International Review of Law and Economics 36 (2013) 12–24

Contents lists available at ScienceDirect

International Review of Law and Economics

Static and dynamic merger effects: A market share based empirical analysis Mikko Packalen ∗ , Anindya Sen University of Waterloo, Department of Economics, 200 University Avenue West, Waterloo, ON N2L 3G1, Canada

a r t i c l e

i n f o

Article history: Received 20 July 2012 Received in revised form 6 March 2013 Accepted 10 April 2013 JEL classification: L41 L71 L2 L1 Keywords: Merger Efficiencies Concentration Market shares Gasoline

a b s t r a c t Merger-specific efficiencies continue to play a relatively small role in merger enforcement and merger retrospectives. Motivated by the paucity of empirical analyses of merger-specific efficiencies, we examine a merger’s market share effects. Standard merger theory predicts that if merger-specific efficiencies are present, the merged firm should regain market share in the long run. We estimate short- and long-run merger effects on market shares from the divestiture of Texaco’s Canadian assets. Using a difference-indifference specification we compare changes for the merging firm against changes for other vertically integrated firms in the same market. A general equilibrium type effect renders our estimates biased but the sign of each effect is consistently estimated. Our approach is a useful complement to across-market comparisons, which are often hindered by the difficulty of finding control markets that experience the same supply and demand shocks as the treatment markets. © 2013 Elsevier Inc. All rights reserved.

1. Introduction The Department of Justice and the Federal Trade Commission in the U.S. have recently issued revised merger guidelines to reflect important developments in industrial organization over the past few decades. For the same reason, the Competition Bureau of Canada also developed new guidelines for its merger review process in 2004, with further modifications to be implemented in 2011. The basic tests, however, remain unchanged. The U.S. agencies focus on consumer surplus and on whether post-merger price increases are likely to be significant. In Canada the emphasis is on short-term (over a two-year period) changes to total surplus. Consistent with these foci in merger enforcement, most empirical merger studies have focused on mergers’ short-run effects and on short-run price effects in particular. When the policy objective is to prevent mergers that reduce consumer welfare, analysis of mergers’ price impacts is sufficient: if a merger increases prices, it will likely harm consumer welfare in the absence of significant impacts on quality and service. By contrast, from the perspective of total welfare, analysis of price effects

∗ Corresponding author. E-mail addresses: [email protected] (M. Packalen), [email protected] (A. Sen). 0144-8188/$ – see front matter © 2013 Elsevier Inc. All rights reserved. http://dx.doi.org/10.1016/j.irle.2013.04.003

alone is not sufficient: a price-increasing merger can still increase total welfare if synergies or other merger-specific efficiencies are present. Therefore, to prevent mergers that cannot increase total welfare relative to what would have occurred without the merger, an antitrust agency needs to know whether merger-specific efficiencies are present. One consequence of merger-specific efficiencies is that the merging firms regain market share as those efficiencies materialize, as we discuss in detail in Section 2.1.1 In this paper, we use this observation on the relationship between market shares and merger-specific efficiencies to conduct an empirical analysis of merger-specific efficiencies. We estimate both short and long-run merger effects on market shares from the acquisition of Texaco’s Canadian crude oil, wholesale, and retail assets by Imperial Oil (in Western and Central Canada) and Ultramar (in the Atlantic region) in July 1989 and October 1990, respectively, using firm-level panel data on 12 cities. Our contribution stems from our use of data that spans a time period that is long enough to detect merger-specific efficiencies at the firm level. The three existing studies (Ashenfelter

1 Some efficiencies associated with mergers can decrease the merging firms’ combined market share but are not specific to mergers – they could be achieved also through unilateral action and competition – and thus should not be considered as an efficiency-enhancing merger rationales (see Section 2.1).

