Alcohol Prohibition and Pricing at the Pump

Fischer, Kai Article — Published Version Alcohol Prohibition and Pricing at the Pump The Journal of Industrial Economics Provided in Cooperation with: John Wiley & Sons Suggested Citation: Fischer, Kai (2023) : Alcohol Prohibition and Pricing at the Pump, The Journal of Industrial Economics, ISSN 1467-6451, Wiley, Hoboken, NJ, Vol. 72, Iss. 1, pp. 548-597, https://doi.org/10.1111/joie.12366 This Version is available at: https://hdl.handle.net/10419/293965 Standard-Nutzungsbedingungen: Die Dokumente auf EconStor dürfen zu eigenen wissenschaftlichen Zwecken und zum Privatgebrauch gespeichert und kopiert werden. Sie dürfen die Dokumente nicht für öffentliche oder kommerzielle Zwecke vervielfältigen, öffentlich ausstellen, öffentlich zugänglich machen, vertreiben oder anderweitig nutzen. Sofern die Verfasser die Dokumente unter Open-Content-Lizenzen (insbesondere CC-Lizenzen) zur Verfügung gestellt haben sollten, gelten abweichend von diesen Nutzungsbedingungen die in der dort genannten Lizenz gewährten Nutzungsrechte. Terms of use: Documents in EconStor may be saved and copied for your personal and scholarly purposes. You are not to copy documents for public or commercial purposes, to exhibit the documents publicly, to make them publicly available on the internet, or to distribute or otherwise use the documents in public. If the documents have been made available under an Open Content Licence (especially Creative Commons Licences), you may exercise further usage rights as specified in the indicated licence. http://creativecommons.org/licenses/by/4.0/ THE JOURNAL OF INDUSTRIAL ECONOMICS 0022-1821 Volume LXXII March 2024 No. 1 ALCOHOL PROHIBITION AND PRICING AT THE PUMP* KAI FISCHER† Firms often sell a transparent base product and a valuable add-on. If only some consumers are aware of the latter, the add-on’s effect on the base product’s price will be ambiguous. Cross-subsidization between products to bait uninformed consumers might lower, intrinsic utility from the add-on for informed consumers might raise the price. We study this trade-off in the gasoline market by exploiting an alcohol sales prohibition at stations as an exogenous shifter of add-on availability. Gasoline margins drop by 5% during the prohibition. The effect is mediated by shop variety and competition. Using traffic data, we unveil sizeable consumer-side reactions. I. MOTIVATION THE LITERATURE ON GASOLINE MARKETS IS broad and has examined many features typical of gasoline competition, such as price dispersion, asymmetries in input cost pass-through and Edgeworth cycles. Most approaches to these topics assume that competition occurs only among gasoline stations, which are usually treated as single-product firms solely selling homogenous gasoline. Only a few papers have dealt with the relation of gasoline prices to stations’ attached services and secondary products such as shops, supermarkets, or carwashes (Doyle et al. [2010]; Haucap et al. [2017a,b]; Wang [2015]; Zimmerman [2012]). However, potential interactions of pricing at the pump and the provision of complementary products have relevant implications for market definition and unveil distributional consequences for heterogeneously informed consumers. If such complementarities distort the signal, that low prices imply the best deal in a homogenous product market like the gasoline market, the matching of consumers, who are uninformed about the availability of complementary products, to suitable stations *I am thankful for helpful comments by the editor, Ryan McDevitt, two anonymous referees, Katharina Erhardt, Justus Haucap, Paul Heidhues, Ulrich Heimeshoff, Andreas Lichter, Simon Martin, José Luis Moraga-González, Andrea Pozzi, Frank Verboven and Biliana Yontcheva. This article also profited from participants’ comments at the DICE PhD Workshop 2021, the Oligo Workshop 2021, the Hohenheimer Oberseminar 2021, and the DICE Empirical IO Group 2022. I gratefully acknowledge funding from the German Research Foundation (DFG) – 235577387/GRK1974. All remaining errors are my own. Open Access funding enabled and organized by Projekt DEAL. †Authors’ affiliation: Düsseldorf Institute for Competition Economics (DICE), Heinrich Heine University Düsseldorf, Universitätsstraße 1, Düsseldorf, 40225, Germany. e-mail: [email protected] © 2023 The Authors. The Journal of Industrial Economics published by The Editorial Board of The Journal of Industrial Economics and John Wiley & Sons Ltd. This is an open access article under the terms of the Creative Commons Attribution License, which permits use, distribution and reproduction in any medium, provided the original work is properly cited. 548 ALCOHOL PROHIBITION AND PRICING AT THE PUMP 549 could deteriorate. Also, common price transparency regulations in gasoline markets, that increase the prominence of stations with cheap gasoline prices, might be misleading then. Whether the existence of a complementary product raises or lowers gasoline prices—relative toa world without the complement—if only some consumers are aware of the complement, is unclear from an ex-ante perspective. On the one hand, better services or a wider product assortment increase the intrinsic utility of some consumers’ shopping. This can cause an outward shift in gasoline demand. Also, consumers will face opportunity costs of traveling if they are not one-stop shoppers but consume gasoline and the complement from different stations. This would explain price increases for gasoline. On the other hand, gasoline stations might use low and transparent gasoline prices as a quasi-loss leader to bait uninformed consumers, who ex-ante do not intend or expect to, in the end, buy additional products in the store. Cross-subsidization could arise (Armstrong and Vickers [2012]; Gabaix and Laibson [2006]; Heidhues et al. [2017]; Lal and Matutes [1994]). Less transparently priced complementary products such as add-on services or shop products might then be purchased by consumers at relatively high prices. Therefore, the overall price effect of complementary products on gasoline prices is ambiguous and a question for empirical research. Similar trade-offs can be found in most markets. In this work, we go into this matter and answer the question of how the introduction of a complementary product affects a firm’s price setting for other products. We provide causal evidence by exploiting a unique setting in the gasoline market, where the availability of a complement is exogenously determined by public policy. In particular, we examine a quasi-experiment, the lifting of a local nightly alcohol ban at gasoline stations in a federal state of Germany, as a shifter of complement availability. The prohibition restricted the shop assortment of stations as it mandated sales of alcohol, an important add-on product for gasoline stations, to be forbidden from 10 pm to 5 am. The policy was implemented in 2010 and lifted in December 2017. It aimed at the reduction of binge alcohol consumption among youths at night. As 60% of all profits of German gasoline stations are linked to the shop, 20% to carwashes, and only 20% to gasoline sales (FAZ [2015]; Ivanov [2019]; Nicolai [2021]; NTV [2015]), the alcohol sales ban reflects a relevant revenue shock. To analyze the effect of the available add-on on the price of the complementary base product gasoline, we use real-time data of all gasoline prices in the German gasoline market at the station level. By means of a difference-in-differences setup, we take advantage of the low menu costs and within-day variation of prices and compare gasoline prices during and after the prohibition as well as between affected and unaffected stations. This allows us to unveil the overall price effect of add-on availability and, hence, the complementarity on the base product’s price. Building on precise © 2023 The Authors. The Journal of Industrial Economics published by The Editorial Board of The Journal of Industrial Economics and John Wiley & Sons Ltd. 550 KAI FISCHER information about stations’ competitive environment and brand affiliation, we further can investigate heterogeneity across firms. Our findings and contributions to the literature are threefold. First, we investigate the effect direction of add-on quality on gasoline prices. We find nightly prices of stations affected by the prohibition to increase by 0.6 Eurocent/l—or 5% of the gross margin—after the lifting of the prohibition. Hence, especially consumers who did not buy alcohol profited from the policy when it was in place. Stations with smaller product variety, where alcohol’s relative importance for shop revenues is higher, reveal even stronger price effects. Similarly, stations with few competitors nearby increase prices more strongly. Opportunity costs of buying alcohol at another station increase with decreasing competition intensity. Thus, a potential cross-subsidization mechanism is overall outweighed by the intrinsic value of additional services. Using detailed, geo-coded traffic counter data, we provide supporting evidence that traffic increases only in the direct vicinity of gasoline stations after the reintroduction of alcohol sales. Our findings add to the literature on the role of station amenities for stations’ pricing behavior. Other papers have shown that stations’ choice to operate convenience stores (Doyle et al. [2010]; Ning and Haining [2003]; Haucap et al. [2017a]) and the proximity to hypermarkets nearby (Zimmerman [2012]) indeed shape pricing behavior. Though, they mainly rely on the endogenous self-selection of stations into lowor high-quality segments while we exploit an exogenous shifter of service and add-on availability. Our results also address the delineation of gasoline markets as price effects vary with the exposure to alcohol sales. Alcohol revenues are also determined by local supermarkets or pubs. This indicates that gasoline stations might not only compete with other stations. Second, while our results address discussions on multi-product competition across most markets, note that the setting studied in this article is unique. It mainly differs from other markets with two price components in three ways: At first, add-on services often are valueless to the consumer and are only jointly bought with the base good such as overdraft fees for financial services (Armstrong and Vickers [2012]; Gabaix and Laibson [2006]). In our setting, consumers are free to opt out of buying alcohol but can still buy other shop products. Beyond that, purchasing alcohol gives positive utility to some consumers. Second, firms often endogenously set the prevailing level of consumer information about prices in the market for the base product by, for example, advertising prices. We consider a price transparency environment that exogenously dictates prices to be equally transparent across firms. By law, gasoline prices of all German stations are published in real-time for consumers. Lastly, we do not just vary add-on revenues but study the add-on existence at the extensive margin. Hence, our results represent an upper bound for fluctuations of add-on revenues in our setting and are helpful in forming benchmarks for other industries. © 2023 The Authors. The Journal of Industrial Economics published by The Editorial Board of The Journal of Industrial Economics and John Wiley & Sons Ltd. ALCOHOL PROHIBITION AND PRICING AT THE PUMP 551 Third, we analyze how active stations are in response to the prohibition. 