Essays on Vertical Restraints and Competition Policy By Chia-Wen Chen B.B.A. (National Taiwan University) 1999 M.A. (National Taiwan University) 2002 ...
5 downloads
20 Views
1MB Size
Essays on Vertical Restraints and Competition Policy By Chia-Wen Chen B.B.A. (National Taiwan University) 1999 M.A. (National Taiwan University) 2002 Dissertation Submitted in partial satisfaction of the requirements for the degree of Doctor of Philosophy in Economics in the Office of Graduate Studies of the University of California Davis Approved:
Christopher Knittel, Chair
Colin Cameron
David Rapson
Victor Stango Committee in Charge 2011
-i-
UMI Number: 3474360
All rights reserved INFORMATION TO ALL USERS The quality of this reproduction is dependent on the quality of the copy submitted. In the unlikely event that the author did not send a complete manuscript and there are missing pages, these will be noted. Also, if material had to be removed, a note will indicate the deletion.
UMI 3474360 Copyright 2011 by ProQuest LLC. All rights reserved. This edition of the work is protected against unauthorized copying under Title 17, United States Code.
ProQuest LLC. 789 East Eisenhower Parkway P.O. Box 1346 Ann Arbor, MI 48106 - 1346
c 2011 by Copyright ⃝ Chia-Wen Chen All rights reserved.
To my grandmothers
-ii-
Contents List of Figures . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
v
List of Tables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
vi
Abstract . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
viii
Acknowledgments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
x
1 The Competitive Effects of Exclusive Dealing: Evidence from Distribution Contract Changes in the U.S. Beer Industry
1
1.1
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
1
1.2
Theoretical Literature on Exclusive Dealing . . . . . . . . . . . . . .
5
1.3
Industry Background . . . . . . . . . . . . . . . . . . . . . . . . . . .
6
1.4
Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
8
1.5
Research Design, Summary Statistics and Empirical Strategy . . . . .
10
1.6
Empirical Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
16
1.6.1
Is Exclusive Dealing Irrelevant? . . . . . . . . . . . . . . . . .
16
1.6.2
Is Exclusive Dealing Anticompetitive? . . . . . . . . . . . . .
18
1.6.3
Is There Any Incentive Conflict under Common Agency? . . .
19
Conclusion and Discussion . . . . . . . . . . . . . . . . . . . . . . . .
22
1.7
2 Estimating the Foreclosure Effect of Exclusive Dealing: Evidence from the Entry of Specialty Beer Producers
35
2.1
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
35
2.2
Industry Background . . . . . . . . . . . . . . . . . . . . . . . . . . .
40
2.3
Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
42
2.3.1
Store Attributes . . . . . . . . . . . . . . . . . . . . . . . . . .
43
2.3.2
Entry Variation . . . . . . . . . . . . . . . . . . . . . . . . . .
44
Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
44
2.4.1
Demand Side . . . . . . . . . . . . . . . . . . . . . . . . . . .
44
2.4.2
Entry Game . . . . . . . . . . . . . . . . . . . . . . . . . . . .
47
Empirical Implementation . . . . . . . . . . . . . . . . . . . . . . . .
52
2.4
2.5
-iii-
2.6
2.7
2.5.1
Market Definition . . . . . . . . . . . . . . . . . . . . . . . . .
52
2.5.2
Pricing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
53
2.5.3
Estimation . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
53
Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
57
2.6.1
Demand Estimates . . . . . . . . . . . . . . . . . . . . . . . .
57
2.6.2
Strategic Entry . . . . . . . . . . . . . . . . . . . . . . . . . .
59
Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
62
3 Vertical Integration and Retail Competition under Capacity Constraints: The Case of California Gasoline Market
75
3.1
Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
75
3.2
Industry Background . . . . . . . . . . . . . . . . . . . . . . . . . . .
78
3.3
Model . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
80
3.3.1
Retail Market . . . . . . . . . . . . . . . . . . . . . . . . . . .
81
3.3.2
Wholesale Market . . . . . . . . . . . . . . . . . . . . . . . . .
81
Numerical Example . . . . . . . . . . . . . . . . . . . . . . . . . . . .
84
3.4.1
Vertical Separation . . . . . . . . . . . . . . . . . . . . . . . .
85
3.4.2
Vertical Integration . . . . . . . . . . . . . . . . . . . . . . . .
85
3.5
Data . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
87
3.6
Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
88
3.7
Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
89
3.4
-iv-
List of Figures
1.1
Number of Stores . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
23
1.2
Average Store Weekly Sales in Six Packs . . . . . . . . . . . . . . . .
24
1.3
InBev Market Share . . . . . . . . . . . . . . . . . . . . . . . . . . .
25
2.1
Locations of Breweries for the California Specialty Beer Producers Observed in the Scanner Data . . . . . . . . . . . . . . . . . . . . . .
63
2.2
Tree Diagram for the Nested Logit Model . . . . . . . . . . . . . . . .
64
2.3
Predicted Demand and Actual Demand: A 5% Random Sample of Data 65
-v-
List of Tables
1.1
Some Typical Entries From CBBD Annual Directory . . . . . . . . .
26
1.2
Product Attributes of Top 30 and InBev Brands By Category . . . .
27
1.3
Summary Statistics . . . . . . . . . . . . . . . . . . . . . . . . . . . .
28
1.4
Diagnosing Trends Using Pre-event Data: InBev Market Share . . . .
28
1.5
Diagnosing Trends Using Pre-event Data: Market Quantity and Price
29
1.6
Benefit from Exclusive Dealing: InBev Market Share . . . . . . . . .
29
1.7
Benefit from Exclusive Dealing: Market Share from Five Individual Brands . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
30
1.8
Market Share from Five Individual Brands: Controlling for Trends . .
30
1.9
Market Quantity and Average Price . . . . . . . . . . . . . . . . . . .
31
1.10 Market Quantity and Average Price: Controlling for Trends . . . . .
31
1.11 Crowd Out Effects: AB Specialty Products . . . . . . . . . . . . . . .
32
1.12 Diagnosing Trends Using Pre-event Data: AB Specialty Products . .
32
1.13 Crowd Out Effects: AB Specialty Products: Controlling for Trends .
33
1.14 Distribution Status Affected by the AB-InBev Distribution Agreement
33
1.15 Crowd Out Effects: Rival Brands . . . . . . . . . . . . . . . . . . . .
34
2.1
Some Typical Entries From the CBBD Annual Directory . . . . . . .
66
2.2
Summary Statistics . . . . . . . . . . . . . . . . . . . . . . . . . . . .
67
2.3
Number of Stores a Firm Entered
. . . . . . . . . . . . . . . . . . .
67
2.4
Pricing for California Specialty Beer Products . . . . . . . . . . . . .
68
2.5
Entry Probability without Controlling for Demand
. . . . . . . . . .
69
2.6
Demand Estimation Results from the Logit Model . . . . . . . . . . .
69
2.7
Demand Estimation Results from the Nested Logit Model
. . . . . .
70
2.8
Price Elasticity Percentiles . . . . . . . . . . . . . . . . . . . . . . . .
70
2.9
Entry Probability with Expected Demand: No Strategic Interactions
71
2.10 Strategic Entry: the Naive Approach . . . . . . . . . . . . . . . . . .
72
2.11 Strategic Entry with Simulated Expected Demand . . . . . . . . . . .
73
-vi-
2.12 Strategic Entry: Regression without Bay Area Counties and the Sacramento County . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2.13 The Effects of Banning Exclusive Dealing
74
. . . . . . . . . . . . . . .
74
3.1
Gasoline Sold by Distribution Methods in Five PADDs . . . . . . . .
91
3.2
Market Outcomes Under Scenario 1 . . . . . . . . . . . . . . . . . . .
92
3.3
Market Outcomes Under Scenario 2 . . . . . . . . . . . . . . . . . . .
92
3.4
Market Outcomes Under Scenario 3 . . . . . . . . . . . . . . . . . . .
93
3.5
Market Outcomes Under Scenario 4 . . . . . . . . . . . . . . . . . . .
93
3.6
Refining Costs Per Gallon . . . . . . . . . . . . . . . . . . . . . . . .
94
3.7
California 2009 Data Used to Set Model Parameters . . . . . . . . . .
94
3.8
Market Outcomes with Double Marginalization . . . . . . . . . . . .
95
3.9
Market Outcomes under Vertical Integration (M =10) . . . . . . . . .
95
3.10 Market Outcomes under Vertical Integration (M =20) . . . . . . . . .
95
-vii-
Abstract of the Dissertation Essays on Vertical Restraints and Competition Policy
Vertical restraints between firms, such as exclusive dealing contracts that forbid a dealer from promoting other manufacturers’ products, are controversial in competition policy because of their potential anticompetitive effects. This dissertation addresses three issues in competition policy and vertical relationships between firms: (1) what are the effects of exclusive dealing on competitiveness of brands? (2) does exclusive dealing foreclose new entrants out of a market? (3) what is the effect of retail competition on market price in a vertically integrated industry when upstream firms face capacity constraints? Chapter 1 examines the impact of allowing more brands access to exclusive distribution networks on brand and market level outcomes. In the U.S., Anheuser Busch is the dominant firm in the beer industry and has exclusive dealing arrangements with many of its distributors. I looked at a recent beer distribution deal between Anheuser Busch and InBev that moved InBev brands into Anheuser Busch distribution networks. I collected beer distributor data before and after the event and matched them with a panel scanner dataset from a major grocery chain in Northern California. Using a difference-in-differences approach, I compared the changes in InBev market shares in markets in which InBev switched to Anheuser Busch distributors, to the changes in market shares in markets in which InBev switched to Anheuser Busch exclusive distributors. The results suggest that exclusive dealing matters in the beer industry: I find InBev market shares to be higher once allowed access to Anheuser Busch exclusive distribution networks. In addition, I do not find overall market quantity to be larger when more brands have access to Anheuser Busch exclusive networks. Instead, the results show cannibalization effects on existing brands’ market share when a distributor acquires more brands. These results are more consistent with an incentive-based explanation for firms preferring exclusive contracts. Chapter 2 examines the effect of exclusive dealing on rival firms’ entry decisions. I estimated an entry model of specialty beer producers in Northern California and
-viii-
tested whether exclusive dealing raises a firm’s fixed costs. I modeled each firm’s entry decision as a static entry game of incomplete information that allows for strategic interactions and employed a new panel scanner dataset from a major grocery chain in Northern California. Given that both firm and location profitability are heterogeneous, I controlled for post-entry demand conditions by estimating the demand for beer using a discrete choice model. Using the demand estimates and the predicted entry probabilities, I recovered a firm’s fixed costs using a two-step estimator. I find some spillover effects on specialty beer producers’ entry decisions. After taking strategic interactions into account, the results indicate that a firm has higher fixed costs at locations with exclusive distributors. The estimates also show that a firm is less likely to enter a location that is farther from its brewery, has lower expected demand or is smaller in store size. Finally, I implemented counter-factual experiments to study the effect of banning exclusive dealing. The results show that the welfare improvement associated with banning exclusive contracts is very small. Chapter 3 (joint with Christopher R. Knittel) considers a model of oligopolistic competition when upstream firms face capacity constraints. We studied the optimality conditions of upstream firms under vertical separation and vertical integration when firms compete on quantity. We illustrated the properties of the equilibrium wholesale and retail prices when the downstream market becomes less competitive with a numerical example. Using data on gasoline demand and refineries’ capacity levels in California, we generated equilibrium wholesale and retail prices when the number of downstream firms varies. We find that whether a higher degree of retail market concentration results in higher retail price depends on market structure and the effectiveness of the capacity constraints. When independent refineries’ capacity constraints are binding, the effect of a decrease in the number of independent retailers on retail gasoline price is very small.
-ix-
Acknowledgments I am indebted to my dissertation chair, Christopher Knittel, for his inspiration, support and encouragement. I have benefited tremendously from his empirical IO class and the advice he has shared with me. Special thanks to Colin Cameron, David Rapson, and Victor Stango, my other committee members, for their valuable advice and guidance. My dissertation would not have been nearly as good without their help. I also thank Joonsuk Lee, Scott Carrell, Douglas Miller, and Hilary Hoynes for helpful comments on my research. I am particularly grateful to John Pauley and The Nielsen Company for allowing me access to the data used in the dissertation. I also thank James Arndorfer for sharing his industry insights. I thank the Applied Microeconomics faculty and my fellow graduate students at UC Davis for providing an extremely stimulating academic environment. I am grateful to my friends for their support throughout graduate school. Especially, I would like to thank Tammy Lu, Nien-Husan Chen, Julia Peng, and Joy Chen for always being there for me. My deepest gratitude goes to my family for their love and support throughout my life.
-x-
1
Chapter 1 The Competitive Effects of Exclusive Dealing: Evidence from Distribution Contract Changes in the U.S. Beer Industry Chia-Wen Chen
1.1
Introduction
Why do firms sign exclusive contracts that forbid their dealers to market other firms’ products? Are these exclusionary contracts anticompetitive or efficiency-enhancing? A long debate has taken place about the nature and the purpose of these contracts. The answers to these questions are important not only for the studies of organization forms but also for the implementation of competition policies. For example, in 1997, Anheuser Busch (AB) launched an incentive program dubbed “100 percent share of mind”, which provided discounts, truck painting allowance and other benefits to dealers that went exclusively with Anheuser Busch. Many microbrewers complained they were being driven out of the market due to this practice.1 In 2009, a similar practice by Intel, one that provided rebates and cash benefits for manufacturers and 1
Many microbreweries exited the market in the late 1990s. However, there is no direct evidence that links Anheuser Busch’s incentive program to the exits of microbreweries at that time.
2
retailers in exchange for purchasing most of their products from Intel, was found to be anticompetitive by the European Commission.2 Foreclosure has always been the main anticompetitive concern against exclusive dealing. In different settings, theoretical works in this area have shown that an upstream firm can sign exclusive contracts with downstream firms to successfully drive potential entrants out of the market. However, incentive-based theories for exclusive dealing argue that exclusive contracts are designed to alleviate the incentive conflicts within vertical relationships, and thus can improve economic efficiency by solving the free-riding problem and other incentive-related issues. Given that theoretical works in this area have different predictions about the effects of exclusionary contracts, it remains an empirical question to what extent these contracts are relevant from a firm’s and the market’s perspective. This paper intends to add empirical evidence to this area by looking at how different vertical arrangements (exclusive dealing versus common agency) affect firm level and market level outcomes in the U.S. beer industry. In the U.S., 50% of the beer is brewed by Anheuser Busch, and 60%-70% of Anheuser Busch’s distribution is exclusive. Previous empirical studies in the beer industry (Sass, 2005; Asker, 2005; Rojas, 2010; Chen, 2011) do not find much anticompetitive effect of these exclusionary contracts. However, the reasons why a brewer may prefer exclusive arrangements with dealers remain unclear. In this paper, I exploit a recent beer distribution deal between Anheuser Busch and InBev that moved InBev brands into Anheuser Busch distribution networks to explore the relationships between distribution status and market outcomes. Because the distribution deal applied to all Anheuser Busch distributors simultaneously across the United States, and given that the exclusivity status of the Anheuser Busch distributors was determined prior to the event, the main idea of this study is to use this event as a quasi-experiment to identify the competitive effects of exclusive dealing: if exclusive dealing has any anticompetitive (or efficiency2
Other recent examples include the exclusionary contracts between Visa/MasterCard and their member banks that prohibited access to American Express and Discover card, and Microsoft’s exclusive dealing practices with computer manufacturers to market its Internet Explorer. See United States v. VISA U.S.A., Inc., 163 F. Supp. 2d 322 (S.D.N.Y. 2001) and United States v. Microsoft Corp., 253 F.3d 34 (D.C. Cir 2001)
3
promoting) effect, then by studying the variations generated by this quasi-experiment (allowing brands access to exclusive dealers), we can identify the effect of exclusive dealing on firm level and market outcomes. The results suggest that exclusive dealing does matter in the beer industry: InBev brands gained market share once they were allowed access to Anheuser Busch’s exclusive distribution networks; however, the benefits were more likely to be incentive based. When more brands were added to the same distribution system, a cannibalization effect occurred on existing brands. These results provide an incentive based explanation as to why firms would prefer to an have exclusive arrangement. Furthermore, market total sales were not affected significantly once more brands were allowed to use Anheuser Busch’s exclusive distribution network, indicating that banning exclusive dealing at the distribution level may not improve economic efficiency. The AB-InBev deal is useful in identifying the effect of exclusive dealing because it affected local markets simultaneously but not identically. According to the deal, Anheuser Busch started to import European InBev brands to the United States after February, 2007. Thus, distributors selling Anheuser Busch products, exclusive to Anheuser Busch or not, would obtain these InBev brands, while non-Anheuser Busch distributors selling InBev brands had to drop these products. In this study, I collected data on brewer-distributor arrangements in California from 2006 to 2007 and combined them with a two year panel scanner dataset from a large grocery retail chain in Northern California. The dataset includes detailed weekly price and sales data for each UPC level malted beverage product at every retail outlet of the grocery chain within the two year time span. Having data from before and after the AB-InBev deal enables me to estimate the exclusive effects using a difference-in-differences approach to control for local market fixed effects and unobserved brand level fixed effects. Most empirical papers that study the vertical restraints between firms focus on the effect of vertical integration on market outcomes. For example, Chipty (2001) and Hastings and Gilbert (2005) found vertical integration led to foreclosure either by carrying fewer rival programs or by raising rivals’ cost. On the contrary, Horta¸csu and Syverson (2007) found little foreclosure effect due to vertical integration in the cement
4
and ready-to-mix concrete industry. Instead, they found productivity explains most of the differences in the market outcomes of their data. They concluded that the foreclosure effect was not quantitatively important in the cement and ready-to-mix concrete industry. The empirical literature on exclusive dealing is limited.3 Sass (2005), Asker (2005) and Rojas (2010) studied exclusive dealing in the beer industry and found no evidence supporting anticompetitive theory of exclusive dealing. However, Chen (2011) estimated a structural entry model of specialty beer producers in Northern California and found that exclusive dealing was associated with higher entry costs for these microbreweries. Nevertheless, Chen (2011) found that the improvements in consumer welfare from removing exclusive dealing contracts were very small. Sass (2005) studied a cross-sectional survey of beer distributors and found that exclusive dealers on average generated higher prices and larger sales for their suppliers, which was more consistent with incentive-based theory. Asker (2005) recovered the costs incurred and the promotional efforts made by distributors in exclusive and less exclusive markets, and found distributors in less exclusive markets were not more efficient than distributors in exclusive markets. His results reject the foreclosure hypothesis. Rojas (2010) looked at a scanner data set from 1988 to 1992 and found that exclusive dealing and exclusive territory agreements were more consistent with the existence of a welfare-enhancing effect. This paper differs from the above studies in several ways. First, I used a new dataset with price and sales data after AB’s mass incentive program in 1997, intending to provide more direct evidence to evaluate the effect of AB’s exclusive program.4 Second, using brand level data allows me to include brand heterogeneity and to study the effect of different distribution status on individual brand level outcomes. Finally, while Sass (2005) and Asker (2005) tested exclusive dealing effects using crosssectional data, my identification strategy is to take advantage of the panel structure of 3
For a thorough review of the empirical studies on vertical integration and vertical restraints, see Lafontaine and Slade (2007) and Lafontaine and Slade (2008). 4 Sass (2005), Asker (2005) and Rojas (2010) all used data prior to AB’s “100 percent share of mind ”program.
5
the data set and to exploit the variations in distribution status generated by the ABInBev distribution agreement, intending to alleviate the concerns regarding potential omitted variables bias. The rest of the paper proceeds as follows. Section 2 gives a brief review of the theoretical literature of exclusive dealing. Section 3 provides an overview of the beer industry. Section 4 describes the data. Section 5 explains the research design and the empirical strategy of the paper. Section 6 provides the results. Section 7 concludes the paper and provides potential future research directions.
1.2
Theoretical Literature on Exclusive Dealing
Economic theories are ambiguous about the effects of exclusive dealing. In general, the gains from exclusive contracts can come from surplus of anticompetitive behavior or from improvement in economic efficiency. Exclusive dealing can be anti-competitive when a monopolist uses exclusive contracts to “lock-in” buyers, with the intent of raising rivals’ costs and creating entry deterrence. However, the above claim was challenged by the Chicago school during the 1970s. Bork (1978) and Posner (1976) argued that if the sole purpose of vertical restraints was to restrict competition, downstream buyers would never sign them in the first place because doing so would only lower the potential total surplus. The above Chicago school critique of the anticompetitive explanation of vertical restraints has led to theoretical work in this area focus on better contracting environments that can rationalize vertical restraints. Bernheim and Whinston (1998) showed that common agency and exclusive dealing are both efficient when there is no contracting externality. Marvel (1982), Klein and Murphy (1988), Besanko and Perry (1993), Martimort (1996), Bernheim and Whinston (1998), and Segal and Whinston (2000b) showed that if we allow for a contracting externality or incomplete contracts, exclusive contracts can provide better incentives for investment or promotional efforts. On the contrary, recent developments in game theoretical models have shown that vertical restraints can be anti-competitive once contracting externality is allowed. Aghion and Bolton (1987), Rasmusen, Ramseyer and Wiley (1991), and Bernheim and
6
Whinston (1998) showed that exclusive dealing can have an entry deterrence effect.5 In the above theoretical settings, a manufacturer can sign an exclusive contract with a buyer to extract rents from potentially more efficient entrants, or it can sign exclusive contracts with a large number of buyers to foreclose the market when entrants face minimum economies of scale or noncoincident markets.6 The above anticompetitive arguments predict that exclusive dealing would restrict market output, while efficiency based arguments suggest that the effect of exclusive dealing on market output would be ambiguous. In the empirical section, I first test whether exclusive dealing is relevant from a firm’s perspective. Then, I test whether open access (allowing more brands to gain access to a dominant firm’s distributor network) helps to improve market level outcomes. Finally, I test whether the results are consistent with the presence of incentive conflicts within distribution networks.
1.3
Industry Background
After Prohibition, the vertical structure of the U.S. beer industry has been heavily regulated by state laws. Basically, the laws require the supply of beer to follow a three-tier system. In the system, direct shipping from brewers to consumers is prohibited. The top tier brewers supply their products to the state-licensed middlemen (distributors). The distributors then store and transport the products to the bottom tier retailers. Distributors not only help store and transport the products, but they are also responsible for point-of-sale promotions and are expected to have sales staff visit or call accounts regularly. During the sample period of the data, there were more than 13,000 labels and 1,500 breweries in the U.S., and the annual total sales of beer were around 100 billion 5
Hart and Tirole (1990) and O’Brien and Shaffer (1992) showed that vertical restraints help to solve a monopolist’ commitment problem and allow it to restore its market power when the monopolist produces an essential input and buyers compete with each other. Rey and Tirole (2007) provided a general overview of market foreclosure theories (including work on vertical foreclosure, horizontal foreclosure and exclusive dealing arrangements) and their policy implications. 6 Segal and Whinston (2000a) pointed out that if manufacturers can take advantage of the coordination problem faced by buyers, they can implement the above “naked exclusion scheme” successfully. Simpson and Wickelgren (2007), Abito and Wright (2008), and Doganoglu and Wright (2010) also provided settings that allow exclusive dealing to achieve inefficient outcomes.