M. Packalen, A. Sen / International Review of Law and Economics 36 (2013) 12–24

& Hosken, 2008; Borenstein, 1990; Coloma, 2002) that examine a merger’s market share impacts all employ data that span only four or fewer years, rendering it unfeasible for those studies to examine whether the mergers in question resulted in merger-specific efficiencies in the long run. Within-market analyses of mergers’ market share impacts are an important complement to empirical merger analyses that focus on price effects and employ across-market comparisons. Withinmarket comparisons of price effects are not always feasible or meaningful (due to data constraints or product homogeneity), and across-market comparisons of price effects come with the caveat that only known supply and demand shocks can be accounted for – a point emphasized by Taylor and Hosken (2007) and Simpson and Taylor (2008) (of course, shocks that are common to all markets and shocks that are constant within a market over time are exceptions). Our ability to study both short- and long-run merger effects at the firm level is important because: (1) the gestation lag in absorbing merger-induced efficiencies may be considerable and (2) the general ambiguity regarding the optimal time span to assess merger impacts. In comparison, most studies have relied on four or fewer years of data and have focused on the price effect. We discuss the related empirical literature in Section 2.2. An additional appealing feature of our analysis is that the motivation for Texaco’s exit was exogenous to local market characteristics as Texaco sold its Canadian assets to finance its long-running legal dispute with Pennzoil in the United States. This natural experiment type feature suggests a relatively clean opportunity to examine merger effects. The mergers were also of considerable scale. The number of national vertically integrated gasoline firms fell from four to three, and the mergers resulted in a significant increase in market concentration in most markets. Moreover, because regular grade gasoline is a relatively homogeneous product, comparisons across firms are more robust than in other industries. Our estimates from firm-level difference-in-difference specifications show that in the long run the mergers resulted in a decline in the merging firms’ combined market share relative to other vertically integrated firms in the same market, and that the magnitude of the long-run impact was much larger than the corresponding shortrun effect. These findings are important as they demonstrate that dynamic (long-run) merger effects can be very different from static (short-run) merger effects, and that dynamic impacts do not necessarily arise through potential efficiencies created by mergers.2 In summary, we view our study to be a useful contribution given the relatively thin empirical literature on merger effects (noted by Ashenfelter, Hosken, & Weinberg, 2009). The benefits stem from developing and using an empirical approach that does not suffer from the same set of same caveats as the existing strategies, and from the parsity of ex-post investigations on how mergers benefit the merging firms in the long run.3

2 We focus entirely on firm-level effects. In an earlier version of the paper, we also examined market level effects, including price effects. However, identification of the market-level effects relied on the comparison of geographically very different treatment and control markets. For this reason we omit these analyses (available upon request) in this version of the paper. 3 A relevant question is – what efficiencies did Imperial Oil and Ultramar hope to obtain from their acquisitions of Texaco’s assets? The acquisitions may have been part of an overall response to significant shocks that transformed the oil industry during 1980s. As documented by the Conference Board of Canada (2001) and Sen and Townley (2010), the gasoline industry significantly rationalized its retail network in response to crude oil price shocks, changes in consumer tastes, and the enhanced fuel efficiency of motor vehicles. Our results, however, suggest that the mergers did not lead to significant outlet rationalization. We return to the largely unknown motivations for these and other mergers at the end of the concluding section.