10% more stations adjust prices during night hours after the prohibition lifting. While this observation could purely represent changes in the Edgeworth cycles, we show that prominent characteristics of price cycles are unaffected by the lifting. Therefore, we believe these findings express changes in opening hours. The remainder of the article is as follows: We start with an explanation of the institutional background and a theoretical motivation in Sections II and III before presenting our data and empirical strategy in Section IV. We then proceed with our analysis in Section Vbefore providing robustness checks and a conclusion in Sections VI and VII. II. INSTITUTIONAL BACKGROUND Particularly, we examine a nightly off-premise alcohol prohibition in Baden-Wuerttemberg, a German federal state with a population of eleven million. This policy primarily affected gasoline stations as the main nightly off-premise places to go for alcohol (Marcus and Siedler [2015]; Baueml et al. [2023]).1From 2010 onwards, Baden-Wuerttemberg prohibited nightly alcohol sales from 10 pm to 5 am via the “Alkoholverkaufsverbotsgesetz” (Alcohol Sales Prohibition Law). As most people do not prestore alcohol, the prohibition was binding (Marcus and Siedler [2015]). This specific legislation ran out on December 08, 2017, as local authorities from then on should have selected specific “hotspots” (e.g., city centres) for bans only. In the three years after the lifting of the policy, there, though, were only rare occasions, when a municipality implemented an alcohol consumption prohibition–mainly during festivals (Landtag von Baden-Wuerttemberg [2020]). Its main intentions were the reduction of binge drinking among youths and of indirect spillovers on crime (Baumann et al. [2019]; Baueml et al. [2023]; Marcus and Siedler [2015]). The policy was effective in several ways indicating a real shock in the volume of alcohol consumed. Up to now, Baueml et al. [2023], Marcus and Siedler [2015] and Baumann et al. [2019] discussed direct effects on health costs (hospital admissions, doctor visits) and crime for this specific case study. All three papers find that the policy had an economically relevant effect. The number and the length of hospital stays among youth binge drinkers and late-night assaults fell due to the policy. The effect is strongest on young adults since they are more price-sensitive, can hardly pre-store alcohol in their parents’ home and are more likely to conduct off-premise pre-drinking (Baueml et al. [2023]). 1During the prohibition, only stations that also ran a diner with an official catering license to sell on-premise alcohol were still allowed to sell alcohol at night (§3a Abs. 1 LadÖG). This mainly concerned highway stations with rest houses, which at the same time were not allowed to sell alcohol due to a highway-specific alcohol prohibition. © 2023 The Authors. The Journal of Industrial Economics published by The Editorial Board of The Journal of Industrial Economics and John Wiley & Sons Ltd. 552 KAI FISCHER As the legislation ran out ahead of time—it was expected that the legislation would not change before 2018 (Mayer [2017])—and because the law was ineffective just a few days after the public announcement of the abolition, anticipatory effects are unlikely. We expect such regulation to have a sizeable impact on the German gasoline market. In Europe, German gasoline stations have one of the lowest net margins on fuels (Scope Ratings [2019]). Therefore, shop sales make up a relevant share of stations’ overall profits. In particular, alcohol and beverage sales account for more than 10% of all in-shop sales (Scope Ratings [2019]). Moreover, consumers coming for alcohol buy other products on the way. Recent years have shown that especially big brands such as ARAL extended their shops by for example integrating shops of supermarket chains. In contrast to other countries, German gasoline stations mostly did not introduce paying at the pump by card, as this would stop consumers from entering the store. Hence, most stations are occupied in person all day long, so that shop sales are possible. Moreover, German gasoline stations often act as “shopping location of last resort” during night times as then German groceries rarely open. Thus, a nightly prohibition impedes a relevant business time. Alcohol revenues may be relevant for gasoline prices. In response, cross-subsidization could plausibly be an optimal pricing strategy next to quality-related price inclines. To show this, we perform a simple, hypothetical back-of-the-envelope calculation based on some assumptions. Following the Statistisches Landesamt Baden-Württemberg [2022], overall annual gasoline and diesel consumption was approximately 7 million tonnes or 9 billion liters in 2017. Admittedly gasoline demand is low at night. But the Federal Cartel Office [2019] documents that still around 5%of all car drivers preferably fuel at night (10 pm to 5 am), which gives a lower bound of the actual demand. This implies that at least around 425 million liters p.a. are sold in Baden-Wuerttemberg at night. Uniformly distributing this over approximately 800 gasoline stations which operate at this daytime, this is slightly more than 0.5 million liters per station and year. If a station followed a cross-subsidization strategy that lowers margins by, for example, only half a Eurocent/l, it would lose around 2500 Euro p.a. This needs to be compensated by additional alcohol sales triggered through lower prices at the pump. Following Scope Ratings [2018], German gasoline stations, on average, earn almost one million Euro shop revenues p.a., of which alcohol products account for approximately a tenth. As alcohol is sold in the evening and night hours for the most part, profits from alcohol sales due to additional attracted consumers could exceed the cost of using gasoline as bait. In the setting studied in this article, consumers’ alcohol demand response to lower gasoline prices is changed from zero to potentially nonzero after lifting the prohibition. © 2023 The Authors. The Journal of Industrial Economics published by The Editorial Board of The Journal of Industrial Economics and John Wiley & Sons Ltd. ALCOHOL PROHIBITION AND PRICING AT THE PUMP 553 III. THEORETICAL SKETCH To get a better understanding of the ex-ante ambiguity of the policy’s effect on gasoline prices, we consider the differences between a gasoline station’s optimization problem before and after the policy lifting. Before the lifting, the station can only sell gasoline. After the lifting, alcohol can be sold in addition. We model a market in which consumers have heterogeneous preferences for alcohol and differ in whether they anticipate buying alcohol at a gasoline station or not. The model is set up in the following way: on the consumer side, a share of 𝛼consumers only want to buy gasoline and no alcohol. 𝜆is the share of informed consumers—among those who do potentially buy alcohol—who are aware of the availability of alcohol products when choosing a gasoline station. (1−𝛼)(1−𝜆)consumers do not consider the existence of alcohol at all. The demand of only-gasoline consumers is given by D𝛼(pt G)with pt Gbeing the gasoline price before (t=b) and after (t=a) the lifting. For consumers who potentially buy alcohol, informed and uninformed consumers’ demand is given by D1−𝛼,𝜆(pt G,𝛾A)and D1−𝛼,1−𝜆(pt G)with t∈{a,b}respectively. 𝛾A∈ {0,1}indicates whether alcohol is available (𝛾A=1) or not (𝛾A=0). Alcohol can only be sold after the lifting. If consumers gain utility from alcohol, then D1−𝛼,𝜆( p,1)>D1−𝛼,𝜆( p,0)∀ p, that is, alcohol availability causes an outward shift in gasoline demand. For simplicity, we assume (marginal) costs of zero.2 We then construct the gasoline station’s profit function before (𝜋b) (1) 𝜋b=pb G⎡⎢⎢⎢⎢⎣ 𝛼D𝛼(pb G) ⏟⏞⏞⏟⏞⏞⏟ (1) +(1−𝛼)(𝜆D1−𝛼,𝜆(pb G,0) ⏟⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏟⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏟ (2) +(1−𝜆)D1−𝛼,1−𝜆(pb G)) ⏟⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏟⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏞⏟ (3) ⎤⎥⎥⎥⎥⎦ and after the prohibition lifting (𝜋a) 𝜋a=(pa G+pA)[(1−𝛼)(𝜆D1−𝛼,𝜆(pa G,1)+(1−𝜆)D1−𝛼,1−𝜆(pa G))] (2) +pa G𝛼D𝛼(pa G). Before the prohibition lifting, the station earns the price pb Gfrom (1) those consumers who are only willing to buy gasoline and from (2) informed and (3) uninformed consumers who would also buy alcohol if available. After the prohibition lifting, the gasoline station is paid pa Gby the same three groups. In addition, they earn alcohol revenues from those informed and uninformed willing to buy it. Also, the station faces an outward shift in the demand for gasoline from informed consumers due to the add-on availability. Maximizing profits and rearranging the first-order conditions yields the policy’s price effect. 