EMPIRICAL STUDIES OF GASOLINE RETAILING: A GUIDE TO THE LITERATURE

Abstract

Abstract This paper surveys the empirical literature on gasoline retailing, which has been growing rapidly over the last three decades, possibly in response to antitrust and regulatory concerns and increased availability of pricing data. Studies of both pricing and non-price decision variables are considered. In general, it is found that crude oil prices are the primary driver of national price movements over time. However, market structure has been identified as playing a role in price dynamics, equilibrium selection and price differentials across markets and stations. The economic literature emphasizes the importance of heterogeneity across stations and coordination problems faced by retailers. Several directions for future work are suggested, including the development of theory and demand estimation using high-frequency station level data.

1. Introduction

Few industries have been subject to as much regulatory and antitrust scrutiny as gasoline retailing. Allegations of collusive behaviour have lead to many government investigations and inquiries in different countries, and ultimately, in some cases, charges and convictions.¹ At the other extreme, allegations of predatory pricing have been common, and have resulted in controversial policies and policy proposals.² Mergers at the retail, wholesale or refiner level have been frequent and have attracted antitrust attention. In the USA, the U.S. Government Accountability Office (2008) reported that from 2000 to 2007, there were more than 1000 mergers in the petroleum industry in the USA, with approximately 18% at the downstream level. In Canada, the Competition Bureau reports that at least 12 merger and acquisition transactions in Canada involving gasoline retailing were subject to antitrust scrutiny since 1986. Petroleum industry mergers have been subject to attention from competition authorities elsewhere.³

Due likely in part to this regulatory and antitrust attention, a large empirical literature studying gasoline retailing has developed. Since 2000 alone, over 75 empirical studies of gasoline retailing have been published in English language academic journals, with many more studies existing in working paper form or as reports issued by governments or other agencies or institutes. Data for many different countries and time periods have been employed to address a large range of questions, focusing both on price and non-price retail decisions. Data sets used have ranged in aggregation from national to station level, and in frequency from annual data to multiple observations per day.

Underlying most of these studies are a small number of broad and related issues about the gasoline industry, which are indeed common to much of the empirical literature in industrial organization. What are the most important determinants of retail gasoline prices? To what extent is market power exercised in retail gasoline markets? Upon what factors does the degree of market power depend? Is the behaviour exhibited by gasoline retailers indicative of anticompetitive conduct? What are the features of appropriate government policy in gasoline markets, and upon what should policy depend?

The purpose of this paper is to provide an overview or roadmap of this literature, in terms of the questions examined, data sets used, methodologies adopted and conclusions reached.⁴ This exercise is motivated by the vast size of the literature, along with the policy focus on the gasoline industry. From this review, a basic economics of gasoline retailing will emerge, which considers the main economic drivers of retail gasoline prices, the efficiency of retail gasoline markets and the role for government intervention and policy.

This paper will proceed as follows. Section 2 discusses the basic structure of the gasoline retailing industry, the data sets employed by researchers and the basic features of the empirical literature. Section 3 discusses the literature on retail gasoline pricing dynamics, whereas Section 4 considers empirical work examining the determinants of price levels and price dispersion. Section 5 examines the literature on non-price choice variables. Section 6 concludes and discusses potential directions for future research.

2. Overview of Gasoline Retailing and the Empirical Literature

2.1 Industry Structure

In general, gasoline is produced by the refining of crude oil, which also yields a range of other products, including diesel, jet fuel, lubricants and heating fuels. The gasoline typically travels (by pipeline, barge or rail) to distribution terminals, and then by truck to individual gasoline stations. See, for example, Borenstein et al. (1997) and Kleit (2005) for details on gasoline production and distribution in the USA, and Conference Board of Canada (2001) for a description of the Canadian industry.

Certain simplicities in gasoline production and distribution and in the nature of the product may contribute to the large number of empirical studies. In particular, researchers often will treat the price of crude oil as the primary input price of interest in the determination of wholesale gasoline prices; retail prices in a market will be then related to an available wholesale price for the market. As well, studies make reference to the homogenous nature of gasoline as a feature that makes it attractive for study; for example, Barron et al. (2004a, p. 1052) state, regarding theories that assume homogeneous products, that ‘gasoline markets appear to be nearly ideal for testing the theories because the physical attributes of regular unleaded gasoline are essentially identical across spatially differentiated sellers’.

Despite this perceived simplicity, complexities exist. Although studies of retail price determination will often make use of a single wholesale price variable, gasoline is in fact supplied to stations at different prices. Gasoline can be purchased at both a branded or unbranded rack price, referring to whether the refiner's brand name can be attached to the product at retail. Certain retailers will be given discounts off of posted rack prices, based on volumes and time commitments. Gasoline may be distributed to stations by the refiners or through independent distributors (jobbers).

As a further complication, gasoline stations have many different contractual arrangements with suppliers. Slade (1998), in the Canadian context, lists four different vertical arrangements between a supplier and a station displaying its brand of gasoline: (1) company owned and operated stations, for which price is set by the company and the station is operated by a salaried employee; (2) commissioned agent stations that are owned by the company which also sets the price, but which are run by a station operator who receives a commission based on volume; (3) lessee dealer stations, owned by the company but leased to a dealer who sets the price and owns the gasoline inventory and (4) dealer owned stations, which set their own price. Because of these different contractual arrangements, it cannot be assumed that a station bearing a particular brand has its retail price directly controlled by the supplier. As well, the incentives of station operators are expected to differ across stations within the same brand according to contractual arrangement.

Finally, although the gasoline sold at different stations may be largely the same, there may be perceived product differences by consumers, and gasoline may differ according to additives. As well, gasoline stations are not single product establishments; rather, gasoline is one of a long list of products sold by most gasoline stations. Other products may include car washes, automotive services and grocery and convenience store items, including tobacco products. In some cases, gasoline stations may be associated with major grocery or big-box retailers. As a result, gasoline stations will be differentiated according to the offerings of other products and services, as well as by geographic location. Further evidence suggests that queuing times influence consumer decisions (see, e.g. Png and Reitman, 1994).

2.2 Data Sources

The availability of data may also contribute to the large number of empirical studies on gasoline retailing. Some government agencies, due likely to ongoing competition concerns, make retail prices, at least at the level of city or state, freely available on a weekly or monthly frequency. Daily city or even station level prices are in some cases available from governments (see Wang, 2009a) or can be purchased for research purposes (see, e.g. Lewis and Noel, 2011). As well, some researchers have taken advantage of the fact that gasoline prices are typically posted on large billboards visible from the street, and have collected their own data sets of high-frequency retail prices (see, e.g. Noel, 2007b; Atkinson, 2009). Prices reported to consumer websites have also been used; Atkinson (2008) discusses the strengths and sample selection issues arising from such data.

Studies of gasoline retailing have also benefited from the availability of necessary explanatory variables. Wholesale (rack) prices and crude oil prices are commonly available. Gasoline station locations and physical characteristics are frequently available for purchase from data-collection agencies, or can be self-collected for small samples. In some areas, information on vertical contracts is available but in other cases such information has not been obtainable. Perhaps the most difficult data to obtain, particularly for studies at the station level, concern output. Although some studies have used high-frequency station level volumes obtained through the cooperation of the firms (see, e.g. Slade, 1992; Wang, 2009b),⁵ other studies at the station level have either estimated reduced form equations, or have used volume data for a longer time period (such as quarterly or bimonthly; see, for example, Houde, 2009).

2.3 Basic Features of the Empirical Literature

Although part of the intent of this paper is to provide a broad overview of the vast empirical literature on gasoline retailing, some limits needed to be placed on the literature being reviewed. First, attention was restricted to papers published since 1980. Papers focused solely on the demand for gasoline were excluded, as were studies considering the impact of gasoline prices on other sectors of the economy; the focus of this survey is on studies of the way in which gasoline is sold at retail, including price and non-price decisions by firms (for overviews of results concerning price elasticities, see Espey, 1998; Brons et al., 2008). There are many interesting government studies and working papers on the subject of gasoline retailing, some of which will be referenced in later sections of the paper as relevant; however for the purposes of this section, I restricted my attention to articles published in academic journals. Practical considerations required me to consider primarily papers published in English (one French language article is included). Finally, studies of upstream oil and gasoline markets were excluded, except for those that focused on the impact of the upstream market on retail prices.

