volatile as a result of periodic shortages in available refining capacity and increased uncertainty in world oil markets. Understanding consumers' ability to respond to such price fluctuations is crucial for predicting the potential macroeconomic impacts of future petroleum supply disruptions and for evaluating the benefits of policy measures intended to mitigate these effects, such as the maintenance and use of strategic petroleum reserves or the use of temporary gasoline tax suspensions.
Interestingly, a number of recent empirical studies (including Hughes et al. 2008, Pock 2016, Small and Van Dender 2007, and Park and Zhou 2010) examining gasoline demand have concluded that demand has become highly price inelastic, at least in the short run. This lack of demand response suggests that more extreme price fluctuations may be necessary to balance markets following supply shocks, and that taxes and other price-based policy mechanisms may not be as effective in achieving desired consumption or pollution reduction goals. For example, Hughes et al. (2008) conclude that alternative measures like the Corporate Average Fuel Economy standards used in the US may be necessary to meet achieve desired reductions while avoiding politically unfeasible tax levels.
Despite the importance of accurate estimates of the price elasticity of gasoline demand, the data generally available for such analysis tend to be highly aggregated and imprecisely measured. Many studies utilise monthly, quarterly, or even annual aggregate proxies of gasoline usage and average prices, often from a single national time series. In reality, individuals make gasoline consumption decisions on a daily basis, responding directly to the gasoline prices observed in their local area on that day. Empirical models relating monthly or annual gasoline volumes to average prices across broad geographic areas necessarily aggregates these different consumption decisions and are likely to mask a significant share of the response by consumers to a local price change. Moreover, the use of highly aggregated data generally requires strong assumptions that restrict the demand relationship from varying across locations or over time. As a result, unobservable location- and time-specific factors in the underlying customer-level demand function have the potential to bias elasticity estimates. Perhaps not surprisingly given such challenges, these aggregate studies have produced a wide range of different estimates of demand elasticity. Academic and government studies evaluating potential policy interventions in gasoline markets frequently rely on estimates from this literature or adopt the same problematic methods to obtain estimates, despite the fact that the elasticity values can often substantially impact predicted policy outcomes.
In a recent paper, we utilise data on citywide gasoline expenditures and daily gasoline prices from 243 US cities to analyse the impact of daily prices on daily gasoline demand (Levin et al. 2016). Our citywide daily gasoline expenditures are obtained by aggregating customer-level credit purchases at gasoline stations during that day. This gives us a direct measure of gasoline consumption that results from customers facing that day’s gasoline price. First, we leverage the higher frequency and greater geographic detail of our consumption and price data to obtain a more robust estimate of gasoline demand, avoiding the potential biases that can arise in more aggregate studies. Second, we derive a decomposition identifying the different sources of bias that arise in more aggregate models and then examine the relative magnitudes of these different biases by estimating demand models at varying levels of data aggregation.
Our findings reveal that gasoline demand may be significantly more elastic than previously thought. We consistently obtain elasticity estimates that are roughly five times more elastic than those reported by other recent studies. We then aggregate our data over time and across cities to varying degrees, to estimate aggregate demand models similar to those commonly used in other studies. The resulting estimates become increasingly less elastic as the level of data aggregation increases. Estimating the model using our data aggregated to a national time series of monthly total expenditures and average prices results in elasticities that are indistinguishable from zero, suggesting that studies using aggregated data may substantially underestimate consumers' price responsiveness.
The results of our decomposition analysis reveal how the primary source of bias differs depending on the dimension and degree of aggregation. Observed biases are largest in time series models where time-period fixed effects can no longer be used to control for demand differences over time. Overall, the sources of bias identified in our decomposition and the magnitudes suggested in our aggregated regressions help to provide a more systematic explanation of why gasoline demand studies using different methodologies have often obtained vastly different price elasticity estimates.
Based on our analysis, we conclude that gasoline demand may be considerably more responsive to short-term price fluctuations than one might conclude based on the recent literature, and the estimates differ by magnitudes large enough to substantially impact subsequent policy evaluation or market analysis.
Consider, for example, studies evaluating the impacts of cap-and-trade policies, like Borenstein et al. (2015), who analyse expected permit prices under California's greenhouse gas emissions cap-and-trade programme. Their analysis relies directly on existing estimates of the price responsiveness of gasoline demand. Partly in response to recent estimates like those of Hughes et al. (2008), they adopt a rather inelastic value for gasoline demand elasticity, which may contribute to their overall prediction that the supply of emissions abatement will be relatively inelastic to permit prices. Acknowledging a greater price elasticity in the demand for gasoline (like that which we obtained in Levin et al. 2016) will reduce predicted GHG permit price levels and volatility.
More accurate elasticity estimates can also substantially impact the inferences one draws when evaluating the macroeconomic costs of gasoline and oil market disruptions and the benefits of policy responses like maintaining a Strategic Petroleum Reserve (SPR) that are intended to reduce these costs. If gasoline demand were significantly more elastic than previously thought, prices would likely increase by substantially less than would otherwise be predicted (by previous estimates) in response to an oil supply disruption, and the quantity of gasoline consumers would purchase at these prices would be substantially smaller. As a result, the overall macroeconomic displacement effect is likely to be much smaller than would have previously been predicted. In addition, if consumers have more elastic demand, the release of a certain volume of fuel from the SPR during a market disruption will not be as effective as a policy lever aimed at reducing price levels. By demonstrating a significantly larger price-responsiveness of gasoline demand, our elasticity results strengthen any argument in favour of eliminating or reducing the size of the SPR and signal an enhanced effectiveness for price-based mechanisms for reducing greenhouse gas emissions.
Having more robust and precise estimates of gasoline demand response and a clearer understanding of the sources of aggregation bias that can arise in this setting should help researchers and policy analysts to more successfully evaluate the reliability of existing estimates and to improve empirical design and identification in future studies.
Hughes, J E, C R Knittel, and D Sperling (2008), “Evidence of a shift in the short-run price elasticity of gasoline demand”, The Energy Journal, 29(1), 93–114.
Levin, L, M S Lewis, F A Wolak (2016), “High frequency evidence on the demand for gasoline”, NBER Working Paper No. 22345, June.
Park, S Y, and G Zhao (2010), “An estimation of US gasoline demand: A smooth time-varying cointegration approach”, Energy Economics, 32, 110–120.
Pock, M (2010), “Gasoline demand in Europe: New insights”, Energy Economics, 32(1), 54-62.
Small, K A, and K Van Dender (2007), “Fuel efficiency and motor vehicle travel: The declining rebound effect”, Energy Journal, 28(1), 25–51.