2We also ignore the add-on’s price pAin D𝜆(pt G,𝛾A)which does not affect the sign of the prohibition’s price effect as long as the intrinsic utility from the add-on is sufficiently high. © 2023 The Authors. The Journal of Industrial Economics published by The Editorial Board of The Journal of Industrial Economics and John Wiley & Sons Ltd. 554 KAI FISCHER Result 1. The price effect of the policy, ΔpG, is implicitly given by the expression (3) ΔpG=pa G−pb G=Δ CS +Δ SQ where ΔCS =− pA[(1−𝛼)(𝜆𝜕D1−𝛼,𝜆(pa G,1) 𝜕pa G+(1−𝜆)𝜕D1−𝛼,1−𝜆(pa G) 𝜕pa G)] 𝛼𝜕D𝛼(pa G) 𝜕pa G+(1−𝛼)(𝜆𝜕D1−𝛼,𝜆(pa G,1) 𝜕pa G+(1−𝜆)𝜕D1−𝛼,1−𝜆(pa G) 𝜕pa G), and ΔSQ =𝛼D𝛼(pb G)+(1−𝛼)(𝜆D1−𝛼,𝜆(pb G,0)+(1−𝜆)D1−𝛼,1−𝜆(pb G)) 𝛼𝜕D𝛼(pb G) 𝜕pa G+(1−𝛼)(𝜆𝜕D1−𝛼,𝜆(pb G,0) 𝜕pb G +(1−𝜆)𝜕D1−𝛼,1−𝜆(pb G) 𝜕pb G) −𝛼D𝛼(pa G)+(1−𝛼)(𝜆D1−𝛼,𝜆(pa G,1)+(1−𝜆)D1−𝛼,1−𝜆(pa G)) 𝛼𝜕D𝛼(pa G) 𝜕pa G+(1−𝛼)(𝜆𝜕D1−𝛼,𝜆(pa G,1) 𝜕pa G+(1−𝜆)𝜕D1−𝛼,1−𝜆(pa G) 𝜕pa G). The expressions ΔCS and ΔSQ represent the two channels that mainly drive price differences for gasoline before and after the policy lifting: First, prices after the lifting are reduced as stations cross-subsidize between alcohol and gasoline revenues. This is expressed in the first addend of (3), ΔCS, which is negative. This term expresses that per-consumer alcohol revenues pAare negatively correlated with the gasoline price pa G. This cross-subsidization channel characterizes gasoline as bait for uninformed consumers. Second, informed consumers increase demand due to the availability of alcohol products. This is expressed in the difference between the two addends of ΔSQ, where alcohol availability (𝛾A=1) increases demand after the policy change (see nominator). This service quality channel, hence, increases demand and prices. Thus, the overall effect on ΔpGis ambiguous. The model further delivers intuitive predictions on how different parameters mitigate the size of the price effect or determine its sign: Result 2. The treatment effect ΔpG • increases in the alcohol-induced demand shift of informed consumers D𝜆(pa G,1)−D𝜆(pa G,0), • is (weakly) negative in perfectly uninformed markets (𝜆=0) and • vanishes in markets with only gasoline buyers: lim𝛼→1ΔpG=0. Result 2states the following: If more consumers are aware of the utility gain from the availability of alcohol, this strengthens the demand expansion of the service quality channel. This is the case when D𝜆(pa G,1)−D𝜆(pa G,0)increases. © 2023 The Authors. The Journal of Industrial Economics published by The Editorial Board of The Journal of Industrial Economics and John Wiley & Sons Ltd. ALCOHOL PROHIBITION AND PRICING AT THE PUMP 555 The channel will be nonexistent if no consumer is aware of alcohol (𝜆=0). However, the cross-subsidization channel is fostered by higher per-consumer alcohol revenues pAin equilibrium, which can, for example, arise from an outward shift in alcohol demand. Both channels will become irrelevant if consumers only buy gasoline (𝛼=1). We use these predictions to guide our empirical analysis of treatment effect heterogeneity across different types of markets and stations later on. This allows us to better understand whether observed prices support the modeled channels. Nevertheless, the observed price effects of the policy might not be purely related to the channels discussed above. For example, we assumed that the share of informed consumers 𝜆among potential alcohol consumers and the share of only-gasoline consumers 𝛼do not change with the policy lifting. Changes in these variables could rationalize positive as well as negative price effects of the policy lifting beyond the channels we discussed above. A higher share of informed consumers, who want to buy alcohol and gasoline, arrive for alcohol and could be less elastic with respect to the gasoline price. A change in the demand elasticity could explain price changes then. We discuss such other potential mechanisms in our empirical analysis later on to clarify the role of the channels modeled above. IV. DATA AND EMPIRICAL STRATEGY Gasoline Price Data. We make use of E5 gasoline prices from all German gasoline stations. The data is collected by the Market Transparency Unit for Fuels (MTU) at the German Federal Cartel Office and accessed via tankerkoenig.de. The data is gathered in real time which allows us to exploit within-day price variation as needed in our setup. We use a full year of price data (mid-September 2017 to mid-September 2018). We construct the time-weighted average daytime (05 am to 10 pm) and nighttime (10 pm to 05 am) price per week and station. Station Characteristics. Further, the MTU provides exact information on station characteristics such as their brand affiliation and geographical location. From this source, we construct several variables that, later on, guide our heterogeneity analysis. First, we derive whether stations open all day long (24/7) and operate at night which is reported in the MTU data.3Our final sample only consists of such 24/7 stations as other stations do not operate all night, which is the prohibition period. Second, we use the location data to match stations to municipalities and counties. This allows us to match detailed information on municipalityand 3We extract this information from the first fully covered opening hours by the MTU being publically available from January 2019, just three months after the end of our sample period. © 2023 The Authors. The Journal of Industrial Economics published by The Editorial Board of The Journal of Industrial Economics and John Wiley & Sons Ltd. 562 KAI FISCHER Figure 2 Dynamic Effects by Hour of Day Notes: This plot gives dynamic estimates of the interaction term BWs×Postwof a simple DID model where one regression is run for each hour separately. The exact timing of the beginning and end of the prohibition is indicated by the black vertical line. Standard errors are clustered at the county level. We provide 90 and 95%confidence intervals for all coefficients [Colour figure canbeviewedatwileyonlinelibrary.com] remaining between 5 am and 6 am. This is likely related to the given timing of the Germany-wide intra-day Edgeworth cycles, where most stations changed prices after 6 am (Federal Cartel Office [2018]). Heterogeneity Analyses. To understand which stations are more prone to react to the prohibition, we study effect heterogeneity across station characteristics such as competition at the pump, variety in the product assortment, or brand affiliation. First, we study competition effects. As described above, our price effect likely originates from the mechanism that alcohol-demanding consumers visit gasoline stations and consume gasoline on the side. Then, the price effect would arise from the opportunity costs of traveling to a different gasoline outlet. This effect should be larger if alternative stations are far away. Similarly, if consumers only have one station nearby, they are more likely to be informed about the add-on which reduces the cross-subsidization incentive. Hence, lower gasoline competition should foster the effect. We study this by splitting the sample at the median number of nightly competitors in a 1 km radius.10 Figure 3reports our results on heterogeneity analyses. Indeed, in the 10 Our results also hold for different radii and sample splits not at the median. © 2023 The Authors. The Journal of Industrial Economics published by The Editorial Board of The Journal of Industrial Economics and John Wiley & Sons Ltd. ALCOHOL PROHIBITION AND PRICING AT THE PUMP 563 Figure 3 Heterogeneity Analyses: Intensive Margin Notes: This plot gives the treatment effect 𝛽3from the baseline regression for subsamples along firm characteristics. The y-axis documents the effect size in Eurocent/l, the x-axis gives the respective subsample. 90% and 95% confidence bands are reported. Standard errors are clustered at the county level [Colour figure can be viewed at wileyonlinelibrary.com] first panel of Figure 3, we find that lower competition is related to a higher nightly price increase after the prohibition lifting. Simultaneously, higher competition is correlated with stations lying in densely populated areas, so that stations in cities do not drive our effect.11 In cities, alcohol consumers may be motorized less often which does not incentivize changes in gasoline prices. Second, we study how ex-ante shop assortments impact the price effect’s size. To sort stations into different shop categories, we follow the definition by the Federal Cartel Office [2011]. Stations are sorted into premium and small assortment stations based on their brand affiliation. Premium stations are known for a wider assortment of products. Alcohol is a very simple product offered by any station, so that the marginal return and relative importance of alcohol revenues is typically higher in smaller shops. At the same time, larger shops imply larger per-consumer revenues from alcohol visitors, which fosters the cross-subsidization effect. Both arguments propose larger shops experience a lower price effect of the policy. In the second panel of Figure 3,we find premium stations with large product variety do not react significantly, while the price effect is especially evident for low assortment stations. Consumers who buy alcohol at gasoline stations may be likely to buy other shop products there as well, so that bigger shops do not experience a comparable 11 This also holds when studying the effect heterogeneity across county differences in the population density. © 2023 The Authors. The Journal of Industrial Economics published by The Editorial Board of The Journal of Industrial Economics and John Wiley & Sons Ltd. 564 KAI FISCHER shock to shops with smaller product varieties. In contrast, the premium station may face consumers who buy more after the prohibition but have visited the station before as well. This then does not lead to more gasoline sold at premium stations. Also, the null effect for premium stations might be a result of stronger cross-subsidization since gasoline purchasers