Papers were identified by searching databases of journals and the websites of the journals themselves, and through the reference lists of articles I had found. Overall, 102 articles published in academic journals were included.⁶ It is to be expected that with a literature of this size, certain articles have been overlooked; however, it is likely that the basic lessons that can be derived from the literature emerge from the database collected.

The articles can be broken down along several dimensions. Much of the literature is very recent, with 79 of the 102 studies collected being published in the 2000s. A review of articles from the 1980s suggests that the focus in that period was on demand estimation as opposed to seller behaviour. The focus of the collected studies is largely on North America; 60 of the 102 studies used data for the USA, whereas Canada was in second place with 25 articles using Canadian data.⁷ The articles are published in 43 different journals, with industrial organization and energy field journals tending to publish the most articles on gasoline retailing per journal.⁸

It is desirable to give a broad description of how this literature breaks down according to topic. Any such breakdown, however, is necessarily subjective, because there are many different categories that can be considered, and many papers will fall under multiple categories. To give a basic description of the breakdown of the literature, Table 1 counts the number of published papers having as a primary purpose to address questions in each of the following categories: crude oil or wholesale price passthrough; Edgeworth cycles; the impacts of mergers or regulation and price dispersion and price differentials across individual stations. Several articles were counted in multiple categories (such as papers examining upstream cost passthrough within Edgeworth cycle markets), and 23 articles were not easily placed in any of these categories. The latter group included articles studying other aspects of price dynamics, or articles looking at average price differences across markets without a regulatory or merger focus, and prices looking at non-price retail decisions.

Table 1.  Articles by Category.

Category	Number of Articles
Passthrough of upstream cost shocks/ response asymmetry	26
Edgeworth cycles	15
Merger/regulatory impact	24
Station level price dispersion and price differentials	24

The first group of papers examines the response of gasoline prices to upstream (crude oil and wholesale gasoline) prices, with many of these papers testing for an asymmetric response to upstream prices. In addition, there were 15 papers with a central focus on Edgeworth cycles in gasoline markets; including the selection of papers examining other aspects of dynamic pricing, we find that almost half of the papers surveyed are focused on issues of price dynamics. Twenty-four papers look at the impacts of mergers or regulations (including studies of tax incidence). Twenty-four articles focus on price differentials across individual stations and price dispersion using station level data.

Finally, the studies can be categorized according to the data sets employed. Although most studies examine retail pricing or use retail price data, some focus on non-price competition, studying, for example, the choice of location, vertical contract or station count. Seventeen of the studies make use of cross-section data, 18 use a single time series and the rest rely on panel data.⁹ Of the time series and panel studies, 23 rely on data with a daily frequency, 17 on weekly and 23 on monthly data; the rest use irregular panels, or other data frequencies. Station level data was employed for 56 studies, with most of the remaining studies employing data at the city, state/province, or country level.

The remainder of this paper will discuss the economic literature on gasoline retailing in more detail. Attention is given first to studies of pricing dynamics, followed by a discussion of price levels, price dispersion and the impacts of mergers and regulation. Finally, the literature on non-price choice variables is considered.

3. Studies of Retail Gasoline Price Dynamics

As demonstrated in Section 2, much of the empirical literature on retail gasoline pricing is concerned with understanding how and why prices change over time. Although much of the focus is on the response of retail prices to upstream price changes, the literature also examines price dynamics driven by other factors.

3.1 Passthrough/Asymmetric Response

The empirical literature examining how shocks to crude oil prices and wholesale gasoline prices are passed through to retail prices is vast. Indeed, these articles are part of an even larger literature examining passthrough of upstream cost shocks in a variety of industries; see Frey and Manera (2007) and Meyer and von Cramon-Taubadel (2004) for surveys of the broader literature which focus on the technical issues of modelling and estimation. Because this literature has been surveyed elsewhere, this paper provides only a general overview.

The primary focus of this literature is on whether gasoline prices respond asymmetrically to positive and negative shocks to upstream (crude oil or wholesale) prices. Many studies use time series data to estimate some variation of an error correction model such as the following:

• 

where p_t and c_t denote the retail and upstream price, Δ denotes first difference, equals the jth lag of the first difference of x if that first difference is positive and zero otherwise, is similarly defined for negative first differences and E_t is an error correction term, which measures the deviation between the retail price and its long run equilibrium in relation to the upstream price. μ_t is a random term. Importantly, such a model permits current changes to the retail price to depend differently on current and past changes to upstream prices, and on past changes to retail prices, depending upon whether such changes were positive or negative. See Borenstein et al. (1997) for an early application of a model of this form to gasoline pricing.¹⁰

Although some variation in results has been observed, and although debate on the appropriate model and econometric approach has raged (see Douglas, 2010, for a recent example challenging past findings), most studies, for different jurisdictions, have found at least some statistical evidence of asymmetry in the response of retail prices to upstream (wholesale or crude oil) prices. In particular, many studies have concluded that, for at least some markets in their sample, retail prices respond differently, and typically faster, to upstream price increases than to decreases. There is evidence that such a pattern is common across other industries as well. See, for example, Peltzman (2000).

Perhaps where many of these studies have been lacking is in their theoretical motivation.¹¹ Few of the studies offer more than a couple of paragraphs about the economic theory that could explain passthrough asymmetry, and almost none make reference to or present formal models. Borenstein et al. (1997) offer two possible explanations for asymmetry at the retail level based on tacit collusion and consumer search, and suggest either could explain the asymmetry they identify in the USA.

The perception of asymmetric movements in gasoline prices, with faster increases than decreases, has often led to claims of collusion, followed by scrutiny from competition authorities.¹² However, little formal theory is offered in support of a connection between collusion and asymmetric response. Borenstein et al. (1997) hypothesize that when upstream prices fall, previous retail prices may act as focal points around which retailers can collude, and suggest that such response asymmetry would be consistent with a price setting variation of the Green and Porter (1984) model of price wars, as presented in Tirole (1988).

The theoretical explanation for response asymmetry that has received the most formal attention has been a consumer search explanation; models in which search results in some form of asymmetric price response are provided in Lewis (2010), Yang and Ye (2008), Tappata (2009) and Cabral and Fishman (2011). In these models, the willingness of consumers to search is influenced by the magnitude of observed price changes, or knowledge of past cost changes.

As an example, in the search model developed by Tappata (2009), marginal cost (which is common to all firms) can be high or low in a given period, and is known to the firm but not to consumers. Each consumer wishes to purchase one unit of the good in a period, provided the total cost does not exceed some reservation price. Consumers each period choose whether to engage in costly search (paying the search cost, which differs by consumer, yields the lowest price in the market), whereas firms each period set their price. In the static model, in equilibrium, some fraction of consumers engages in search, whereas firms exhibit price dispersion. Notably, a higher fraction of consumers search when cost is low than when it is high, because the gains to search are greater under low costs.

Now consider the dynamic model, and suppose that marginal cost exhibits persistence – if costs are low, they are more likely to be low next period, and vice versa. Suppose that costs were originally high, so that consumers expect them to be high in the current period. As a result they are expected to engage in little search. If costs do indeed fall, little of that reduction will be passed through immediately into price, because consumer demand is inelastic (because most consumers choose not to search, and accept a randomly chosen firm). On the other hand, suppose that marginal cost is originally low, so that many consumers search. If costs increase, much of this increase will be passed on. The key is that because the search intensity of consumers, and hence the elasticity of demand faced by firms, depends on past marginal costs, the incentive to pass through cost change will depend upon the direction of change. Although in other models, the mechanism leading to asymmetry differs, the basic principle is the same.