13

2. Related literature 2.1. Related theoretical literature Salant, Switzer, and Reynolds (1983) derive the surprising result that in the standard Cournot model with homogenous goods, linear demand, and constant marginal costs, mergers are unprofitable if the merging firms’ combined pre-merger market share is less than 80%. In response, Perry, Perry, and Porter (1985) note that models with constant average costs do not yield reliable analyses of mergers. They construct a model based on fixed industry capital stock, and show that mergers are often profitable in this setup. McAfee and Williams (1992) build on Perry et al. (1985) to explore the welfare effects and testable implications of mergers. A robust finding is that the market share of the merged firm is less than the merging firms’ combined pre-merger market share. Farrell and Shapiro (1990) demonstrate that absent synergies or other merger efficiencies, prices should increase. The possibility that synergies from mergers can be significant was raised by Williamson (1968) and Demsetz (1973), though their presence should not be taken for granted (e.g. White, 1987; Fisher, 1987). When synergies or other merger-induced efficiencies are present, they generally enable the merging firms’ to regain their combined market share (relative to a merger with no efficiencies) and can lead to an improvement in total economic welfare (e.g. Farrell & Shapiro, 1990). The prediction that merger efficiencies will be reflected in market shares is useful as it is often difficult to pin down a merger’s cost-impacts at the firm-level. Because the realization of synergies (or other efficiencies) is likely to occur mostly only in the longrun (e.g. Focarelli & Panetta, 2003), the relevant prediction is thus that a comparison of a merger’s short- and long-run market share impacts will yield an indication of the presence of merger-induced efficiencies. Of course, it is true that one form of efficiency sometimes associated with mergers, namely post-merger rationalization (the elimination of under performing assets), may instead lead to a further decrease in the merging firms’ combined market share. However, rationalization is an important category of non-synergy efficiencies, and non-synergy efficiencies can in general be achieved also through unilateral action and competition (Farrell & Shapiro, 2001). Therefore, if the merging firms do not regain market share post merger (in the long run), it is statistical evidence that the merger did not have any merger-specific efficiencies. Accordingly, to provide guidance for a merger policy that aims to block mergers which necessarily lead to a decrease in total economic welfare (i.e. have no potentially redeeming efficiency rationale) it is sufficient to determine which type of mergers does not lead to the merging firms regaining market share in the long run. In summary, standard merger theory indicates that (1) mergers may increase or decrease prices; (2) synergies and other efficiencies may be present but should not be taken for granted; (3) when efficiencies are present price effects alone are not sufficient to determine the merger impact on total economic welfare; and (4) if merger-specific efficiencies are present the merged firm regains market share in the long run.4 The ambiguity of the merger price

4 This ambiguity in predicted merger effects extends to the possibility of tacit collusion. Compte, Jenny, and Rey (2002), Kuhn and Motta (1999) and Vasconcelos (2005) examine this when a firm’s assets are defined in terms of its product variety, when firm assets are product capacities, and when assets owned by a firm determine its cost function, respectively. These three studies arrive at the same conclusion: a merger makes tacit collusion more difficult if the merger increases the asymmetry (in terms of firm size) among potentially colluding firms. This result is contrary to conventional wisdom that is based on theories ranging from conscious parallelism to explicit agreements and according to which mergers increase the likelihood of collusive activities.

14

M. Packalen, A. Sen / International Review of Law and Economics 36 (2013) 12–24

Table 1 Outcome variables and time span in empirical merger analyses. Research paper

Mergers

Time span

Outcome variable(s)

Mueller (1985) Borenstein (1990) Kim and Singal (1993) Prager and Hannan (1998) Vita and Sacher (2001) Mark (2002) Coloma (2002) Focarelli and Panetta (2003) Dranove and Lindrooth (2003) Hastings (2004) Hastings and Gilbert (2005) Taylor and Hosken (2007) Chouinard and Perloff (2007) Simpson and Taylor (2008) Ashenfelter and Hosken (2008) Dafny (2009) Winston, Mahesri, and Dennis (2009) Our analysis

Various Airlines Airlines Banks Hospitals Publishers Gasoline Banks Hospitals Gasoline Gasoline Gasoline Gasoline Gasoline Various Hospitals Railroads gasoline

1950 and 1972 3 years (1985–1987) 4 years (1985–1988) 3 years (1991–1994) 11 years (1986–1996) 7 years (1988–2001) 3 years (1998–2000) 9 years (1990–1998) 11 years (1988–1998) 1 year (1997) 2.5 years (1996–1998) 3 years (1997–1999) 9 years (1989–1997) 6 years (1997–2002) 2–4 years (1995–2000) 16 years (1985–2000) 18 years (1989–2006) 7–10 years (1988–1998)