might buy more products beyond alcohol when entering the store. Third, we study the role of market power. In the German gasoline market, market power is associated with vertical integration to oil refinery firms as these also supply competitors and have been determining the daily Edgeworth cycles for years (Federal Cartel Office [2011]; Siekmann [2017]). Vertically integrated, so-called “oligopolistic” brands are, for example, Shell, Aral (BP), or Total. We study whether the effect differs across oligopolistic and nonoligopolistic brands. We find that especially nonoligopolistic brands increase nightly prices after the prohibition lifting. Our results in the third panel of Figure 3show that oligopolistic stations’ price level was not lower before the prohibition lifting, so that a price drop during the prohibition did not occur at stations with market power. Fourth, we study a sample of only highway stations in the fourth panel of Figure 3. Highway stations have been subject to an alcohol prohibition throughout night hours, independent of the discussed alcohol prohibition. Hence, as these stations were still not subject to the opportunity to sell alcohol from December 08, 2017, onwards, we expect to observe a zero treatment effect. In terms of our model, both channels are switched off. That is why this analysis might be interpreted as a “quasi-placebo” test. Indeed, at highway stations, no price effect is found. Fifth, based on stations’ names and brand affiliations, we define a group of stations that likely do not sell any alcohol-related products at night, so that the policy should not affect the outcome. In terms of the model in Section III, this reflects a situation where no consumer is interested in alcohol or where alcohol does not give any utility to consumers. For means of econometric power, the respective group of stations pools supermarket stations, unmanned stations and car dealer stations. Supermarket stations most of the time do not have a shop at all as they typically are owned by the supermarket nearby (Haucap et al. [2017b]). Though, supermarkets are closed during the nightly prohibition (10 pm to 5 am), so that supermarket stations are not affected by the policy lifting. Unmanned gasoline stations (e.g., by the brand AVIA Xpress) do not operate a shop (at night). Similarly, car dealer stations’ main purpose is to provide fuel for the main business. As expected, we find that such stations, indeed, do not change prices in response to the policy (see fifth panel of Figure 3). Sixth, we investigate whether the price effect is mitigated by the fact whether stations are located in urban counties or the periphery. We follow the county-level definition of urbanity by Federal Institute for Research on Building, Urban Affairs and Spatial Development. We find stations in urban © 2023 The Authors. The Journal of Industrial Economics published by The Editorial Board of The Journal of Industrial Economics and John Wiley & Sons Ltd. ALCOHOL PROHIBITION AND PRICING AT THE PUMP 565 vicinities to increase prices more strongly (see sixth panel of Figure 3). This likely reflects the higher share of youths in urban regions, which Marcus and Siedler [2015] found to increase their alcohol demand. Hence, in urban stations, more consumers should receive utility from the newly available product after the prohibition lifting. In the last panel of Figure 3, we study heterogeneity in the local share of youths (18–25-year-olds). This traces back to Marcus and Siedler [2015], who find that the discussed alcohol prohibition especially reduced alcohol binge consumption among young adults. We investigate whether a higher share of youths proxies a demand shock for gasoline as well. With regard to the model, youths might reflect a consumer group, who is aware of the alcohol product (high 𝜆in the model above). As they gain most from the availability of alcohol, the alcohol-driven demand shift should cause higher prices for stations with a high local share of youths (see Result 2 in Section III). Our estimates do not reveal a clear treatment effect heterogeneity when comparing stations from municipalities above and below the median youth share. Though, when zooming in on the heterogeneity of the youth share more intensively, a clear relation between a higher youth share and a higher treatment effect is evident. For example, see Figure A1 in the appendix for treatment effect heterogeneity across terciles and quartiles of the distribution. Note that we, as a robustness check, also ran our heterogeneity analysis in a single regression instead of separate regressions. This should ensure that the different heterogeneity results are not driven by one and the same factor which correlates with several station characteristics. Table A2 in the appendix presents these results. Qualitatively our results do not change. Especially stations with few competitors at night and in municipalities with a high share of youths experience higher treatment effects. Also, small assortment stations increase prices more strongly. Station Activity. As we find that gasoline prices at stations in BadenWuerttemberg during the prohibition have been lower, there likely is an unambiguous effect on the overall revenues of stations: Alcohol revenues vanish and gasoline prices drop. Hence, it is a natural question whether some stations change how actively they participate in the market in response to the policy lifting. To study stations’ activity, we use the real-time price data to elicit whether a gasoline station in a certain week changed prices at night or not. If stations change prices, this will be indicative of whether they open at night. Due to data availability, we cannot fully exclude that effects on price changes are shaped by Edgeworth cycle adaptions due to the policy instead of operating times. Though, in the appendix, we provide some evidence in Table A3 on whether gasoline stations in Baden-Wuerttemberg show different Edgeworth cycle characteristics after the policy lifting. The number of price changes over the day as well as cycling frequency and asymmetry remain unaffected. © 2023 The Authors. The Journal of Industrial Economics published by The Editorial Board of The Journal of Industrial Economics and John Wiley & Sons Ltd. 566 KAI FISCHER Figure 4 Dynamic Effects on Likelihood to be Active at Night Notes: This plot gives dynamic estimates of the leads and lags from equation (4). The left plot defines 1[Active at Night]sw with changing prices between 10 pm and 5 am, the right plot takes a more restricting definition of price changes between 11 pm and 4 am. Standard errors are clustered at the county level. The exact timing of the beginning and end of the prohibition is indicated by the black vertical line. We provide 90 and 95%confidence intervals for all coefficients from a linear probability model [Colour figure can be viewed at wileyonlinelibrary.com] We determine whether a station has changed its price between 10 pm and 5 am and, for a second measure, whether there have been changes between 11 pm and 4 am. We apply a standard dynamic DID estimator in a two-way fixed effects model to studystations’ propensity to operate at night. Again, flat pre-trends will be indicative of whether the parallel trend assumption holds: (4) 1[Active at Night]sw =𝛼s+𝜆w+𝜏 ∑ t=𝜏,t≠−1,−2𝛾t1[BWs×Liftingw−t]+𝜖sw. 1[Active at Night]sw is a dummy which will turn one if a station shas operated at night in week w. We apply two definitions for this outcome: First, the variable will turn one if a station is active/changes the price at least once a week between 10 pm and 5 am. Second, the variable will turn one if a station is active/changes the price between 11 pm and 4 am at least once a week. Figure 4gives the dynamic estimates for both outcomes. It appears that the share of stations being active at night increases substantially after the lifting of the prohibition. In fact, stations in Baden-Wuerttemberg are 8.7 percentage points (or 10% respectively) more likely to operate/change prices at some point between 10 pm and 5 am than when the prohibition was active. In contrast to the price effect, which arises after 5–7 weeks, the reaction in night activity takes about twice as long until reaching a constant treatment effect level. This is very much in line with lower menu costs for price level changes than structural changes in a station’s activity at night. When investigating heterogeneous responses across stations with small or large assortment, we find heterogeneity, which corresponds to the price effects found above. Stations with a small assortment typically sell fewer products, © 2023 The Authors. The Journal of Industrial Economics published by The Editorial Board of The Journal of Industrial Economics and John Wiley & Sons Ltd. ALCOHOL PROHIBITION AND PRICING AT THE PUMP 567 Figure 5 Dynamic Effects on Station Activity: Heterogeneity Along Assortment Variety Notes: This plot gives dynamic estimates of the leads and lags from equation (4) for two subsamples of stations with heterogeneous store assortment. The outcomes 1[Active at Night]sw is defined as the weekly share on which prices have been changed between 10 pm and 5 am. Standard errors are clustered at the county level. The exact timing of the beginning and end of the prohibition is indicated by the black vertical line. We provide 90 and 95%confidence intervals for all coefficients from a linear probability model [Colour figure can be viewed at wileyonlinelibrary.com] so that a restriction on alcohol might hit them more strongly. Indeed, we find that such stations react more pronouncedly in activity during prohibition hours (see Figure 5). We also checked again, whether highway stations do not react to the policy in means of nightly activity and, indeed, that is observed. A concern is that price changes might not perfectly reflect opening hours. For example, stations that open at night but only start to change prices after the prohibition lifting, are implicitly understood to extend opening hours due to the policy. Hence, this would likely upward bias the estimated