Some empirical studies of response asymmetry in gasoline markets have examined theoretical explanations. Johnson (2002) argues that search costs are lower for diesel purchasers than for gasoline purchasers, and so according to a search argument, greater asymmetric response should be expected in gasoline prices than diesel prices, which is indeed what he finds. Lewis (2010), in a study of weekly station level data in San Diego and city level prices in Los Angeles, finds evidence in favour of a search-based explanation for response asymmetry. In contrast, Verlinda (2008), Balmaceda and Soruco (2008) and Deltas (2008), using station level data in Southern California and Chile, and state level US data, respectively, all present evidence consistent with market power being associated with response asymmetry, as may be expected under a collusive explanation. Borenstein and Shepard (1996) use a monthly panel of prices for US cities to estimate a asymmetric-adjustment model of retail margins and test predictions taken from the collusive models of Rotemberg and Saloner (1986) and Haltiwanger-Harrington (1991). The authors find evidence consistent with tacit collusion.

In summary, an area within the response asymmetry literature where productive work is still to be done is in developing further the theoretical explanations of response asymmetry, and in determining which explanations best fit observed pricing in retail gasoline markets. At the moment, evidence exists consistent with both of the general theories discussed above.

3.2 Edgeworth Cycles

A smaller but rapidly growing area of the literature on gasoline retailing has also been focused on price dynamics, but has been interested in dynamics not driven by upstream prices. This literature dates back at least to Castanias and Johnson (1993), who document a peculiar phenomenon observed in retail prices in Los Angeles over the 1968–1972 period; this pattern was also observable in other cities over the period (Allvine and Patterson, 1974). Using weekly city level price data, Canstanias and Johnson demonstrate an asymmetry in retail price movements: retail gasoline prices would rise quickly during a single week by large amounts, declining slowly to the original level over several weeks, in an asymmetric cycle not observed in the relevant wholesale price. Castanias and Johnson argued that this pattern closely resembled the ‘Edgeworth cycle’ dynamic pricing equilibrium proposed by Edgeworth (1925) and formalized in Maskin and Tirole (1988).

Since Castanias and Johnson (1993), there has been an explosion of empirical work documenting Edgeworth cycles in gasoline pricing, and testing certain hypotheses about them. Such cycles have now been shown to exist in the Midwest USA (Lewis 2009a; Doyle et al., 2010; Zimmerman et al., 2010), Canada (see, e.g. Eckert, 2002, 2003; Noel, 2007a, 2007b; Atkinson, 2009), Australia (Wang, 2008, 2009a and Europe (Foros and Steen, 2008). An overview of methods of identifying Edgeworth cycles can be found in Zimmerman et al. (2010). These pricing patterns have also been the subject of intense scrutiny by antitrust authorities and other government agencies, and have even resulted in enforcement action (see Wang, 2008, for an overview of an Australian case).

As an example of the asymmetric cycle observed in certain cities, Figure 1 plots daily retail and wholesale prices for the city of Guelph, Ontario, over the month from 24 October to 24 November, 2005.¹³ The retail price is the mode price for all Guelph stations each day at noon. The wholesale price used is the average rack price for the city of London, Ontario, the closest terminal with posted prices. The existence of a retail price cycle not present in the rack price is clear, as is its asymmetric nature. The average magnitude of retail price restorations is 7.5 cents per litre. Other graphs illustrating Edgeworth cycles in different cities can be found in most of the papers listed in the previous paragraph.

Figure 1. Edgeworth Cycles in Guelph Ontario Mode Retail Prices: 24 October to 24 November, 2005.

With little exception, the theoretical foundation used in empirical studies of the asymmetric cycles observed in gasoline pricing has been the alternating move price setting model of Maskin and Tirole (1988). In this model, two firms produce homogeneous products, and have identical marginal costs that are constant over output. The model is one of infinite horizon and discrete timing. The demand function is the same every period. Firms alternate setting price, with one firm setting price in even periods and the other setting price in odd periods. A firm is committed to its price for two periods.

As an infinite horizon model, this setup permits many equilibria. The authors focus on Markov perfect equilibria, and identify two different classes of equilibria. In the first, the firms converge on a focal price, each matching the other at this price (which is bounded around the monopoly price) forever. The other class of equilibria, and the class of interest here, consists of what is referred to as Edgeworth cycle equilibria because of the resemblance to the cycles proposed by Edgeworth in settings with capacity constraints (although there are no such constraints in Maskin and Tirole, 1988). In such an equilibrium, there is no stable equilibrium price. Rather, in the Edgeworth cycle equilibrium of Maskin and Tirole (1988), firms alternate undercutting each other until eventually marginal cost is reached. At this point, a war of attrition is initiated, with each firm randomizing between setting marginal cost and restoring price to initiate a new cycle.

Intuitively, several features of the model contribute to the existence of such an equilibrium. The equilibrium relies on the assumption that a firm is committed to its price for some period of time, permitting its rival to undercut it and serve the entire market before such an undercut can be responded to. As well, product homogeneity (or at least limited product differentiation) is needed to create the undercutting incentive.

Some attempts to extend this model have been made, but, possibly because of its complexity, these attempts have been limited. Wallner (1999) considers a finite-horizon version of the Maskin and Tirole model, and demonstrates that asymmetric cycles are exhibited in equilibrium. Eckert (2003) relaxes the tie-breaking rule when firms match price, and found that existence of the focal price equilibrium and the nature of the Edgeworth cycle equilibrium are sensitive to the tie breaking rule. The result is interpreted as providing intuition on the roles of large and small firms, respectively. Noel (2008) uses computational methods to examine Edgeworth cycle equilibria in a setting in which marginal costs fluctuate randomly over time. In this setting, he uses the model to explore robustness of the cycle equilibria to changes in demand elasticities and discount factors, to introduce capacity constraints and product differentiation, and to add a third firm. He finds that Edgeworth cycles continue to exist in equilibria provided product differentiation is not too great, and capacity constraints are not too tight, although particular features of the cycle can change. As well, he finds that Edgeworth cycle equilibria continue to exist with three firms, but that new equilibrium features emerge: in equilibrium, one can observe delayed price restorations and even false starts, where a restoration is attempted but subsequently abandoned.

As mentioned above, a primary contribution of the empirical literature has been to document that asymmetric cycles in prices, not related to observable upstream costs, exist in gasoline markets.¹⁴ Beyond that, the empirical literature has focused on the following questions:

• (a) 
  In What Market Settings are Such Asymmetric Cycles Most Likely to be Observed, and how do Market Characteristics Change the Nature of the Cycle?
  A focus of much of the empirical gasoline literature on this topic has been to identify what market conditions are most likely to yield Edgeworth cycles. Eckert (2003) and Noel (2007a), in examinations of a selection of Canadian cities, find that the existence of cycles is associated with lower market concentration or a greater prevalence of small retailers. Doyle et al. (2010), in an examination of US cities, find that cycles are most likely to be observed in cities with intermediate levels of concentration, and in markets with more independent stations that operate convenience stores. This result is argued to be consistent with stations who sell secondary goods having a greater incentive to undercut rival gasoline prices. Zimmerman et al. (2010), in another study of Edgeworth cycles in the USA, conclude that cycling is associated with a higher concentration of vertically integrated stations.
  How market characteristics affect the nature of the cycle is examined by Noel (2007a). The author examines weekly data for 12 Canadian cities, and employs Markov-switching regression techniques to model the switching of prices from price restorations to undercutting regimes, and from cyclic to non-cyclic regimes, as well as the determination of prices within the regimes. He finds that in markets with more firms, the undercutting phases are shorter, as suggested by theory. An increase in the number of small firms is also associated with an increase in the amplitude of the cycle. Faster cycles with less asymmetry are found in larger markets.
• (b) 
  How do Edgeworth Cycles Affect the Passthrough of Upstream Cost Shocks to Retail Gasoline Prices?
  Some researchers tie empirical work on Edgeworth cycles directly to the literature on asymmetric cost passthrough, by considering how the passthrough of shocks to wholesale gasoline prices is affected by the presence of Edgeworth cycles. Eckert (2002) argues that in the presence of an Edgeworth cycle, how and when a cost shock is reflected in retail prices will depend upon the current position of the cycle. If we suppose that price restorations are triggered when retail prices come within a certain proximity of marginal cost, then an increase in the wholesale price may be rapidly met with a retail price increase if it pushes the retail price below the threshold to initiate a new cycle. On the other hand, if the wholesale price falls, and if the magnitude of retail price undercuts is independent of the wholesale price, there would be no impact of the wholesale price drop on retail prices until the new and lower threshold was reached.¹⁵ As a result, an asymmetric response pattern can be observed.
  Noel (2009) decomposes the asymmetric passthrough into the component that is explainable by the existence of Edgeworth cycles, and the component that is not, using twice-daily station level data for Toronto. He finds that Edgeworth cycles explain some but not all of the observed asymmetry. Lewis (2009a) discusses how the presence of asymmetric retail price cycles influenced how retail prices in a city responded to the shocks from Hurricane Rita. Lewis and Noel (2011) examine daily price data for cities in the USA, and conclude that upstream price changes are passed through more quickly into retail prices in cities with cycles than in cities without Edgeworth cycles. They argue that Edgeworth cycles, with their constant price adjustments, facilitate a faster response to upstream shocks.
• (c) 
  What are the Strategies Employed by Individual Stations in a Market with Edgeworth Cycles?
  Some authors have employed high-frequency data for individual stations to analyse the behaviour of individual stations engaged in Edgeworth cycles. Noel (2007b) uses a Markov switching model to examine individual station behaviour, using a data set of retail prices observed twice daily for stations in the Toronto area. The author finds, as expected, that although small firms tend to lead price reductions, price restorations are lead primarily by larger firms.
  Atkinson (2009) uses individual station price data observed eight times daily in the Canadian city of Guelph over 100 consecutive days in summer and fall of 2005 to examine price leadership in Edgeworth cycles. He finds that two specific national major brands tend to lead price restorations. Heterogeneity is identified in the behaviour of independents; although one particular independent station is found to set the lowest price in the market, this behaviour is observed much less frequently by other independents. Further, Atkinson's (2009) description of price restorations using high-frequency data sheds some light on the coordination problem that may exist in price restorations (recall Noel's (2009) finding that with more than two firms, price restorations take longer and can fail). In Guelph, Atkinson (2009) identifies a tendency of the major brand leaders to establish the new restoration price mid afternoon, but to reduce price back to original levels until late evening. Such a pattern may allow the leaders to signal the beginning of a new cycle, without being priced above the market for an extended period.
  The coordination problem is addressed further by Lewis (2009b), who studies station level prices for 67 cities in the US Midwest, along with 2 month's worth of prices for a single brand collected three times daily. His findings point to restorations in a city being led by a single firm with a high degree of vertical integration. The two firms identified as restoration leaders are also strongly associated with the presence of cycles in a market. Foros and Steen (2008) consider recommended prices and price support systems as coordinating restorations in Norway.
  Wang (2009a) uses station level price data for Perth, Australia, to examine behaviour in Edgeworth cycles before and after the introduction of a government policy that requires retailers to set prices once per day, and to notify the government the day before of the next day's prices, effectively meaning that stations set prices simultaneously on a daily basis. Wang (2009a) concludes that whereas before the regulation a single larger firm leads restorations, once firms were forced to set prices daily and simultaneously, behaviour is consistent with firms playing mixed strategies, randomizing between restoring and remaining at the low price. A further interesting observation to be made, in light of the fact that the main theory of Edgeworth cycles assumes staggered price setting, is that Edgeworth cycles are observed even after firms are required to price simultaneously.
• (d) 
  Is Edgeworth Cycle Theory the Best Available Explanation for Observed Pricing?
  The question of alternative theories to explain observed asymmetric cycles has received only limited attention. Noel (2007b) considers, and provides arguments against, several possible alternatives: cycles in consumer demand, station level inventory explanations, unobserved cycles in discounts at the wholesale level and collusion. More recently, however, Houde and Clark (2011) argue that cyclic behaviour in certain Quebec markets, where retailers have pleaded guilty to price fixing, can be explained as the outcome of collusion among asymmetric retailers facing a price floor. One alternative, considered, for example, by Foros and Steen (2008) is that retail price cycles represent a form of intertemporal price discrimination. An example of a theoretical model in which such price discrimination results in price cycles can be found in Conlisk et al. (1984). One hurdle that the price discrimination explanation may face is the finding that Edgeworth cycles are less likely to be observed in highly concentrated markets, which may be the opposite of what is expected if Edgeworth cycles are a collusive strategy to facilitate intertemporal price discrimination.
• (e) 
  What are the Average Price Effects of Price Cycles?
  From a policy standpoint, perhaps the most interesting question is whether retail price cycles in gasoline markets are associated with higher prices, controlling for other factors. Maskin and Tirole (1988) would suggest that in fact, industry profits under equilibria without cycles can be higher; the sole renegotiation proof equilibrium involves firms matching each other at the monopoly price forever along the equilibrium path. Although the Edgeworth cycle equilibrium constructed by the authors will occasionally have prices at the start of the cycle that exceed the monopoly price, average profits over the cycle will be lower. However, because of the multiplicity of equilibria, more precise statements about prices and profits under cyclic and non-cyclic equilibria are difficult to make. Note as well that Edgeworth cycles involve prices above marginal cost; a marginal cost pricing equilibrium cannot exist in the Maskin and Tirole (1988) framework because of the timing of the model.*
  Overall, because most studies that examine Edgeworth cycles across markets are limited to a small cross section (e.g. Canadian studies have focused on a sample of 19 cities; see Eckert, 2003; Noel, 2007a), strong price comparisons are not made. However, Doyle et al. (2010) and Zimmerman et al. (2010), using data on US cities, both find evidence that average prices under Edgeworth cycles are no greater than in cities without cycles. Doyle et al. (2010, p 658) find that ‘a cycling city is found to have lower prices by 1 to 2 cents per gallon on average…It appears that the cycling cities have cycles that spend roughly equal time above and below the price levels in non-cycling cities’. Lewis (2009a) finds that, as a result of differences in passthrough patterns, consumers in US cycling cities faced lower prices in the period following Hurricane Rita than in non-cycling cities holding other factors constant.
  A finding that retail prices tend to be lower in cycling cities than elsewhere might seem to be at odds with antitrust cases involving conspiracy among retailers in cyclic cities, in which it is suggested that retailers communicated or coordinated over the restoration phase of the price cycle (see Wang, 2008, for an Australian case and Clark and Houde, 2011, for a discussion of a Canadian case). However, as suggested above, the theory of Edgeworth cycles is not well developed beyond the simple duopoly case. Noel's (2008) finding that increasing the number of players may result in less effective restorations and failed restorations may point towards incentives to coordinate on restorations, even remaining in a cyclic equilibrium overall.

3.3 Other Studies of Pricing Dynamics

Beyond studies estimating the passthrough of upstream prices or those examining Edgeworth cycles there exist some studies looking at other aspects of retail price dynamics. Some of these studies simply test conventional wisdom regarding retail price adjustment. Hall et al. (2007), Davis (2010) and Erutku (2007) all consider empirically the belief that retail prices increase heading into weekends or holidays, which if true suggests at least to some consumers the existence of price fixing. The consensus of these studies is that there is no holiday/weekend effect.¹⁶ It should be noted, however, that because of data limitation, these studies have been forced to rely on price data for only a few stations (in the former two papers) or city level prices observed twice weekly (in the latter paper). Neilson (2009) examines the perception that gasoline retailers engaged in ‘price gouging’ (defined by Neilson as prices temporarily higher that what would result from a firm's ‘standard pricing formula’) following Hurricane Rita. It is concluded that in fact retail prices were abnormally lower than expected after Hurricane Rita.

Some authors have considered the question of whether the dynamic pricing strategies employed at individual gasoline station can be boiled down to simple ‘rules of thumb’. Here the evidence is somewhat mixed. Atkinson et al. (2009) describe the pricing behaviour exhibited by stations during an Edgeworth cycle, using prices observed eight times per day from 14 August to 24 November, 2005, for stations in Guelph, Ontario. It is found that many stations appear to follow simple pricing rules, such as simply matching or setting a fixed differential with the prices set by certain key competitors, which are often geographically close. In contrast, Hosken et al. (2008, p. 1426) conclude from a study of weekly prices for 272 stations in suburban Washington DC that ‘stations do not appear to use simple static pricing rules: stations do not charge a fixed markup over their wholesale costs, nor do they maintain their relative position in the pricing distribution over time’.