7
dollars.7 Nevertheless, the brewing industry is highly concentrated even with such a large number of brands: Anheuser Busch, Miller and Coors collectively claimed nearly 80% of the market. There are mainly three market segments at the brewing level: domestic macro brands, domestic specialty brands and imported brands.8 Anheuser Busch, Miller and Coors, the dominant players in the industry, enjoy economies of scale to produce macro products which are lower-priced, have large package size options, and are supported by national advertising campaigns. Specialty beer producers entered the brewing industry during the 1980s and 1990s. They started out as microbreweries and produce specialty products that emphasize favor and taste. Sierra Nevada Brewing and the Boston Beer company are the pioneers of the microbrewery movement and are the most successful and nationally known companies in this segment. The last segment includes imported brands that are usually well-established products from foreign countries. Most imported brands have higher prices than domestic macro brands. Intra-brand competition at the distribution level is not common: brewers tend to adopt exclusive territory systems to prevent competition between their distributors.9 In contrast, most distributors are common agencies that intensify inter-brand competition at the distribution level. Macro brewers’ (AB’s, Miller’s, and Coor’s) distribution networks are often viewed by the beer press as a superior promotional vehicle due to economies of scale in distribution.10 It is probably why Anheuser Busch’s exclusive campaign may have potential raising rivals’ costs effect in this setting. Finally, terminating existing distribution 7
Statistics are obtained from the National Beer Wholesalers Association and the Brewers Association web-sites. 8 Mergers and acquisitions are quite common in the industry. For example, AB has acquired microbreweries such as Widmer Brothers and Redhook in the 1990s. More recently, Miller and Coors merged to a new company in 2008 and AB also merged with InBev in 2009. For a general introduction to the U.S. brewing industry, see Tremblay and Tremblay (2005). 9 In some states, it is even required by law to use an exclusive territory system for beer distributors. For empirical studies of the effect of exclusive territory in the U.S. beer industry, see Culbertson and Bradford (1991), Sass and Saurman (1993), Sass and Saurman (1996), Brenkers and Verboven (2006), and Rojas (2010). 10 For example, Bump Williams, an IRS industry analyst, said “there’s nobody better than these three networks. They can get these beers on shelves overnight.” See Kesmodel (2007).
8
contracts can be costly for brewers. Some states have franchise laws that prevent brewers from terminating their distribution contracts without a good cause.11 Due to the difficulties in switching distributors at will, the distribution agreement between AB and InBev, which happened at the national scale, provides a great opportunity to look into how distribution status affects market outcomes.
1.4
Data
The California Beer and Beverage Distributors (CBBD) trade association publishes an annual directory containing its member distributors in California. This a list of each distributor’s representing brewers and its operating counties. Table 1.1 lists some typical entries in the 2006 CBBD directory.12 Most distributors represented at least one of the domestic macro brewers (Anheuser Busch, Miller and Molson Coors) and also carried other imported brands or domestic specialty brands. There were also independent distributors that did not carry brands from any domestic macro firms at all but collected a large number of specialty or imported brands. Well-established domestic specialty and imported brands usually were carried by distributors that also carried at least one domestic macro brands. For example, most InBev distributors also carried AB or Miller products. This indicates that there may be economies of scale at the distribution level. Additional evidence to support the existence of economies of scale at the distribution level is that all independent distributors carried a large number of specialty brands. Demand scanner data are provided by Nielsen. The data set contains weekly, UPC level, price and sales data of malt beverage category for a major grocery chain in Northern California and Nevada and includes 258 stores in 150 cities. The twoyear data begin at April 15, 2006 and end at April 5, 2008, with a total of 104 weeks. In the dataset there are six major categories: lager, stout/porter, light beer, pale ale, malt liquor, and flavored alcoholic beverage. Within each brand there are many package sizes, different volumes and bottling methods. For example, Budweiser has 11
Such practices make sure a distributor’s promotional contribution to a brand does not get ripped off easily by another distributor. However, these practices may make a brewer vulnerable to termination of a bad distributorship, for example, see Day (2006). 12 The 2006 and 2007 trade directories are provided by local distributors.
9
7 package sizes (1, 4, 6, 12, 18, 20 and 30), sold in bottles or cans. For each brand, I added up sales for all its package sizes and containers to calculate its aggregate sales. Therefore, sales data are at brand level instead of at UPC level. There were 237 brands with positive sales during the two-year time frame; most of them had less than 1% market share. Figure 1.1 provides a histogram of the average number of store listings for those 237 brands. Nearly 30 percent of brands were carried by fewer than 20 stores. In fact, there were 25 brands carried by fewer than three stores. This suggested that a typical brand may not need to acquire a large store base in order to enter to market, so the foreclosure mechanism implied in a naked exclusion theory may not work properly in this industry. Table 1.2 lists the top 30 beer brands’ market share, price, and their basic characteristics (alcohol by volume, ABV, and calories per 12oz) in the dataset.13 On average, domestic macro brands were the lightest in both ABV and calories, and were usually priced lower than other products. Top domestic specialty brands were almost all rich beer with 5.0% or higher ABV and had at least 150 calories per 12 oz bottle. There were basically five InBev brands in my dataset, including Bass, Beck’s, Boddingtons, Beck’s Premier Light and Stella Artois.14 Their market share, average price, and product characteristics can also be found in Table 1.2. The InBev brands were all imports. Because domestic macro brands had very different attributes from other brands, I recalculated market shares using total sales from specialty brands and imports in the following empirical exercise. Most brands have their own price promotion schedule that prevails across stores. However, within a product there are still pricing differences in the data, probably due to demographic differences. For example, a product can have a price range from $15 to $16.5 across different stores within a week, but has similar promotion pattern throughout a year. 13
Beer prices were for average six-pack 12 oz containers across packages. I adjusted volume and package sizes to find out how many units sold for each brand in terms of a regular 12oz six-pack. I divided total sales revenue by adjusted sales to find out adjusted average 12oz six-pack price for each brand. Market shares in sales were defined as total brand sales divided by total beer sales. 14 Beck’s Oktoberfest was another InBev product in my dataset. However, it was a seasonal product that had only 11 observations with positive sales (carried by just 4 stores). Therefore, I excluded Beck’s Oktoberfest when I calculated InBev’s market share.
10
1.5
Research Design, Summary Statistics and Empirical Strategy
The Event In November, 2006, AB announced a distribution agreement with InBev (henceforth the event) which allowed it to distribute InBev European products in the U.S. market after February, 2007.15 According to the 2006 CBBD directory, prior to the event, there were 39 AB distributors in California. Out of these 39 distributors, 11 were exclusive dealers, and of the other 28 AB nonexclusive dealers, 11 were already carrying InBev products. Moreover, in the 2007 CBBD directory, nearly all AB distributors were listed as having acquired InBev products. Therefore, the event provided a discrete change in brand distributorship that affected local markets differently: at the distribution level, some local markets were never affected by this deal; in other markets, however, InBev brands switched to new distributors. Given this event, I assigned distribution status to each store in the dataset based on its AB distributor’s exclusivity and the AB distributor’s ownership of InBev distribution rights prior to the event. If an AB distributor carried InBev brands prior to the event, the store was assigned to the control group (no brand movement at the distribution level). If an AB distributor carried no InBev brands prior to the event and was a nonexclusive dealer, the store was assigned to the treatment group 1. Finally, if an AB distributor carried no InBev brands prior to the event and was exclusive, the store was labeled as treatment group 2. Summary Statistics I then matched the brewer-distributor relationship data to the scanner data set. Because sales territory data are only available at the county level from the CBBD directory, I contacted each exclusive AB distributor and each AB distributor of stores in control group to gain more information about their sales territories. I dropped a store 15
Prior to the AB-InBev deal, AB also obtained Grolsch’s distribution rights in April and acquired Rolling Rock and Rock Green Light in August, 2006. It was widely believed by many industry analysts that AB intended to use those deals to expand its product portfolio and to have more control over its distributors. For example, see Kesmodel (2008). AB later dropped Grolsch from its distribution network in mid-2008 because the brewery was acquired by SABMiller.
11
out of my sample if it was not in California, if it did not operate during the entire time span, or if I did not have enough information to identify its AB distributor. My final sample size is 197 stores, having started with 256 stores. Table 1.3 gives summary statistics of store and distributor attributes for the three assigned groups during the entire two year time span of the data. Store characteristics differ across the three groups. In general, AB exclusive stores had the highest store sales and revenue, the lowest average price, and carried the most brands. For distributor attributes, AB exclusive distributors have on average smaller market share (26%) than the other AB distributors, which is reasonable since they carried fewer brands. Note that non-AB distributors obtained higher market share (44%) in AB exclusive markets. If economies of scale are important in distribution, running an exclusive network seems to be a disadvantage for AB. Another point to note from Table 1.3 is that the mean total market shares for AB nonexclusive distributors in control group and in treatment group 1 are similar, suggesting that the prior ownership of InBev brands may not be correlated with overall distributor abilities. Because exclusivity is associated with higher store sales, lower prices and a larger number of entrants, these preliminary results seem to go against the anticompetitive theory of exclusive dealing. However, the differences may also come from permanent unobserved covariates associated with exclusivity, such as demographics differences. If this is the case, panel data on a fixed effects model can help to parse out these differences and can provide more credible results on the competitive effects of exclusive dealing. Empirical Strategy Ideally, to see to what extent exclusive contracts affect market outcomes, a researcher would randomly assign brands to different distributors. For example, to test anticompetitive based theories such as foreclosure or raising rivals’ costs, the researcher would assign a foreclosed product to an exclusive distribution network in one market, and assign the same product to an alternative, not foreclosed network in another market. If gaining access to an exclusive distribution network stimulates more competition
12
and leads to higher market quantity and lower market price, then there are potential benefits to removing exclusive dealing. Similarly, from a firm’s perspective, to test whether exclusive dealing is relevant, we would randomly add products to distribution networks and test how distribution status affects a firm’s market share. Finally, to test incentive theories, the researcher would randomly add products to different distribution networks. If receiving more products makes existing products in the same distribution network perform worse in the marketplace, there is evidence of incentive conflicts within a distribution network. Our thought experiment is best carried out by random assignment because it guarantees that the differences in outcomes are not caused by omitted factors. If the assignment is not random, unobserved factors that cause a distributor to become exclusive may confound the final results. For example, suppose an AB exclusive distributor expects a boom in the specialty beer segment and decides to carry nonAB specialty brands. In this case, because distribution status is endogenous, the positive relationship between higher market quantity and a shared distribution house is misleading. Even though running a controlled experiment has many advantages, it is not feasible in most cases. However, if a researcher can find natural experiments that generate variations that are likely to be exogenous to potential outcomes, it is still possible to identify the relationship between the dependent and the independent variable. In this paper, I take advantage of the distribution reassignment of the AB-InBev deal and use the variations generated by the event to carry out the above thought experiment. Given the panel structure of my data, I use a “difference-in-differences” approach to estimate changes in market outcomes due to the event. The advantage of this approach is that it not only controls for national shocks, but it also eliminates unobserved permanent fixed effects in each local market. Also, because the variations in distribution status are generated by the AB-InBev agreement, which applied to the entire U.S. at the same time, this approach reduces the potential omitted variable bias problem due to the interactions of local demand unobservables and time variables. In other words, the AB-InBev deal can be treated as
13
a quasi-experiment to test how distribution networks matter in the beer industry. First, I obtained the “Anheuser Busch exclusive effect” by comparing the changes in market outcomes in markets in which InBev switched to Anheuser Busch (nonexclusive) distributors, to the changes in market outcomes in markets in which InBev switched to Anheuser Busch exclusive distributors. Second, with because I have a natural control group (where InBev brands were already using the AB distribution network prior to the event), I also compared the changes in market outcomes in markets in which InBev switched to Anheuser Busch (nonexclusive or exclusive) distributors, to the changes in market outcomes in markets in which InBev brands did not switch between different distributors. Therefore, I am able to identify “Anheuser Busch effect,” too. The first effect (Anheuser Busch exclusive effect) is estimated by comparing the differences between treatment 1 and treatment 2 stores, which measures how exclusivity affects the market outcomes and is the main focus of this paper. With the first effect, the second effect (Anheuser Busch effect) can be obtained by comparing differences between control and treatments (both treatment 1 and treatment 2) due to the event. The comparison shows how adding brands to the AB network changes market outcomes. Because AB is the biggest and the only brewer that runs a massive exclusive incentive program, it will be interesting to see whether AB dealers are inherently different from the others. I then fit the data using the following baseline specification for store i in week t: Ait = µ + αi + αm + β1 ait + β2 bit + ϵit
14
where: Ait = dependent variable µ = constant αi = store fixed effects αm = month dummies ait = indicator whether InBev products moved into an AB exclusive distribution network; ait = 1 for treatment 2 stores after the event bit = indicator whether InBev products moved into an AB distribution network; bit = 1 for treatment 1 and treatment 2 stores after the event Therefore, after the event, we have Ait =µ + αi + αm + ϵit Ait =µ + αi + αm + β2 bit + ϵit Ait =µ + αi + αm + β1 ait + β2 bit + ϵit
for stores in control group, for stores in treatment group 1, for stores treatment group 2.
In this case, once we remove the store fixed effects, we can obtain an estimate of β1 by comparing the results of stores in treatment group 1 to the results of stores in treatment group 2; and given β1 , we can obtain an estimate of β2 by comparing the results of stores in control group to the results of stores in treatment groups (treatment 1 and treatment 2 stores). Depending on the context, the dependent variable can be market level or brand level market outcomes, such as market shares or prices. For example, consider the dependent variable being Beck’s market share. In this case, store dummies αi control for fixed, store specific unobservables that affect Beck’s average market share within a store, and month dummies αm control for monthly fluctuations in demand across stores. Coefficient β1 measures how important it is for Beck’s to be distributed by an AB exclusive distribution network, and coefficient β2 measures the effect on Beck’s market share once it is carried by an AB distributor. Because products in the same
15
store tend to face the same local demand shocks, the standard errors are clustered at the store level for all regressions. Similarly, when the dependent variable is market quantity, coefficient β1 measures how allowing more brands to use AB exclusive distributors affected a store’s average weekly sales; while coefficient β2 provides the effect on sales once AB distributors acquired more brands in their portfolios. The identification assumption here is that after controlling for monthly shocks across the stores and the permanent unobserved covariates within the stores, the error term will be uncorrelated with distribution status. This assumption may be invalid if the trends in market outcomes vary with distribution status. For example, if AB exclusive stores have lower growth rates in total sales than AB nonexclusive stores, then my estimates of β1 (the exclusive effect) will pick up these differences in trends and will be biased downward. Similarly, if AB distributors tend to acquire products with higher growth rates in their territories, the coefficient β2 will be underestimated. Figure 1.2 and Figure 1.3 graph average weekly store sales and InBev market shares for the three groups. In general, it seems that most of the variation in outcome variables comes from monthly shocks, which affect all the three groups. Also, from the graphs, it seems that these three groups follow similar trends for both outcome variables. Figure 1.2 and Figure 1.3 also provide general pictures of the econometric analysis provided in the next section. From Figure 1.2, average store sales were not much affected by the event. This indicates exclusive contracts in the beer industry may not have anticompetitive effects to the extent that they harness competition. Also, from Figure 1.3, after InBev had access to AB distribution network, the vertical gap between treatment 1 and treatment 2 lines became smaller. This finding suggests that exclusive dealing contracts do serve a purpose in the beer industry. Given the panel structure of the data, I am able to test whether there are preexisting differences in trends by distribution status using data prior to the event with the following specification:
16
Ait = µ + αi + αm + t + γ1 ci t + γ2 di t + ϵit where: Ait = dependent variable µ = constant αi = store fixed effects αm = month dummies ci = indicator of belonging to AB exclusive stores (AB exclusive stores) di = indicator of belonging to treatment stores (either AB exclusive or AB nonexclusive stores) t = trend variable Table 1.4 and Table 1.5 provide estimated trend coefficients for InBev market share, weekly store sales and average store prices. For InBev market shares, I find γ1 not statistically significant and γ2 negative and statistically significant. In this case, the effect on InBev market shares when acquired by AB may be biased. Similarly, I find weekly store sales have a significant negative trend coefficient for the AB exclusive stores (γ1 < 0), so the effect on relaxing the exclusive constraint on market sales may also be biased without controlling for trends. To show how empirical results may be affected by trends across different groups, in the next section, I provide both the baseline results and the results controlling for group specific trends whenever there is evidence of significant pre-existing group specific trends.
1.6 1.6.1
Empirical Results Is Exclusive Dealing Irrelevant?
Bernheim and Whinston (1998) showed that exclusive dealing is not anticompetitive when there is no contracting externality. However, exclusive dealing turns out to be irrelevant in this setting. Empirically, Asker (2005) finds no evidence to support the view that exclusive dealers are more efficient than others, either in terms of cost advantage or promotional aptitude. As a result, it remains a question why firms ever
17
want to adopt exclusive dealing. To see whether there is any benefit to exclusive dealing, I first looked at the market outcomes of InBev brands before and after they were transferred to AB’s exclusive distribution network. In addition, I looked at the effect of joining AB’s distribution network on InBev’s market share to see whether AB dealers are indeed more efficient. If AB dealers are more efficient than others, denying access to some of these distributors may lead to anticompetitive results. Table 1.6 displays the estimated changes in market share for InBev brands. On average, InBev’s market share increased by 5 to 6 percentage points after joining AB’s exclusive network. However, there was no significant impact on InBev’s market share when it moved to AB’s nonexclusive distribution network. Without controlling for trends, the AB coefficient is small and insignificant. The magnitude of this coefficient becomes larger when we add treatment group specific trends into the regression, although the coefficients are much less precisely estimated when trends are included. The above results indicate that while exclusive dealing helped InBev to capture more market share, joining AB’s distribution network did not provide an additional advantage (in terms of market share) for InBev brands. Table 1.7 displays individual InBev product market share regression results.16 Individual brand level results show that not every brand benefited from using AB’s exclusive network: popular brands such as Stella Artois and Beck’s were the least affected, both in terms of statistical significance and magnitude. Also, with the baseline specification, there was no statistically significant impact on a brand’s market share when it joined an AB distribution network. However, as shown in Table 1.8, when we controlled for pre-existing trends, Stella Artois did improve its market share when it had access to the AB distribution network. These results imply some underlying product heterogeneity within InBev brands: low market share brands improved significantly once they were allowed to use exclusive networks while the effect of joining a dominant distribution network was relatively small for these products. In any case, exclusive contracts did play a role in determining a product’s market performance in 16
Given that the market share results are similar regardless of whether the market shares are calculated based on quantity or revenue, for the following analysis, I provide results with quantity based market shares.
18
the beer industry. Moreover, the fact that simply joining AB’s distribution network did not increase a brand’s market share for most of the InBev brands suggests that denying access to the AB distribution network did not have much impact on InBev.17 The foreclosure theory and the incentive based theory interpret the above results differently. On one hand, raising rivals’ costs theory suggests that exclusive dealers are more efficient per se: AB may only sign exclusive contracts with more efficient dealers, with the intention to foreclose the market. On the other hand, the incentive theory suggests that exclusive dealers are more efficient because exclusive dealing is a mechanism designed to solve incentive conflict problems: exclusive dealing networks allow for more investment in the brewer-distributor relationship. As a result, both theories predict that brands would perform better using an exclusive network. The two theories do have different predictions regarding market level outcomes. Based on the raising rivals’ cost theory, excluding rivals from using superior exclusive networks has an adverse effect on market outcomes, while the effects under incentive theories are ambiguous (Besanko and Perry 1993, Bernheim and Whinston 1998). In the next section, I test whether there was favorable effect on market outcomes once more brands were allowed to use AB exclusive networks.
1.6.2
Is Exclusive Dealing Anticompetitive?
Suppose a firm that previously had been denied access to an exclusive network suddenly had access to an exclusive network. If denying access to the exclusive network had anticompetitive effect, all other things being equal, we should see an increase in total market quantities. The AB-InBev distribution agreement provides a chance to test the above counterfactual implied by the raising rivals’ cost theory. Table 1.9 presents the baseline results for changes in market quantities and prices due to the event. Basically, allowing InBev to use the AB exclusive distribution network had no significant impact on general market outcomes. In terms of magnitude, total mar17
Nevertheless, one needs to be careful to generalize this result. InBev is the dominant firm in Europe and its brands are popular enough in the U.S. so that denying access to the AB distribution network could hardly shut InBev out of a market. In fact, Chen (2011) provides empirical evidence to show that exclusive dealing did raise the entry barrier for California microbreweries.)
19
ket quantities increased by only 2 six-packs per store when the distribution channel became less exclusive, and average market prices decreased by less than 1 cent per six-pack. When we looked only at non-macro products, relaxing the exclusive constraint decreased the market quantities by 8 six-packs per store and increased the average non-macro price by 2.5 cents, although the results are not significant. Table 1.10 provides the results when we controlled for trends specific to exclusive stores. In this specification, market quantities even decreased significantly by an average of 44 six packs when more products were using AB exclusive networks. For stores where AB distributors gained InBev brands, total market sales dropped significantly. Compared to the exclusive effect, the reduction in sales is large. On average, when AB distributors gained InBev brands, a store’s total sales decreased by 86 six-packs, and the price increased by 4 cents per six-pack. This implies that there may be some adverse effects when the dominant distribution networks gain more market power. If this is the case, this effect could offset the anticompetitive effect of exclusive dealing: exclusive dealing restricts a dealer’s external trading opportunities and its market power at the distribution level. In sum, there is no evidence to support the view that allowing more brands to use exclusive networks will lead to higher total market sales or lower market price. The above results thus are not consistent with the raising rivals’ costs theory.
1.6.3
Is There Any Incentive Conflict under Common Agency?
If exclusive dealing is not used to foreclose rivals, why do firms employ these arrangements? One possible reason for upstream firms to prefer exclusive dealing is that they are not able to sign complete contracts specifying distributors’ promotional efforts. Because dealer promotional investments in each brand are substitutes, the incentive problem can get worse when distributors are responsible for promoting more and more brands. I then tested whether there is any cannibalization effect in beer distribution by looking at changes in market share for AB products and non-AB products in the non-macro (specialty) segment after their distributors gained InBev brands. By default, all AB products were subject to potential cannibalization effects when their distributors gained InBev brands. However, non-AB products may also have been
20
affected as long as they shared the same distributors with AB products. I focused on non-macro brands because these products, compared to macro brands, are not able to run advertising campaign at the national level and rely more on local promotion efforts. Crowd-Out Effect for AB Specialty Products Table 1.11 presents the potential cannibalization effect for AB’s specialty products. Column 1 gives the effects for all AB’s specialty products, and the rest of the columns provide regression results with brands grouped according to their overall rank in market shares.18 On average, there was no significant cannibalization effect on the AB specialty products: most coefficients are not statistically significant. However, low performance brands’ market share decreased significantly after their distributors acquired InBev brands. The cannibalization effect was less severe when the a low-performance brand’s distributor was an exclusive dealer. I also tested whether there were pre-existing trends across different groups. Table 1.12 shows that there were significant negative trends specific to exclusive stores for middle market share brands. For low market share brands, there were significant negative trends of treatment stores in general. Table 1.13 presents regression results that control for the above trends. Once trends are controlled for, middle market share brands were crowded out significantly at AB exclusive stores. To conclude, I find evidence of crowd-out effects for AB specialty brands at the distribution level for middle and low performance brands when their dealers started to carry more brands. Crowd-Out Effect for Non-AB Products For non-AB products, I estimated two effects: the effect when a brand’s distributor received more brands and the effect when a brand’s distributor carried fewer brands. For example, product S be carried by an AB distributor in County A and a Miller distributor in County B prior to the event. Suppose that prior to the event the AB distributor in County A did not have InBev and the Miller distributor in County B had InBev; then after the event, product S is more crowded out in County A and is 18
I ranked each brand by its overall sales in the dataset.
21
less crowded out in County B at the distribution level. Table 1.14 shows the variations I use to identify these crowd-out effects. I limited my sample to brands that were carried by at least one AB distributor in Northern California. Other than AB, there were 8 product families that were affected by the event. In Table 1.14, In (Out) represents whether the product’s local distributor added or dropped InBev from its brand portfolio. From Table 1.14, there was no specific pattern regarding to a firm’s tendency to seek or evade InBev brands at the distribution level (except for Crown Imports, which tended to be in different distribution houses from InBev prior to the event). This observation provides us with more confidence in the research design. Table 1.15 presents the regression results for all non-AB specialty products and separates the results by the products’ market share. Overall, for the eight firms that hired at least one AB distributor in the dataset, their market share increased after the event. When the distribution channel was less crowded, all brands benefited from the change. However, when the distribution channel was more crowded, only brands with existing low market shares were crowded out significantly. In all four columns, the coefficients of dropping InBev brands from the common agency are larger than the ones of adding InBev brands. The above results provide evidence of incentive conflicts at the distribution level. The empirical results show that InBev brands gained more market share when they had access to the AB exclusive distribution network. However, AB’s specialty brands’ market share was lower after the event. The evidence of cannibalization effects on AB specialty products suggests that the promotional efforts for different brands within the same distribution house may be substitutes, at least in the short run. Distributors shifted their promotional efforts toward incoming InBev brands. One explanation for this substitution is that brands with higher market share, such as InBev, generated larger volumes and needed more staff and marketing support. Another explanation is that the promotional effort at the distribution level and at the manufacturing level are complementary. If so, we can expect a distributor to allocate more of its promotional effort to brands that were able to receive more support from their own brewery.