Market share Price, market share, load factor, capacity Price Price (interest rate) Price, costs Price Price, quantity, market share Price (interest rate) Costs Retail price Wholesale price Retail, wholesale prices (mkt level) Retail, wholesale prices (mkt level) Retail price (mkt level) Price, market share Price Price, quantity (mkt level) Market shares, outlets (firm level)

effect and the ambiguity of whether efficiencies are present motivate empirical merger analyses. Our data does not allow us to identify the mergers’ price effects but does enable us to examine whether merger-specific efficiencies are present through an analysis of how the mergers impacted market shares over time (implication 4 above) 2.2. Related empirical literature Most empirical merger analyses examine transactions in the airline, banking, gasoline, and hospital industries. Borenstein (1990) evaluates the effects of two airline mergers in 1986. Kim and Singal (1993) examine 14 airline mergers from 1985 to 1988. Peters (2006) constructs merger simulation models, which successfully predict subsequent price and quantity changes following five airline mergers in 1986 and 1987. With respect to banking, Prager and Hannan (1998) examine horizontal U.S. bank mergers. Focarelli and Panetta (2003) study the short- and long-run deposit interest rate impacts of bank mergers in Italy in the 1990s. A number of recent papers examine merger impacts in the gasoline industry. Coloma (2002) examines the price and quantity impacts of a merger in Argentina. Hastings (2004) estimates the effect of a vertically integrated firm’s long-term lease of 260 gasoline stations in Southern California from an independent retailer in 1997 (see also Taylor, Kreisle, & Zimmermann, 2010). Hastings and Gilbert (2005) study the wholesale price effects of the merger between an upstream firm and a firm possessing both upstream and downstream operations. Taylor and Hosken (2007) examine the retail and wholesale price effects of the Marathon–Ashland joint venture in four cities. Simpson and Taylor (2008) estimate the effects of Marathon–Ashland’s acquisition of Ultramar’s assets in 1999 in Michigan. Chouinard and Perloff (2007) examine the relative price impacts of mergers, crude oil prices, taxes and regulations. Research most similar to ours is Mueller (1985), which uses data on 1000 companies in 1950 and 1972 to estimate the merger effect on market shares from 209 acquisitions. While Mueller (1985) uses national market share data in various industries, we use cityspecific market share data on retail gasoline and compare market shares for merging vs. non-merging vertically integrated firms in the same city. Moreover, because the data in Mueller (1985) covers just two points in time, only our study can compare short- vs. longrun outcomes to examine whether merged firms regain market share as efficiencies materialize. Our study and Mueller (1985) are

nevertheless similar relative to the rest of the literature, being the only studies that use market share data to examine merger-specific efficiencies. Table 1 summarizes the related literature except the contributions that focus mostly on the development of merger simulation strategies (such as Nevo, 2000; Pinske & Slade, 2001) and some contributions in the extensive literature on hospital mergers. Column 3 shows that with the exception of hospital merger analyses, the time span is short in most merger studies. Analysis of merger-specific efficiencies is thus typically infeasible. Column 4 in turn reveals that only five of these studies have examined merger impacts on variables other than price. Moreover, only four studies have examined mergers’ market share impacts, and with the exception of Mueller (1985) the time-span in these studies is 4 or fewer years whereas the time-span in our study is 7–10 years. 3. The divestiture of Texaco’s Canadian assets The immediate motivation for the divestiture of Texaco’s Canadian assets was Texaco Incorporator’s filing for bankruptcy in April 1987. The underlying reason for the bankruptcy was Texaco’s loss in its legal battle against Pennzoil.5 As part of its effort to emerge from bankruptcy, Texaco sold $6.8 billion dollars worth of assets between April 1988 and January 1989, including its ownership share of Texaco Canada.6 Texaco Incorporated owned 78% of Texaco Canada. These shares were sold to Imperial Oil Limited, the Canadian subsidiary of the Exxon Corporation, in January 1989. Imperial Oil subsequently purchased all shares of Texaco Canada. The total cost of the acquisition was approximately $4.96 billion.7 Immediately following the acquisition of Texaco Canada by Imperial Oil, the Director of Research and Investigation of the Canadian Competition Bureau (“Director”) and Imperial Oil agreed that until the Director had identified assets that needed to be divested for the merger to be otherwise approved, Imperial Oil would keep its own downstream operations (the Esso branded retail stations) separate from the Texaco branded downstream operations that it