treatment effect. Therefore, we provide additional robustness checks on the effect of opening hours (see Section Cin the Appendix for an in-depth analysis). Using historical opening hours for a subset of gasoline stations (≈25% of all stations), which we obtained from the internet archive web.archive.org, we find opening hour reactions in line with our results above. Again opening hours are increased significantly in Baden-Wuerttemberg after the policy lifting—especially at stations with smaller shop assortment. Though, our robustness check identifies smaller treatment effects. This is likely due to the potential upward bias in the analysis based on price changes as explained above. © 2023 The Authors. The Journal of Industrial Economics published by The Editorial Board of The Journal of Industrial Economics and John Wiley & Sons Ltd. 568 KAI FISCHER Finally, the extended opening hours likely cause our baseline price effect to be downward-biased as more competitors have been found to correlate with lower prices in gasoline markets (Haucap et al. [2017a]; Martin [2023]; Pennerstorfer et al. [2020]). In the appendix, we show how a change in the number of nighttime competitors affects nighttime prices. Table A4 reports the results of an interaction term analysis in columns (1) and (2). We find that the treatment effect is larger in concentrated markets. In Figure A3, we exploit the staggered timing of competitors’ nighttime entry across incumbents and show that nighttime entry in a 1km radius decreases prices by up to 1 Eurocent/l.12 The effect size is very similar to Fischer et al. [2023] who estimate the causal effect of station entry on incumbent prices to be around 0.5ct/l. This indicates, that, indeed, our baseline results are downward-biased. As nighttime entry decreases prices by up to 1 Eurocent/l, it absorbs the policy-induced price effect completely in markets where nighttime entry takes place. However, nighttime entry is costly and might not be possible or profitable in all markets, so that positive price effects remain in the majority of markets, which are not entered. This leads to the on average positive price effect of the policy found above. In columns (3) and (4) of Table A4,wetryto quantify by how much nighttime entry decreases the price effect which would have been observed absent nighttime entry. For this, we include the entry of competitors as “bad control” in the price regressions. We show that the price effect changes only slightly in comparison to the estimated baseline effect. This indicates that entry only marginally decreases the policy’s average price effect. Traffic Flow Analysis. To better understand the mechanism underlying the observedpriceeffects, westudy trafficflow reactions to thepolicy.Theanalysis is twofold: First, we analyze whether nightly traffic increases in response to the policy lifting in Baden-Wuerttemberg and especially near open gasoline stations. Secondly, we study how traffic at the federal state’s border is affected by the shock. We start by running the triple difference-in-differences regression from above on the logged number of counted cars for traffic counters near gasoline stations open at night (≤2km linear distance). Figure 6’s blue estimates report the dynamic effect of the policy lifting on traffic near gasoline stations in Baden-Wuerttemberg in four-week bins. After the policy lifting, nightly traffic in Baden-Wuerttemberg persistently increases by up to 5%–10%. This is indicative of more cars traveling near and, hence, likely also to gasoline stations. This is in line with a demand expansion through the service quality channel as alcohol is available after the policy. To show that this effect really reflects an increasing interest in gasoline stations, we run this analysis 12 Similar procedures can be found in the reduced-form entry literature as in Arcidiacono et al. [2020], Goolsbee and Syverson [2008] and Matsa [2011]. © 2023 The Authors. The Journal of Industrial Economics published by The Editorial Board of The Journal of Industrial Economics and John Wiley & Sons Ltd. ALCOHOL PROHIBITION AND PRICING AT THE PUMP 569 Figure 6 Dynamic Effects on Traffic Flows Notes: This plot gives dynamic estimates of the leads and lags from equation (4)wherethe outcome variable is logged traffic flows. The blue estimates give the effect of the policy on traffic counts in Baden-Wuerttemberg near gasoline stations (≤2km linear distance) in a subsample of traffic counters of maximum 2km linear distance to gasoline stations open at night. The red estimates give the effect of the policy of traffic counts near the border (≤2km linear distance) to Baden-Wuerttemberg at non-Baden-Wuerttemberg counters in a subsample of non-Baden-Wuerttemberg traffic counters. To account for the logarithm of very few zero traffic observations, we use the hyperbolic sine transformation of the outcome variable. Standard errors are clustered at the county level. The exact timing of the beginning and end of the prohibition is indicated by the black vertical line. We provide 90 and 95%confidence intervals for all coefficients [Colour figure can be viewed at wileyonlinelibrary.com] separately for groups of traffic counters that have different distances to the nearest open gasoline station. The traffic effect should be highest for counters near gasoline stations if traffic increases really relate to more visits to gasoline stations. Figure 7, indeed, shows that this is the case. While traffic in Baden-Wuerttemberg overall increases by around 5%, this effect is strongest for counters right next to gasoline stations (≤1 km linear distance). There is no significant effect on traffic flows for counters more than 2 km away from open gasoline stations. We take this as support for our demand expansion channel. In addition, Figure A2 in the appendix reveals that the increase in traffic is especially high for traffic counters in municipalities with a high share of youths. This fits the story that especially youths respond to the policy change. © 2023 The Authors. The Journal of Industrial Economics published by The Editorial Board of The Journal of Industrial Economics and John Wiley & Sons Ltd. 570 KAI FISCHER Figure 7 Effects on Traffic Flow by Counter Distance to Station Notes: This plot gives the estimates from the static version of the equation (4) where the outcome variable is logged traffic flows. Stations are grouped by the minimum distance to a gasoline station open at night. To account for the logarithm of very few zero traffic observations, we use the hyperbolic sine transformation of the outcome variable. Standard errors are clustered at the county level. The exact timing of the beginning and end of the prohibition is indicated by the black vertical line. We provide 90 and 95%confidence intervals for all coefficients [Colour figure canbeviewedatwileyonlinelibrary.com] We further study border traffic. Before the policy lifting, consumers living in Baden-Wuerttemberg had to leave the federal state to get alcohol at night at off-premise locations. This border traffic should have been reduced after the policy lifting. For this, we compare traffic at traffic counters outside of Baden-Wuerttemberg but near the border (≤2 km linear distance) to all other non-Baden-Wuerttemberg traffic counters before and after the policy. Figure 6’s red estimates report the results of the triple difference-in-differences regression. Indeed, trafficnearthebordertoBaden-Wuerttembergbutoutside of Baden-Wuerttemberg falls in response to the policy. This can be interpreted as a demand shift to gasoline stations in Baden-Wuerttemberg. Also, this result indicates that alcohol consumption has a sufficiently high value to consumers to induce border travel. Note that we also tried out other distance thresholds up to 5 km distance to the border and our results qualitatively remain the same. We, further, reproduce the heterogeneity analysis from Figure 3with traffic flows as an outcome to support the mechanisms described above. To conduct © 2023 The Authors. The Journal of Industrial Economics published by The Editorial Board of The Journal of Industrial Economics and John Wiley & Sons Ltd. ALCOHOL PROHIBITION AND PRICING AT THE PUMP 571 heterogeneity analysesalonggasolinestation characteristics(brand,shop size, etc.), we match counters to the nearest station. The results in Figure A4 in the appendix show that traffic increases more strongly at counters with many stations nearby and also is stronger in urban areas with a high youth share. We complement the traffic data results on a demand expansion mechanism with an analysis of geo-coded traffic accidents with personal damage in Germany.13 In Table A5, we, at the extensive margin, do not find an effect of the policy lifting on the overall number of accidents with personal damage in Baden-Wuerttemberg.14 However, we show that the likelihood of accidents being very near (≤1 km) to open gasoline stations increases by 3% after the policy lifting. On average, the distance of accidents to the nearest gasoline station at night decreases by 7% after the policy lifting. This shows that traffic flows likely shift toward areas surrounding gasoline stations. Bite of the Policy. To quantify the consequences of the policy for gas stations as well as consumers, it is not sufficient to show that the price effect is around 5% of an average station’s margin. We need to understand how many consumers visit gasoline stations at night. To approximate daytime-specific demand, we rely on Google Popularity data,15 which we scraped for all stations available once in July 2019 (≈85% of all German stations). Figure 8 plots the average distribution of gas station visits over the course of the day. Non-negligible 7%–8% of visits lie in the treatment time between 10 pm and 5am. 16 Moreover, stations do not only use revenues from gasoline sales during the prohibition but also lose alcohol revenues. Industry surveys (Scope Ratings [2018]) show that the annual alcohol revenues of an average station are approximately 100,000 Euro. Furthermore, consumers potentially switching away from stations, which increase prices more strongly after policy lifting, are a concern when discussing the exposure of consumers to the policy. Especially informed consumers would not be affected by the policy then and distributional implications would arise. While we do not observe actual transactions—so where consumers fuel—we can show that consumers can hardly avoid being affected by the policy effect as long as the policy shifts the complete price 13 The data comes from the “Unfallatlas” (https://unfallatlas.statistikportal.de/) of the Federal Statistical Office and the Statistical Offices of the German States and covers traffic accidents with personal damage for 12 out of 16 federal states. 