22
1.7
Conclusion and Discussion
This paper provides empirical evidence to show that exclusive dealing does provide benefits to firms, and that these benefits are not completely anticompetitively based. As a distributor acquires more brands, some of its brands’ performance declines. This cannibalization effect suggests that incentive conflicts exist under common agency, which explains why firms would prefer to use exclusive dealers in the beer industry. More importantly, the competitive effects of exclusive dealing on overall market sales and prices are not consistent with foreclosure theory. Given that fixed costs are the main concern for distributors, consolidation at the distribution level is another important issue. The industry has become much more concentrated after the Miller/Coors and the AB/InBev merger events. When Miller and Coors integrated their distribution system after the merger, some of their existing distributors lost representation of these macro brands. If economies of scale are important at the distribution level, this may result in fewer distributors and aggravate the incentive conflict problem at the distribution level. Future research on the dynamics of the vertical structure of the industry will be highly valuable. Finally, due to data limitation, the results discussed in this paper are based solely on brewerdistributor arrangements. More data on contracting details (terms on distributor margins and how contracts can be terminated) between brewers and distributors will be extremely useful in studying the vertical relationships between firms.
0
5
10
Percentage 15 20
25
30
35
23
0
20
40
60
80
100 120 140 160 180 200 220 240 260 Number of Stores
Figure 1.1: Number of Stores
1800
2300
2800
Average Store Weekly Sales
3300
24
1300
AB−InBev Deal
2006Apr
2006Aug
2006Dec
2007Apr
2007Aug
2007Dec
Baseline Stores Treatment 1: AB Nonexclusive Treatment 2: AB Exclusive Figure 1.2: Average Store Weekly Sales in Six Packs
2008Apr
InBev Market Share 4 6 8
10
25
2
AB−InBev Deal
2006Apr
2006Aug
2006Dec
2007Apr
2007Aug
Baseline Stores Treatment 1: AB Nonexclusive Treatment 2: AB Exclusive Figure 1.3: InBev Market Share
2007Dec
2008Apr
26
Table 1.1: Some Typical Entries From CBBD Annual Directory Distribution Status
Brand Portfolio
Exclusive AB distributor
180 Energy Drink, Anheuser-Busch, Redhook Ale Brewery, Rolling Rock, Widmer Brothers Brewing
Nonexclusive AB distributor
Anheuser Busch, Boston Beer, Gambrinus, Gordon Biersch Brewing, Heineken USA, InBev USA, Redhook Ale Brewery, Scottish & Newcastle Importers, Sierra Nevada Brewing, Spaten North America, Widmer Brothers Brewing
Non-AB distributor 1
Anchor Brewing, Asahi Breweries, Boston Beer, Diageo-Guinness USA, Gambrinus, Heineken, Hornell Brewing, InBev USA, Mark Anthony Brands, McKenzie River, Miller Brewing,Pabst Brewing, Pyramid Breweries, Sapporo USA, Scottish & Newcastle Importers, Sierra Nevada Brewing, Sierra Springs Water, US Beverage
Non-AB distributor 2
Alaska Brewing, Barton Beers, Boston Beer, Diageo-Guinness USA, InBev USA, Lake Tahoe Brewing, Mark Anthony Brands, Miller Brewing, Molson Coors Brewing, New Belgium Brewing, Pyramid Breweries, Redhook Ale Brewery, Sapporo USA, Sierra Nevada Brewing
Independent distributor
Alaska Brewing, Allagash Brewing, Anderson Valley Brewing, Arizona Beverage, Asahi Breweries USA, Bear Republic, Binding International, Bison Brewing, California Cider, Costancia Brewery Firestone-Walker Brewing, Friedlin Imports Full Sail Brewing Co., Humboldt Brewing, InBev USA, Mad River Brewing, Mendocino Brewing Marin Brewing, Moylan’s Brewery, Nestle Beverage, Ommegang Brewery, Pabst Brewing, Panorama Brewing, Sapporo USA, Scottish & Newcastle Importers, Sierra Nevada Brewing, Spaten North America, Speakeasy, Stone Brewing, Sudwerk Privatbrauerei, Thames America Trading, U.S. Beverage, Wyder’s Beverage
Notes: 180 Energy Drink, Redhook, Rolling Rock, and Widmer Brothers are all affiliated Anheuser Busch products listed on Anheuser Busch company website.
27
Table 1.2: Product Attributes of Top 30 and InBev Brands By Category Rank
Brand
Market Shares
ABV
Calories (per 12oz)
Average Price (per 6 pack)
Coors Light Bud Light Budweiser Miller Genuine Draft Miller Lite Coors Banquet Pabst Blue Ribbon Michelob Ultra Light O’doul’s Near Beer Miller High Life
10.40% 7.29% 7.02% 3.48% 2.51% 2.00% 0.91% 0.87% 0.81% 0.80%
4.2% 4.2% 5.0% 4.7% 4.2% 5.0% 5.0% 4.2% 0.4% 4.7%
102 110 145 143 96 142 153 95 70 143
4.15 3.94 3.98 4.30 4.12 3.99 3.25 5.82 5.07 3.68
Sierra Nevada Pale Ale Fat Tire Amber Ale Samuel Adams Boston Lager Widmer Brothers Hefeweizen Anchor Steam Redhook ESB Ale Pyramid Hefeweizen Ale Gordon Biersch Marzen Larger
4.31% 1.39% 1.07% 1.04% 0.77% 0.77% 0.76% 0.74%
5.6% 5.2% 4.9% 4.9% 5.8% 5.2% 5.2% 5.6%
190 164 170 156 152 179 145 174
7.10 7.65 6.67 6.69 8.02 6.44 6.91 6.57
Corona Extra Heineken Corona Light Tecate Smirnoff Ice Pacifico Newcastle Brown Ale Guinness Draught Ale Stella Artois (InBev) Heineken Light Beck’s (InBev) Amstel Light Bass Pale Ale (InBev) Boddingtons (InBev) Beck’s Premier Light (InBev)
8.05% 5.28% 2.22% 2.10% 2.03% 1.76% 1.72% 1.46% 1.25% 1.09% 1.02% 0.77% 0.56% 0.26% 0.09%
4.6% 5.0% 3.7% 4.5% 5.0% 4.8% 4.7% 4.2% 5.2% 3.2% 5.0% 3.5% 5.1% 4.6% 2.3%
148 150 109 142 228 146 140 125 145 110 143 95 150 155 64
7.02 7.01 7.17 4.84 7.90 6.47 7.45 7.91 7.85 7.06 6.93 7.05 6.81 7.50 7.01
Domestic Macro 1 3 4 7 8 12 22 23 24 25 Domestic Specialty 6 16 19 20 27 28 29 30 Imports 2 5 9 10 11 13 14 15 17 18 21 26 40 65 107
Notes: ABV (alcohol by volume) measures the percentage of alcohol as a fraction of the total volume.
28
Table 1.3: Summary Statistics Control group
Treatment group 1: AB nonexclusive
Treatment group 2: AB exclusive
Number of stores
18
139
40
Weekly store revenue
1.14
0.94
1.14
Average store price per six pack Number of brands per store
5.59
5.43
5.29
125.75
129.19
134.20
AB distributors:
4
9
3
Number of non-macro brands carried Market share from AB macro products Market share from AB non-macro products Market share from non AB products Total market share
29
27.29
9.67
21.58
22.19
22.47
3.23
3.49
3.30
18.38
16.15
0
41.98
43.48
25.77
Non-AB distributors:
6
10
7
Number or non-macro brands carried Total market share
24.2
25.25
51.66
37.7
26.96
43.84
Store characteristics
Note: “Weekly store revenue” are normalized to have mean equal to zero.
Table 1.4: Diagnosing Trends Using Pre-event Data: InBev Market Share (1) InBev
(2) Bass
(3) Beck’s
(4) Beck’s Light
(5) Boddingtons
(6) Stella Artois
Trend for all groups
0.0148∗∗∗ (0.0029)
-0.0507∗∗∗ (0.0053)
0.0985∗∗∗ (0.0065)
0.0060∗ (0.0025)
0.0047∗ (0.0022)
0.0139 (0.0100)
Trend for exclusive stores
0.00076 (0.00103)
-0.00014 (0.00166)
-0.00033 (0.00265)
0.00059 (0.00124)
0.00101 (0.00074)
0.00319 (0.00263)
Trend for treatment stores
-0.00647∗∗ (0.00208)
0.00647∗ (0.00270)
0.00232 (0.00250)
0.00130 (0.00210)
-0.00073 (0.00109)
-0.04040∗∗∗ (0.00882)
0.932∗∗∗ (0.012)
1.004∗∗∗ (0.026)
1.691∗∗∗ (0.037)
0.217∗∗∗ (0.013)
0.476∗∗∗ (0.014)
1.173∗∗∗ (0.030)
39774 0.077
8232 0.258
8190 0.189
7224 0.047
7938 0.057
8190 0.243
Constant Observations Adjusted R2
Notes: All regressions control for store and monthly fixed effects. Standard errors, clustered at the store level, are shown in parentheses. ∗
significant at 10%,
∗∗
significant at 5%,
∗∗∗
significant at 1%.
29
Table 1.5: Diagnosing Trends Using Pre-event Data: Market Quantity and Price (1) Quantity
(2) Quantity
(3) Price
(4) Price
Trend for all groups
-16.870∗∗∗ (2.564)
-7.140∗∗∗ (1.405)
0.0258∗∗∗ (0.0011)
0.0358∗∗∗ (0.0009)
Trend for exclusive stores
-5.606∗∗∗ (1.370)
-2.883∗∗∗ (0.763)
-0.0009 (0.0005)
0.0003 (0.0004)
Trend for treatment stores
-2.445 (1.610)
-1.514 (0.951)
-0.0022∗∗ (0.0007)
-0.0013∗∗ (0.0004)
1778.5∗∗∗ (12.3)
859.7∗∗∗ (6.8)
5.265∗∗∗ (0.005)
6.579∗∗∗ (0.004)
Constant
Drop macro products? Observations Adjusted R2
No
Yes
No
Yes
8274 0.341
8274 0.470
8274 0.235
8274 0.338
Notes: All regressions control for store and monthly fixed effects. Standard errors, clustered at the store level, are shown in parentheses. ∗
significant at 10%,
∗∗
significant at 5%,
∗∗∗
significant at 1%.
Table 1.6: Benefit from Exclusive Dealing: InBev Market Share (1) Quantity
(2) Revenue
(3) Quantity
(4) Revenue
0.0603∗∗∗ (0.0174)
0.0488∗∗ (0.0181)
0.0603∗∗∗ (0.0174)
0.0488∗∗ (0.0181)
InBev moved to AB (exclusive or nonexclusive)
0.0219 (0.0285)
-0.0154 (0.0286)
0.1020 (0.0634)
0.0575 (0.0645)
Constant
0.963∗∗∗ (0.012)
1.013∗∗∗ (0.013)
0.886∗∗∗ (0.021)
0.951∗∗∗ (0.021)
No
No
Yes
Yes
97188 0.067
97188 0.048
97188 0.070
97188 0.050
InBev moved to AB exclusive
Trends Observations Adjusted R2
Notes: All regressions control for store and monthly fixed effects. Standard errors, clustered at the store level, are shown in parentheses. ∗
significant at 10%,
∗∗
significant at 5%,
∗∗∗
significant at 1%.
30
Table 1.7: Benefit from Exclusive Dealing: Market Share from Five Individual Brands (1) Bass
(2) Beck’s
(3) Beck’s Light
(4) Boddingtons
(5) Stella Artois
0.1800∗∗∗ (0.0391)
-0.0187 (0.0549)
0.0469∗ (0.0190)
0.0515∗ (0.0211)
0.0411 (0.0541)
InBev moved to AB (exclusive or nonexclusive)
0.0828 (0.0727)
0.0730 (0.0989)
-0.0190 (0.0230)
-0.0409 (0.0413)
0.0096 (0.1473)
Constant
0.916∗∗∗ (0.025)
1.891∗∗∗ (0.040)
0.240∗∗∗ (0.012)
0.485∗∗∗ (0.016)
1.124∗∗∗ (0.037)
20332 0.310
20332 0.182
16536 0.219
19604 0.084
20384 0.135
InBev moved to AB exclusive
Observations Adjusted R2
Notes: All regressions control for store and monthly fixed effects. Standard errors, clustered at the store level, are shown in parentheses. ∗
significant at 10%,
∗∗
significant at 5%,
∗∗∗
significant at 1%.
Table 1.8: Market Share from Five Individual Brands: Controlling for Trends (1) Bass
(2) Beck’s
(3) Beck’s Light
(4) Boddingtons
(5) Stella Artois
0.1800∗∗∗ (0.0391)
-0.0187 (0.0549)
0.0469∗ (0.0190)
0.0515∗ (0.0211)
0.0411 (0.0541)
InBev moved to AB (exclusive or nonexclusive)
-0.0995 (0.0792)
0.0730 (0.0989)
-0.0190 (0.0230)
-0.0409 (0.0413)
0.6130∗∗ (0.2031)
Constant
0.939∗∗∗ (0.035)
2.229∗∗∗ (0.060)
0.186∗∗∗ (0.017)
0.506∗∗∗ (0.023)
0.934∗∗∗ (0.046)
Yes
Yes
Yes
Yes
Yes
20332 0.317
20332 0.196
16536 0.221
19604 0.091
20384 0.159
InBev moved to AB exclusive
Trends Observations Adjusted R2
Notes: All regressions control for store and monthly fixed effects. Standard errors, clustered at the store level, are shown in parentheses. ∗
significant at 10%,
∗∗
significant at 5%,
∗∗∗
significant at 1%.
31
Table 1.9: Market Quantity and Average Price Quantity
Price
(1) All Products
(2) Drop Macro Products
(3) All Products
(4) Drop Macro Products
InBev moved to AB exclusive
1.90 (22.15)
-8.19 (10.51)
-0.0090 (0.0123)
0.0247 (0.0174)
InBev moved to AB (exclusive or nonexclusive)
-85.98∗∗ (29.97)
-57.54∗∗ (19.39)
-0.0339 (0.0213)
-0.0155 (0.0157)
1738.1∗∗∗ (10.7)
841.5∗∗∗ (6.1)
5.313∗∗∗ (0.005)
6.648∗∗∗ (0.004)
20488 0.354
20488 0.377
20488 0.280
20488 0.519
Constant Observations Adjusted R2
Notes: All regressions control for store and monthly fixed effects. Standard errors, clustered at the store level, are shown in parentheses. ∗
significant at 10%,
∗∗
significant at 5%,
∗∗∗
significant at 1%.
Table 1.10: Market Quantity and Average Price: Controlling for Trends (1) Quantity
(2) Quantity
(3) Price
(4) Price
-77.85 (42.27)
-44.35∗ (20.63)
-0.0090 (0.0123)
0.0247 (0.0174)
InBev moved to AB (exclusive or nonexclusive)
-85.98∗∗∗ (17.53)
-57.54∗∗∗ (8.55)
0.0446∗ (0.0187)
0.0199 (0.0184)
Constant
1213.5∗∗∗ (36.8)
684.0∗∗∗ (18.0)
5.296∗∗∗ (0.007)
6.462∗∗∗ (0.006)
Drop macro products
No
Yes
No
Yes
Trends
Yes
Yes
Yes
Yes
20488 0.395
20488 0.418
20488 0.308
20488 0.562
InBev moved to AB exclusive
Observations Adjusted R2
Notes: All regressions control for store and monthly fixed effects. Standard errors, clustered at the store level, are shown in parentheses. ∗
significant at 10%,
∗∗
significant at 5%,
∗∗∗
significant at 1%.
32
Table 1.11: Crowd Out Effects: AB Specialty Products (1) All Brands
(2) High Market Share Brands
(3) Middle Market Share Brands
(4) Low Market Share Brands
InBev moved to AB exclusive
0.0335 (0.0221)
0.0274 (0.0258)
-0.0424 (0.0246)
-0.0288 (0.0378)
InBev moved to AB (exclusive or nonexclusive)
0.0197 (0.0260)
0.0428 (0.0387)
-0.0033 (0.0214)
-0.191∗∗∗ (0.051)
Constant
0.744∗∗∗ (0.012)
0.979∗∗∗ (0.015)
0.290∗∗∗ (0.014)
0.188∗∗∗ (0.032)
105664 0.034
70668 0.066
25272 0.196
9724 0.154
Observations Adjusted R2
Notes: All regressions control for store and monthly fixed effects. Standard errors, clustered at the store level, are shown in parentheses. ∗
significant at 10%,
∗∗
significant at 5%,
∗∗∗
significant at 1%.
Table 1.12: Diagnosing Trends Using Pre-event Data: AB Specialty Products (1) All Brands
(2) High Market Share Brands
(3) Middle Market Share Brands
(4) Low Market Share Brands
0.0134∗∗∗ (0.0020)
0.0149∗∗∗ (0.0026)
0.0093∗∗∗ (0.0021)
0.0147∗∗∗ (0.0026)
Trend for exclusive stores
-0.0010 (0.0008)
0.0003 (0.0012)
-0.0055∗∗∗ (0.0016)
-0.0036 (0.0025)
Trend for treatment stores
0.0021 (0.0011)
0.0041∗∗∗ (0.0012)
-0.0014 (0.0016)
0.0004 (0.0022)
Constant
0.703∗∗∗ (0.012)
0.950∗∗∗ (0.017)
0.282∗∗∗ (0.013)
0.156∗∗∗ (0.029)
43722 0.028
28224 0.044
11760 0.077
3738 0.275
Trend for all groups
Observations Adjusted R2
Notes: All regressions control for store and monthly fixed effects. Standard errors, clustered at the store level, are shown in parentheses. ∗
significant at 10%,
∗∗
significant at 5%,
∗∗∗
significant at 1%.
33
Table 1.13: Crowd Out Effects: AB Specialty Products: Controlling for Trends (1) All Brands
(2) High Market Share Brands
(3) Middle Market Share Brands
(4) Low Market Share Brands
InBev moved to AB exclusive
0.0335 (0.0221)
0.0274 (0.0258)
-0.0789∗∗ (0.0239)
-0.0288 (0.0378)
InBev moved to AB (exclusive or nonexclusive)
0.0197 (0.0260)
-0.1101 (0.0632)
-0.0026 (0.0216)
-0.1910∗∗∗ (0.0507)
Constant
0.741∗∗∗ (0.012) Yes
0.980∗∗∗ (0.026) Yes
0.192∗∗∗ (0.019) Yes
0.218∗∗∗ (0.035) Yes
105664 0.034
70668 0.068
25272 0.207
9724 0.157
Trends Observations Adjusted R2
Notes: All regressions control for store and monthly fixed effects. Standard errors, clustered at the store level, are shown in parentheses. ∗
significant at 10%,
∗∗
significant at 5%,
∗∗∗
significant at 1%.
Table 1.14: Distribution Status Affected by the AB-InBev Distribution Agreement County/Firm
AB
Anchor
Crown
Guinness
Heineken
Newcastle
Pyramid
Samuel Adams
Sierra Nevada
Del Norte, Humboldt San Jose (city) Alameda Monterey, San Benito, Santa Cruz Sacramento Yolo Solano Lassen San Francisco Stanislaus
In
*
*
*
*
*
In
In
*
In In In
Out * In
In In In
In Out In
In Out In
Out – In
Out Out *
In In Out
In In Out
In In In – – –
Out Out * – – –
– – In – – –
In In Out – – –
Out Out Out – – –
Out Out Out * – –
Out Out Out – – –
Out Out Out – – –
In In In – – –
Notes: I include products which use at least one AB dealer for distribution. –: not affected by the AB-InBev distribution agreement In: distributor acquires InBev after the event ∗:
cannot identify the distributor in the area
34
Table 1.15: Crowd Out Effects: Rival Brands (1) All Brands
(2) High Market Share Brands
(3) Middle Market Share Bands
(4) Low Market Share Brands
InBev in
0.119∗∗∗ (0.031)
0.307∗∗∗ (0.064)
-0.015 (0.025)
-0.041∗∗ (0.014)
InBev out
0.308∗∗∗ (0.027)
0.626∗∗∗ (0.058)
0.096∗∗∗ (0.017)
0.042∗∗ (0.016)
Constant
2.290∗∗∗ (0.009)
4.379∗∗∗ (0.022)
0.937∗∗∗ (0.012)
0.230∗∗∗ (0.010)
248092 0.007
105456 0.021
104208 0.012
38428 0.030
Observations Adjusted R2
Notes: All regressions control for store and monthly fixed effects. Standard errors, clustered at the store level, are shown in parentheses. ∗
significant at 10%,
∗∗
significant at 5%,
∗∗∗
significant at 1%.
35
Chapter 2 Estimating the Foreclosure Effect of Exclusive Dealing: Evidence from the Entry of Specialty Beer Producers Chia-Wen Chen
2.1
Introduction
An exclusive dealing contract is a vertical agreement between a manufacturer and a dealer that forbids the dealer from promoting other manufacturers’ products. Such an incentive contract is controversial in competition policy because of the potential foreclosure effects. For example, in 1997, Anheuser Busch launched an incentive program that provided discounts and other benefits to dealers who went exclusively with Anheuser Busch. At that time, many microbreweries complained that they were being dropped by distributors due to this practice. Theory suggests that exclusive dealing can be efficiency-enhancing because it encourages investment in manufacturer-dealer relationships; however, theory also suggests that exclusive dealing can be anticompetitive because it raises the cost of entry for rivals. In practice, the extent to which exclusive dealing enhances incentives for dealers or dampens competition remains an
36
empirical question to be explored.1 This paper examines the relationship between exclusive dealing and specialty beer producers’ entry decisions.2 I model each firm’s entry decision for a location using a static entry game that allows post-entry profits to depend on the presence of rivals. I employ a panel scanner dataset and estimate the demand for beer using a discrete choice model to control for heterogeneity across firms and locations. After controlling for expected post-entry sales, a location’s proximity to its brewery, the size of a store’s selling area, and the expected number of rivals, I find that exclusive dealing reduces a firm’s entry probability. Nevertheless, when I implement counter-factual experiments to study the effect of banning exclusive dealing, the results show very limited improvement in consumer welfare from such banning. This paper contributes to the existing literature in several ways. First, empirical studies have tested the potential anticompetitive effects of exclusive dealing by estimating the changes in equilibrium prices or the implied competitive advantage gained due to these exclusive contracts (Asker, 2005; Sass, 2005). I extend these studies by looking directly at the entry patterns of specialty beer producers. Because these firms are smaller in scale, their entry decisions are more likely to be impeded by concerns about higher entry costs due to exclusive dealing. Second, because beer products are differentiated, counter-factual analysis based solely on prices may underestimate the foreclosure effects on consumer welfare. With the new dataset, I estimate the demand for these specialty products and study the counter-factuals when a consumer’s choice set is expanded to include more products with different attributes. Finally, I study 1
The foreclosure argument led the courts to condemn exclusive dealing contracts between dominant firms and their distributors after the enactment of the Clayton Act in 1914. However, the Chicago school’s defense of exclusive dealing started to prevail in the 1970s, and since then the courts have emphasized a rule of reason approach to such provisions. For earlier cases where the courts were against exclusive dealing, see Standard Fashion Co. v. Magrane-Houston Co., 258 U.S. 346 (1922) and Standard Oil Co. of California et al. v. United States, 337 U.S. 293 (1949). For decisions favoring exclusive contracts, see Tampa Electric Co. v. Nashville Coal Co., 365 U.S. 320 (1961) and Beltone Electronics. Corp., 100 F.T.C. 68 (1982). In Europe, a recent practice by Intel, one that provided rebates and cash benefits to manufacturers and retailers in exchange for purchasing most of their products from Intel, was found to be anticompetitive by the European Commission and resulted in a e1.06 billion fine in 2009. 2 Most specialty beer producers are microbreweries. However, some of the specialty beer producers, such as the Boston Beer Company and the Sierra Nevada Brewing Company, have grown successfully and are no longer microbreweries.