Attempts to formally estimate the dynamic pricing strategies of individual gasoline retailers have been limited, perhaps because of data requirement. Barron et al. (2008) use a unique opportunity to impose exogenous price changes on stations of a firm in certain Californian areas to estimate the reactions of competitors. The authors find that rival firms adjust price in response to price changes, but by a lesser amount than the original price change, and find that the degree of price response depends inversely on station density. Slade (1992) uses daily price and sales data for 10 gasoline stations in a specific region of Vancouver B.C., during a price war that erupted in the summer of 1983 to estimate a system of station level demand functions and intertemporal reaction functions. Asymmetric strategies are identified, in which independents lead price reductions but major refinery brand stations lead price restorations.¹⁷ A conclusion similar to that of Atkinson et al. (2009) is reached (p. 273): ‘The empirical regularities that have been uncovered are consistent with the notion that rather simple strategies capture the essence of station behaviour in the Vancouver retail-gasoline market’. This is argued to be expected in a tacitly collusive market.¹⁸

4. The Determinants of Price Levels: Mergers, Regulations and Price Dispersion

The previous section reviewed the literature on retail price movements over time. However, a large empirical literature exist examining determination of price levels, and how and why prices differ across markets and across stations. Many of the questions addressed have policy motivations. What are the main determinants of retail gasoline prices? What degree of market power do firms exercise? What impact do mergers and regulations have on retail prices? Do firms engage in anticompetitive behaviour, and what is the impact of policy designed to prevent such conduct? This section reviews the existing evidence on these questions.

4.1 Reduced Form Studies of Prices as a Function of Market Structure, at the City or State Level

The approach taken by some authors has been to work with weekly or monthly data at the level of the city or state, and estimate reduced form price equations to describe how market structure is associated with price levels and to identify the primary determinants of retail prices. Sen (2003, 2005) uses a panel data set of monthly city level prices for 11 Canadian cities over the 1991–1997 period to estimate reduced form equations in which prices are functions of current and lagged wholesale prices, and measures of market structure. Sen (2003) concludes that wholesale prices play a larger role in price determination than city level market structure, whereas Sen (2005) focuses on the difference between major refinery brand retailers and independent retailers, and concludes that increases in the market share of smaller firms are associated with lower retail prices. Chouinard and Perloff (2007), using a panel of monthly observations at the state level for the 1989–1997 period, estimate reduced form pricing equations in which prices are associated with crude oil prices, demographics, refinery outages, environmental regulations and information on mergers, vertical contracts and taxes. The authors come to a stronger conclusion (p. 22) than Sen (2003): ‘Virtually the entire variation in national gasoline prices over the last decade is due to a rise in the price of crude oil’. However, they find that across-state variation in prices can be attributed to demand variation, as well as differences in taxes, environmental regulations and market power.

4.2 The Effects of Mergers and Regulations

Several studies consider the effect of particular events on retail pricing.¹⁹ Analyses of the impacts of specific mergers on retail gasoline prices are given in Coloma (2002), Hastings (2004), Silvia and Taylor (2010), Simpson and Taylor (2008), Taylor and Hosken (2007)²⁰ and Taylor et al. (2010).²¹ These mergers have involved both upstream and downstream operations. Methodology in these studies often involves some form of difference-in-difference estimation. Although results have been mixed, the majority of these studies find little effect on retail gasoline prices.²²

Other types of events have also been analysed. Erutku and Hildenbrand (2010) examine a price fixing cartel in the province of Quebec, and use a difference in difference approach to consider whether the announcement of the criminal investigation caused a price reduction. Data used are weekly city-average prices for Sherbrooke, one of the cities involved in the conspiracy. Montreal and Quebec City are used as controls. The authors conclude that the announcement caused a price reduction of 1.75 cents per litre.²³

Several studies examine the impact of regulations in specific markets, again typically using difference in difference approaches. Carranza et al. (2011) consider the impact of a price floor introduced in Quebec in 1997, using data on cities in Ontario as controls, and find that the long run effect of the regulation was to lower prices and productivity. Sales below costs laws are also studied by Anderson and Johnson (1999) using a panel data set of weekly city level prices for the March 1992 to December 1993 period. They conclude that sales below cost laws result in higher retail margins. In contrast, Skidmore et al. (2005) find that sales below cost laws in the USA were associated with a 1 cent reduction in price after 5 years, along with an increase in outlets. A discussion of sales-below cost laws in Wisconsin can be found in Brannon (2003). Doyle and Samphantharak (2008) used a temporary gasoline tax moratorium in Illinois and Indiana to examine the passthrough of gasoline taxes to consumers, and find that 70% of the initial tax reduction and up to 100% of the tax reinstatement were passed on.²⁴Johnson and Romeo (2000) examine the price and station characteristic impact of self-service bans in New Jersery and Oregon, concluding that these bans have a positive but small impact on retail margins.

Regulations regarding vertical integration and restraints have received considerable attention, as have regulations regarding ‘zone pricing’ at the wholesale level, whereby refiners offer different wholesale prices to retailers in different zones but uniform prices within zones. Vita (2000) studies the impact of divorcement regulations that restrict vertical integration of refining and retailing, often with the motivation of preventing predatory behaviour by refiners. The author applies reduced form methods to a panel data set of monthly retail prices, by state, for the 1995–1997 period. Controlling for the presence of local divorcement and sales below cost regulations using dummy variables, the author finds that divorcement regulations increase retail prices. Barron and Umbeck (1984) examine the impact of a 1974 divorcement law in Maryland, using data on 99 stations that were franchised by refiners after the regulation was imposed, and the stations identified as their competitors. The authors find that prices at the newly franchised stations increased, and service hours decreased, relative to competitors. Weak evidence suggests that competitor prices increased.

A number of other papers discuss the potential impacts of proposed regulations concerning vertical restrictions and integration, without analysing the impact of such restrictions were already in place. Blass and Carlton (2001), in an analysis of non-price data on newly constructed stations from 10 integrated refiners for the year 1988, conclude that national divorcement regulations could have costs greater than $1 billion, and that integration decisions are efficiency driven, rather than motivated by predation. Discussions of the possible implications of branded open supply and the removal of zone pricing can be found in Comanor and Riddle (2003), Keely and Elzinga (2003), Barron et al. (2004b) and Langenfeld et al. (2003). The general consensus is that such regulation would lead to higher retail prices.

4.3 Station Level Prices, Price Dispersion and Uniformity

Much attention has been paid to cross sectional price variation at the station level. Empirical studies of station level price variation have attempted to explain why observed prices are diverse or uniform, and to identify the main variables associated with price differentials across stations and products, and price variation within markets.

Several empirical studies involve regression analysis in which the price or margin at a particular station is a function of station and market characteristics. Although price control variables vary by study, variables can typically be divided into different broad categories: (1) local demographics and station location; (2) physical station characteristics; (3) brand and contractual arrangement and (4) station density and local concentration. As a general overview of this literature, Table 2 provides a list of recent papers presenting regression analysis on station or local retail prices or margins (Clemenz and Gugler (2006) use price data for Austrian districts), indicating which categories of explanatory variables are controlled for and found to be associated with price levels. Findings regarding specific variables in each category, and the signs of the associations, will be discussed below. Note that Table 2 focuses on regressions that relate price levels or margins to the variables discussed; related studies not reported include Barron et al. (2000) and Shepard (1991), which examine price differentials by grade or service, and Eckert and West (2005a), which examines whether individual stations match the market mode price. As noted in the literature, interpretation of the findings of these studies can be difficult because local market structure may be jointly determined. See, for example, Hosken et al. (2008) and Barron et al. (2004a) for discussions on this point.