14 This is in line with the results in Baueml et al. [2023] who do not find the policy’s introduction in 2010 to affect alcohol-related traffic accidents. 15 On Google Maps, it is reported how crowded and popular a business is for every hour of the day. Popularity is based on measures such as mobile phone mobility and traffic and is reported in an index between 0 and 100 at the station level. 16 In a telephone survey of the German Ministry for Economic Affairs and Energy from 2016, the share of respondents who fuel at this time is of similar magnitude (Bundesregierung [2018]). © 2023 The Authors. The Journal of Industrial Economics published by The Editorial Board of The Journal of Industrial Economics and John Wiley & Sons Ltd. 578 KAI FISCHER Figure A1 Price Effect: Heterogeneity Along “Youth Share” Distribution Notes: Heterogeneity analysis based on sample splits along the distribution of the variable “Youth Share”. 90% and 95% confidence bands are reported. Standard errors are clustered at the county level [Colour figure can be viewed at wileyonlinelibrary.com] TABLE A2 HETEROGENEITY ANALYSIS:ROBUSTNESS CHECK Gasoline price in Euro/l (1) (2) BW ×Post ×1[Street Station] −0.0064 (0.0079) BW ×Post ×1[Urban] 0.0054 (0.0051) BW ×Post ×1[Below median competition] 0.0064∗∗∗ (0.0024) BW ×Post ×1[Below median youth share] −0.0048∗ (0.0027) BW ×Post ×1[Large assortment] −0.0127∗∗∗ (0.0046) BW ×Post ×1[Oligopolistic] −0.0024 (0.0042) BW ×Post ×1[No shop sales] −0.0102∗ (0.0062) BW ×Post ×Night ×1[Street station] −0.0059 (0.0076) BW ×Post ×Night ×1[Urban] 0.0056 (0.0050) © 2023 The Authors. The Journal of Industrial Economics published by The Editorial Board of The Journal of Industrial Economics and John Wiley & Sons Ltd. ALCOHOL PROHIBITION AND PRICING AT THE PUMP 579 TABLE A2 Continued Gasoline price in Euro/l (1) (2) BW ×Post ×Night ×1[Below median competition] 0.0063∗∗ (0.0024) BW ×Post ×Night ×1[Below median youth share] −0.0045∗ (0.0026) BW ×Post ×Night ×1[Large assortment] −0.0131∗∗∗ (0.0046) BW ×Post ×Night ×1[Oligopolistic] −0.0024 (0.0042) BW ×Post ×Night ×1[No shop sales] −0.0110∗ (0.0062) Sample Night prices All prices Observations 296,598 593,193 Adjusted R20.894 0.911 Notes: All results are based on OLS regressions with standard errors clustered at the county level. The regression setup extends the regression equation from the “Data and Empirical Strategy” section by additional interactions. Model (1) only uses night prices and a triple difference-in-differences estimator, while model (2) uses quadruple interactions to extend the baseline triple difference-in-differences estimator to account for effect heterogeneity. Other interactions not reported in the regression table. *p<0.1; **p<0.05; ***p<0.01. Figure A2 Heterogeneity in Traffic Response Along Youth Share Distribution Thisplot givesthe estimates from thetriple DiDmodel presented in the Section“Dataand Empirical Strategy” with trafficflows as outcome. The analysis is run for all stations, only stations in a radius of 2 km linear distance to gasoline stations open at night or a 4 km radius. 90% and 95% confidence bands are reported. Standard errors are clustered at the county level Notes: [Colour figure can be viewed at wileyonlinelibrary.com] © 2023 The Authors. The Journal of Industrial Economics published by The Editorial Board of The Journal of Industrial Economics and John Wiley & Sons Ltd. 580 KAI FISCHER TABLE A3 EDGEWORTH CYCLE CHARACTERISTICS Median price change ln(# Price Changes) Price spread (1) (2) (3) BW ×Post 0.0004 0.0225 0.0018 (0.0003) (0.0152) (00017) Approach DID DID DID Observations 2,155,817 2,156,356 2,118,970 Adjusted R20.189 0.753 0.591 Notes: All results are based on OLS regressions with standard errors clustered at the county level. The regression setup follows a simple DID. *p<0.1; **p<0.05; ***p<0.01. TABLE A4 REGRESSIONS ON MITIGATING ENTRY EFFECT Gasoline price in Euro/l (1) (2) (3) (4) BW ×Post 0.0097∗∗∗ 0.0082∗∗∗ (0.0027) (0.0024) BW ×Post ×(# Competitors ∈ [0,1] km) −0.0043∗∗∗ −0.0039∗∗ (0.0015) (0.0015) BW ×Post ×(# Competitors ∈ (1,2] km) −0.0004 −0.0002 (0.0011) (0.0011) # New competitors active ∈[0,1] km −0.0101∗∗∗ −0.0091∗∗∗ (0.0033) (0.0031) # New competitors active ∈(1,2] km −0.0042 −0.0033 (0.0028) (0.0028) Station FE ✓✓✓✓ Week FE ✓×✓× State ×Week FE ×✓×✓ Observations 296,598 296,598 296,598 296,598 Adjusted R20.878 0.882 0.876 0.880 Notes: All results are based on OLS regressions with standard errors clustered at the county level. The regression equation is a triple difference-in-differences regression for nighttime prices comparing prices across federal states, before and after the policy and across competition environments. The other interaction terms of triple difference-in-differences estimator in columns (1) and (2) are omitted. *p<0.1; **p<0.05; ***p<0.01. © 2023 The Authors. The Journal of Industrial Economics published by The Editorial Board of The Journal of Industrial Economics and John Wiley & Sons Ltd. ALCOHOL PROHIBITION AND PRICING AT THE PUMP 581 Figure A3 Effects of Nighttime Entry on Prices Notes: This plot gives estimates from an event study regression of nightly prices on leads and lags of the nighttime entry of competitors in a 1 km or 1–2 km radius around incumbents. Station and state-week fixed effects are included. Nighttime entry of stations is identified in the week after which a station operates two consecutives weeks at night for the first time. Endpoints are binned and not reported due to an unbalanced panel in event time (Fuest et al. [2018]). Standard errors are clustered at the county level. The exact timing of entry is indicated by the black vertical line. We provide 90% and 95%confidence intervals for all coefficients [Colour figure can be viewed at wileyonlinelibrary.com] Figure A4 Heterogeneity Analyses: Traffic Notes: This plot gives the treatment effect of the policy on traffic flows for subsamples along counter and station characteristics. The y-axis documents the effect size in %, the x-axis gives the respective subsample. 90% and 95% confidence bands are reported. Standard errors are clustered at the county level. To be able to conduct heterogeneity analyses along station characteristics, we match counters to the nearest stations operating 24/7. We only include counters in the analyses that are closer than 5km to a gasoline [Colour figure can be viewed at wileyonlinelibrary.com] © 2023 The Authors. The Journal of Industrial Economics published by The Editorial Board of The Journal of Industrial Economics and John Wiley & Sons Ltd. 582 KAI FISCHER TABLE A5 POLICY LIFTING’SEFFECT ON ACCIDENTS ln(# Accidents) 1[Distance Station ≤1] log(Distance Station) (1) (2) (3) (4) (5) (6) (7) (8) (9) BW ×Post ×Night −0.017 0.031∗∗ −0.069∗∗ (0.022) (0.015) (0.035) BW ×Post −0.027 −0.043 0.004 0.036∗∗ −0.003 −0.069∗∗ (0.023) (0.030) (0.004) (0.014) (0.010) (0.034) County FE ✓✓✓✓✓✓ ✓ ✓ ✓ Month × Hour FE ✓✓✓✓✓✓ ✓ ✓ ✓ Approach TDID Only Only TDID Only Only TDID Only Only Day Night Day Night Day Night Observations 182,016 128,928 53,088 393,445 368,544 24,901 393,445 368,544 24,901 Adjusted R20.512 0.426 0.241 0.170 0.158 0.187 0.169 0.148 0.273 Notes: Regressions (4) to (9) are at the individual accident-level and hence also include controls for whether the accident included bicycles, motorbikes and pedestrians. Regressions (1) to (3) are at the county-month-hour level. The distance to the nearest open station depends on the time of the day. The post-treatment period is the first full month after the policy lifting and beyond. Standard errors are clustered at the county level. *p<0.1; **p<0.05; ***p<0.01. TABLE A6 ROBUSTNESS CHECKS:PROPENSITY SCORE MATCHING —BALANCING CONDITION Before Matching After Matching ΔΔ Control BW (p-value) Control BW (p-value) Outcomes ln[E5 gasoline price (Day)] 0.320 0.314 0.00∗∗∗ 0.315 0.314 0.74 ln[E5 gasoline price (Night)] 0.368 0.364 0.03∗∗ 0.364 0.364 0.99 ln[Margin (Day)] -2.385 -2.473 0.00∗∗∗ -2.463 -2.473 0.59 ln[Margin (Night)] -1.943 -1.989 0.01∗∗ -1.988 -1.989 0.97 Competition # Competitors 0.5 km Radius (Day) 0.471 0.452 0.51 0.467 0.452 0.70 # Competitors 0.5 km Radius (Night) 0.258 0.229 0.17 0.231 0.229 0.96 # Competitors 1 km Radius (Day) 1.078 1.097 0.71 1.079 1.097 0.79 # Competitors 1 km Radius (Night) 0.546 0.538 0.79 0.502 0.538 0.41 Stations characteristics Share of Youths (18-25-year-old, Munic. Level) 0.075 0.083 0.00∗∗∗ 0.083 0.083 0.99 Premium station 0.439 0.415 0.21 0.394 0.415 0.43 Oligopolistic station 0.373 0.276 0.00∗∗∗ 0.247 0.276 0.21 Notes: Matching was done in a sample of observations from the last pre-treatment week only. Matching was conducted with nearest neighbor matching without replacement. Only stations, which set a price (i.e., which were active) in the respective week, were included in the matching regression. Only observations with positive margins included in the matching regression. *p<0.1; **p<0.05; ***p<0.01. © 2023 The Authors. The Journal of Industrial Economics published by The Editorial Board of The Journal of Industrial Economics and John Wiley & Sons Ltd. ALCOHOL PROHIBITION AND PRICING AT THE PUMP 583 TABLE A7 ROBUSTNESS CHECKS:PROPENSITY SCORE MATCHING —DIDRESULTS Gasoline price in Euro/l ln(Gross Margin) (1) (2) (3) (4) BW ×Night ×Post 0.0051∗∗ 0.0702∗∗∗ (0.0025) (0.0219) BW ×Post 0.0068∗∗ 0.0018 (0.0028) (0.0020) Approach TDID DID DID TDID Sample