37
the specialty beer producers’ joint entry decisions based on a strategic framework to explore the interactions between firms. Inferences about the foreclosure effect through direct comparisons of entry patterns can suffer from omitted variable bias because post-entry market outcomes can differ across locations and firms. Thus, I estimate the demand for beer to construct counter-factual expected post-entry sales. With the panel structure of the dataset, I can control for invariant brand and location fixed effects. I estimate the demand system using a nested logit model. The model allows the substitution patterns to vary based on product segments (e.g., country origin or the style of beer) and has more reasonable substitution patterns than a simple logit model.3 This paper also builds on empirical studies that estimate static entry games. Bresnahan and Reiss (1991) show how to estimate an equilibrium model of entry for symmetric firms with data on market characteristics and the number of firms. Similar to Bresnahan and Reiss (1991), I allow both variable profits and fixed costs to depend on the number of firms. Moreover, given that I observe a pool of global potential entrants, along with their actual sales and entry patterns, my model allows heterogeneous firms to produce differentiated products with different fixed costs.4 Following recent developments in estimating strategic games (Seim, 2006; Augereau, Greenstein and Rysman, 2006; Ellickson and Misra, 2008; Sweeting, 2009; Bajari, Hong, Krainer and Nekipelov, 2010), I model the entry behavior of specialty beer producers using an incomplete game framework that helps to incorporate a large number of players in the game. In this setup, a firm’s entry decision depends on the expected market profitability and a piece of private information. I estimate the model following a two-step estimation procedure similar to that of Ellickson and Misra (2008) and Bajari, Hong, Krainer and Nekipelov (2010). In addition, I use the demand estimates to control for post-entry sales. The estimation is done in three steps. 3
The nested logit model has wide applications in estimating transportation and energy demand, and is also applied to other industries such as automobiles, movies, home videos, and banking. (Goldberg, 1995; Einav, 2007; Chiou, 2008; Dick, 2008). 4 Berry (1992) first estimates a model of entry game that allows for firm heterogeneity in fixed costs. This paper exploits data on beer prices and sales and allows a firm’s variable profits to also depend on rivals’ identities.
38
In the first step, I estimate the equilibrium entry probabilities implied by the model. In the second step, I estimate the demand for beer. Using the demand estimates and the beliefs about rivals’ entry probabilities, I construct expected post-entry sales. Then, in the third step, I plug the above estimates into the likelihood function to recover a firm’s fixed costs. Economic theories vary in their explanations of exclusive contracts. Traditionally, the Chicago school has argued that exclusive dealing cannot be used as a device for monopolization (Posner, 1976; Bork, 1978): if the sole purpose of exclusive contacts were to restrict competition, downstream buyers would never sign them in the first place because doing so would only lower the potential total surplus. Bernheim and Whinston (1998) support this claim by showing that common agency and exclusive dealing are both efficient when there is no contracting externality. Moreover, incentive theories show that exclusive dealing enhances incentives for investment or promotional efforts when contracting externality or incomplete contracts are allowed for (Marvel, 1982; Klein and Murphy, 1988; Besanko and Perry, 1993; Martimort, 1996; Bernheim and Whinston, 1998; Segal and Whinston, 2000b). Anticompetitive arguments focus on the potential foreclosure effect of exclusive dealing. In this literature, manufacturers either sign exclusive contracts with lowercost buyers, trying to raise their rivals’ costs, or with a large number of buyers, trying to foreclose the market directly when facing minimum economies of scale or noncoincident markets.5 In addition, Segal and Whinston (2000a) point out that whether manufacturers can successfully carry out the above “naked exclusion” scheme depends on how well they can exploit the coordination problem faced by buyers.6 The empirical literature on exclusive dealing is limited.7 Heide, Dutta and Bergen (1998) conduct a survey of managerial distribution decisions in manufacturing indus5
Salop and Scheffman (1983), Aghion and Bolton (1987), Rasmusen, Ramseyer and Wiley (1991), and Bernheim and Whinston (1998) provide theoretical foreclosure arguments for exclusive dealing. 6 Simpson and Wickelgren (2007), Abito and Wright (2008), and Doganoglu and Wright (2010) also provide settings that allow exclusive dealing to achieve inefficient outcomes. Simpson and Wickelgren (2007) and Abito and Wright (2008) consider the case when buyers compete, and Doganoglu and Wright (2010) study exclusive contracts under the network effect. 7 For a thorough review of the empirical studies on vertical integration and vertical restraints, see Lafontaine and Slade (2007) and Lafontaine and Slade (2008).
39
tries. They find that the main reason for managers to consider exclusive contracts is fear of the free riding problem presented in common agency. More recently, Sass (2005), Asker (2005) and Rojas (2010) study exclusive dealing in the beer industry and find no evidence supporting anticompetitive theory of exclusive dealing. Sass (2005) studies a cross-sectional survey of 381 beer distributors in 1997 and finds that exclusive dealers on average generate higher prices and larger sales for their suppliers, which is more consistent with incentive-based theory. Asker (2005) exploits brewer-distributor data and the scanner data of a retail grocery chain in Illinois in 1994 to test the raising rivals’ costs theory. By recovering the costs incurred and the promotional efforts made of distributors in exclusive and less exclusive markets, Asker (2005) finds that distributors in less exclusive markets are not more efficient than distributors in exclusive markets and rejects the foreclosure hypothesis. Rojas (2010) looks at a scanner data set from 1988 to 1992 and finds that exclusive dealing and exclusive territory agreements are more consistent with the existence of a welfare-enhancing effect. Another related paper is Slade (1998), which studies whether the United Kingdom Monopolies and Mergers Commission (MMC) recommended measures that broke the tie between brewers and public houses led to lower prices as the MMC claimed would happen. Slade (1998) finds that the recommended measures had the opposite effect and suggests that it is important to recognize the strategic aspects of beer ownership structure when considering banning these contracts.8 I find that the demand for beer is elastic. Moreover, the price of a specialty beer product is lower at locations that are closer to its producer’s establishment. Controlling for prices, I also find that consumers enjoy a product more if it is locally brewed. These findings explain why most specialty beer producers are not present in every location. I then take the demand estimates to the model of entry. When strategic interactions are not allowed in the model, I find that exclusive dealing has 8
The exclusive arrangements in Slade (1998) are between brewers and pubs in the United Kingdom. The industry background is very different in the United States: most states require brewers to find distributors to bridge them to retailing outlets. The brewer, distributor and retailer structure is called the three-tier system and is heavily regulated by state laws in the United States.
40
no impact on a producer’s entry decision. Once strategic interactions are allowed, I find that a store with an Anheuser Busch exclusive distributor is associated with a reduction of 3.5 percentage points (12.5 percent) in a specialty beer producer’s entry probability, suggesting some foreclosure effects due to exclusive dealing. To identify the strategic effects among specialty beer producers, I use the distances from a store to rival firms’ breweries as exclusion restrictions. I find that a firm has lower fixed costs at a location where the number of rivals is larger. The result implies that firms can benefit from clustering their strategic decisions, which is similar to the findings in previous studies that estimate strategic effects, such as Ellickson and Misra (2008), Sweeting (2009), Bajari, Hong, Krainer and Nekipelov (2010), and Vitorino (2010). Finally, I use demand estimates to carry out counter-factual experiments that remove exclusive dealing. I do not find any welfare changes due to banning exclusive dealing. Moreover, I find that adding more specialty brands does not provide much benefit to consumers: when all California specialty brands are included in a market, the change in consumer welfare is at most 4 cents per market. The above results show that banning exclusive dealing will not have much impact on consumer welfare. This paper proceeds as follows. I begin by discussing the industry and the data. Then, I describe the model of demand and entry behavior. Next, I discuss the corresponding estimating procedures and identification issues. Finally, I present the empirical results and discuss the implications and potential future research.
2.2
Industry Background
The beer industry has always been an important industry in the United States. During the sample period of the data, there were more than 13,000 labels and 1,500 breweries in the U.S., and the total sales of beer were around 100 billion dollars.9 However, the industry is highly concentrated even with such a large number of brands: Anheuser Busch, Miller and Coors collectively claim nearly 80% of the market. It is common to divide the industry into three segments: domestic macro brands, 9
Statistics obtained from the National Beer Wholesalers Association and the Brewers Association websites.
41
domestic specialty brands and imported brands.10 Domestic macro brands refer to brands with mass production scale and advertising campaigns. These are usually lower-priced products with large package size options. Anheuser Busch, Miller and Coors are the main competitors in this category. Domestic specialty brands, or craft beer products, refer to brands that emphasize flavor and taste. Most of these specialty brands are produced locally at a smaller scale. Sierra Nevada Brewing and the Boston Beer company are the most successful and nationally known companies in this segment. They are also pioneers that led the microbrewery movement during the 1980s and 1990s; during that time a lot of entrepreneurs entered the specialty beer segment and the boom ended with a shakeout in the late 1990s. The last category, imported brands (usually well-established ones), includes products from foreign countries.11 Most of the firms focusing on specialty brands are local. While a microbrewer can help local customers understand its brands by operating a local pub or by directly contacting retail outlets, it can become very difficult to promote its brands to other markets. Building and maintaining good relationships with distributors is thus vital to a brewer’s success, especially when it comes to entering a new market. Distributors not only help store and transport the products, but they are also responsible for point-of-sale promotions and are expected to have sales staff visit or call accounts regularly. While a brewery would prefer its dealers to be exclusive or to devote as much promotional care to its products as possible, it may not be in the best interest of dealers to build a relationship with just one brewery. To maintain a stable cash stream, a distributor is more likely to prefer having all the best-selling products in its house, even though there may be some conflicts of interest handling brands from competing breweries at the same time. Given the incentive problem described above, Anheuser Busch launched an incentive program during the late 1990s, called “100 percent share of mind.” The program 10
For example, see Tremblay and Tremblay (2005). It is a common practice for firms to gain market share or to enhance brand portfolios through mergers and acquisitions. For example, Anheuser Busch acquired two microbreweries in the late 1990s in response to consumers’ demand for specialty brands, and recently merged with InBev, the biggest brewer in the Europe. 11
42
provided discounts and other benefits for dealers in exchange for exclusivity with Anheuser Busch. As a result, many brands from other breweries were dropped by their Anheuser Busch distributors and some of them had to turn to smaller or independent distributors. The exclusive program raises concerns of potential anticompetitive and foreclosure effects because Anheuser Busch has been the biggest competitor in the industry and its distribution networks are often viewed as a superior promotional vehicle due to economies of scale in distribution.12 The potential foreclosure effect of exclusive dealing on specialty beer producers’ entry decisions is thus the main focus of this paper.
2.3
Data
The demand scanner data set is provided by Nielsen. The data set contains weekly price and sales data of the malt beverage category for all stores of a major grocery chain in Northern California from April, 2006 to April, 2008. The original dataset comes at the UPC level, which includes all sales records from all packaging options for all brands. I collapse the data to quarterly brand level to take into account that some specialty brands may have very little, or even no sales, within a week or a month at the UPC level, and that the demand estimates from quarterly data are more likely to be suitable for policy analysis. I search a product’s website for information on the product’s country of origin (domestic or foreign) and product ownership. For domestic specialty beer producers, I calculate the distance from a store to the firm’s nearest establishment (brewery or beer pub) using the Google map service. I also collect data on local contract rents at the zip code area level (“Gross Rent”) from Census 2000 as a further control for a location’s fixed costs.13 The scanner dataset also includes a product category variable, which includes light, lager, ale, stout/porter, malt liquor, non-alcoholic (alcohol by volume less than 12
Economies of scale are important in distribution. For example, Bump Williams, an IRS industry analyst, said “there’s nobody better than these three networks (Anheuser Busch, Coors, and Miller). They can get these beers on shelves overnight.” See Kesmodel (2007). 13 In Census 2000, “Gross Rent” is defined to include “contract rent and estimated average monthly cost of utilities (electricity, gas, water and sewer) and fuels (oil, coal, kerosene, wood, etc.) if these are paid by the renter.”
43
0.5%), and three categories for alternative malt beverage. These are basically the beer styles that are used in estimating the demand system. Due to product similarity, I assign all alternative malt beverages to one style (alternative), and generate a new style specifically for domestic mainstream lager products.14 In this way, I end up with eight different product styles. Data on Anheuser Busch exclusive distributors and their territories are from the California Beer and Beverage Distributors (CBBD) annual member directories. Each directory contains a list of each member distributor’s representing brewers and its operating counties.15 Table 2.1 lists some typical entries from the 2006 CBBD directory, with California specialty beer producers that can be matched to the scanner data set denoted by bold type. Most distributors represent at least one of the domestic macro brewers (Anheuser Busch, Miller and Molson Coors) and also carry other imported brands and domestic specialty brands. There are in addition independent distributors that do not carry any brands from domestic macro firms but collect a large number of specialty or imported brands.
2.3.1
Store Attributes
Table 2.2 shows the means and standard deviations for store attributes. I provide summary statistics for stores with Anheuser Busch non-exclusive distributors and stores with Anheuser Busch exclusive distributors. About 18% of the stores are located in counties in which Anheuser Busch has exclusive distributors. On average, these stores have more California specialty beer producers entering the markets and generate more sales.16 Therefore, it appears that exclusive dealing has no anticompetitive effect on a firm’s entry decision. However, from Table 2.2, we also see that store attributes and demographics differ across the two types of stores. Inferences about the foreclosure effect directly from Table 2.2 can thus suffer from omitted variable bias. For example, suppose Anheuser Busch is more likely to hire exclusive distributors 14
I define a product to be domestic mainstream lager if it is a lager product from one of the three biggest domestic competitors in the industry and has large package size options. 15 The 2006 and 2007 trade directories were provided by local distributors. 16 I normalized total units of sales per store. Therefore, the overall mean is zero and the standard deviation is one for “Store total sales”.
44
where the demand for beer is higher. If higher demand also leads to more entrants, our simple comparison will lead the exclusive effect to be biased upward (less foreclosure). Similarly, if a dealer is more likely to give up all the other profitable brands and to become exclusive to Anheuser Busch when the local rents are lower, and if lower local rents raise a firm’s entry probability, then the exclusive coefficient will also be biased upward. To deal with the potential problem from omitted variable bias, the empirical strategy to identify the exclusive effect in this paper is to control for firm specific post-entry sales by estimating the demand for beer, and to control for unobserved fixed costs at a location with the expected number of rivals.
2.3.2
Entry Variation
There are 32 specialty beer producers in the data, and 26 of them are California based firms. Figure 2.1 provides the locations of the California based specialty beer producers’ establishments (breweries or pubs). Most of these firms are located along Highway 101. Also, many firms cluster their breweries around the Bay Area and Sacramento. I tabulate their entry patterns in Table 2.3 to show variations in the data. From the table, we can see firms vary by the number of stores they enter. There are some firms that enter all 218 locations. However, there are also firms that are more local and have sales for fewer than 10 stores. Moreover, breweries from other states are more likely to be the well-established ones and on average enter more stores than local firms. As will become clearer in the empirical estimation, I use the distance from breweries to stores as exclusion restrictions to identify the strategic entry effect. Because the distances from a store to the main breweries of non-California firms are extremely large, I only consider the entry decisions of California specialty beer producers in the empirical section.
2.4 2.4.1
Model Demand Side
A consumer’s decision problem for purchasing a beer product is modeled using a discrete choice model. Each consumer is presented with different types of beer with various quality levels, and is assumed to decide whether or not to purchase one unit of
45
beer.17 Consumers are differentiated by the markets they live in, and their preferences for different types of beer. In particular, consumer c purchasing product j in market t will receive a mean utility term δjt and an idiosyncratic term νcjt . The utility from consuming beer is given by ucjt = δjt + νcjt . When νcjt is independently distributed with Type 1 extreme value distribution, the model is a logit model that offers easy tractability. Nevertheless, the logit specification is well-known to produce unrealistic substitution patterns that are driven by market shares instead of product similarities. To allow for more plausible substitution patterns, the independence assumption for νcjt can be relaxed by introducing random coefficients on product attributes. In particular, a nested logit model allows νcjt to include correlated shocks from different market segments, so a consumer who purchases a certain type of beer will be more likely to substitute for products with similar attributes. Similar to Chiou (2008), I use a four-level nested structure, which is shown in Figure 2.2. The nesting is consistent with the data that have a category variable (beer style), and the industry practice that uses products’ country of origin (domestic/imported) for market segmentation, I allow the correlated shocks for consumer c to come from three possible sources: a shock, ζb , for all beer products; a shock, ζd , for all products in domestic group d; and a shock, ζg , for all products within a specific domestic/foreign beer style g. Then, following Cardell (1997), the error structure of a nested logit model can be decomposed as νcjt = ζcb + λ3 ζcd + λ2 λ3 ζcg + λ1 λ2 λ3 ωcjt , where ω is independent and identically distributed with Type 1 extreme value, and the correlated shocks ζ are from a distribution such that if ω is distributed with Type 1 extreme value, ζ + λω will also be distributed with Type 1 extreme value when 0 < λ ≤ 1. The λ terms measure how products in the same category are independent of each other: when all λs approach to one, all shocks are independent and the model reduces 17
A unit is defined as a six-pack of beer (72 oz).
46
to a simple logit model. The nested logit model is basically a random coefficients model with coefficients on group dummy variables that allows correlation to be higher within a group and still preserves the closed form tractability similar to a logit model. The mean utility of product j in market t is specified to be: δjt = xj β − α pjt + γAjt + ξjt , where xj includes observed fixed product attributes such as alcohol by volume and calories, pjt is the price, and ξjt is the unobserved (by an econometrician) advantage for brand j in market t. To allow preference for local products, I include a dummy variable Ajt , which is equal to one if the product is locally brewed. In the estimation, I include brand, store, and quarter dummies to absorb invariant product and market attributes. Consumers are allowed to not purchase any product presented in the model. The mean utility from purchasing an outside good is normalized to zero. The mean utility can be recovered using Berry (1994)’s inversion technique: ln(sjt ) − ln(s0t ) = xj β − α pjt + γAjt + (1 − λ1 λ2 λ3 ) ln(sj|g ) + (1 − λ2 λ3 ) ln(sg|d ) (2.1) + (1 − λ3 ) ln(sd|b ) + ξjt , where sjt and s0t are market shares for product j and the outside good, sj|g is the market share for brand j as a fraction of the market share of group g, sg|d is the market share for group g as a fraction of the market share of the domestic/foreign product group, and sd|b is the market share for domestic/foreign product group as a fraction of the market share of all beer products. The market share of product j is the probability that product j is chosen by consumers. In a nested logit model, there is a closed form solution for the choice probability. First, we can decompose the choice probabilities (market shares) into conditional probabilities (omitting the t subscript): sj = sj|g × sg|d × sd|b × sb ,
(2.2)
where sb is the probability of choosing the inside good. Following McFadden (1978),
47
define the inclusive values for each nest as (∑ ) I1g ≡ log exp( (x β − α p + γA )/λ λ λ ) j jt jt 1 2 3 j∈Gg I2d ≡ log I3 ≡ log
(∑
) exp(λ I ) 1 1g g∈Dd
(∑ d∈B
) exp(λ2 I2d ) .
We can then express the conditional probabilities as exp( (xj β − α pjt + γAjt )/λ1 λ2 λ3 ) sj|g = ∑ exp( (xk β − α pkt + γAkt )/λ1 λ2 λ3 )
(2.3)
k∈Gg
sg|d
sd|b
( exp λ1 I1g ) ( = ∑ exp λ1 I1l ) l∈Dd ( exp λ2 I2d ) ( = ∑ exp λ2 I2m ) m∈B
sb =
exp(dq zq + ds zs + λ3 I3 ) , 1 + exp(dq zq + ds zs + λ3 I3 )
where zq and zs are dummy variables for quarters and stores, and dq and ds are the corresponding coefficients which do not vary for all inside goods. Gg , Dd , B are the sets of all products in domestic/foreign beer style group g, in domestic/foreign group d and in the inside good group B, respectively. Combining (2.2) and (2.3) gives the choice probability of product j as a function of model parameters.
2.4.2
Entry Game
Following the setup in Bajari, Hong, Krainer and Nekipelov (2010), I model a specialty beer producer’s entry decision using a static discrete choice model with private information. In each period, N potential entrants make entry decisions to different locations simultaneously.18 Let ai denote firm i’s entry decision, where ai is equal to 18
It is worth noting that in reality, an entry decision is never determined solely by the brewers, but is an outcome jointly determined by the actions of brewers, distributors and retailers. In fact, when a brewer is interested in a market, the brewer needs to contact distributors in that area to see whether a distribution contract can be reached. Even though a brewer assigns a distributor in an area, it does not guarantee its products’ penetration to all of the distributor’s retail accounts. A brewer still needs to constantly work with the distributor and different retailers to make sure the
48
one if it enters, and ai is equal to zero if it does not enter. Also, let a−i denote rivals’ action profile (a1 , ...ai−1 , ai+1 ..., aN ). Let s denote public state variables, commonly known by all firms and econometricians, and let ϵi denote firm i’s private information. Following the literature, I assume the private information shocks are independent and identically distributed from a known distribution. Firm i’s period utility function is a function of its own action, ai ; rival firms’ action profile, a−i ; state variables, s; its private information, ϵi ; and model parameters, θ. The period utility function can be expressed as: ui (ai , a−i , s; θ) = πi (ai , a−i , s; θ) + ϵi (ai ), where
V (a , s; θ) − F (a , s; θ) if a = 1 i −i i −i i πi (ai , a−i , s; θ) = 0 if ai = 0,
and Vi (a−i , s; θ) and Fi (a−i , s; θ) are firm i’s variable profits and fixed costs, respectively. In the specification, both variable profits and fixed costs depend on rivals’ actions.19 Since firm i does not observe its rivals’ private shocks, firm i’s decision rule is a function of state variables and its own private information. Therefore, the entry probability of firm i conditional on state variables is an integral of the decision rule over all possible values of ϵi weighted by the density of ϵi . Let σi (ai = 1|s) denote firm i’s entry probability conditional on state variables, ∏ and let σ−i (a−i |s) = j̸=i σj (aj |s) denote the probability of an action profile from rivals. Firm i chooses to enter a location if ∑
πi (ai = 1, a−i , s; θ)σ−i (a−i |s) + ϵi (ai = 1) > ϵi (ai = 0).
a−i
products are properly promoted. Because a full-fledged model of the search and negotiation process between manufacturers, distributors and retailers is out of the scope of this paper, I assume that manufacturers know the demand conditions and fixed costs involved entering a market prior to its entry decision, and make decisions accordingly. 19 I assume demand shocks are realized after a firm’s entry decisions. Also, in the current static setup, I cannot distinguish between sunk entry costs and fixed costs that are not sunk. The estimated fixed costs will be a combination of both types of costs.
49
If we assume the private information shocks are distributed with Type 1 extreme value distribution, the equilibrium entry probabilities from a Bayesian Nash Equilibrium are: ∑ exp( a−i πi (ai = 1, a−i , s; θ)σ−i (a−i |s)) ∑ σi (ai = 1|s) = , for i = 1, ..., N. (2.4) 1 + exp( a−i πi (ai = 1, a−i , s; θ)σ−i (a−i |s)) Equation (2.4) implies that given any set of parameters, the optimal entry probabilities need to satisfy a fixed point condition. In his work that studies a dynamic programming model, Rust (1987) shows one can estimate an equation similar to equation (2.4) with a nested fixed point algorithm (NFXP). Specifically, for any set of parameters, a NFXP algorithm searches for the implied fixed point (conditional choice probabilities) in (2.4) to construct the likelihood. Thus it requires solving a fixed point for a lot of values of parameters. Moreover, constructing the likelihood under the case of multiple equilibia requires specifying an equilibrium selection mechanism. To reduce the computation burden from a NFXP algorithm, Aguirregabiria and Mira (2002) propose a two-step approach. Instead of assuming the equilibrium is unique or employing a specific equilibrium selection mechanism, the two-step approach only requires that the equilibrium played in the data does not switch. In the first step, a researcher estimates the consistent choice probabilities conditional on observed states. Then, in the second step, the researcher recovers the parameters of the utility function conditional on these choice probabilities.20 I now turn to the specifications of the variable profits and the fixed costs in order to estimate the model. For each rival action profile, there is a corresponding variable profit Vi (a−i , s; θ) for firm i when it enters. I specify Vi (a−i , s; θ) as: Vi (a−i , s; θ) = mi Si (a−i , s; θ), where Si (a−i , s; θ) is the total units of sales of firm i (in terms of the number of sixpacks), and mi is firm i’s variable profits per unit of sales. Total units of sales of firm 20 The two-step approach led to many applications in dynamic games such as Bajari, Benkard and Levin (2007), Pakes, Ostrovsky and Berry (2007).