5

See “Texaco Files to End Bankruptcy,” New York Times, December 22nd, 1987. See “Texaco Plans Sale to Exxon,” New York Times, January 21st, 1989. See paragraphs 9 and 10 in Director of Investigation and Research v. Imperial Oil Ltd. (29 June 1989), CT8903, Notice of Application. http://www.ct-tc.gc. ca/CMFiles/CT-1989-003 0001 38LBI-532004-3652.pdf (Last accessed 25 August 2010). 6 7

M. Packalen, A. Sen / International Review of Law and Economics 36 (2013) 12–24

had just acquired.8 This agreement was in place until the Consent Interim Order was issued on July 4th in 1989.9 In Western and Central Canada the Consent Interim Order no longer required Imperial Oil to keep it’s pre-existing downstream ‘Esso’ branded operations separate from the downstream operations that it had acquired from Texaco. Given the fact that the Consent Interim Order was issued on July 4th, 1989, we assume that Texaco Canada’s operations in Western and Central Canada were merged with Imperial Oil’s operations from July 1989 onwards. The Consent Interim Order still explicitly prohibited Imperial Oil from combining its operations with assets previously owned by Texaco in Atlantic Canada or even exchanging several types of information, including information on pricing.10 The Consent Interim Order remained in place until the Consent Order was issued on February 6th, 1990. The Consent Order directed Imperial Oil to divest many of the assets it had acquired in Atlantic Canada as part of its acquisition of Texaco Canada, including all of the Texaco branded retail stations. The divestiture was to be completed within 12 months of the issuance of the Consent Order and the sale was subject to approval of the Director. In August 1990 it was announced that Ultramar had agreed to purchase Texaco’s operations in Atlantic Canada from ImperialOil.11 The Director approved the deal sometime between

8 See paragraph 11 in Director of Investigation and Research v. Imperial Oil Ltd. (29 June 1989), CT8903, Notice of Application. http://www.ct-tc.gc.ca/CMFiles/ CT-1989-003 0001 38LBI-532004-3652.pdf (Last accessed 25 August 2010). The Director of Investigation and Research of the Competition Bureau of Canada filed a notice of application under Sections 92 and 105 under the Competition of Canada (R.S.C. 1984, C – 34) alleging that the acquisition of Texaco by Imperial Oil would result in a substantial lessening of competition in the wholesale and retail markets for refined petroleum products, particularly gasoline. The Competition Tribunal agreed with some of the Director’s concerns and allowed the merger to proceed under the several conditions. The most substantial of these conditions was the divestiture of most of Texaco’s retail outlets in Atlantic Canada and the Eastern Passage Refinery. The other conditions were (1) the divestiture of 68 retail outlets in Quebec; (2) a further divestiture of 346 outlets across Western and Central Canada; (3) the divestiture of nine wholesale terminals in Ontario, Quebec, and Atlantic Canada; (4) the divestiture of the Eastern Passage Refinery and (5) a guarantee that the merged entity would not refuse to supply independent retailers with no direct access to refining in Ontario and Quebec. (See Director of Investigation and Research v. Imperial Oil Ltd. (6 February 1990), CT8903, Consent Order. http://www.ct-tc.gc.ca/CMFiles/CT-1989-003 0397c 38KPP-1142005-9533.pdf (Last accessed 25 August 2010)). In principle the additional divestitures required by the consent order might imply that our estimates of merger effects on outlets and market shares would merely reflect these consent order required additional divestitures rather than post-merger competition. However, our results show that the mergers had little impact on outlets. Hence, it is unlikely that the additional divestiture requirement in the consent order had a significant impact on outlets either in the short-term or in the longterm. Moreover, according to the consent order, these additional divestitures were to be completed by February 1991 (within 12 months of the consent order). Consequently, if our estimates of the merger effects on market shares are driven by these consent order required additional divestitures, the merger effects should be already present for the period subsequent to February 1991. However, additional analyses (see the end of Section 6.1) show that this is not the case – the merger effects are not yet present in the period subsequent to the period when the consent order required additional divestitures were to be completed. Additionally, our main result (an eventual decrease in the merging firms’ combined market share) is not limited to the markets affected by these consent order required additional divestitures as our market-specific analyses (see Section 6.1) indicate that the result applies also to the two markets in Atlantic Canada in which a non-trivial merger occurred. 9 See paragraph 11 in Director of Investigation and Research v. Imperial Oil Ltd. (7 July 1989), CT8903, Consent Order Impact Statement. http://www.ct-tc.gc. ca/CMFiles/CT-1989-003 0013 38LCS-532004-8052.pdf (Last accessed 25 August 2010). 10 See paragraph 6 in Director of Investigation and Research v. Imperial Oil Ltd. (4 July 1989), CT8903, Consent Interim Order. http://www.ct-tc.gc.ca/ CMFiles/CT-1989-003 0010 38NNQ-532004-7536.pdf (Last accessed 25 August 2010).

