How do Consumers Respond to Gasoline Price Cycles?

• Author: Byrne, David P.

1. INTRODUCTION

   The continued rise in gasoline prices over the past decade has seen anti-trust authorities increase their scrutiny over how companies set prices, and has heightened interest over the nature of competition in gasoline markets. A key finding from various policy and academic studies on gasoline pricing is that retail prices exhibit asymmetric cycles in many markets. (1) Figure 1 depicts an example of a cycling market from Kingston, Canada. In cycling markets like Kingston, retail prices experience large infrequent price jumps or restorations, and an undercutting phase that consists of small daily price cuts between consecutive price restorations.

   While a number of researchers have examined retailers' pricing behavior in coordinating restorations and undercutting prices, there are virtually no empirical studies on whether or how consumers respond to price cycles. More generally, there is little evidence on how consumers respond to daily fluctuations in gasoline prices. The disproportionate amount of supply-side research reflects the fact that while daily retail price data is available to researchers, daily data on volumes of fuel sold at a station or in a market are generally confidential and unavailable to researchers. The lack of demand-side research is unfortunate given its relevance for policymakers concerned with consumer welfare and demand-side sources of market power in retail gasoline markets.

   In this paper, we develop a novel empirical study of how consumers respond to gasoline price cycles. The study proceeds in two parts. We first examine firms' pricing behavior in cycling gasoline markets and provide new evidence that dominant retailers engage in price leadership in these markets. We then use a novel measure of daily market-level consumer responsiveness to study how consumers respond to daily fluctuations in price levels and dispersion arising from firms' pricing behavior, and to test for consumer stockpiling and search behavior.

   Our analysis uses a new dataset of daily, station-level prices from Ontario, Canada from 2007-2008. The owners of GasBuddy Organization Inc. (GasBuddy), who run the most popular online gasoline price reporting websites in North America, provided us with these data. Users of these websites upload stations' prices from their local markets via the internet using mobile devices and computers. The dataset consists of every station-level price report submitted to GasBuddy's websites over the sample period.

   In the first part of the paper, we study firm-specific pricing behavior, and its impact on daily retail price levels and dispersion in cycling markets. We show that stations run by major vertically-integrated brands systematically take on a leadership role in coordinating price restorations. In contrast, we find independent (non-branded) stations do not engage in such price leadership. We further show that branded retailers not only influence the level of prices when coordinating restorations, but also station-level price dispersion. In particular, market-level price dispersion collapses after a restoration price jump, and gradually rises as price undercutting ensues.

   In the second part of the paper, we analyze how consumers respond to these daily fluctuations in price levels and dispersion. Like previous researchers, we do not have access to data on volumes of fuel sold for our analysis. Instead we exploit the fact that our data contain every price report submitted to GasBuddy's websites over the sample period. This allows us to construct a daily, market-specific measure of demand responsiveness in terms of the number of price reports these websites receive from price spotters for a given date and market. This price reporting-based measure of responsiveness essentially assumes that consumers are more actively shopping and sampling gasoline stations' prices on days when GasBuddy's websites receive more price reports for a given market.

   The main advantage of this measure is we can match it to daily, market-level changes in price levels and dispersion. Such disaggregation is particularly helpful in studying consumer demand responses in cycling markets where retail prices exhibit large fluctuations on a daily basis. The main shortcoming of the measure is it only captures demand responses of consumers who actively report gasoline prices to websites. These individuals are unlikely to be typical consumers, and as such we interpret our results as corresponding to the behavior of highly informed/attentive consumers.

   We use regressions to study how price reporting intensity varies with the level and dispersion of prices over the cycle. We find reporting intensity experiences a dramatic, statistically significant increase just before and during restorations, a period where both price levels and dispersion also rise dramatically. (2) The baseline estimates further establish that daily price dispersion is the main determinant of price reporting intensity across different days of the cycle.

   We consider two non-mutually exclusive mechanisms in interpreting these patterns: stockpiling and consumer search. (3) The baseline results alone, while interesting, are not overly helpful in determining whether one or both of these mechanisms drive price reporting behavior. Forward-looking consumers could monitor daily changes in price dispersion, which varies in predictable ways over the cycle, to time their fuel purchases at the bottom of the cycle. Similarly, a costly search model would predict a rise in price reporting around restorations since the marginal benefit to searching for and reporting prices is higher when prices are more dispersed, and consumers are more likely to find a deal.

   We try to disentangle stockpiling from search-based incentives in price reporting by investigating heterogeneous relationships between reporting intensity, price dispersion, and positive and negative price changes by day of the cycle. Interestingly, we find reporting intensity is strongly linked to price dispersion on days just prior to and during restorations; however, no such relationship exists the day immediately after a restoration. This can be explained by a model of stockpiling and demand accumulation: if GasBuddy's price spotters tend to time their purchases and fill their cars' fuel tanks at the bottom of the cycle, then one day after a restoration they will not be actively shopping for fuel, and hence not monitoring price dispersion to make another well-timed fuel purchase. In contrast, a search-based explanation of price reporting alone has a difficult time explaining why reporting intensity and price dispersion has a strong positive relationship on some days of the cycle but not others.

   Our analysis also reveals heterogeneous relationships between price reporting and lagged retail price changes that further point to dynamic demand behavior. On most days of the cycle, price reporting is largely unrelated to lagged positive and negative retail price changes. There are, however, two key exceptions: on days when restorations are initiated, reporting intensity rises when lagged retail prices start to rise, or when lagged price cuts become increasingly small. We document that both small price increases and smaller price cuts during the undercutting phase of the cycle tend to signal a restoration price jump is about to occur. As such, we see these findings as being consistent with forward-looking consumers who use this information to anticipate restorations, and time their purchases at the bottom of the cycle.

   As a final test for distinguishing between stockpiling and search-based explanations of price reporting, we examine price reporting in the context of rural gasoline markets. These markets are well-suited for testing stockpiling behavior for two reasons. First, price restorations and undercutting phases are present in rural markets, as well as periods of price rigidity following restorations; in short, these markets exhibit "slow" price cycles. Thus, consumers in these markets have particularly strong incentives to time their purchases and stockpile fuel prior to restorations to avoid paying higher prices in the future. Second, the majority of rural markets in the sample have five or fewer stations that tend to be located in close proximity to each other. As such, there is little scope for search-based incentives to drive price reporting in these markets. Therefore, if stockpiling plays no role in price reporting, we should not find the jump in reporting intensity around restorations in rural markets that we found in the sample's (faster) cycling markets. Our empirical findings show this is not the case: price reporting exhibits a large, statistically significant rise when restorations are initiated in rural markets. The evidence further supports the hypothesis that dynamic demand incentives drive price reporting.

   As we discuss in the paper's conclusion, our results have implications for price transparency policies that provide consumers with web-based information on daily, market-level retail price fluctuations; such policies have recently been considered and enacted by anti-trust authorities and policy makers in Australia and Canada. By providing some of the first "hard" empirical evidence of stockpiling behavior, this paper emphasizes the importance of policies that help consumers make well-timed retail fuel purchases. In this way, these policies can yield consumer welfare gains in gasoline markets, particularly in those with price cycles (Noel, 2012).

   1.1 Related Literature

   The paper contributes to a large empirical literature on consumer demand in retail gasoline markets. Previous researchers have examined a number of issues such as identifying price elasticities with low frequency (monthly, quarterly) data (Small and Van Dender, 2007; Hughes, Knittel, and Sperling, 2008) and high (daily) frequency data (Lewis, Levin, and Wolak, 2012), spatial differentiation (Houde...