50
i are: Si (a−i , s; θ) =
∑
Ms sj (a−i , s; θ),
j∈Fi
where Fi is the set containing all products from firm i, sj is the market share for product j, and Ms is the potential market size of store s. Note that in this specification a firm’s variable profits do not vary according to the identity of rivals or locations. The underlying assumptions are that the number of specialty beer producers does not have immediate effects on the equilibrium prices upon entry and a firm’s variable profit per unit is a constant across stores. Given that specialty beer producers are fringe firms in each market, it is not unreasonable to assume that the presence of rivals in this segment does not have much effect on other firm’s pricing strategy and that these firms have relatively the same margins across stores. Similar to Bresnahan and Reiss (1991), I allow later entrants to have different fixed costs than incumbents and use the following specification for the fixed costs: ′
Fi (a−i , s; θ) = di + s η + τ
∑
1{aj = 1}.
(2.5)
j̸=i
As Bresnahan and Reiss (1991) discussed in their paper, τ is positive when later entrants are less efficient or face entry barriers, and they imposed τ to be positive in their estimation. However, since the industry context in this paper is very different from theirs, I do not restrict τ to be positive.21 The state variables s entering in equation (2.5) include factors that affect access to distributors and to shelf space after controlling for post-entry variable profits. In estimation, the above state variables include each firm’s distance from a store to its brewery, the size of a store’s selling area, and the presence of an Anheuser Busch exclusive distributor. Firm fixed effects are also included to allow for firm heterogeneity. Using the specifications for the variable profits and the fixed costs, equation (2.4) 21
Bresnahan and Reiss (1991) looked at the entry behaviors of doctors, dentists, druggists, plumbers, and tire dealers.
51
becomes: ( ∑ ) ∑ ′ exp mi Si (a−i , s; θ)σ−i (a−i |s) − (di + s η + τ σj (aj = 1|s) ) a j̸=i ( −i∑ ), ∑ σi (ai = 1|s) = ′ 1 + exp mi Si (a−i , s; θ)σ−i (a−i |s) − (di + s η + τ σj (aj = 1|s) ) a−i
j̸=i
for i = 1, ..., N.
(2.6)
For each location, the state variables s are the store’s potential market size, each product’s attributes, each firm’s distance from the store to its brewery, the size of the store’s physical selling area, and whether the Anheuser Busch distributor serving the store is exclusive. Parameters θ include the demand side parameters and the supply side parameters. Demand side parameters are marginal utility of income, α; marginal utility of fixed product attributes, βj ; marginal utility of locally brewed product, γ; and the inclusive values from the nested logit model, λ1 , λ2 , λ3 . Supply side parameters are each firm’s variable profit per unit (mi ), and the fixed cost parameters (di , η and τ ) are firm fixed effects, store size, distance from stores to breweries, the presence of an Anheuser Busch exclusive distributor, and the strategic effect on fixed costs. Ellickson and Misra (2008) and Bajari, Hong, Krainer and Nekipelov (2010) both estimate a static game using a two-step approach. They first estimate firms’ beliefs about the choice probabilities, and then maximize the pseudo likelihood function. Following the same approach, I first estimate the equilibrium entry probabilities σi (ai = 1|s). Moreover, I exploit the data on price and quantity to estimate the predicted total units of sales, Sˆi (a−i , s; θ). Finally, I plug the above estimates into 2.6 to maximize the pseudo likelihood. The estimation is done in three steps: 1. Estimate the consistent choice probabilities conditional on state variables. This produces σ ˆi (ai = 1|s) for i = 1, ..., N . 2. Estimate the demand system. Given the estimates of the demand system, predict units of sales conditional on each rivals’ action profile and construct ∑ˆ Si (a−i , s)ˆ σ−i (a−i |s). a−i
52
3. Plug the estimated
∑ˆ Si (a−i , s)ˆ σ−i (a−i |s) and σ ˆj (aj = 1|s) into equation (2.6) a−i
and estimate the supply parameters by maximizing the pseudo likelihood. The final estimating equations are: ( ∑ ) ∑ ′ exp mi Sˆi (a−i , s; θ)ˆ σ−i (a−i |s) − (di + s η + τ σ ˆj (aj = 1|s) ) σi (ai = 1|s) =
(
a−i
j̸=i
), ∑ ∑ 1 + exp mi Sˆi (a−i , s; θ)ˆ σ−i (a−i |s) − (di + s′ η + τ σ ˆj (aj = 1|s) ) a−i
for i = 1, ..., N.
j̸=i
(2.7)
I discuss the estimation procedure and the identification assumptions in detail in the below section.
2.5 2.5.1
Empirical Implementation Market Definition
An entry is made when a firm has access to a store’s shelf space. I study the entry decisions of specialty beer producers at the store-quarter level. Because the scanner data only include sales data that actually occur, the entry variable is defined to be one when a firm has positive sales in a store during a quarter. Since there are 218 stores and 8 quarters in the data, store s is from 1 to 218, quarter q is from 1 to 8, and market t is from 1 to 1744. I calculate the market share for brand j using the total quantities sold (adjusted for different package sizes) divided by the potential market size. Because I never observe the true size of a potential market, I need to construct a variable that measures it. One way to accomplish this is to define the outside good as beer consumption through the supermarket channel, and use the per capita consumption and the legal drinking age population in the zip code area to construct the market size (Hellerstein (2008)). However, directly applying this approach to the dataset in this paper produces unreasonable market shares for the inside good for some markets.22 Therefore, 22
Since beer consumption per capita per year in California is 25 gallons, and the volume of beer sold through the supermarket channel is around 16%, the per capita beer consumption through the supermarket channel per quarter is 25 ∗ 0.16/4 ≈ 1 gallon, which is around 1.78 units of six-packs per capita per quarter. The potential market size is thus the legal drinking age population in the zip code area multiplied by 1.78. However, there are 80 stores in which the sales of inside good
53
I instead define the potential market size of beer in a store to be the store’s total sales in the alcoholic beverage segment. In the industry, the combined retail dollar sales of wine and spirits are roughly the same as dollar sales of beer, so I multiply a store’s maximum total sales of beer by two to form the potential market size.
2.5.2
Pricing
I predict a firm’s post-entry prices using a reduced form approach.23 I regress a brand’s price on the distance between the brand’s brewery to a store, and the number of rivals at a store, controlling for brand, store and quarter fixed effects. Table 2.4 presents the results from regressions for brands from California specialty beer producers. Controlling for brand, store and quarter fixed effects, a brand has higher prices at stores that are farther away from the firm’s brewery. Moreover, prices are not responsive to strategic concerns. Neither the number of California firms nor the number of total firms in a market has a statistically significant (at 5% level) impact on a specialty beer product’s price. One potential explanation for these results is that many of the specialty beer producers are fringe firms that carry only an average of two products. I use the baseline results in column (1) of Table 2.4 to construct predicted prices. The predicted prices are then used to construct expected sales.
2.5.3
Estimation
Entry Probability
The equilibrium entry probability of firm i at store s can be estimated using its empirical counterpart. Given the panel structure of the data, a simple frequency ∑Q
estimator is σ ˆis (ais = 1) =
q=1
aiqs
Q
exceed the potential market size defined above. Therefore, I need to either adjust the per capita beer consumption or drop some stores to eliminate erroneous shares of the inside good. The above adjustments provide estimates that are very close to the ones presented in the empirical section and are thus not reported. 23 Equilibrium prices can also be recovered by modeling the vertical relationships between retailers, distributors, and manufacturers and solving the pricing game backward. However, because we do not observe the wholesale prices for the retailer chain or prices set by other grocery chains, and given that the retailer chain is clearly not a monopolist at the retail level, we are not able to use demand estimates to pin down retail margins without making further assumptions. Therefore, I do not pursue this approach and instead use a reduced form approach to construct predicted prices.
54
where aiqs is equal to one if firm i chooses to enter store s in quarter q. The expected number of rivals of firm i at store s is the sum of all firms’ entry probabilities subtracted by i’s own entry probability.24 Demand Side Parameters
Recall the estimating equation for the demand system in (2.1) is ln(sjt ) − ln(s0t ) = xj β − α pjt + γAjt + (1 − λ1 λ2 λ3 ) ln(sj|g ) + (1 − λ2 λ3 ) ln(sg|d ) + (1 − λ3 ) ln(sd|b ) + ξjt . There is a potential endogeneity problem in estimating the coefficients on pjt , ln(sj|g ), ln(sg|d ) and ln(sd|b ) because these variables may be correlated with unobserved product quality in a specific market, ξjt . For example, if a firm has higher unobserved quality level in market t and is able to price higher and earn a large market share, then the price and the λ coefficients will be biased. To deal with this endogeneity problem, note that we can decompose ξjt = ξj +ξt +∆ξjt , in which ξj is the unobserved brand fixed effect, ξt is the unobserved market fixed effect, and ∆ξjt is the unobserved deviation from brand and market fixed effects. Given this, I include brand, store and quarter dummies in the estimation to control for ξj and ξt . Because beer tastes better when it is fresh, and consumers usually value a product more if it is a local product, I construct a dummy variable “locally brewed” when the brewery is located within a 10 mile radius of a store. This will help to explain some of the variations in ∆ξjt . However, unobserved promotional activities at the brand level are still a concern. Therefore, I pursue an instrumental variable approach that requires finding instrumental variables that are correlated with the explanatory variables but are uncorrelated with ∆ξjt . To this end, I instrument for pjt and ln(sj|g ) using the number of brands a firm carries and the number of rival brands in group g. The identification assumptions are that the number of brands a firm carries and the number of rival brands in group g are determined prior to local promotional activities. 24 We can also estimate the entry probabilities by fitting a linear probability model. I run a regression of entry decisions on store, quarter, and firm fixed effects. I also include firm by store interactions in the regression. The predicted entry probabilities are basically the same as the ones calculated using the simple frequency estimator described above. However, there are predictions out of the [0, 1] interval, thus I do not use results from the linear probability model.
55
Similarly, for a domestic (foreign) product, I use the number of rival domestic (foreign) brands with different styles and the number of total foreign (domestic) brands in the market to instrument for ln(sg|d ) and ln(sd|b ). The estimation is done using the two stage least squares method with standard errors clustered at store by product style level. Using the demand estimates and the predicted prices, I can predict sales for each firm at each location under any rivals’ action profile using the nested logit probability formula. Ideally, we calculate the predicted demand conditional on each potential rivals’ action profile, and use the corresponding probabilities of rivals’ action profile to construct the expected sales. However, given that we have 26 players in the entry game, to calculate the exact expected sales for any firm in any market will need 225 permutations of rivals’ action profiles. To reduce the number of permutations, the expected sales are calculated two different ways: first by a “naive” approach, and second by simulation. The naive approach uses the exact identity of rival firms observed in the market to calculate the expected sales. The advantage of this approach is that it only takes one evaluation for each firm in each market. I also use simulation to approximate expected sales. Using the first stage entry probabilities, for each store, I form draws of rivals’ action profiles, calculate the sum of predicted sales from these draws, and divide the predicted sales by the number of draws. For each location, I use the base quarter (January to March) to form the predicted sales for the store. Identification of the Supply Side Parameters
As discussed in previous sections, the identification strategy of the effect of exclusive dealing on a firm’s entry decision is to control for post-entry sales and the number of expected rivals. ∑
I now discuss the assumptions used to identify the strategic effect. Let Πi (ai , s, θ) =
a−i
πi (ai , a−i , s; θ)σ−i (a−i |s) be the deterministic part of the expected profit. It has
been shown that we are not able to identify Πi (ai , s, θ) without imposing distribution assumptions on the stochastic shocks ϵ. Following the standard treatment for this problem, I assume ϵ is identically and independently distributed with a Type 1 ex-
56
treme value distribution. Furthermore, we are only able to identify the deterministic part of the expected profit up to the difference between Πi (a = 1, s, θ) − Πi (a = 0, s, θ), so we normalize Πi (a = 0, s, θ) to zero in order to identify Πi (a = 1, s, θ). Therefore, I assume that a firm’s payoff from staying out of the market is zero in the model. This assumption is similar to the outside good assumption made in standard discrete choice models. In addition, there can be multiple equilibria in the game. Following the literature on two-step estimation methods, the estimates are obtained under the assumption that in each location, firms play the same strategy and do not switch to others. The estimating equations are a system of simultaneous equations. Bajari, Hong, Krainer and Nekipelov (2010) discuss that one needs exclusion restrictions to nonparametrically identify a static game. Even though nonparametric identification is not the main purpose of this paper, having variables that satisfy exclusion restriction conditions better explains where the identification comes from. Basically, the exclusion restriction condition says we need to have state variables that affect rival firms’ entry probabilities but do not enter into a firm’s payoff directly. In this paper, I use the distances between rival breweries to a store to satisfy the exclusion restriction condition. If a firm is more likely to enter stores closer to its breweries, which is the case in the beer industry, and the locations of rival firms’ breweries do not enter into the firm’s profit function directly, then the distances of rival firms’ breweries to a store can serve as exclusion restrictions. One concern of using distances to satisfy the exclusion restriction is that the proximity to rivals’ brewery may affect a firm’s profitability through demand. If consumers prefer to purchase locally brewed products, a firm’s market share can be lower when it enters a store closer to a rival’s brewery. To deal with this problem, I include a “locally brewed” variable in the demand estimation to capture these effects, and the maintained assumption will be that after controlling for demand, the distance variables of rival firms do not enter into a firm’s profit function. Another concern is that a firm’s choice of brewery location may not be exogenous. As previously shown in Figure 2.1, there are several breweries clustered around
57
the Bay Area and Sacramento. If firms chose their breweries’ locations based on unobserved common factors in these areas, then the estimated strategic effect will be biased upward. I address this concern by doing robustness checks. I show that dropping all the stores in Bay Area Counties and Sacramento County does not affect the main results. Standard Errors
Since the estimation is done in two steps, I bootstrapped the standard errors across stores by resampling. The assumption is that after controlling for brand, quarter and store dummies, the error terms across stores are independent. Moreover, given that I estimate a static model with eight quarters of data, I cluster the standard errors at the store level to control for the likely positive correlation over time.
2.6
Results
I first estimate an entry model without controlling for demand or strategic effects. The estimation results are shown in Table 2.5. Column (1) provides the baseline regression results. Column (2) further controls for gross rent, and columns (3) and (4) provide results with squared distances as robustness checks. All coefficients on store size, distance and gross rent have the expected signs: a location that has a larger selling area, is closer to a firm’s brewery, or has lower gross rent is associated with higher entry probability. All coefficients on the Anheuser Busch exclusive dealer dummy are positive, though only those coefficients in (3) and (4) are statistically significant. As discussed in the previous section, the coefficient of the exclusive effect is likely to be biased upward. In fact, in column (2), we can see the coefficient of the exclusive effect goes down once we control for gross rent. To explore more rigorously the effect of exclusive dealing on a firm’s entry decision, I proceed by estimating the demand for beer and the entry game with strategic interactions.
2.6.1
Demand Estimates
The demand system is estimated to achieve two goals. First, it provides predictions for counter-factual post-entry sales, and second, it enables welfare analysis. Table 2.6 provides demand estimation results using a logit model. Column (1)
58
gives the OLS regression results, and column (2) provides the results from the instrumented regression. All specifications include brand, store, and quarter fixed effects. As can be seen in column (2), the magnitude of the negative coefficient on price increases when prices are instrumented for. Also, in both specifications, a consumer enjoys a product more if it is locally brewed. The implied mean price elasticity is -9.88, suggesting that the demand is quite price elastic.25 Table 2.7 presents the demand estimation results using a nested logit model, with prices instrumented in column (2). Columns (3) to (6) show the first stage results, where the various instruments were presented in section 2.5.3. In column (3), the instrument for ln sj|g , the number of rival brands in group g, is negatively correlated to ln sj|g and is significant. The results in columns (4) and (5) are similar to those in column (3) and have the expected signs for the instruments: as the number of the brands in other rival groups increases, the market share of a product’s own group will decrease. In column (6), the number of brands a firm carries in group g is associated with lower prices for its brands in group g. This association is probably due to the fact that big domestic firms are more likely to enjoy economies of scale and are more capable to provide products at very low prices. The first stage F-statistics, with standard errors clustered at store level, are 118.16, 43.52, 27.67, and 119.89, suggesting that the instruments are not weak. A comparison of the first two columns in Table 2.7 shows that demand becomes more elastic once we instrument for prices. As in the logit model, consumers also prefer to have local products in both specifications. The mean own price elasticity of the nested logit model is -8.41. Table 2.8 provides the percentiles of price elasticities based on the logit and the nested logit models using the estimates in columns (2) in Table 2.6 and Table 2.7. The nested logit model provides less elastic demand estimates across percentiles. The coefficients on ln(sj|g ), ln(sg|d ), and ln(sd|b ) represent the similarities within 25
Hellerstein (2008) estimated a model of beer demand using a random-coefficients model with data from 1991 to 1994. The estimated demand for beer is also very elastic. The own elasticities of her selected beer products are between -5.71 to -6.37 (the data used in (Hellerstein, 2008) are monthly data and contain no specialty beer products).
59
nests. As McFadden (1981) shows, a nested logit model is consistent with a utility maximization model for any values if all of the inclusive value coefficients are within the [0, 1] interval. A negative estimate of inclusive value indicates a violation of utility maximization. The implied inclusive value coefficients (λ1 , λ2 , and λ3 ) in the nested logit model are 1.51, 0.58, and 0.30, respectively. These results suggest that the model is consistent with some values of explanatory variables, but is not consistent with all values of explanatory variables.26 Also, λ1 is greater than one means consumers substitute more often across different beer styles than within the same style. Still, given that the similarity coefficients are not jointly zero, I can reject a simple logit model, so I will use the nested logit model to be my preferred setting to predict post-entry sales. With the demand estimates, Figure 2.3 provides a scatter plot of predicted sales and actual sales for California specialty beer producers in markets that they actually entered. The graph shows that the predicted sales are highly correlated with the observed ones. Table 2.9 presents the results that control for demand but not strategic entry decisions. The first two columns restrict the variable profits per unit of sales across firms to be the same. Column (3) allows variable profits per unit of sales to differ across firms, and columns (4), (5) and (6) add the squared distance as robustness checks. The coefficients on expected sales have the expected positive signs and are significant. Compared to the results in Table 2.5, the exclusive coefficients are smaller and are no longer statistically significant in all specifications.
2.6.2
Strategic Entry
To allow for strategic interactions between firms, I estimate the fixed costs using equation (2.7), along with the demand estimates and the beliefs about entry probabilities constructed before. Tables 2.10 and 2.11 provide the results from the naive approach and from simulation (with 50 draws), respectively. The two methods provide similar estimation results. The coefficients for store size, distance, and expected sales are all 26
Nevertheless, I cannot reject the hypothesis that λ1 is equal to one at the 5% level. I also estimate equation (2.1) with the constraint that λ1 is equal to one, and the results are similar to the ones reported above.
60
statistically significant, and have the expected signs. Compared to the results in Table 2.9, the magnitude of the coefficients for store size and gross rent are smaller in Tables 2.10 and 2.11. The coefficients for gross rent are now positive, although the estimated coefficients are not statistically significant. When the expected number of rivals is controlled for, the coefficients of exclusive dealing become negative and statistically significant in all specifications, suggesting that there is a foreclosure effect due to Anheuser Busch’s exclusive dealing program. To interpret the magnitude of the foreclosure effect, I use the estimates from column (1) in Table 2.10 to calculate the marginal effects. I find that a store with an Anheuser Busch exclusive distributor reduces a firm’s entry probability by approximately 3.5 percentage points. The effect does not appear large at first glance; however, given that the mean entry probability of the 23 specialty beer producers in California is only 28 percentage points (excluding the three California firms that enter all 218 stores), the exclusive effect plays an important role in a firm’s entry decisions. The results also show that a decrease in sales of 100 six-packs a quarter reduces a firm’s entry probability by 3 percentage points, a reduction in 10,000 square feet of store size reduces a firm’s entry probability by 1.2 percentage points, and a 100 mile increase in the distance between a firm’s brewery and a store reduces the firm’s entry probability by 14 percentage points. The coefficients for the expected number of rivals are also positive and significant: the presence of an additional specialty beer producer raises a firm’s entry probability by 1.4 percentage points. As discussed in the previous section, the identification strategy for the strategic effect relies on using the distances from a store to rival firms’ breweries as exclusion restrictions, and the fact that many breweries are located around the Bay Area and the Sacramento area raises a concern about whether these exclusion restrictions are valid. To ease this concern, I provide results that drop all the stores in the Bay Area and Sacramento County in Table 2.12.27 The strategic coefficients remain positive and significant in all specifications, and the marginal 27
The Bay Area counties are: Alameda County, Contra Costa County, Marin County, Napa County, San Francisco County, San Mateo County, Santa Clara County, Solano County and Sonoma County.
61
effect from having an additional rival in the specialty beer segment raises a firm’s entry probability by 3 percentage points. The above results suggest spillover effects for specialty beer producers at the cost side. One potential reason for this spillover effect is that it takes many brands for a store to create a specialty beer category. When a store prefers to build a specialty beer segment to attract certain types of consumers (and is currently carrying many specialty brands), it is easier for a firm or a distributor to persuade the store to add additional specialty brands, compared to persuading another store that has no interest in building a specialty beer category (and is currently carrying very few specialty brands) to have more brands. Finally, I conduct counter-factual experiments to study the effect of banning exclusive dealing. I simulate demand using the predicted entry probabilities with and without exclusive dealing. Following Small and Rosen (1981), the changes in consumer welfare can be found by [ (∑ ) ( ∑ )] 1 δj ln ∆EVt = e − ln eδ j , α 1 0 j∈J
j∈J
where J 0 and J 1 are choice sets before and after removing exclusive dealing.28 Table 2.13 provides the results of counter-factual experiments based on 100 draws from the predicted entry probabilities. Removing exclusive dealing does not have much impact on entry behavior: at most only one additional firm will enter a market, and the impact on consumer welfare is almost zero. In fact, given the demand structure, even if we allow all the California specialty beer producers to enter a store with an Anheuser Busch exclusive distributor, the implied increase in consumer welfare is at most 4 cents per market (a store in a quarter). There are several reasons for getting such small estimates. First, as shown in Table 2.4, adding specialty brands does not have much impact on pricing. Second, most of the specialty brands that do not have a presence in many markets have very little market shares even in stores where they have a presence. Therefore, even though exclusive dealing increases the fixed costs 28
We can find
∑ j∈J
( ) eδj by using the 1 + exp(dq zq + dt zt + λ3 I3 ) term in equation (2.3).
62
for some specialty beer producers, the welfare improvement from banning exclusive dealing is very limited.
2.7
Conclusion
This paper estimates the entry decisions of California specialty beer producers to retail stores to test the foreclosure effect of exclusive dealing. I find that a specialty beer producer’s entry decisions have spillover effects on rival firms’ entry decisions. I also find that exclusive dealing between Anheuser Busch and its distributor is associated with a 3.5 percentage point (12.5%) reduction in a specialty beer producer’s entry probability, controlling for the expected post-entry sales, the location’s proximity to the firm’s brewery, the size of a store’s selling area, and the expected number of rivals. Given the static setting in this paper, it is difficult to disentangle whether the adverse effect of exclusive dealing comes from higher sunk costs or higher fixed costs in every period. One potential extension of this study is to consider a firm’s entry decision in a dynamic setting. Finally, even though exclusive dealing raises a specialty beer producer’s fixed costs, when I implement counter-factual experiments to study the effect of banning exclusive dealing, the results show very limited improvement in consumer welfare.