Baseline Only Night Only Day Baseline Observations 147,818 73,909 73,909 147,796 Adjusted R20.887 0.865 0.952 0.768 Notes: All results are based on OLS regressions with standard errors clustered at the county level. The regression setup follows the regression equation from the “Data and Empirical Strategy” section. The sample is based on a propensity score matching estimator with nearest-neighbor matching without replacement within the last pre-treatment period. *p<0.1; **p<0.05; ***p<0.01. TABLE A8 ROBUSTNESS CHECKS:TDIDSETUP Gasoline price in Euro/l (1) (2) (3) (Baseline) (5) (6) (7) BW ×Night ×Post 0.0055∗∗ 0.0055∗∗ 0.0055∗∗ 0.0056∗∗ 0.0056∗∗ 0.0056∗∗ 0.0056∗∗ (0.0022) (0.0022) (0.0022) (0.0023) (0.0023) (0.0023) (0.0023) Approach TDID TDID TDID TDID TDID TDID TDID BW dummy ✓×× × ××× Post dummy ✓✓× × ××× Night dummy ✓✓✓ × ××× BW ×Post ✓✓✓ ✓ ××× BW ×Night ✓✓✓✓✓×× Post ×Night ✓✓✓ × ××× Station FE ×✓✓✓✓✓✓ Week FE ××✓ ✓ ✓✓✓ Night ×Week FE ××× ✓ ✓✓✓ BW ×Week FE ××× × ✓✓✓ Night ×Station FE ××× × ×✓✓ State Trends ××× × ××✓ Observations 593,193 593,193 593,193 593,193 593,193 593,193 593,193 Adjusted R20.072 0.529 0.868 0.889 0.890 0.912 0.914 Notes: All results are based on OLS regressions with standard errors clustered at the county level. The regression setup follows the regression equation from the “Data and Empirical Strategy” section. The models provide different specifications of a TDID setup. *p<0.1; **p<0.05; ***p<0.01. © 2023 The Authors. The Journal of Industrial Economics published by The Editorial Board of The Journal of Industrial Economics and John Wiley & Sons Ltd. 584 KAI FISCHER TABLE A9 ROBUSTNESS CHECKS:STATE BORDER Gasoline price in Euro/l (1) (2) BW ×Night ×Post −0.0033 −0.0034 (0.0215) (0.0077) Approach TDID TDID Robustness check Border (≤1 km) Border (≤2.5 km) Observations 1,682 7,310 Adjusted R20.874 0.879 Notes: All results are based on OLS regressions with standard errors clustered at the county level. The regression setup follows the regression equation from the “Data and Empirical Strategy” section. We subsample stations near the policy border. *p<0.1; **p<0.05; ***p<0.01. Figure A5 Treatment Effect for Individual Federal State as Control Group Notes: This plot gives the estimates from the triple DiD model presented in the Section “Data and Empirical Strategy” for different control groups. In particular, each estimate uses a different federal state as control group. Standard errors are clustered at the county level. We provide 90 and 95%confidence intervals for all coefficients. States are as follows: Schleswig-Holstein (1), Hamburg (2), Lower Saxony (3), Hamburg (4), Northrhine-Westphalia (5), Hesse (6), Rhineland-Palatinate (7), Baden-Wuerttemberg (8), Bavaria (9), Saarland (10), Berlin (11), Brandenburg (12), Mecklenburg-Hither Pomerania (13), Saxony (14), Saxony-Anhalt (15), Thuringia (16) [Colour figure can be viewed at wileyonlinelibrary.com] © 2023 The Authors. The Journal of Industrial Economics published by The Editorial Board of The Journal of Industrial Economics and John Wiley & Sons Ltd. ALCOHOL PROHIBITION AND PRICING AT THE PUMP 585 TABLE A10 INFERENCE OF BASELINE REGRESSION Coefficient baseline 0.0056 p-value One-way clustering Station level (Baseline) (0.0014)∗∗∗ County level (Baseline) (0.0022)∗∗ Two-digit postcode level (0.0029)∗ Two-way clustering Station level +week (0.0005)∗∗∗ County level +week (0.0009)∗∗∗ Two-digit postcode level +week (0.0010)∗∗∗ Wild bootstrap (999 rep.) Station level (0.0015)∗∗∗ County level (0.0028)∗∗ Two-digit postcode level (0.0029)∗ Cluster size N(Stations) 6,144 N(Counties) 401 N(Postcode areas) 92 N(Week) 52 *p<0.1; **p<0.05; ***p<0.01. TABLE A11 DIFFERENT EFFECT WINDOWS Gasoline Price in Euro/l (Baseline) (2) (3) (4) (5) (6) BW ×Night ×Post 0.0056∗∗ 0.0039∗∗∗ 0.0049∗∗ 0.0066∗∗∗ 0.0068∗∗∗ 0.0070∗∗∗ (0.0023) (0.0014) (0.0019) (0.0020) (0.0022) (0.0023) Effect window (in Weeks) [-13, 38] [-10, 10] [-10, 20] [-20, 20] [-20, 30] [-20, 40] Observations 593,193 239,037 353,416 467,986 582,423 696,355 Adjusted R20.889 0.833 0.846 0.833 0.865 0.888 Notes: All results are based on OLS regressions with standard errors clustered at the county level. The outcome variable gives where a station chages the price at least one per week in the time period between 10 pm and 5 am or 11 pm and 4 am. The independent variable gives whether a station opens 24/7 in a certain week or not. Observations are at the station ×week level. *p<0.1; **p<0.05; ***p<0.01. APPENDIX B QUANTILE TREATMENT EFFECTS–ROBUSTNESS CHECK To show that the policy lifting shifts the gasoline price distribution in a first-order stochastic manner, that is, the unconditional quantile treatment effects are positive at all quantiles, we elicit the counterfactual price distribution—so prices in Baden-Wuerttemberg absent the policy lifting after December 08, 2017—in the style of Chernozhukov et al. [2013]. To be precise, we estimate by how much the policy lifting increases/decreases the likelihood of price observations to lie below/above certain price thresholds. Formally, the value of the empirical distribution function (ECDF) of the counterfactual distribution at price pis given through the following © 2023 The Authors. The Journal of Industrial Economics published by The Editorial Board of The Journal of Industrial Economics and John Wiley & Sons Ltd. 586 KAI FISCHER Figure B1 Distributional Effects of the Policy: Counterfactual Price Distribution Notes: This plot gives the empirical distribution function of observed nighttime post-lifting prices in Baden-Wuerttemberg (blue) and the counterfactual distribution for a scenario without policy lifting (red). The counterfactual distribution comes from distribution regression in the style of Chernozhukov et al. [2013] in one Eurocent/l steps. We provide 95%confidence intervals. Standard errors are clustered at the county level. The distributions are trimmed at the 5th and 95th percentile [Colour figure can be viewed at wileyonlinelibrary.com] “distribution regression” which estimates the change in the propensity of a price to be below pdue to the treatment: (B1) 1[PE5 sw <p]=𝛽BWs×Postw+𝛼s+𝜆w+esw. The value of the counterfactual ECDF is given by the ECDF of the observed prices minus 𝛽. Repeating the procedure for multiple pconstructs the full counterfactual distribution. Figure B1 visually compares observed and counterfactual prices. The policy lifting shifted the price distribution in a first-order stochastic manner to the right. This is indicative of all consumers being affected as the effect is not just driven by one part of the distribution. Instead, consumers at all quantiles of the distribution are affected. Note that we use weekly average prices and hence we do not fully show that there is no switching opportunity for consumers at a certain point in time which avoids price increases. Admittedly, weekly average prices are indicative of a lack of such switching opportunities. Nevertheless, we encounter this concern by running the same distribution analysis for daily station prices at midnight. We chose midnight as timing as there are barely any price changes after midnight until the end of the prohibition (5 am) (see Figure B2). Hence, the price distribution at midnight likely reflects the price distribution at 1 am, 2 am and, hence, most parts of the nightly prohibition period between 10 pm and 5 am. Results do not change qualitatively (see Figure B3). © 2023 The Authors. The Journal of Industrial Economics published by The Editorial Board of The Journal of Industrial Economics and John Wiley & Sons Ltd. ALCOHOL PROHIBITION AND PRICING AT THE PUMP 587 Figure B2 Timing of Price Changes Notes: This plot gives the timing of price changes of stations in the sample [Colour figure can be viewed at wileyonlinelibrary.com] Figure B3 Distributional Effects of the Policy: Counterfactual Price Distribution—Midnight Notes: This plot gives the empirical distribution function of observed midnight post-lifting prices in Baden-Wuerttemberg (blue) and the counterfactual distribution for a scenario without policy lifting (red). The counterfactual distribution comes from distribution regression in the style of Chernozhukov et al. [2013] in one Eurocent/l steps. We provide 95%confidence intervals. Standard errors are clustered at the county level. The distributions are trimmed at the 5th and 95th percentile [Colour figure can be viewed at wileyonlinelibrary.com] © 2023 The Authors. The Journal of Industrial Economics published by The Editorial Board of The Journal of Industrial Economics and John Wiley & Sons Ltd. 594 KAI FISCHER TABLE E1 EFFECT OF POLICY LIFTING ON LIKELIHOOD OF RANK REVERSALS rrct (1) (2) (3) (4) 1[1 ≥Distance in km ≤1.5] −0.0109∗∗ (0.0043) 1[0.5 ≥Distance in km ≤1] −0.0108∗∗ (0.0047) 1[0.15 ≥Distance in km ≤0.5] −0.0159∗∗∗ (0.0060) 1[Distance in km ≤0.15] −0.0260∗∗∗ (0.0083) BW ×Post −0.0040 0.0101 0.0090 (0.0194) (0.0122) (0.0084) Couple Distance ≤2km ≤0.5km ≤1km ≤2km Couple FE ×✓✓✓ Observations 7,073 991 2,513 7,073 Adjusted R20.002 0.408 0.359 0.345 Notes: Rank reversal measures based on daily prices at midnight. Standard errors are clustered at the station couple level. Post dummy included in the regressions. *p<0.1; **p<0.05; ***p<0.01. for all observations before and once for all observations after the policy lifting. Only data from dates on which both stations operate is used. We use data from midnight prices. As prices hardly change during nighthours (see Figure B2), this analysis likely holds for all other points in time during the nightly prohibition. We then run the following regressions for a subsample of couples with a maximum linear distance between the two stations of 1 or 2 km: rrct =𝜆c+𝛾Postt+𝛽BWc×Postt+𝜖ct, where 𝛽gives the change in rank reversals related to the policy lifting. A couple is considered to belong to Baden-Wuerttemberg (BWc=1) if both stations are located in Baden-Wuerttemberg but our results also hold when a couple is also treated in the case that only one station lies in Baden-Wuerttemberg. Table E1 shows two results: First, column (1) shows that frictions decrease