Table 2.  A Summary of Statistical Significance Conclusions in Retail Price or Margin Determination, by Study and Variable Category.*

Study	Geographic Area	Level of Observation	Brand or Vertical Structure	Local Demography or Location**	Station Characteristics	Local Competition
*Y indicates that some variables in this category were found to be significantly associated with retail price or margin at the 10% level. N indicates variables were included, but either coefficients were not significantly different from zero at the 10% significance level, or were not reported. Blanks indicate that control variables in this category were not included. **Includes regional dummy variables.
Barron et al. (2004a)	California, Arizona	Station	Y	N	Y	Y
Clemenz and Gugler (2006)	Austria	Districts	Y	Y		Y
Cooper and Jones (2007)	Lexington Kentucky	Station	Y	Y		Y
Eckert and West (2004)	Vancouver, British Columbia	Station	Y	Y	Y	Y
Hosken et al. (2008)	Northern Virginia	Station	Y	Y	Y	Y
Ning and Haining (2003)	Sheffield, United Kingdom	Station	N	Y	Y	Y
Pennerstorfer (2009)	Austria	Station	Y	Y	Y	Y
Shepard (1993)	Massachusetts	Station	Y	N	Y	Y
Van Meerbeeck (2003)	Belgium	Station	Y	Y		Y

A sizeable literature has arisen considering the role of geographic space on retail gasoline prices, and the degree to which market power is localized; the distinct (and much smaller) literature on the determinants of location will be discussed in Section 6. The focus of this literature is on station level prices and the relationship between pricing and local market structure. As well, studies using station level data that are focused on non-spatial questions will frequently control for or examine the role of local market power.

Reduced form studies of individual station prices frequently ask whether station level price is associated with local concentration or station density. Typically, regressions will be estimated in which station level price or station level margin over wholesale cost is a function of station level characteristics, local market characteristics and region or time period fixed effects. Often, local station density is defined according to a fixed and arbitrary radius; Hosken et al. (2008) and Barron et al. (2004a), for example, consider the number of stations within a 1.5 mile radius, whereas Eckert and West (2005a) employ a 2 km radius. In other cases, ‘local’ is determined according to political boundaries (Van Meerbeeck, 2003, looks at station density within a municipalty, whereas Clemenz and Gugler, 2006, employ Austrian administrative districts), or other considerations. Cooper and Jones, 2007 account for the number of competing stations along the same commuter route; further evidence that interactions between stations are associated with routeways is presented in Haining (1983). Alternatively or in addition, some authors control for distance to the nearest competitor or some measure of local concentration within a certain radius.

In general, the findings have been mixed. Barron et al. (2004a), van Meerbeek (2003), Eckert and West (2004), Clemenz and Gugler (2006) and Shepard (1993) all report that station density is negatively associated with price. Cooper and Jones (2007) find an asymmetric effect; although increased density on a commuter route decreases prices, the count of competitors between a station and the central business district is more strongly related to price than the density of competitors further from the central business district. In many cases, however, the effects of station density are small; Barron et al. (2004a) report, for example, that a 50% increase in the station count within 1.5 miles is associated with price decreases of approximately 0.5%. Hosken et al. (2008) find no association between local station density and price when all brands are included in the regression; distance to the closest station becomes significant with a positive coefficient when one brand that systematically undercuts rivals is dropped.

Studies employing local concentration measures [such as the fraction of outlets within a certain radius that are associated with major refiners, or an Herfindahl-Hirschman Index (HHI)] as opposed to simple station density variables or competitor counts also find mixed results. Hosken et al. (2008) find no impact of local concentration measures, whereas Clemenz and Gugler (2006) find evidence that concentration within a station's ZIP code is associated with higher margins. Pennerstorfer (2009), applying a spatial lag model to station level price data for 400 Austrian stations, concludes that the competition-increasing effect of independent retailers is muted by a ‘composition effect’: as the local presence of unbranded independents increases, there will be fewer stations offering branded products of perceived higher quality, which can result in higher prices at major branded stations.

One important question in modelling the role of space in retail gasoline markets is how to measure proximity. As mentioned above, many studies have used distance bands for this purpose, whereas in spatial studies in other industries nearest point sets have been employed (see Siem and Waldfogel, 2010, for a recent application to liquor retailing). However, Houde (2009) demonstrates the importance of commuting routes (see also Cooper and Jones, 2007, above), by developing a structural spatial model of gasoline retailing that is applied to station level data for Quebec City. The author develops a spatial model of demand in which, instead of being located at a single point as in a standard Hotelling model, a consumer's ‘home’ is a commuting path, so that transportation costs are incurred when a consumer must leave her commuting path to make purchases. Houde (2009) finds that such a model better fits store level sales data than the standard Hotelling model. Different ways of incorporating geographic space into gasoline price competition are considered by Ning and Haining (2003) in a study of pricing in Sheffield, England. They find that for most time periods considered there is a statistically significant positive association between a station's price and prices of stations in the same local cluster.

A small number of spatial studies of gasoline retailing have attempted to estimate the willingness of consumers to travel in order to obtain lower prices, a question that has been of particular policy relevance in jurisdictions where it has been alleged that consumers will travel large distances to save tenths of a cent per litre, resulting in price uniformity across wide areas. An overview of these claims in a Canadian context can be found in Atkinson et al. (2009). Manuszak and Moul (2009), using data on tax regions near Chicago, estimate that the average consumer is willing to incur an additional mile round trip for savings exceeding approximately 7 cents per gallon.

Studies of price levels at individual stations have also concluded that associations exist between price levels and station characteristics, including whether the station has a car wash or service bay, whether it offers full service and the size of the station; see, for example, Eckert and West (2004, 2005a), Ning and Haining (2003), Hosken et al. (2008) and Barron et al. (2004a).²⁵ In some cases, however, the impact of characteristics on price is quite small; for example, in Hosken et al. (2008), the only significant characteristic, whether the station provides repair service, raises price by less than 1 cent. Ning and Haining (2003) also report associations between retail price and certain locational characteristics, such as proximity to a supermarket. A firm's decision of whether to charge for the use of credit (through a ‘discount for cash’) is analysed by Barron et al. (1992). Some evidence has also been presented that station level prices vary by contractual form (see, e.g. Shepard, 1993; Hosken et al., 2008). Finally, brand specific effects are found to important in a range of studies.

Some attention has been paid to the determinants of the degree of equilibrium price dispersion or uniformity. Data from different studies suggest that the degree to which individual stations set uniform or dispersed prices can vary dramatically across markets. Eckert and West (2005a) use data reported to a consumer website for the Vancouver market, and find that on average 61% of stations with prices reported on a given day report the same price. Atkinson et al. (2009), using a panel of self-collected prices for Guelph, Ontario, find that on average 30% of the 27 stations exhibit the mode price at a given point in time, and that averaged across stations, a station exactly matches the price of its closest rival in 41% of observations. In contrast, in other markets, considerable price dispersion has been observed (see, e.g. Lewis, 2008).

Different theoretical frameworks are offered to explain the degree of price variation or uniformity observed in a market. Iyer and Seetharaman (2008) model the price and service quality decisions of competing gasoline retailers, and find that whether the observed pricing equilibrium exhibits identical or dispersed prices and station characteristics depends on the relative importance of geographic product differentiation versus dispersion in consumer quality valuation; when horizontal differentiation is relatively high, price uniformity is observed, whereas when a local area serves a population with a highly diverse valuation of service quality, price and quality dispersion is expected. In an empirical examination of price and quality dispersion within individual census tracts in the St. Louis Metropolitan area, the authors find support for these basic predictions.

Png and Reitman (1994) consider whether price dispersion across nearby stations may result from stations differentiating according to waiting times, based on theoretical models developed by Luski (1976) and Reitman (1991). Using station level data for four counties in Massachusetts, the authors find, as predicted by the theory, that price dispersion across stations increases among stations facing more direct competition.

Barron et al. (2004a) find that, controlling for station characteristics, price dispersion decreases in station density. Lewis (2008), in an examination of prices in San Diego, finds as well that price dispersion is associated with the number of local competitors, but also with the types of those competitors. He concludes that consumer search may play an important role in retail prices.

4.4 Structural Estimation of Market Power

Few papers estimate structural models of retail gasoline markets that can yield estimates of the degree of market power being exercised. This may be a result of the difficulty in obtaining volume data at the station level. Those papers that have attempted to estimate such models have come to mixed conclusions.