63
Figure 2.1: Locations of Breweries for the California Specialty Beer Producers Observed in the Scanner Data
64
Figure 2.2: Tree Diagram for the Nested Logit Model
0
500
Predicted Sales 1000 1500
2000
65
0
500
1000 Observed Sales Sales
1500
2000
45 degree line
Figure 2.3: Predicted Demand and Actual Demand: A 5% Random Sample of Data
66
Table 2.1: Some Typical Entries From the CBBD Annual Directory Distribution Status
Brand Portfolio
Exclusive AB distributor
180 Energy Drink, Anheuser-Busch, Redhook Ale Brewery, Rolling Rock, Widmer Brothers Brewing
Nonexclusive AB distributor
Anheuser Busch, Arizona Beverage, Gambrinus, Gordon Biersch Brewing, Heineken USA, InBev USA, Redhook Ale Brewery, Scottish & Newcastle Importers, Sierra Nevada Brewing, Spaten North America, Widmer Brothers Brewing
Non-AB distributor 1
Anchor Brewing, Asahi Breweries, Boston Beer, Diageo-Guinness USA, Gambrinus, Heineken, Hornell Brewing, InBev USA, Mark Anthony Brands, McKenzie River, Miller Brewing, Pabst Brewing, Pyramid Breweries, Sapporo USA, Scottish & Newcastle Importers, Sierra Nevada Brewing, Sierra Springs Water, US Beverage
Non-AB distributor 2
Alaskan Brewing, Barton Beers, Boston Beer, Diageo-Guinness USA, InBev USA, Lake Tahoe Brewing, Mark Anthony Brands, Miller Brewing, Molson Coors Brewing, New Belgium Brewing, Pyramid Breweries, Redhook Ale Brewery, Sapporo USA, Sierra Nevada Brewing
Independent distributor
Alaskan Brewing, Allagash Brewing, Anderson Valley Brewing, Arizona Beverage, Asahi Breweries USA, Bear Republic, Binding International, Bison Brewing, California Cider, Constancia Brewery, Firestone-Walker Brewing, Friedlin Imports Full Sail Brewing Co., Humboldt Brewing, InBev USA, Mad River Brewing, Mendocino Brewing, Marin Brewing, Moylan’s Brewery, Nestle Beverage, Ommegang Brewery, Pabst Brewing, Panorama Brewing, Sapporo USA, Scottish & Newcastle Importers, Sierra Nevada Brewing, Spaten North America, Speakeasy, Stone Brewing, Sudwerk Privatbrauerei, Thames America Trading, U.S. Beverage, Wyder’s Beverage
Notes: AB (Anheuser Busch). 180 Energy Drink, Redhook, Rolling Rock, and Widmer Brothers are all affiliated Anheuser Busch products listed on the Anheuser Busch company website. California specialty beer producers that can be matched to the scanner data set are denoted by bold type.
67
Table 2.2: Summary Statistics Anheuser Busch Non-exclusive Stores Number of stores
Anheuser Busch Exclusive Stores
179
39
Number of CA specialty beer producers
14.73 (2.43)
15.87 (1.58)
Store total sales
-0.095 (0.981)
0.434 (0.972)
Store selling area
27926 (9775)
30370 (8806)
Population
33305 (16840)
26399 (13094)
Household income
65671 (22325)
48491 (18358)
997 (278)
795 (257)
Gross rent
Notes: All entries reported are means with standard deviations shown in parentheses. “Store total sales” are normalized to have mean equal to zero. “Store selling area” is measured in square feet.
Table 2.3: Number of Stores a Firm Entered Firm Non-CA Firm Non-CA Firm Non-CA Firm Non-CA Firm Non-CA Firm Non-CA Firm CA Firm 1 CA Firm 2 CA Firm 3 CA Firm 4 CA Firm 5 CA Firm 6 CA Firm 7 CA Firm 8 CA Firm 9 CA Firm 10 CA Firm 11 CA Firm 12 CA Firm 13 CA Firm 14 CA Firm 15 CA Firm 16 CA Firm 17 CA Firm 18 CA Firm 19 CA Firm 20 CA Firm 21 CA Firm 22 CA Firm 23 CA Firm 24 CA Firm 25 CA Firm 26 Total
1 2 3 4 5 6
2006q2
2006q3
2006q4
2007q1
2007q2
194 218 181 184 205 218 218 166 122 6 2 3 4 2 11 184 218 3 210 3 175 9 1 198 3 44 1 216 5 218 9 7 101
194 218 186 184 206 218 218 166 110 6 91 3 4 2 3 187 218 3 211 3 171 8 1 201 2 39 1 214 6 218 9 7 103
193 218 185 182 204 218 218 162 107 6 167 3 4 2 1 189 218 3 211 3 168 8 1 205 2 33 1 214 7 218 9 7 105
194 218 190 184 203 218 218 159 105 7 19 3 4 2 0 191 218 3 212 3 168 8 1 209 2 35 1 214 6 218 9 6 101
193 218 190 185 199 218 218 157 104 7 5 3 4 2 1 189 218 3 212 3 166 9 1 208 2 36 1 214 7 218 8 6 100
2007q3 199 218 201 191 206 218 218 169 143 7 3 3 4 2 0 196 218 4 218 3 172 8 1 210 2 30 0 218 16 218 8 6 103
2007q4
2008q1
mean
201 218 199 188 145 218 218 159 145 5 1 2 4 2 0 191 218 5 218 3 192 8 1 204 0 15 0 218 6 218 7 1 100
200 218 195 188 135 218 218 154 142 8 0 2 4 2 0 189 218 4 218 2 195 8 1 204 0 11 0 218 2 218 7 1 99
196 218 191 186 188 218 218 162 122 7 36 3 4 2 2 190 218 4 214 3 176 8 1 205 2 30 1 216 7 218 8 5 102
Notes: All entries reported are the counts of “enter” for a firm across stores during a quarter. The variable “enter” is equal to one if a firm has positive sales in a store during a quarter.
68
Table 2.4: Pricing for California Specialty Beer Products
Distance
(1)
(2)
(3)
(4)
(5)
(6)
0.082∗∗
0.082∗∗
0.082∗∗
0.305∗∗
0.305∗∗
0.305∗∗
(0.035)
(0.035)
(0.035)
(0.105)
(0.105)
(0.105)
Number of CA micro firms
-0.005
-0.005
(0.004)
(0.004)
Number of total firms
-0.003+
-0.003+
(0.002)
(0.002)
Distance Squared
Constant
-0.102∗∗
-0.102∗∗
-0.102∗∗
(0.034)
(0.034)
(0.034)
7.886∗∗
7.919∗∗
7.979∗∗
7.816∗∗
7.848∗∗
7.910∗∗
(0.065)
(0.074)
(0.087)
(0.048)
(0.058)
(0.074)
Observations
39176
39176
39176
39176
39176
39176
R2
0.880
0.880
0.880
0.881
0.881
0.881
Adjusted
Notes: “Distance” is measured in hundreds of miles. All prices are volume adjusted six-pack prices. Data are collapsed to quarter/store/brand level. All regressions control for brand, quarter, and store fixed effects. Standard errors, clustered at store level, are shown in parentheses. +
significant at 10%,
∗
significant at 5%,
∗∗
significant at 1%.
69
Table 2.5: Entry Probability without Controlling for Demand (1)
(2)
(3)
(4)
AB exclusive distributor
0.256 (0.164)
0.180 (0.183)
0.387∗ (0.168)
0.303+ (0.182)
Store size
0.624∗∗ (0.081)
0.616∗∗ (0.081)
0.637∗∗ (0.081)
0.629∗∗ (0.081)
Distance
-1.836∗∗ (0.163)
-1.852∗∗ (0.161)
-3.417∗∗ (0.335)
-3.463∗∗ (0.332)
Gross rent
-0.040 (0.029)
Distance2
-0.046 (0.029) 0.645∗∗ (0.118)
0.659∗∗ (0.117)
Constant
-4.657∗∗ (0.509)
-4.246∗∗ (0.607)
-4.156∗∗ (0.533)
-3.697∗∗ (0.627)
Observations Log likelihood
40112 -7018.445
40112 -7007.021
40112 -6939.857
40112 -6925.032
Notes: Data are collapsed to quarter/store/firm level. All regressions control for firm fixed effects. Standard errors, clustered at store level, are shown in parentheses. AB (Anheuser Busch). “Distance” is measured in hundreds of miles. “Store size” is measured in ten thousands of square feet. “Gross rent” is measured in hundreds of dollars. +
significant at 10%,
∗
significant at 5%,
∗∗
significant at 1%.
Table 2.6: Demand Estimation Results from the Logit Model (1) OLS
(2) 2SLS
Locally brewed
0.647∗∗ (0.058)
0.531∗∗ (0.052)
Price
-0.316∗∗ (0.010)
-1.496∗∗ (0.174)
Constant
-4.218∗∗ (0.202) 195688 0.806
6.757∗∗ (1.620) 195688 0.695
Observations Adjusted R2
Notes: Data are collapsed to quarter/store/brand level. All regressions control for brand, quarter, and store fixed effects. Standard errors, clustered at store level, are shown in parentheses. “Locally brewed” is equal to one when a store is located within a 10 mile radius of a firm’s brewery. +
significant at 10%,
∗
significant at 5%,
∗∗
significant at 1%.
70
Table 2.7: Demand Estimation Results from the Nested Logit Model First stage results OLS
2SLS
ln(sj|g )
ln(sg|d )
ln(sd|b )
Price
(1)
(2)
(3)
(4)
(5)
(6)
0.002∗
0.129
0.515∗∗
0.187∗∗
-0.038∗∗
-0.090
(0.001)
(0.085)
(0.062)
(0.024)
(0.009)
(0.056)
Price
0.000 (0.001)
-0.320 (0.241)
ln(sj|g )
0.998∗∗ (0.000)
0.735∗∗ (0.174)
ln(sg|d )
0.996∗∗ (0.001)
0.825∗∗ (0.124)
ln(sd|b )
1.000∗∗ (0.001)
0.698∗∗ (0.230)
Price
0.035∗∗ (0.008)
0.025∗∗ (0.007)
0.007∗∗ (0.002)
-0.049∗∗ (0.004)
ln(sj|g )
-0.022∗∗ (0.002)
0.023∗∗ (0.004)
0.004∗∗ (0.001)
-0.002∗∗ (0.001)
ln(sg|d )
-0.002∗∗ (0.001)
-0.004∗∗ (0.001)
0.003∗∗ (0.001)
-0.002∗∗ (0.000)
ln(sd|b )
-0.003∗∗ (0.001)
-0.002∗∗ (0.001)
-0.005∗∗ (0.001)
0.005∗∗ (0.001)
Locally brewed
Instrumental variables for:
Constant
Observations
-0.782∗∗ (0.015)
0.833 (1.330)
-2.295∗∗ (0.199)
-3.438∗∗ (0.031)
-0.382∗∗ (0.024)
9.388∗∗ (0.121)
195688
195688
195688
195688
195688
195688
118.16
43.52
27.67
119.89
0.853
0.940
0.914
0.936
First stage F-stat Adjusted
R2
0.996
0.979
Notes: Data are collapsed to quarter/store/brand level. All regressions control for brand, quarter, and store fixed effects. Standard errors, clustered at store level, are shown in parentheses. “Locally brewed” is equal to one when a store is located within a 10 mile radius of a firm’s brewery. +
significant at 10%,
∗
significant at 5%,
∗∗
significant at 1%.
Table 2.8: Price Elasticity Percentiles 10%
25%
median
75%
90%
Logit
-12.32
-11.34
-10.36
-8.70
-5.85
Nested Logit
-10.93
-9.64
-8.59
-7.33
-4.99
Notes: Price elasticities are calculated using columns (2) in Table 2.6 and Table 2.7.
71
Table 2.9: Entry Probability with Expected Demand: No Strategic Interactions (1)
(2)
(3)
(4)
(5)
(6)
AB exclusive distributor
0.093 (0.163)
0.036 (0.185)
0.001 (0.185)
0.231 (0.165)
0.166 (0.182)
0.116 (0.181)
Store size
0.524∗∗ (0.075)
0.520∗∗ (0.076)
0.450∗∗ (0.080)
0.536∗∗ (0.075)
0.532∗∗ (0.075)
0.460∗∗ (0.079)
Distance
-1.731∗∗ (0.163)
-1.745∗∗ (0.161)
-1.700∗∗ (0.161)
-3.340∗∗ (0.333)
-3.372∗∗ (0.330)
-3.014∗∗ (0.345)
Expected sales
0.007∗∗ (0.001)
0.007∗∗ (0.001)
0.007∗∗ (0.001)
0.007∗∗ (0.001)
Gross rent
-0.031 (0.029)
-0.019 (0.030)
Distance2
Constant
Margins vary by firm? Observations Log likelihood
-0.036 (0.029)
-0.024 (0.030)
0.655∗∗ (0.113)
0.663∗∗ (0.111)
0.534∗∗ (0.126)
-4.537∗∗ (0.492)
-4.223∗∗ (0.598)
-6.656∗∗ (0.787)
-4.019∗∗ (0.515)
-3.663∗∗ (0.614)
-6.161∗∗ (0.821)
No
No
Yes
No
No
Yes
40112 -6892.916
40112 -6886.330
40112 -6531.007
40112 -6814.020
40112 -6805.082
40112 -6482.735
Notes: Data are collapsed to quarter/store/firm level. All regressions control for firm fixed effects. Standard errors, clustered at store level, are shown in parentheses. AB (Anheuser Busch). “Distance” is measured in hundreds of miles. “Store size” is measured in ten thousands of square feet. “Gross rent” is measured in hundreds of dollars. +
significant at 10%,
∗
significant at 5%,
∗∗
significant at 1%.
72
Table 2.10: Strategic Entry: the Naive Approach (1)
(2)
(3)
(4)
(5)
(6)
-0.527∗∗
-0.518∗∗
-0.546∗∗
-0.359∗∗
-0.360∗∗
(0.108)
(0.117)
(0.123)
(0.107)
(0.113)
-0.398∗∗ (0.116)
Store size
0.156∗∗ (0.044)
0.156∗∗ (0.044)
0.096∗ (0.048)
0.167∗∗ (0.042)
0.167∗∗ (0.042)
0.102∗ (0.046)
Distance
-1.872∗∗ (0.148)
-1.869∗∗ (0.148)
-1.768∗∗ (0.143)
-3.602∗∗ (0.326)
-3.602∗∗ (0.326)
-3.290∗∗ (0.354)
Expected sales
0.004∗∗ (0.001)
0.004∗∗ (0.001)
0.004∗∗ (0.001)
0.004∗∗ (0.001)
Expected number of rivals
0.187∗∗ (0.010)
0.187∗∗ (0.010)
0.187∗∗ (0.010)
0.188∗∗ (0.009)
0.188∗∗ (0.010)
0.188∗∗ (0.010)
0.005 (0.014)
0.017 (0.015)
-0.000 (0.013)
0.010 (0.014)
0.707∗∗ (0.093)
0.707∗∗ (0.093)
0.622∗∗ (0.107)
AB exclusive distributor
Gross rent
Distance2
Constant
-10.135∗∗ (0.588)
-10.196∗∗ (0.611)
-12.974∗∗ (0.912)
-9.597∗∗ (0.581)
-9.593∗∗ (0.602)
-12.393∗∗ (0.910)
No
No
Yes
No
No
Yes
40112 -5838.991
40112 -5838.841
40112 -5536.737
40112 -5753.890
40112 -5753.889
40112 -5482.001
Margins vary by firm? Observations Log likelihood
Notes: Data are collapsed to quarter/store/firm level. All regressions control for firm fixed effects. Standard errors, clustered at store level, are shown in parentheses. AB (Anheuser Busch). “Distance” is measured in hundreds of miles. “Store size” is measured in ten thousands of square feet. “Gross rent” is measured in hundreds of dollars. +
significant at 10%,
∗
significant at 5%,
∗∗
significant at 1%.
73
Table 2.11: Strategic Entry with Simulated Expected Demand (1)
(2)
(3)
(4)
(5)
(6)
-0.521∗∗
-0.512∗∗
-0.513∗∗
-0.352∗∗
-0.353∗∗
(0.108)
(0.117)
(0.120)
(0.107)
(0.113)
-0.367∗∗ (0.113)
Store size
0.156∗∗ (0.044)
0.157∗∗ (0.044)
0.095∗ (0.047)
0.168∗∗ (0.042)
0.168∗∗ (0.042)
0.102∗ (0.045)
Distance
-1.868∗∗ (0.148)
-1.866∗∗ (0.148)
-1.737∗∗ (0.141)
-3.596∗∗ (0.326)
-3.596∗∗ (0.326)
-3.229∗∗ (0.356)
Expected sales
0.004∗∗ (0.001)
0.004∗∗ (0.001)
0.004∗∗ (0.001)
0.004∗∗ (0.001)
Expected number of rivals
0.187∗∗ (0.010)
0.187∗∗ (0.010)
0.186∗∗ (0.010)
0.188∗∗ (0.010)
0.188∗∗ (0.010)
0.187∗∗ (0.010)
0.005 (0.014)
0.016 (0.015)
-0.001 (0.013)
0.009 (0.014)
0.705∗∗ (0.093)
0.705∗∗ (0.093)
0.609∗∗ (0.109)
AB exclusive distributor
Gross rent
Distance2
Constant
Margins vary by firm? Observations Log likelihood
-10.119∗∗ (0.589)
-10.176∗∗ (0.611)
-13.332∗∗ (0.953)
-9.586∗∗ (0.582)
-9.578∗∗ (0.603)
-12.761∗∗ (0.952)
No
No
Yes
No
No
Yes
40112 -5840.314
40112 -5840.183
40112 -5533.519
40112 -5755.739
40112 -5755.736
40112 -5481.699
Notes: Expected demand are calculated by taking 50 draws of rivals’ action profile in a given market. Data are collapsed to quarter/store/firm level. All regressions control for firm fixed effects. Standard errors, clustered at store level, are shown in parentheses. AB (Anheuser Busch). “Distance” is measured in hundreds of miles. “Store size” is measured in ten thousands of square feet. “Gross rent” is measured in hundreds of dollars. +
significant at 10%,
∗
significant at 5%,
∗∗
significant at 1%.
74
Table 2.12: Strategic Entry: Regression without Bay Area Counties and the Sacramento County (1)
(2)
(3)
(4)
(5)
(6)
-0.890∗∗
-0.851∗∗
-0.895∗∗
-0.713∗∗
-0.654∗∗
(0.159)
(0.189)
(0.210)
(0.144)
(0.169)
-0.679∗∗ (0.179)
Store size
0.084 (0.067)
0.082 (0.067)
0.061 (0.071)
0.089 (0.064)
0.085 (0.064)
0.050 (0.071)
Distance
-1.882∗∗ (0.218)
-1.879∗∗ (0.218)
-1.733∗∗ (0.208)
-3.836∗∗ (0.490)
-3.843∗∗ (0.491)
-3.561∗∗ (0.588)
Expected sales
0.004∗ (0.001)
0.004∗ (0.001)
0.004∗∗ (0.001)
0.004∗∗ (0.001)
Expected number of rivals
0.162∗∗ (0.013)
0.163∗∗ (0.014)
0.167∗∗ (0.016)
0.165∗∗ (0.013)
0.166∗∗ (0.013)
0.168∗∗ (0.016)
0.022 (0.041)
0.020 (0.040)
0.032 (0.039)
0.026 (0.035)
0.713∗∗ (0.138)
0.716∗∗ (0.138)
0.668∗∗ (0.180)
AB exclusive distributor
Gross rent
Distance2
Margins vary by firm?
-7.920∗∗ (0.829) No
-8.124∗∗ (0.941) No
-11.269∗∗ (1.209) Yes
-7.345∗∗ (0.779) No
-7.634∗∗ (0.887) No
-10.783∗∗ (1.191) Yes
Observations Log likelihood
8352 -1729.786
8352 -1729.488
8352 -1591.893
8352 -1688.995
8352 -1688.377
8352 -1565.733
Constant
Notes: Data are collapsed to quarter/store/firm level. All regressions control for firm fixed effects. Standard errors, clustered at store level, are shown in parentheses. AB (Anheuser Busch). “Distance” is measured in hundreds of miles. “Store size” is measured in ten thousands of square feet. “Gross rent” is measured in hundreds of dollars. The Bay Area counties are: Alameda County, Contra Costa County, Marin County, Napa County, San Francisco County, San Mateo County, Santa Clara County, Solano County and Sonoma County. +
significant at 10%,
∗
significant at 5%,
∗∗
significant at 1%.
Table 2.13: The Effects of Banning Exclusive Dealing Variable
Mean
Min
Max
With exclusive dealing
10.12
5.39
15.36
Without exclusive dealing
10.64
5.99
16.89
Remove exclusive dealing
0.0006
-0.0001
0.0016
Include all CA specialty brands
0.0165
0.0071
0.0398
Predicted number of CA specialty beer producers
Changes in consumer welfare
Notes: Based on estimation results of column (2) in Table 2.10. Welfare changes are evaluated at locations with Anheuser Busch exclusive distributors.
75
Chapter 3 Vertical Integration and Retail Competition under Capacity Constraints: The Case of California Gasoline Market Chia-Wen Chen and Christopher R. Knittel
3.1
Introduction
In a conventional Cournot framework as the number of firms decreases, equilibrium price increases. In this paper, we show that adding capacity constraints and vertical structure (realistic features of many markets) can reverse this result. We develop a theoretical model to study how concentration in the downstream market affects equilibrium outcomes in a vertical industry with capacity constraints. We test our theoretical predictions on the gasoline industry. This setting offers a starting point for analysis of the effects of potential policy interventions that affect the downstream market structure in the gasoline industry. We find that when independent upstream refineries capacity constraintare binding, the effect of a decrease in the number of independent retailers on retail gasoline price is small. We first show the optimality and the equilibrium conditions of the theoretical model under both vertical separation and vertical integration. Then, we provide a
76
numerical example to illustrate the properties of the equilibrium wholesale and retail prices when the downstream market becomes more or less competitive. Finally, we take our theoretical model to data. Using our model of oligopolistic competition with capacity constraints, along with demand and refinery capacity data in California, we calibrate equilibrium wholesale and retail prices for California when the number of retailers varies under different vertical structures. We find that wholesale and retail prices are higher once an upstream firm’s capacity constraint is binding. Moreover, upstream firms’ total profits increase when a firm’s capacity constraint is binding. In our numerical example, vertical integration lowers wholesale and retail prices even though the vertically integrated firm refuses to deal with both its upstream and downstream rivals. We also find that whether a higher degree of retail market concentration results in higher retail price depends on market structure and the effectiveness of the capacity constraints. Under vertical integration, a decrease in the number of independent retailers has limited effect on retail market price when independent upstream firms’ capacity constraints are binding. When we use the model to study market equilibrium in California’s gasoline industry, we find that independent firms’ capacity tends to be binding. The results suggest that the effect of a decrease in the number of independent retailers on retail gasoline price may be small. It is well known that California has the highest wholesale and retail gasoline prices in the U.S. For example, the average retail gasoline price in San Francisco in 2009 was $2.74 per gallon, while the U.S. average in the same year was only $2.41 per gallon. Moreover, differences in prices cannot be completely explained by differences in production costs and taxes. Therefore, these differences in prices receive wide attention from policy makers, researchers and the public. Many factors contribute to the observed differences in gasoline prices across regions, including capacity constraint at the refining level, product differentiation due to regulations, and vertical restraints between refineries and retailers. Capacity constraints at the refining level allow refineries to exercise market power and to charge a higher wholesale gasoline price. Without capacity constraints, if a firm intended to
77
raise the wholesale price by reducing its production unilaterally, rivals could expand production in response and the practice would hurt the firm’s profits. However, when rivals face capacity constraints, a unilateral decrease in production can be profitable and can result in higher wholesale prices.1 In addition, California has stricter environmental regulations on the content of gasoline than other states. These regulations weaken the ability of out-of-state refineries to meet the supply shortages in California, and further increase a refinery’s ability to exercise its market power at the refining level. In California, approximately 66% of the wholesale gasoline is supplied through vertical relationships with refineries in 2009, while the national average is only 27%. Vertical controls, such as pricing or distribution restrictions imposed by refineries on retail outlets, often raise concerns about whether refineries can use their market power at the wholesale level to impede competition at the retail level (Borenstein, Bushnell and Lewis, 2004; Borenstein and Bushnell, 2005). Moreover, it is often argued that vertically integrated firms have low incentive to carry renewable fuels and may slow down the diffusion of renewable fuels carried in gasoline stations. In general, the effect of vertical controls on prices is ambiguous. On one hand, vertical controls lower retail prices because they align the incentives of a refinery with those of a retailer. Without vertical controls, a retailer has incentive to exercise its market power at the retail level by charging a higher retail price, even though the increase in retail price leads to lower consumption and hurts the upstream refinery’s profit. In this regard, vertical controls can solve this classic double marginalization problem and result in a lower retail price. On the other hand, vertical controls also allow for strategic pricing behavior. It has been shown that under some circumstances, a vertically integrated firm can increase its profit by raising the wholesale price to increase its rivals’ costs (Salop and Scheffman, 1987; Ordover, Saloner and Salop, 1990; Hendricks and McAfee, 2010). Several empirical studies look at gasoline pricing dynamics in response to demand 1
Moreover, if firms can coordinate on their capacities, they can increase their collusive profits. For pricing dynamics when capacity constraints are endogenous, see Staiger and Wolak (1992), Fabra (2006) and Knittel and Lepore (2010).