for a lower distance between stations which is in line with findings in the literature (Chandra and Tappata [2011]; Martin [2023]; Pennerstorfer et al. [2020]). Second, we show that the policy does not affect rank reversals in columns (2) to (4). Hence, there is no indication for a change in consumer information caused by the policy. APPENDIX F CORRELATION OF PRICES In this section of the appendix, we show that prices of neighboring stations do not comove more or less in response to the policy lifting. Demand could have become less elastic in Baden-Wuerttemberg after the policy lifting as more consumers visit the gasoline station for alcohol and, hence, might care less about the gasoline price. If this was the case, prices of competitors would become © 2023 The Authors. The Journal of Industrial Economics published by The Editorial Board of The Journal of Industrial Economics and John Wiley & Sons Ltd. ALCOHOL PROHIBITION AND PRICING AT THE PUMP 595 TABLE F1 EFFECT OF POLICY LIFTING ON NEIGHBORING STATIONS’PRICE CORRELATION corrct Actual Prices Residualized Prices (1) (2) (3) (4) (5) (6) BW ×Post 0.0341 0.0157 0.0288∗∗ −0.0343 −0.0200 −0.0114 (0.0464) (0.0240) (0.0145) (0.0524) (0.0367) (0.0217) Couple Distance ≤0.5km ≤1km ≤2km ≤0.5km ≤1km ≤2km Couple FE ✓✓✓✓✓✓ Observations 957 2476 7017 990 2510 7062 Adjusted R20.449 0.496 0.521 0.652 0.622 0.623 Notes: Correlations calculated based on daily, station-level midnight prices. Standard errors are clustered at the station couple level. Post dummy included in the regressions. *p<0.1; **p<0.05; ***p<0.01. less important (less elastic cross-price elasticity). At the extreme, demand could be sufficiently inelastic so that stations become quasi-monopolists. Then, neighboring stations’ prices will not be strategic responses. To examine whether neighboring stations’ price comovement is affected by the policy lifting, we setup the following difference-in-differences regression equation: corrct =𝜆c+𝛾Postt+𝛽BWc×Postt+𝜖ct, where corrct is the correlation between midnight prices of neighboring stations for two time periods t—before and after the treatment—separately. We later on use the correlation of actual prices as well as prices residualized from state-date specific price effects. The latter accounts for correlation purely caused by input price fluctuations. 𝛽gives the treatment effect. Table F1 gives the estimation results. There is no robust evidence for a change in the correlation of neighboring stations’ prices. This supports our claim that the price effect is not caused by less elastic demand in response to the policy. This would have likely resulted in a lower correlation of prices after the policy lifting in Baden-Wuerttemberg. REFERENCES Arcidiacono, P.; Ellickson, P. B.; Mela, C. F. and Singleton, J. D., 2020, ‘The Competitive Effect of Entry: Evidence from Supercenter Expansion,’ American Economic Journal: Applied Economics, 12(3), pp. 175–206. Armstrong, M. and Vickers, J., 2012, ‘Consumer Protection and Contingent Charges,’ Journal of Economic Literature, 50(2), pp. 477–493. Assad, S.; Clark, R.; Ershov, D. and Xu, L., 2023, ‘Algorithmic Pricing an Competition: Empirical Evidence from the German Retail Gasoline Market,’ Journal of Political Economy, forthcoming. Baueml, M.; Marcus, J. and Siedler, T., 2023, ‘Health Effects of a Ban on Late-Night Alcohol Sales,’ Health Economics, 32(1), pp. 65–89. Baumann, F.; Buchwald, A.; Friehe, T.; Hottenrott, H. and Mechtel, M., 2019, ‘The Effect of a Ban on Late-Night Off-Premise Alcohol Sales on Violent Crime: Evidence from Germany,’ International Review of Law and Economics, 60, 105850. © 2023 The Authors. The Journal of Industrial Economics published by The Editorial Board of The Journal of Industrial Economics and John Wiley & Sons Ltd. 596 KAI FISCHER Bundesregierung, 2018, ‘Bericht über die Ergebnisse der Arbeit der Markttransparenzstelle für Kraftstoffe und die hieraus gewonnenen Erfahrungen,’ Drucksache, 19/3693, pp. 1–40. Chandra, A. and Tappata, M., 2011, ‘Consumer Search and Dynamic Price Dispersion: An Application to Gasoline Markets,’ RAND Journal of Economics, 42(4), pp. 681–704. Chernozhukov, V.; Fernandez-Val, I. and Melly, B., 2013, ‘Inference on counterfactual distributions,’ Econometrica, 81(6), pp. 2205–2268. Doyle, J. J. J.; Muehlegger, E. and Samphantharak, K., 2010, ‘Edgeworth Cycles Revisited,’ Energy Economics, 32(3), pp. 651–660. FAZ, 2015, Tankstellen ärgern sich über Benzinpreise (Frankfurter Allgemeine Zeitung, Frankfurt, Germany). Federal Cartel Office, 2011, Sektorenuntersuchung Kraftstoffe (Federal Cartel Office, Bonn). Federal Cartel Office, 2018, Markttransparenzstelle für Kraftstoffe (MTS-K): Jahresbericht 2017 (Federal Cartel Office, Bonn). Federal Cartel Office, 2019, Markttransparenzstelle für Kraftstoffe (MTS-K): Jahresbericht 2018 (Federal Cartel Office, Bonn). Fischer, K.; Martin, S. and Schmidt-Dengler, P., 2023, ‘The Heterogeneous Effects of Entry on Prices,’ CESifo,Working Paper 10566, pp. 1–59. Fuest, C.; Peichl, A. and Siegloch, S., 2018, ‘Do Higher Corporate Taxes Reduce Wages? Micro Evidence from Germany,’ American Economic Review, 108(2), pp. 393–418. Gabaix, X. and Laibson, D., 2006, ‘Shrouded Attributes, Consumer Myopia, and Information Suppression in Competitive Markets,’ The Quarterly Journal of Economics, 121(2), pp. 505–540. Garthwaite, C.; Gross, T. and Notowidigdo, M., 2014, ‘Public Health Insurance, Labor Supply and Employment Lock,’ Quarterly Journal of Economics, 129, pp. 653–696. Goolsbee, A. and Syverson, C., 2008, ‘How Do Incumbents Respond to the Threat of Entry?’ Quarterly Journal of Economics, 123(4), pp. 1611–1633. Haucap, J.; Heimeshoff, U. and Siekmann, M., 2017a, ‘Fuel Prices and Station Heterogeneity on Retail Gasoline Markets,’ Energy Journal, 38(6), pp. 81–103. Haucap, J.; Heimeshoff, U. and Siekmann, M., 2017b, ‘Selling Gasoline as a By-Product: The Impact of Market Structure on Local Prices,’ DICE Discussion Paper, 240, pp. 1–43. Heidhues, P.; Koszegi, B. and Murooka, T., 2017, ‘Inferior Products and Profitable Deception,’ Review of Economic Studies, 84, pp. 323–356. Isphording, I. E.; Lipfert, M. and Pestel, N., 2021, ‘Does Re-Opening School Contribute to the Spread of SARS-CoV-2? Evidence from Staggered Summer Breaks in Germany,’ Journal of Public Economics, 198, pp. 104426. Ivanov, A., 2019, ‘Lohnt es sich noch, eine Tankstelle zu betreiben?’Handelsblatt,(Düsseldorf, Germany). Korff, A., 2021, ‘Competition on the Fast Lane: The Price Structure of Homogeneous Retail Gasoline Stations,’ DICE Discussion Paper 359, pp. 1–42. Lal, R. and Matutes, C., 1994, ‘Retail Pricing and Advertising Strategies,’ Journal of Business, 67(3), pp. 345–370. Lichter, A. and Schiprowski, A., 2021, ‘Benefit Duration, Job Search Behavior and Re-Employment,’ Journal of Public Economics, 193, pp. 104326. Marcus, J. and Siedler, T., 2015, ‘Reducing Binge Alcohol? The Effect of a Ban in Late-Night Off-Premise Alcohol Sales on Alcohol-Related Hospital Stays in Germany,’ Journal of Public Economics, 123, pp. 55–77. Martin, S., 2023, ‘Market Transparency and Consumer Search–Evidence from the German Retail Gasoline Market,’ RAND Journal of Economics, forthcoming. © 2023 The Authors. The Journal of Industrial Economics published by The Editorial Board of The Journal of Industrial Economics and John Wiley & Sons Ltd. ALCOHOL PROHIBITION AND PRICING AT THE PUMP 597 Matsa, D. A., 2011, ‘Competition and Product Quality in the Supermarket Industry,’ Quarterly Journal of Economics, 126(3), pp. 1539–1591. Mayer, N., 2017, Land Kippt Verkaufsverbot Erst 2018 (Stuttgarter Nachrichten, Stuttgart, Germany). Montag, F.; Mamrak, R.; Sagimuldina, A. and Schnitzer, M., 2023, ‘Imperfect Price Information, Market Power, and Tax Pass-Through,’ Working Paper. Nicolai, B., 2021, Alkohol, Izgaretten, Klopapier–was die Tanke jetzt über unser Leben verrät (Welt, Berlin, Deutschland). Ning, X. and Haining, R., 2003, ‘Spatial Pricing in Interdependent Markets: A Case Study of Petrol Retailing in Sheffield,’ Environment and Planning A: Economy and Space, 35(12), pp. 2131–2159. Noel, M. D., 2007, ‘Edgeworth Price Cycles, Cost-based Pricing and Sticky Pricing in Retail Gasoline Markts,’ Review of Economics and Statistics, 89(2), pp. 324–334. NTV, 2015, ‘Tankstellen-Pächter zahlen für Jo-Jo-Preise,’ (NTV.de, Berlin, Deutschland). Olden, A. and Moen, J., 2022, ‘The Triple Difference Estimator,’ Econometrics Journal, 25(3), pp. 531–553. Pennerstorfer, D.; Schmidt-Dengler, P.; Schutz, N.; Weiss, C. and Yontcheva, B., 2020, ‘Information and Price Dispersion: Theory and Evidence,’ International Economic Review, 61(2), pp. 871–899. Scope Ratings, 2018, Branchenstudie Tankstellenmarkt Deutschland, vol. 2017. (Scope Ratings, Berlin, Germany). Scope Ratings, 2019, Branchenstudie Tankstellenmarkt Deutschland, vol. 2018. (Scope Ratings, Berlin, Germany). Siekmann, M., 2017, ‘Characteristics, Causes, and Price Effects: Empirical Evidence of Intraday Edgeworth Cycles,’ DICE Discussion Paper 252. Statistisches Landesamt Baden-Württemberg, 2022, ‘Kraftfahrzeugbestand und Kraftstoffverbrauch,’ (Statistisches Landesamt Baden-Württemberg, Stuttgart, Germany). https://www.statistik-bw.de/Verkehr/KFZBelastung//LRt1506.jsp?path=Energie /Energiebilanz/. Varian, H. R., 1980, ‘A Model of Sales,’ American Economic Review, 70(4), pp. 651–659. von Baden-Wuerttemberg, L., 2020, ‘Folgen der Aufhebung des nächtlichen Alkoholverkaufsverbots und Umsetzung der Einführung von Alkoholkonsumverboten in Baden-Würrtemberg,’. Wang, Z., 2015, ‘Supermarkets and Gasoline: An Empirical Study of Bundled Discounts,’ Resources for the Future Discussion Paper, pp. 15–44. Wildenbeest, M. R., 2011, ‘An Empirical Model of Search with Vertically Differentiated Products,’ RAND Journal of Economics, 42(4), pp. 729–757. Zimmerman, P. R., 2012, ‘The Competitive Impact of Hypermarket Retailers on Gasoline Prices,’ The Journal of Law and Economics, 55(1), pp. 27–41. © 2023 The Authors. The Journal of Industrial Economics published by The Editorial Board of The Journal of Industrial Economics and John Wiley & Sons Ltd.