Manuszak (2010) uses data on volumes, prices and characteristics for 1350 stations in Maui and Kauai to estimate a model designed for the evaluation of upstream mergers. Demand is based on a discrete choice model, and equations for both upstream and downstream price setting are estimated. Regarding downstream market power, the author concludes that market power is low at both the upstream and downstream levels, with downstream markups of approximately 10 cents per gallon. Houde (2009) comes to a similar conclusion using station level data for Quebec City. Houde's estimation results suggest that market power is low, and reject cooperative pricing.

A different conclusion is reached by Slade (1987), who uses station level data on volumes and prices in a certain Vancouver region to estimate the degree of market power being exercised and to test different oligopolistic models. She concludes that outcomes in the market are less profitable than perfect collusion, but more so than non-cooperative equilibria.

5. Non-Price Choice Variables

Some empirical studies of gasoline retailing have focused on non-price choice variables, including outlet counts, network sizes, station locations and contractual arrangements.

Some of these studies have been motivated by certain stylized facts. An important trend in the evolution of gasoline retailing has been the rationalization of stations that occurred broadly over the last 30 to 40 years. The Conference Board of Canada (2001) reports that in Canada, the total number of gasoline stations fell by 40% from 22,000 to 13,250 over the period 1989 to 2000. Scherer (1996) reports that the number of gasoline stations in the USA fell by 37% from 1972 to 1982, and by a further 9% from 1982 to 1992, whereas Kleit (2005) reports a 21% reduction from 1994 to 2003.²⁶ Gasoline stations in Austria fell by more than one third from 1983 to 1999 (Gotz and Gugler, 2006).²⁷ Along with a reduction in stations came a movement towards larger capacity self-service stations with fewer automotive repair facilities (see, e.g. Scherer, 1996; Carranza et al., 2011).

Several explanations for this reduction have been put forward, focusing largely on demand and cost changes. The Conference Board of Canada (2001) and Scherer (1996) both identify price increases by OPEC in the early 1970s as a starting point. Changes in consumer preferences away from stations with automotive service and full service stations have been suggested. Yin et al. (2007) and Eckert and Eckert (2010) consider the role of environmental regulations on station shutdown, whereas Carranza et al. (2011) examine the effect of a price floor in Quebec on station shutdown. Eckert and West (2006) examine the surge in shutdown that followed the entry of Atlantic Richfield Company (ARCO) into the Vancouver area in the late 1990s.

Studies of station shutdown in gasoline have also considered whether shutdown has been asymmetric across station categories, and the impact of rationalization on market concentration and market power. Yin et al. (2007) find that environmental regulations have a greater impact on small outlets. Sen and Townley (2010), in an examination of city level aggregate statistics, find that Canadian rationalization resulted in increased prices and concentration.

In addition to the number of stations in a market, empirical attention has turned to the location of those stations. A basic question of spatial economics concerns the relative sizes of the market share and market power (or direct and strategic) effects of distance between competitors, and whether retailers will tend to maximally or minimally differentiate. See, for example, Tirole (1988, chapter 7) for an overview of the economic theory of retail location. Netz and Taylor (2002), in a study of station locations in the Los Angeles area, find evidence to indicate that as competition is increased, stations locate further from competitors. This is taken to indicate that the market power effect causes stations to differentiate spatially. A related observation, made in Clemenz and Gugler (2006) for Austrian districts, is that there exists a negative association between market concentration and the total number of stations in a market. This is taken by the authors as consistent with the hypothesis that in equilibrium, less variety is offered to consumers in less competitive markets. Eckert and West (2008) find that, in the aftermath of the entry of ARCO into the Vancouver area, there was an association between whether a station shut down and its proximity to ARCO stations. Likewise, Eckert and West (2005b), in a study of the Toronto area, find an association between whether a major brand station shuts down and the local presence of independent stations.

Some empirical work has examined the contractual arrangement between stations and their suppliers and the degree of separation between suppliers and retailers. Slade (1998), using station level data for an area of Vancouver, finds evidence consistent with strategic motivations for vertical separation. Pinkse and Slade (1998) examine spatial patterns of contract types, and find some evidence that market-share considerations result in clustering of similar contract types. Blass and Carlton (2001), as discussed above, conclude from their study of retail outlets in the USA that observed vertical integration and separation is consistent with efficiency as opposed to predatory motivations. Taylor (2000) uses station level data from Los Angeles to test predictions from principal-agent theory, and concludes that the degree of vertical control is related to station characteristics that determine the importance of unobserved effort. Agency problems are also examined in Slade (1996).

Little empirical work has been done on the choice of station characteristics. One exception is the work of Iyer and Seetharaman (2008), which examines the determinants of quality differentials across stations. Some attention has been paid to the impact of self-service bans on station size and characteristics: Vandegrift and Bisti (2001), in a study of New Jersey, concludes that a self-service ban resulted in smaller stations, less vertical integration, different services and reduced price discrimination. Johnson and Romeo (2000), studying self-service bans in New Jersey and Oregon, find that such bans result in a reduced presence of service stations with convenience stores. However, the determination of station characteristics is an area where more work could likely be done.

6. Conclusions

At this point, we can return to the broad questions asked at the beginning of this paper, and consider what general lessons about gasoline retailing can be gleaned from the literature. A first basic question concerns which variables are most strongly associated with retail gasoline prices. As highlighted by large reduced form panel studies, and by the literature on upstream shock passthrough, the most important driver of highly aggregated (e.g. national) gasoline prices is the price of crude oil. That said, however, it has also been demonstrated that crude oil prices and retail gasoline prices can deviate substantially from their long run equilibrium relationship in the short term. Evidence of asymmetric response in many markets suggests that this deviation could be more pronounced when crude oil prices fall.

The relationship between retail prices and market structure variables is less clear. Reduced form panel studies have identified an association between market concentration and price levels, although it is found that crude oil prices play a larger role in the determination of retail prices. Evidence exists to suggest that market concentration does play a role in response asymmetry. Market structure appears to be associated with the determination of the nature of the equilibrium played in a given market (such as cycling versus non-cycling), although the relationship is complex, and may depend upon the precise characteristics of retailers in a market as opposed to simple concentration. Prices do appear to be associated with market structure at the local level, and space does indeed appear to matter for station level pricing; however, the importance of local market structure (within 1 or 2 miles of a station) seems to be small. The evidence is mixed regarding the impacts of mergers on retail prices, with difference-in-difference style studies generally reaching different conclusions from broad panel studies. As well, structural estimation suggests that, at least in those markets considered, market power is low or at least less than would be expected under perfect collusion.

Further complicating the questions of the impact of market structure and the degree to which gasoline retailers exercise market power is antitrust action against the industry and evidence of collusion. Two recent cases concerning price fixing have resulted in guilty pleas in Canada and Australia. Indeed, the incentive for collusion may be high in gasoline retailing; Houde (2009) notes for Quebec City that collusion could increase retail margins by a factor of four. Several studies have suggested that retailers may be adopting pricing practices intended to facilitate price coordination. Lewis (2009b), Wang (2008, 2009a) and Atkinson (2009) discuss apparent price leadership to facilitate price restorations in Edgeworth cycle markets. Slade (1992) points to simple dynamic reaction functions as possibly designed to facilitate tacit collusion, although one explanation that has been identified (and for which some support exists) for response asymmetry involves tacit collusion. Atkinson et al. (2009) consider whether price uniformity across spatially differentiated retailers may arise as a facilitating device; see also Slade (1986, 1992) on this point.

At the other end of the spectrum, however, little evidence exists of exclusionary or predatory strategies being exercised in gasoline retailing. With some exceptions, the general consensus of the empirical literature is that regulations designed to prevent anticompetitive practices, such as divorcement regulations, have or could lead to price increases and welfare reductions (see, e.g. Anderson and Johnson, 1999; Blass and Carlton, 2001, and other references in Section 4.2 above).

Certain features of consumer demand seem to play important roles. Consumer search has been identified as an important feature of retail gasoline markets, being identified as a factor in both dynamic response asymmetries and cross-sectional price dispersion. Whether or not consumer search should play a greater role in gasoline retailing than in other retail markets is unclear – whereas on the one hand, comparing gasoline prices may require geographic travel (in contrast with price comparison of grocery items or items available within a single large department or big box store), commuting routes plus prices being displayed on large billboards mean that most consumers may costlessly collect prices of multiple stations as part of their daily travel. The high sensitivity of consumers to price differentials across stations has been pointed to as a contributor to the existence of Edgeworth cycles and price uniformity, although limited empirical evidence exists regarding this sensitivity (see Wang, 2009b, for an exception).