78
and cost shocks. For example, Borenstein, Cameron and Gilbert (1997) provides evidence on price transmissions from crude oil to wholesale and retail gasoline prices. Borenstein and Shepard (1996) and Borenstein and Shepard (2002) look at retail and wholesale pricing dynamics to further explore the relationship between industry conduct, market power and gasoline prices. Many empirical papers provide explanations for regional gasoline price differences. For example, Brown, Hastings, Mansur and Villas-Boas (2008) show that gasoline content regulations are associated with higher wholesale prices. In particular, they use variations in gasoline content regulations across markets generated by the 1990 Clean Air Act Amendments to investigate the effect of gasoline content regulations on wholesale prices. They find that wholesale gasoline price increased by 3 cents per gallon in metropolitan areas that were subjected to regulations compared to the ones that were not. In addition, Muehlegger (2006) estimates the cost structure of the U.S. refineries and finds that price spikes due to shortages in markets with stricter gasoline content regulations could be mitigated if these markets were subjected to similar federal standards on gasoline content regulations. To test whether vertical controls are associated with higher wholesale prices, Hastings and Gilbert (2005) studied the changes in wholesale prices in markets that were affected by the Tosco-UnoCal acquisition. They showed that Tosco’s wholesale price was higher in markets where Tosco gained more market presence due to the acquisition and had more interactions with independent gasoline stations. They argue that the evidence is consistent with the strategic incentive to increase the refinery’s profit by raising rivals’ costs. Finally, Hastings (2004) studied how retail prices respond to changes in vertical structure. She finds that the presence of independent retailers helps to decrease local retail prices and suggests that a model of product differentiation and brand loyalty is consistent with her findings.
3.2
Industry Background
Gasoline is a product of petroleum refining. Through refining, crude oil is turned into various end products (gasoline, jet fuel, diesel, etc.). The refining process is
79
flexible to respond to fluctuations in consumer demand. A refinery can increase its gasoline output via two methods. It can adjust its end product mix by configuring the refining process in advance or it can raise its capacity utilization rate.2 Nevertheless, a refinery’s total production of gasoline is still bounded by its capacity.3 Once a refinery reaches its capacity limits, the marginal cost for extracting one additional unit of gasoline from crude oil becomes extremely high. After refining, the end products are shipped to regional wholesale terminals through pipelines or by barge. For example, gasoline refined in Gulf Coast states can be shipped to the East Coast or West Coast by barge and to the Mideast through pipelines. Unlike electricity, which can be transmitted in a short time, shipping gasoline across regions can take weeks or even months. Vertical controls play an important role in gasoline refining and marketing. First, a gasoline station can be branded or unbranded. For example, a Chevron gas station sells gasoline from the Chevron (refining) company.4 In addition, within the branded category, the degree of vertical control varies by station ownership and distribution method. Company-owned stations (operated directly by the company or by a lessee-dealer) mostly receive gasoline directly from the refinery. Independently owned gasoline stations can also be branded and receive gasoline directly from a refiner, paying Dealer Tank Wagon (DTW) prices, or by purchasing it at wholesale terminals (self-distributed or distributed by jobbers), paying the rack price plus transportation costs. Similarly, unbranded stations can also purchase gasoline through the above distribution methods. Table 3.1 shows the volume and the percentage of total gasoline sold by different distribution methods across five Petroleum Administration for Defense Districts 2
On average, motor gasoline accounts for around 46% of output produced from crude oil refining. However refineries are able to increase their gasoline yield by adding more blendstocks and oxygenates into the refining process. Borenstein, Bushnell and Lewis (2004) discuss how California refineries raised their gasoline share of crude oil plus blendstock inputs from 51% to 55% from 1995 to 2002 to meet the increasing demand for California reformulated gasoline (CaRFG). 3 The national utilization rate was above 90% from 1995 to 2005, reached its peak at 95.6% in 1998 and declined to 83% in 2009. 4 According to National Petroleum News, there were 155,662 gasoline stations in the U.S. in 2009. Shell had the largest number of stations (9.3%), followed by BP (7.4%), ExxonMobil (6.6%), Chevron (6.2%) and ConocoPhillips (5.5%).
80
(PADD) in 2009. Retailers who purchase gasoline from the spot market (Bulk Sales and Rack Sales) are subjected to less vertical control from upstream refineries. Nationwide, 75% of sales are from terminals (bulk and rack sales), 13.5% are by direct retail and 11.2% are by Dealer Tank Wagon sales. However, there are some regional differences. Notably, compared to the national average, California has a lower share of sales from terminals and a higher share of direct and DTW sales.5 This suggests that the vertical controls of refineries in the downstream market is an important attribute in California’s gasoline market. As for prices, wholesale prices are lowest in PADD 3 (Gulf Coast) and are highest in PADD 5 (West Coast). Although retail prices are not shown in Table 3.1, the West Coast also has the highest retail gasoline prices. In the next section, we review the theoretical and empirical work on explaining regional wholesale and retail gasoline price differences.
3.3
Model
In this section, we provide a model of oligopolistic competition under both vertical separation and vertical integration when producers face capacity constraints. The industry is composed of L upstream firms (refiners), N downstream firms (retailers) and, among the N firms, M vertically integrated downstream firms.6 Throughout the paper, we will use i to indicate downstream firms and j to indicate upstream firms. An upstream firm j faces capacity constraint kj , and has marginal production cost cj . Demand for gasoline consumption is a function of price with constant elasticity, so that Q = Ap−ϵ , where p is the retail price, Q is the quantity demanded, ϵ is the price elasticity of gasoline, and A is other factors that shift the demand curve.7 The model contains two stages of production. First, given the market structure, 5
Some data for PADD 4 (Rocky Mountain) and PADD 5 (West Coast) are withheld by the EIA to avoid disclosure of individual company data. To show the regional differences, we present the data from California as an example. 6 For example, suppose there are three refineries and two gasoline stations, then L=3, and N =2. If only one gasoline station is vertically integrated, then M =1. If both stations are vertically integrated (either both with one refinery or with different refineries), then M =2. 7 The functional form assumption follows previous studies in estimating the demand for gasoline. For example, see Dahl (1982) and Hughes, Knittel and Sperling (2008).
81
upstream firms compete in a Cournot fashion with capacity constraints. Then, unintegrated downstream firms take wholesale price as given, and order the intermediate good in the wholesale market, while integrated downstream firms have the option to order the intermediate good from their own production subsidiary. Finally, downstream firms compete in the retail market. The downstream equilibrium is again Cournot, and we solve the game backward.
3.3.1
Retail Market
A downstream firm i produces qi to maximize its profit πi given its marginal cost ci : qi = arg max(P (Q) − ci )qi . q
The first order condition for a profit maximizing firm i under quantity competition can be written as:
p − ci qi si = = , p ϵQ ϵ
(3.1)
where si is the market share of firm i. The left hand side of the equation measures a firm’s ability to charge a retail price over its marginal cost and is known as the Lerner index, which is often used as a measure of market power. Solving for the retail price, we have:
N ∑ ϵ ci , p= ϵN − 1 i=1
where N is the number of downstream firms. For simplicity, in the following analysis, we assume that the only marginal cost for a retailer is the cost of purchasing its input. We then discuss the output decision of upstream firms under vertical separation and vertical integration.
3.3.2
Wholesale Market
3.3.2.1
Vertical Separation
We first consider the case when none of the upstream firms has vertical controls over downstream firms through vertical integration or long term contracts. In this case, the wholesale price, pw , is the marginal cost of a downstream firm. The derived
82
demand curve faced by an upstream firm is: Q = Ap
−ϵ
N ∑ ϵN ϵN −ϵ −ϵ ϵ ′ = A( cri )−ϵ = A( pw )−ϵ = A( ) pw = A p−ϵ w . ϵN − 1 i=1 ϵN − 1 −1 } | ϵN{z A
′
(3.2) We first discuss the optimality conditions without any capacity constraint, which is a classic case of double marginalization. Then we turn to the case when upstream firms face capacity constraints. No Capacity Constraint
The derived demand curve faced by an upstream firm in equation (3.2) has the same functional form as consumers’ demand for gasoline, so the first order condition for an upstream firm j under Cournot competition can also be written as: pw − cj sj = , pw ϵ or qj =
(3.3)
pw − c j ϵQ, pw
(3.4)
where sj is upstream firm j’s market share. One result from equation (3.4) is that under vertical separation, when upstream firms face constant elastic demand and do not have capacity constraint, the downstream competition (measured by N ) has no effect on wholesale price. With Capacity Constraint
When an upstream firm j has capacity kj , the optimality conditions become: 0 ≤ qj ⊥ cj + λj ≥ pw +
∂pw (Q) qj , ∂qj
0 ≤ λj ⊥ kj ≥ qj ,
∀j
(3.5)
∀j
(3.6)
where 1 pw ∂pw (Q) =− , ∂qj ϵQ and ⊥ represents the complementary condition, meaning that at least one inequality within an equation must hold as an equality, and λj is the shadow price of the capacity constraint of firm j.
83
Equation (3.5) suggests that firm j should keep producing until its marginal cost (marginal production cost plus the shadow price of the capacity constraint) is greater than its marginal benefit. Equation (3.6) suggests that when the capacity constraint is not binding, the shadow price of the capacity constraint should be zero. 3.3.2.2
Vertical Integration
Here we consider the optimal conditions of upstream firms when vertical integration is allowed. Salinger (1988) analyzed vertical integration when upstream firms have a symmetric cost structure and compete in quantity.8 He showed that an integrated upstream firm has no incentive to participate in the wholesale market because it can make more profit by selling the intermediate good to its own subsidiary rather than to another retailer. Similarly, he showed that a vertically integrated downstream firm will not purchase in the wholesale market because it can acquire the input internally with lower cost. Given these results, a vertically integrated firm j’s total cost from production to retail is cj = c and an independent downstream firm i’s total cost is pw . Recall that M is the number of vertically integrated downstream firms. Furthermore, assume that all upstream firms have the same marginal costs cj . Then the downstream market equilibrium is an asymmetric Cournot equilibrium such that
p=
N ∑ ϵ ϵ ci = (M c + (N − M )pw ), ϵN − 1 i=1 ϵN − 1
and Q = Ap−ϵ = A
(
)−ϵ ϵ (M c + (N − M )pw ) . ϵN − 1
(3.7)
(3.8)
Let q denote each independent retailer’s output. The total output from all indepen˜ is therefore (N −M )q. Using equation (3.1) and the demand equation dent retailers Q (3.8), the derived demand curve faced by an independent upstream firm j becomes ˜ = (N − M )q = ϵ(N − M )Q p − pw = Aϵ(N − M )(p − pw )p−(1+ϵ) , Q p
(3.9)
where retail price p is a function of wholesale price pw as shown in (3.7). No Capacity Constraint 8
For a general treatment and a thorough literature review of the theory of vertical foreclosure, see Rey and Tirole (2007).
84
From the previous discussion, in our model, only independent upstream firms participate in the wholesale market. The corresponding first order condition for an unintegrated firm is: (pw − c) +
˜ ∂pw (Q) qj = 0, ∂qj
for all independent upstreams firm j.
(3.10)
We can calculate the partial derivative in equation (3.10) using equation (3.9).9 With Capacity Constraint
When an upstream firm j faces capacity constraint kj , the optimality conditions are: ˜ ∂pw (Q) qj , ∂qj ∂p(Q) qj , 0 ≤ qj ⊥ c + λ j ≥ p + ∂qj 0 ≤ qj ⊥ c + λ j ≥ p w +
if j is not vertically integrated;
(3.11)
if j is vertically integrated;
(3.12)
for all upstream firm j.
(3.13)
0 ≤ λj ⊥ kj ≥ qj ,
The interpretations of the complementary conditions are the same as before. Finally, the market equilibrium is a combination of market quantities (q1 , ...qN ), wholesale price pw , retail price p, shadow prices of the capacity constraints (λ1 ...λL ) that solve the the market clearing condition and the optimality conditions for the upstream firms. The equilibrium conditions represent a mixed complementary problem. We program the mixed complementary problem in GAMS, and use the PATH algorithm (Dirkse and Ferris, 1995) as the solution solver.10
3.4
Numerical Example
Before we apply the theoretical model to the gasoline market, we first study a simple numerical example when there are two upstream firms (L=2) and N downstream 9
We apply the implicit function theorem to find ˜
˜ ∂pw (Q) ∂qj . ∂F ∂qj ∂F ∂pw
˜ − If we define F (Q, pw ) as Q
∂F ∂F w (Q) , where ∂q Aϵ(N − M )(p − pw )p−(1+ϵ) = 0, then we have ∂p∂q =− =1 and ∂p =−Aϵ(N − j j w ] [ (p−p ) ∂p ∂p M )p−(ϵ+1) ( ∂pw − 1) − p w (1 + ϵ) ∂pw . 10 GAMS is a computer language that specializes in mathematical programming and optimization. To use the PATH solver, the program can be uploaded to the NEOS server, which allows users to submit mixed complementary problems written in GAMS or in AMPL (another computer language for mathematical programming models). For more details, see Ferris and Munson (1998).
85
firms. Let Q = 250 p2 , cj = c = 3. We consider four scenarios that vary the number of vertically integrated downstream firms (M ) and the upstream firms’ capacity constraints (k1 and k2 ): 1. Vertical separation with nonbinding capacity constraint. M =0, k1 =20, k2 =20. 2. Vertical separation with a binding capacity constraint. M =0, k1 =20, k2 =7. 3. Vertical integration with nonbinding capacity constraint. M =1, k1 =20, k2 =20. 4. Vertical integration with a binding capacity constraint. M =1, k1 =20, k2 =7. Tables 3.2 through 3.5 provide the market outcomes under, respectively, each of the above scenarios.
3.4.1
Vertical Separation
Table 3.2 presents the equilibrium outcomes when no firms are vertically integrated and the capacity constraints are not binding. As discussed in the previous section, given the functional form, wholesale price pw is not responsive to the downstream market structure in this scenario. We find that the wholesale price (pw ) is higher than the marginal production cost (c), and the retail price is higher than the wholesale price. The results reflect the standard double marginalization problem. As the downstream market becomes more competitive, the downstream markup becomes lower, the total profit of downstream firms becomes lower, and the total profit of upstream firms becomes higher. Table 3.3 shows the results when we restrict upstream firm 2’s capacity at k2 = 7. Once firm 2 reaches its capacity limit, allowing for more entrants in the downstream market drives up the wholesale price, while the retail price still falls when the market gets more competitive. Comparing Table 3.2 and 3.3, the results suggest that the wholesale price, retail price, and firm 1 and firm 2’s profits are all higher when firm 2’s capacity constraint is binding.
3.4.2
Vertical Integration
We provide the results when upstream firm 1 is vertically integrated with downstream firm 1 in Tables 3.4 and 3.5. In Table 3.4, we find that when both firms’ capacity
86
constraints are not binding, having a vertically integrated firm in the model generates a lower wholesale price and retail price than those in Table 3.2. More importantly, the vertically integrated firm’s profit is actually lower than that of the independent firm when the retail market becomes more competitive (N =20 and N =100). This is because when a vertically integrated firm considers whether or not to serve the wholesale market, it finds that its marginal benefit from participating in the wholesale market, pw − cj , is always lower than its marginal cost for doing so, p − cj , which is its marginal profit if it sells through its own retail outlets. However, when more and more retailers enter the market, refusing to deal with independent retailers turns out to deliver a lower profit compared to that of the independent upstream firm. For example, when N = 100, the vertically integrated firm 1 owns 1 retailer and has a profit of 6.02 while the independent firm 2 supplies 99 retailers and has a profit of 7.91. In this case (N = 100), an upstream firm to be vertically integrated with only one downstream firm is not an equilibrium outcome, and we should expect few firms to become vertically integrated in the first place when they have little presence in the downstream market. Finally, Table 3.5 displays the results when upstream firm 1 is vertically integrated and firm 2’s capacity constraint is binding in some cases. In this scenario, we find that the wholesale price starts to increase once firm 2 reaches its capacity. Moreover, once firm 2 reaches its capacity, allowing for more independent retailers provides no further downward pressure on retail price. We can compare the wholesale price, the retail price, and firm 1 and firm 2’s profits under the four scenarios when N is fixed at 100: p2w > p4w > p1w > p3w , p2 > p 1 > p 4 > p 3 , π14 > π12 > π11 > π13 , π23 > π21 > π22 > π24 . To sum up, our numerical exercise shows that when firm 2’s capacity constraint is not binding, firm 1’s incentive to become vertically integrated with a downstream firm
87
depends on the downstream firm’s market presence in the retail market. However, when firm 2’s capacity constraint is binding, it provides firm 1 more incentive to become vertically integrated even though the downstream firm’s market presence is low. Wholesale price is the highest when firm 1 is vertically integrated and firm 2’s capacity constraint is binding. Nevertheless, retail prices are always lower when firm 1 is vertically integrated. We then apply our model to study the equilibrium outcomes in California’s gasoline market.
3.5
Data
The U.S. Energy Information Agency (EIA) collects information on the production and the distribution of several selected petroleum products. Data on consumption, wholesale prices, and retail prices are at the Petroleum Administration for Defense Districts (PADD) level or at state level. Capacity is reported at the refinery level, and raw materials and refinery operating costs are reported at the national level. Table 3.6 lists the U.S. refining costs from 2000 to 2009.11 Costs fluctuate by year and were especially high in 2008. To capture the marginal costs of refining, we used the costs that are directly associated with the refining process (the sum of raw material costs and energy costs).12 We searched each refinery’s website to find information on its vertical relationships with retailers. The refinery capacity data provides the identity of each California’s refinery. We coded a refinery as vertically integrated if it had branded gasoline stations in California. Eight refineries of fifteen refineries were found to have branded gasoline stations in California. Variables used in this paper include: • Q (consumption): annual consumption of motor gasoline products (billions of gallons). • p (retail gasoline price): the weighted retail price (tax excluded) of all three 11
Data of refining costs are only available at the U.S. level. Given the data availability, we assume that the production cost remains constant up to a firm’s capacity level. 12
88
grades of gasoline (regular, midgrade, premium). • pw (wholesale gasoline price): resale gasoline prices. • k (refinery capacity): annual total operable capacity of each refinery (billions of gallons). • c (refinery costs): the sum of raw material costs and energy costs. Table 3.7 presents the means of the above variables in California in 2009. The refining cost is the average national refining costs in 2009 because data are only available at the U.S. level. For comparison purposes, we also list the means by PADD in 2009.13 Fifteen firms in California has refining capacity and eight of them have affiliated branded gasoline stations. We use data on consumption and retail price to fit the demand equation with constant elasticity. In order to calibrate the model, given any elasticity ϵ, we find the parameter A in the demand equation by solving the equation using consumption and price data from 2009. We set the elasticity at −0.3 to fit the demand function of gasoline in California.14 We use the data on California refineries’ capacity levels and their relationships with retailers to calibrate equilibrium wholesale and retail prices by varying retail market structure parameters N and M .15
3.6
Results
Tables 3.8, 3.9, and 3.10 provide the results when the vertical integration parameter M is set at 0, 10, and 20, respectively. N represents t.16 First, Table 3.8 presents the 13
Consumption of the East Coast exceeds the total capacity of the refineries in the district because the East Coast receives shipments from other districts (the net receipts from PADD 1 to 5 are 12.22, 5.45, −18.62, 0.16 and 1.11, respectively.) 14 Espey (1998) provides estimates of gasoline demand in studies published between 1966 and 1997 and finds the median short-run and long-run price elasticities of gasoline are -0.23 and -0.43, respectively. 15 Even though we proceed with our abstract model that treats California as a single market to simplify our analysis, it is important to note that shipments from refineries in other states is possible. Nevertheless, as discussed in the previous section, the ability of out-of-state refineries to meet the supply shortages in California is weaken by transportation costs and the stricter environmental regulations on the content of gasoline in California. 16 In this exercise, we allowed the number of downstream gasoline retailers in California, N , to vary between 40 and 100. And we restrict the number of vertically integrated firms, M , to be either 0, 10, or 20. In future work, we plan to collect detailed gasoline station ownership data to provide results that are directly based on observed vertical relationships in California’s gasoline market.
89
results when none of the downstream firms is vertically integrated (M =0). We find that a decrease in retailers has little impact on wholesale price. However, a decrease in retailers raises the retail price. The resulting competitive (N =100) wholesale price ($2.36) and retail price ($2.44) are much higher than those observed in 2009—only $1.94 and $2.12, respectively. Given that these results are obtained by assuming vertical separation, we consider the cases when 10 downstream firms are vertically integrated (M =10) and 20 downstream firms are vertically integrated (M =20). All of the independent firms’ capacity constraints turn out to be binding in all cases. When we allow 10 retailers to become vertically integrated, holding N fixed at 100, the wholesale and retail prices fall to $2.018 and $2.036, respectively. Moreover, Table 3.9 suggests that when there is no excess capacity for independent refineries, reducing the number of retailers lowers the wholesale price and has no impact on retail price. The results in Table 3.9 are that when we increase the number of vertically integrated firms from 10 to 20, the wholesale and the retail prices drop to $1.728 and $1.745, respectively (when N is fixed at 100). Similarly, given that the capacity constraints for independent refineries are binding in all cases, changing the number of retailers has no impact on retail price. Given the results in Tables 3.9 and 3.10, the effects of a reduction in the number of gasoline retailers depend on whether the retailers have vertical relationships with upstream refineries. For example, suppose we currently have 90 retailers and 20 of them have vertical relationships with refineries. In this case, M =20, N =90, pw =1.725, and p=1.744. Suppose that we remove 10 independent retailers, so that M =20 and N =80. Because the capacity constraints for the independent refineries are binding, the impact on retail price is small (pw =1.722, p=1.744). On the contrary, if we remove 10 vertically integrated retailers, such that M =10 and N =80, wholesale and retail prices both increase by 29 cents (pw =2.013, p=2.036).
3.7
Conclusion
In this paper, we study a model of two-stage Cournot competition when upstream firms face capacity constraints. We show that the extent to which changes in retail
90
market structure affect the retail prices depend on the vertical relationships between upstream firms and downstream firms and on the effectiveness of upstream firms’ capacity constraints. We find that in California’s gasoline market, the independent refineries’ capacity levels tend to be binding, while vertically integrated firms’ capacity levels are not. The results suggest that the effect of a decrease in the number of independent retailers on retail gasoline price may be small. More work remains to be done to improve our understanding of market equilibrium under vertical integration when firms face upstream capacity constraints. Our model takes firms’ capacity levels as given and assumes that gasoline products are homogeneous, which is suitable here given the limitation of the EIA data. Future work on collecting detailed data on retail gasoline prices, station attributes and vertical relationships between firms, and work on modeling product differentiation and endogenous capacity level decisions in a vertical market structure would be very useful.
91
Table 3.1: Gasoline Sold by Distribution Methods in Five PADDs PADD
1
2
3
4
5
California
U.S. Total
Direct Retail
Volume Share Price
15,961 14.63 1.88
13,122 13.44 1.87
9,405 12.20 1.83
1,131 NA 1.88
8,351 NA 2.09
6,672 16.97 2.128
47,969 13.50 1.89
DTW
Volume Share Price
12,318 11.29 1.81
1,729 1.77 1.79
170 0.22 1.78
NA NA 1.90
NA NA 1.95
19,189 48.81 1.973
39,896 11.23 1.84
Bulk Sales
Volume Share Price
7,543 6.91 1.66
6,994 7.16 1.67
24,706 32.06 1.65
NA NA NA
NA NA 1.82
1,485 3.78 1.817
41,380 11.65 1.67
Rack Sales
Volume Share Price
73,270 67.16 1.76
75,811 77.63 1.75
42,792 55.52 1.73
11,130 NA 1.74
23,075 NA 1.89
11,966 30.44 1.903
226,077 63.63 1.77
Notes: Unit: thousands of gallons per day. NA: data withheld to avoid disclosure of individual company data by EIA. All data are in 2009. Direct Retail: sales through company-owned outlet. DTW (Dealer Tank Wagon Sales): wholesale sales of gasoline priced on a delivered basis to a retail outlet. Bulk Sales: wholesale sales of gasoline in individual transactions which exceed the size of a truckload. Rack Sales: wholesale truckload sales or smaller of gasoline where title transfers at a terminal. Definitions follow EIA Energy Glossary.
92
Table 3.2: Market Outcomes Under Scenario 1 ∑2
∑N
M
N
q1
q2
pw
λ1
λ2
p
π1
π2
0
2
4.395
4.395
4
0
0
5.333
4.395
4.395
8.79
11.72
0
5
6.328
6.328
4
0
0
4.444
6.328
6.328
12.656
5.625
0
10
7.051
7.051
4
0
0
4.211
7.051
7.051
14.102
2.969
0
20
7.427
7.427
4
0
0
4.103
7.427
7.427
14.854
1.523
0
100
7.735
7.735
4
0
0
4.020
7.735
7.735
15.47
0.311
j
πj
i
πi
Notes: M represents the number of vertically integrated downstream firms. N represents the number of downstream firms. In this example, the industry is vertically ∑2 separated. Q = 250 p2 , k1 = 20, k2 = 20, c1 = c2 = 3. j πj represents the total ∑N profits of upstream firms. i πi represents the total profits of downstream firms.
Table 3.3: Market Outcomes Under Scenario 2 ∑2
πj
∑N
πi
M
N
q1
q2
pw
λ1
λ2
p
π1
π2
0
2
4.395
4.395
4
0
0
5.333
4.395
4.395
8.79
11.72
0
5
6.328
6.328
4
0
0
4.444
6.328
6.328
12.656
5.625
0
10
7.061
7
4.006
0
0.009
4.217
7.103
7.042
14.145
2.965
0
20
7.508
7
4.047
0
0.071
4.151
7.861
7.329
15.190
1.505
0
100
7.871
7
4.08
0
0.119
4.101
8.501
7.56
16.061
0.305
j
i
Notes: M represents the number of vertically integrated downstream firms. N represents the number of downstream firms. In this example, the industry is vertically separated. Upstream firm 1 has higher capacity k1 . Q = 250 p2 , k1 = 20, k2 = 7, c1 = c2 = 3. ∑2 ∑N i πi represents the total profits of j πj represents the total profits of upstream firms. downstream firms.
93
Table 3.4: Market Outcomes Under Scenario 3 ∑2
N
q1
q2
pw
λ1
λ2
p
π1
π2
1
2
8.183
3.092
4.063
0
0
4.709
13.983
3.287
17.270
5.284
1
5
8.016
6.668
3.892
0
0
4.126
9.027
5.948
14.975
1.877
1
10
7.728
8.219
3.846
0
0
3.959
7.414
6.953
14.368
0.876
1
20
7.532
9.076
3.824
0
0
3.880
6.627
7.479
14.105
0.420
1
100
7.352
9.789
3.808
0
0
3.819
6.021
7.910
13.931
0.081
j
πj
∑N
M
i
πi
Notes: M represents the number of vertically integrated downstream firms. N represents the number of downstream firms. In this example, there are two upstream firms. Upstream firm 1 integrates with downstream firm 1 and has the same capacity k as firm 2. Q = ∑2 250 p2 , k1 = 20, k2 = 20, c1 = c2 = 3. j πj represents the total profits of upstream firms. ∑N i=2 πi represents the total profits of downstream firms (excluding vertically integrated firm 1).
Table 3.5: Market Outcomes Under Scenario 4 ∑2
πj
∑N
πi
M
N
q1
q2
pw
λ1
λ2
p
π1
π2
1
2
8.183
3.092
4.063
0
0
4.709
13.983
3.287
17.270
5.284
1
5
8.016
6.668
3.892
0
0
4.126
9.027
5.948
14.975
1.877
1
10
7.962
7
3.981
0
0.142
4.087
8.656
6.870
15.527
0.846
1
20
7.962
7
4.037
0
0.221
4.087
8.657
7.262
15.919
0.401
1
100
7.962
7
4.078
0
0.278
4.088
8.660
7.546
16.206
0.077
j
i
Notes: M represents the number of vertically integrated downstream firms. N represents the number of downstream firms. In this example, there are two upstream firms. Upstream firm 1 integrates with downstream firm 1 and has higher capacity k1 . Q = 250 p2 , k1 = 20, k2 = 7, ∑2 ∑N c1 = c2 = 3. j πj represents the total profits of upstream firms. i=2 πi represents the total profits of downstream firms (excluding vertically integrated firm 1).
94
Table 3.6: Refining Costs Per Gallon Raw Materials Energy Marketing Other Raw Materials, Total Costs and Costs Costs Operating Product Purchases and Costs Product Purchases Costs Energy Costs (1) (2) (3) (4) (1)+(2) (1)+(2)+(3)+(4) 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009
0.746 0.616 0.638 0.763 1.005 1.352 1.600 1.766 2.344 1.487
0.032 0.031 0.026 0.034 0.035 0.043 0.041 0.046 0.062 0.031
0.033 0.038 0.039 0.032 0.037 0.036 0.034 0.040 0.042 0.035
0.047 0.052 0.062 0.057 0.059 0.073 0.076 0.092 0.124 0.108
0.778 0.647 0.664 0.797 1.040 1.395 1.642 1.812 2.406 1.519
0.857 0.737 0.766 0.887 1.136 1.504 1.751 1.944 2.572 1.661
Notes: Units: dollars per gallon (not adjusted for inflation). 2000-2008 data are calculated from EIA survey Form EIA-28 (Financial Reporting System). 2009 data are calculated from EIA’s Performance Profiles of Major Energy Producers report. Other Operating Costs include Other Raw Material Supply Expense (raw material transportation, raw material exchange differentials, and inventory change) and Other Refining Operating Expenses (purchases of non-petroleum blending stocks, any canning and blending operation expenses not assigned to other marketing expenses).
Table 3.7: California 2009 Data Used to Set Model Parameters PADD
1 East
2 Midwest
Coast
3 Gulf
4 Rocky
5 West
Coast
Mountain
Coast
California
Q
49.55
38.62
21.59
4.58
23.41
14.79
p
1.88
1.87
1.83
1.88
2.08
2.12
pw
1.76
1.76
1.72
1.78
1.92
1.94
c ∑
1.52
1.52
1.52
1.52
1.52
1.52
27.87
60.60
136.16
10.13
51.63
32.94
j
k
Notes: Data are obtained from EIA website. All data are in 2009. Q is the consumption of motor gasoline products (billions of gallons). p the weighted retail price (tax excluded) of all three grades of gasoline (regular, midgrade, premium). pw is the the gasoline price for resale. c ∑ is the sum of raw material costs and energy costs. j k is the aggregate refining capacity of all refineries in a Petroleum Administration for Defense District (billions of gallons).
95
Table 3.8: Market Outcomes with Double Marginalization ∑15
∑N
M
N
pw
p
Q
0
100
2.36
2.44
14.18
11.90
1.15
0
90
2.36
2.45
14.17
11.88
1.29
0
80
2.36
2.46
14.15
11.86
1.45
0
60
2.36
2.49
14.09
11.78
1.95
0
40
2.35
2.57
13.97
11.63
2.99
j
πj
i
πi
Notes: M represents the number of vertically integrated downstream ∑15 firms. N represents the number of downstream firms. j πj represents ∑N the total profits of upstream firms. i πi represents the total profits of downstream firms. Table 3.9: Market Outcomes under Vertical Integration (M =10) ∑
πj1
∑
πj2
∑15
πj
∑N
πi
M
N
pw
p
Q
10
100
2.018
2.036
14.975
5.876
1.787
7.663
0.206
10
90
2.016
2.036
14.974
5.881
1.778
7.658
0.231
10
80
2.013
2.036
14.974
5.879
1.768
7.647
0.264
10
60
2.004
2.036
14.974
5.884
1.733
7.617
0.370
10
40
1.982
2.036
14.974
5.878
1.657
7.535
0.617
j
j
j
i
Notes: M represents the number of vertically integrated downstream firms. N repre∑15 sents the number of downstream firms. j πj represents the total profits of upstream ∑N firms. i πi represents the total profits of downstream firms. Table 3.10: Market Outcomes under Vertical Integration (M =20) ∑
πj1
∑
πj2
∑15
N
pw
p
Q
20
100
1.728
1.745
15.685
2.720
0.743
3.463
0.201
20
90
1.725
1.744
15.686
2.709
0.737
3.446
0.230
20
80
1.722
1.744
15.686
2.712
0.725
3.437
0.268
20
60
1.711
1.744
15.686
2.713
0.685
3.398
0.402
20
40
1.678
1.744
15.686
2.716
0.566
3.282
0.803
j
j
j
πj
∑N
M
i
πi
Notes: M represents the number of vertically integrated downstream firms. N repre∑15 sents the number of downstream firms. j πj represents the total profits of upstream ∑N firms. i πi represents the total profits of downstream firms.
96
References Abito, Jose Miguel and Julian Wright, “Exclusive Dealing with Imperfect Downstream Competition,” International Journal of Industrial Organization, 2008, 26 (1), 227 – 246. Aghion, Philippe and Patrick Bolton, “Contracts as a Barrier to Entry,” The American Economic Review, 1987, 77 (3), 388–401. Aguirregabiria, Victor and Pedro Mira, “Swapping the Nested Fixed Point Algorithm: A Class of Estimators for Discrete Markov Decision Models,” Econometrica, 2002, 70 (4), pp. 1519–1543. Asker, John, “Diagnosing Foreclosure Due to Exclusive Dealing,” 2005. Working paper, New York University. Augereau, Angelique, Shane Greenstein, and Marc Rysman, “Coordination versus Differentiation in a Standards War: 56K Modems,” The RAND Journal of Economics, 2006, 37 (4), pp. 887–909. Bajari, Patrick, C. Lanier Benkard, and Jonathan Levin, “Estimating Dynamic Models of Imperfect Competition,” Econometrica, 2007, 75 (5), pp. 1331– 1370. , Han Hong, John Krainer, and Denis Nekipelov, “Estimating Static Models of Strategic Interactions,” Journal of Business and Economic Statistics, 2010, 28 (4), 469–482. Bernheim, B. Douglas and Michael D. Whinston, “Exclusive Dealing,” The Journal of Political Economy, 1998, 106 (1), 64–103. Berry, Steven T., “Estimation of a Model of Entry in the Airline Industry,” Econometrica, 1992, 60 (4), pp. 889–917. , “Estimating Discrete-Choice Models of Product Differentiation,” The RAND Journal of Economics, 1994, 25 (2), 242–262. Besanko, David and Martin K. Perry, “Equilibrium Incentives for Exclusive Dealing in a Differentiated Products Oligopoly,” The RAND Journal of Economics, 1993, 24 (4), 646–667. Borenstein, Severin, A. Colin Cameron, and Richard Gilbert, “Do Gasoline Prices Respond Asymmetrically to Crude Oil Price Changes?,” The Quarterly Journal of Economics, 1997, 112 (1), pp. 305–339. and Andrea Shepard, “Dynamic Pricing in Retail Gasoline Markets,” The RAND Journal of Economics, 1996, 27 (3), pp. 429–451.
97
and , “Sticky Prices, Inventories, and Market Power in Wholesale Gasoline Markets,” The RAND Journal of Economics, 2002, 33 (1), pp. 116–139. and James B. Bushnell, “Retail Policies and Competition in the Gasoline Industry,” 2005. Center for the Study of Energy Markets Working paper 144, University of California Energy Institute. , , and Matthew S. Lewis, “Market Power in California’s Gasoline Market,” 2004. Center for the Study of Energy Markets Working paper 132, University of California Energy Institute. Bork, Robert H., The Antitrust Paradox: A Policy at War with Itself, New York: Basic Books, 1978. Brenkers, Randy and Frank Verboven, “Liberalizing A Distribution System: The European Car Market,” Journal of the European Economic Association, 03 2006, 4 (1), 216–251. Bresnahan, Timothy F. and Peter C. Reiss, “Entry and Competition in Concentrated Markets,” The Journal of Political Economy, 1991, 99 (5), pp. 977–1009. Brown, Jennifer, Justine Hastings, Erin T. Mansur, and Sofia B. VillasBoas, “Reformulating competition Gasoline content regulation and wholesale gasoline prices,” Journal of Environmental Economics and Management, 2008, 55 (1), 1 – 19. Cardell, N. Scott, “Variance Components Structures for the Extreme-Value and Logistic Distributions with Application to Models of Heterogeneity,” Econometric Theory, 1997, 13 (2), pp. 185–213. Chen, Chia-Wen, “Estimating the Foreclosure Effect of Exclusive Dealing: Evidence from the Entry of Specialty Beer Producers,” 2011. Working paper, University of California at Davis. Chiou, Lesley, “The Timing of Movie Releases: Evidence from the Home Video Industry,” International Journal of Industrial Organization, 2008, 26 (5), 1059 – 1073. Chipty, Tasneem, “Vertical Integration, Market Foreclosure, and Consumer Welfare in the Cable Television Industry,” The American Economic Review, 2001, 91 (3), 428–453. Culbertson, W. Patton and David Bradford, “The Price of Beer: Some Evidence from Interstate Comparisons,” International Journal of Industrial Organization, 1991, 9 (2), 275 – 289. Dahl, Carol A., “Do Gasoline Demand Elasticities Vary?,” Land Economics, 1982, 58 (3), pp. 373–382.
98
Day, Nicholas, “Bye-Bye Bell’s: How Illinois Beer Distribution Laws, Fiercely Protected by a Powerful Industry, Drove Away One of Chicago’s Favorite Small Brewers,” 2006. Chicago Reader, December 14, 2006. Dick, Astrid A., “Demand estimation and consumer welfare in the banking industry,” Journal of Banking & Finance, 2008, 32 (8), 1661 – 1676. Dirkse, Steven P. and Michael C. Ferris, “The PATH Solver: A Non-Monotone Stabilization Scheme for Mixed Complementarity Problems,” Optimization Methods and Software, 1995, 5, 123–156. Doganoglu, Toker and Julian Wright, “Exclusive Dealing with Network Effects,” International Journal of Industrial Organization, 2010, 28 (2), 145 – 154. Einav, Liran, “Seasonality in the U.S. Motion Picture Industry,” The RAND Journal of Economics, 2007, 38 (1), pp. 127–145. Ellickson, Paul B. and Sanjog Misra, “Supermarket Pricing Strategies,” Marketing Science, 2008, 27 (5), 811–828. Espey, Molly, “Gasoline Demand Revisited: An International Meta-analysis of Elasticities,” Energy Economics, 1998, 20 (3), 273 – 295. Fabra, Natalia, “Collusion with Capacity Constraints over the Business Cycle,” International Journal of Industrial Organization, 2006, 24 (1), 69 – 81. Ferris, Michael C. and Todd S. Munson, “Complementarity Problems in GAMS and the PATH Solver,” Journal of Economic Dynamics and Control, 1998, 24, 2000. Goldberg, Pinelopi Koujianou, “Product Differentiation and Oligopoly in International Markets: The Case of the U.S. Automobile Industry,” Econometrica, 1995, 63 (4), pp. 891–951. Hart, Oliver and Jean Tirole, “Vertical Integration and Market Foreclosure,” Brookings Papers on Economic Activity. Microeconomics, 1990, 1990, pp. 205–286. Hastings, Justine S., “Vertical Relationships and Competition in Retail Gasoline Markets: Empirical Evidence from Contract Changes in Southern California,” The American Economic Review, 2004, 94 (1), pp. 317–328. and Richard J. Gilbert, “Market Power, Vertical Integration and the Wholesale Price of Gasoline,” The Journal of Industrial Economics, 2005, 53 (4), 469–492. Heide, Jan B., Shantanu Dutta, and Mark Bergen, “Exclusive Dealing and Business Efficiency: Evidence from Industry Practice,” Journal of Law and Economics, 1998, 41 (2), 387–407. Hellerstein, Rebecca, “Who Bears the Cost of a Change in the Exchange Rate? Pass-through Accounting for the Case of Beer,” Journal of International Economics, 2008, 76 (1), 14 – 32.
99
Hendricks, Kenneth and R. Preston McAfee, “A Theory of Bilateral Oligopoly,” Economic Inquiry, 2010, 48 (2), 391–414. Horta¸csu, Ali and Chad Syverson, “Cementing Relationships: Vertical Integration, Foreclosure, Productivity, and Prices,” The Journal of Political Economy, 2007, 115 (2), 250–301. Hughes, Jonathan E., Christopher R. Knittel, and Daniel Sperling, “Evidence of a Shift in the Short-Run Price Elasticity of Gasoline Demand,” The Energy Journal, 2008, 29 (1), 113–134. Kesmodel, David, “To Trump Small Brewers, Beer Makers Get Crafty,” Wall Street Journal - Eastern Edition, 2007, 250 (99), B1–B3. , “Beer Distributors Want More Than One Best Bud,” Wall Street Journal - Eastern Edition, 2008, 251 (30), B1–B2. Klein, Benjamin and Kevin M. Murphy, “Vertical Restraints as Contract Enforcement Mechanisms,” Journal of Law and Economics, 1988, 31 (2), pp. 265–297. Knittel, Christopher R. and Jason J. Lepore, “Tacit Collusion in the Presence of Cyclical Demand and Endogenous Capacity Levels,” International Journal of Industrial Organization, 2010, 28 (2), 131 – 144. Lafontaine, Francine and Margaret E. Slade, “Vertical Integration and Firm Boundaries: The Evidence,” Journal of Economic Literature, 2007, 45 (3), pp. 629–685. and , “Exclusive Contracts and Vertical Restraints: Empirical Evidence and Public Policy,” in Paolo Buccirossi, ed., Handbook of Antitrust Economics, MIT Press, 2008. Martimort, David, “Exclusive Dealing, Common Agency, and Multiprincipals Incentive Theory,” The RAND Journal of Economics, 1996, 27 (1), 1–31. Marvel, Howard P., “Exclusive Dealing,” Journal of Law and Economics, 1982, 25 (1), 1–25. McFadden, Daniel L., “Modeling the Choice of Residential Location,” in A. Karlqvist, L. Lundqvist, F. Snickers, and J. Weinull, eds., Spatial Interaction Theory and Planning Models, North-Holland, 1978. Muehlegger, Erich J., “Gasoline Price Spikes and Regional Gasoline Content Regulations: A Structural Approach,” 2006. Working paper, John F. Kennedy School of Government, Harvard University. O’Brien, Daniel P. and Greg Shaffer, “Vertical Control with Bilateral Contracts,” The RAND Journal of Economics, 1992, 23 (3), pp. 299–308.
100
Ordover, Janusz A., Garth Saloner, and Steven C. Salop, “Equilibrium Vertical Foreclosure,” The American Economic Review, 1990, 80 (1), pp. 127–142. Pakes, Ariel, Michael Ostrovsky, and Steven Berry, “Simple Estimators for the Parameters of Discrete Dynamic Games (With Entry/Exit Examples),” The RAND Journal of Economics, 2007, 38 (2), pp. 373–399. Posner, Richard A., Antitrust Law: An Economic Perspective, Chicago: University of Chicago Press, 1976. Rasmusen, Eric B., J. Mark Ramseyer, and Jr. Wiley John S., “Naked Exclusion,” The American Economic Review, 1991, 81 (5), 1137–1145. Rey, Patrick and Jean Tirole, “A Primer on Foreclosure,” in M. Armstrong and R. Porter, eds., , Vol. 3 of Handbook of Industrial Organization, Elsevier, 2007, pp. 2145 – 2220. Rojas, Christian, “The Competitive and Welfare Effects of Exclusive Agreements: Evidence from US brewing,” 2010. Working paper, University of Massachusetts at Amherst. Rust, John, “Optimal Replacement of GMC Bus Engines: An Empirical Model of Harold Zurcher,” Econometrica, 1987, 55 (5), pp. 999–1033. Salinger, Michael A., “Vertical Mergers and Market Foreclosure,” The Quarterly Journal of Economics, 1988, 103 (2), pp. 345–356. Salop, Steven C. and David T. Scheffman, “Raising Rivals’ Costs,” The American Economic Review, 1983, 73 (2), 267–271. and , “Cost-Raising Strategies,” The Journal of Industrial Economics, 1987, 36 (1), pp. 19–34. Sass, Tim R., “The Competitive Effects of Exclusive Dealing: Evidence from the U.S. Beer Industry,” International Journal of Industrial Organization, 2005, 23 (3-4), 203–225. and David S. Saurman, “Mandated Exclusive Territories and Economic Efficiency: An Empirical Analysis of the Malt-Beverage Industry,” Journal of Law and Economics, 1993, 36 (1), pp. 153–177. and , “Efficiency Effects of Exclusive Territories: Evidence from the Indiana Beer Market,” Economic Inquiry, 1996, 34 (3), 597–615. Segal, Ilya R. and Michael D. Whinston, “Exclusive Contracts and Protection of Investments,” The RAND Journal of Economics, 2000, 31 (4), 603–633. and , “Naked Exclusion: Comment,” The American Economic Review, 2000, 90 (1), 296–309.
101
Seim, Katja, “An Empirical Model of Firm Entry with Endogenous Product-Type Choices,” The RAND Journal of Economics, 2006, 37 (3), pp. 619–640. Simpson, John and Abraham L. Wickelgren, “Naked Exclusion, Efficient Breach, and Downstream Competition,” The American Economic Review, 2007, 97 (4), 1305–1320. Slade, Margaret E., “Beer and the Tie: Did Divestiture of Brewer-Owned Public Houses Lead to Higher Beer Prices?,” The Economic Journal, 1998, 108 (448), 565–602. Small, Kenneth A. and Harvey S. Rosen, “Applied Welfare Economics with Discrete Choice Models,” Econometrica, 1981, 49 (1), pp. 105–130. Staiger, Robert W. and Frank A. Wolak, “Collusive Pricing with Capacity Constraints in the Presence of Demand Uncertainty,” The RAND Journal of Economics, 1992, 23 (2), pp. 203–220. Sweeting, Andrew, “The Strategic Timing Incentives of Commercial Radio Stations: An Empirical Analysis Using Multiple Equilibria,” The RAND Journal of Economics, 2009, 40 (4), pp. 710–742. Tremblay, Victor J. and Carol Horton Tremblay, The U.S. Brewing Industry : Data and Economic Analysis, MIT Press, 2005. Vitorino, Maria A., “Empirical Entry Games with Complementarities: An Application to the Shopping Center Industry,” 2010. Working paper, University of Pennsylvania.