Several questions concerning gasoline retailing have received little attention in the economic literature. Industry studies, including LECG Canada (2006) and Conference Board of Canada (2001), indicate that non-gasoline products and services represent an important part of a gasoline retailer's revenues. Although there is some evidence that these products and services influence gasoline pricing decisions, little is understood about the interaction of the demand for gasoline and non-gasoline products and services, the decision to adopt non-gasoline products and services, and the precise way that they influence equilibrium pricing; for a recent exception, see Wang (2010), who examines bundled discounts for gasoline and groceries in Australia. Understanding the relationship between gasoline prices and ancillary revenues can be particularly important for policy questions concerning potential predatory pricing as well as for understanding the incentives to undercut rivals in Edgeworth cycle markets.

Likewise, other than a small number of studies on station rationalization, the evolution and determination of market structure in gasoline markets has received little attention. This may in part be due to the difficulties in analysing retail location in oligopolistic markets, but recent developments may result in greater attention to these questions. The difficulties in, and currently available methodologies for, analysing these problems are well surveyed in Berry and Reiss (2007) and Doraszelski and Pakes (2007).

Overall, but particularly in the context of pricing dynamics, the literature indicates a need for a renewed attention to the development of theory. Efforts towards theories of asymmetric response have already begun and are ongoing. Important work is likely yet to be done in the development of economic theories of Edgeworth price cycles. Although the existing extensions of the Maskin and Tirole (1988) framework, particularly through simulations in Noel (2008), are important and have increased our understanding of where we might expect cycles and what they might look like, several interesting questions remain, including the role of conspiracy within a cycle and the existence of Edgeworth cycles in spatial markets.

Much of the existing work also points towards a need to understand station level demand at a high frequency. Although station level data on pricing have gone a long way towards identifying the forces that drive retail prices, ultimately a firm understanding of precisely why retailers price in the ways that they do may require estimating demand equations for individual stations at the daily frequency or better. Such efforts would contribute to our understanding of the precise causes of Edgeworth cycles and response asymmetry, price dispersion and price uniformity/dispersion in spatial markets. Data limitation, not surprisingly, have meant that the number of such studies is currently very small, and that those studies that do exist make use of small samples of stations. It is unclear that we should expect the number of such studies to increase rapidly in the near term, because station level volumes at a high frequency would likely require participation of the firms.

Acknowledgements

The author thanks Doug West, Gwillym Allen and Heather Eckert for comments and suggestions and Narinder Sharma for research assistance.

Notes

• * 

  Note: This sentence has been changed/updated on 30 November 2011 after first publication online on 22 July 2011.

• 1

  Competition Bureau activity in the gasoline industry in Canada is described at http://www.competitionbureau.gc.ca/eic/site/cb-bc.nsf/eng/00235.html. Details of a recent conspiracy case in Australia are given in Wang (2008); see also Australian Competition and Consumer Commission (2001) for a discussion of Australian competition concerns. Office of Fair Trading (1998) deals with competition concerns in the United Kingdom.

• 2

  See, for example, Blass and Carlton (2001) and Barron et al. (1985) for discussions on divorcement laws, and Anderson and Johnson (1999) on sales-below cost laws.

• 3

  See, for example, Hyde (2002) for a discussion of petroleum mergers in Australia.

• 4

  A shorter overview of certain aspects of this literature can be found in Houde (2010).

• 5

  In another example of firm co-operation, Barron et al. (2008) were given the opportunity to impose exogenous price changes on stations of a major retailer in certain California cities over a 3-month period.

• 6

  Note that in the interests of manuscript length, the reference section for this paper includes only those papers explicitly cited in this literature review. The full bibliography is available from the author upon request.

• 7

  This North American focus may partly result from the restriction to English language articles. In database searches, key words such as ‘petrol’ were used in an attempt to pick up studies outside of North America.

• 8

  The six most frequent journals represented (with the number of articles) were Energy Economics (9), Review of Economics and Statistics (8), Journal of Industrial Economics (7), Review of Industrial Organization (7), International Journal of Industrial Organization (6) and Journal of Law and Economics (6).

• 9

  Note that in some studies, multiple data sets are used for different sections of the paper. For this categorization, I looked at the data set most relevant to gasoline retailing (as opposed to upstream competition), or the data set used for econometric analysis as opposed to descriptive analysis. Otherwise, the highest-frequency data were recorded.

• 10

  Other approaches have been taken. See, for example, Godby et al. (2000) for an application of a threshold autoregressive model to Canadian retail prices. A survey of different methods that have been used to measure response asymmetry can be found in, for example, Frey and Manera (2007).

• 11

  This observation is also made by Geweke (2004).

• 12

  As an example, expert reports estimating response asymmetry were commissioned by the Canadian Competition Bureau in response to price fixing allegations made in 1996. See Hendricks (1996) and Lermer (1996).

• 13

  I thank Ben Atkinson for providing the Guelph price data. To provide a clear example of price cycles, the earlier part of Atkinson's sample is excluded because the cycle is obscured by the cost shocks associated with Hurricanes Katrina and Rita. Analysis of pricing over the entire sample can be found in Atkinson (2009).

• 14

  The only other settings to my knowledge where Edgeworth cycling behaviour has been studied are bidding for online advertising positions (see Zhang 2006), and in experimental settings (see Kruse et al., 1994).

• 15

  Note that this is a simplification, because it is possible that distance to marginal cost may influence the size of price undercuts. For example, in Maskin and Tirole (1988), once retail prices become sufficiently low, a large undercut will be observed down to marginal cost.

• 16

  There is, however, evidence that retail gasoline prices are more likely to adjust upward or downward on specific days of the week; see, for example Davis (2010), Foros and Steen (2008) and Asplund et al. (2000). However, because these studies tend to find that price increases are more likely early to mid-week, it suggests a different cause than a demand change on weekends and holidays.

• 17

  The theme that different types of retailers (majors versus independents, traditional stations versus big box retailers and grocery chains) play different roles in leading price increases and decreases is familiar from the Edgeworth cycle literature. See, for example, Atkinson (2009), Atkinson et al. (2009), Zimmerman et al. (2010) and Lewis and Noel (2011).

• 18

  Slade (1986, p. 367) also argues that observed price uniformity in Vancouver is consistent with attempted tacit collusion: ‘The substantial price uniformity that was observed prior to 1981, therefore, cannot have been a joint-profit-maximizing strategy. Instead, it probably resulted from an attempt to discipline a market that is difficult to coordinate’.

• 19

  There has also been some attention to the impact of regulations on wholesale prices. See, for example, Chakravorty et al. (2007) for an analysis of the effects of environmental regulations.

• 20

  Taylor and Hosken (2007) consider the impact of a joint venture.

• 21

  Mergers are also often included in reduced form studies of large panel data sets; see, for example, Chouinard and Perloff (2007). The impact of Australian mergers is examined in Hyde (2002).

• 22

  Exceptions are Coloma (2002) and Hastings (2004). However, Taylor et al. (2010), using a different data set to analyse the same transaction as Hastings (2004), find only a minor impact.

• 23

  Further information on this cartel, including information on guilty pleas, can be found at http://www.competitionbureau.gc.ca/eic/site/cb-bc.nsf/eng/00235.html#2009 (last accessed August 5, 2010). See also Houde and Clark (2011).

• 24

  Gasoline tax incidence is also examined using panel data for US states in Chouinard and Perloff (2004) and Alm et al. (2009).

• 25

  Evidence suggests that stations offering both full and self serve gasoline are able to successfully price discriminate, separating consumers with high and low marginal valuations for service (see, e.g. Shepard, 1991; Barron et al., 2001).

• 26

  Counting only outlets focused on retail gasoline sales, with paid employees (Scherer 1996, p. 133).

• 27

  The phenomenon of rationalization of retail outlets has been studied in other industries. See, for example, Allen et al. (2008) on the Canadian banking industry.