15

September 24th 1990 and October 3rd 1990.12 Based on this information we set October 1st 1990 as the date of merger between Ultramar and the Atlantic assets previously owned by Texaco Canada. The exception is St. John’s where we determine the data that Texaco’s assets were acquired by G.E.O. on October 1st 1990.13 These series of events have important implications for the empirical analysis. The fact that Texaco exited the Canadian market because of legal disputes in the United States indicates that the divestiture of Texaco’s Canadian assets was by and large an event that was exogenous to local market conditions. Moreover, both the difference in the identity of the acquiring firm across cities and the difference in the timing of the merger across cities provide opportunities to control for unobserved market, time, and firm-specific shocks.

4. Data 4.1. Data sources Data on retail gasoline sales of regular grade gasoline for each firm and the number of retail outlets owned by them in 12 Canadian cities (Calgary, Charlottetown, Halifax, Montreal, Ottawa, Quebec, Regina, Saint John, St. John’s, Toronto, Vancouver and Winnipeg) from January 1988 to August 1998 were obtained from Kent Marketing.14 The 12 cities represent a geographically comprehensive sample of Canadian cities. The 12 cities also capture over 65% of the Canadian population. To alleviate concerns that the findings are dependent on the selection of these 12 cities we also present market-specific difference-in-difference estimates of the merger impacts on market shares and outlets. Consistent with previous studies on the Canadian gasoline industry (Conference Board of Canada, 2001) we assume that each city forms its own market. Depending on the city and time period, the sales and outlet data are either monthly or bimonthly. We convert these data into monthly data. While the conversion to monthly data introduces correlation across observations within a market, it does not artificially inflate the results’ statistical significance because we cluster standard errors by market From the data on firm-level sales we calculate the market share of each firm in each

11

See “Company Briefs” in the New York Times on August 18th, 1990. See page 9 in Director of Investigation and Research v. Imperial Oil Ltd. (10 November 1993), CT8903, Reasons for Decision Regarding Jurisdiction over undertakings. http://www.ct-tc.gc.ca/CMFiles/CT-1989-003 463 38OII-532004-7977.pdf (Last accessed 25 August 2010). 13 Our data on gasoline retail sales (see Section 4) indicate that the former Texaco retail stations in the Atlantic Region were re-branded as Ultramar stations sometime between November 1990 and February 1991 except in St. John’s where Texaco retail stations were rebranded as G.E.O. stations. 14 This data set on market shares in these 12 Canadian cities was obtained in 1999 for the Canadian Competition Bureau by one of the authors. At the time only data for these 12 cities was available. Kent marketing is a private firm in London, Ontario. Kent Marketing collects this information during onsite visits to each retail outlet located in a Municipal Statistical Area (MSA) in Canada. We have no data on firm sales or number of outlets for Ottawa in September 1995. For each firm in this market we impute these values to equal the average value of the corresponding variables in August and October of the same year for the firm in this city. We do not have observations on firm sales or the number of outlets for Saint John from April 1990 to November 1990. For the missing observations preceding the merger (April 1990 to September 1990 in Saint John) we impute the corresponding values of the last pre-merger value (March 1990) of the variable. For the missing observations following the merger (October 1990 to November 1990) we impute the first post-merger value (December 1990) of the corresponding variable. 12

16

M. Packalen, A. Sen / International Review of Law and Economics 36 (2013) 12–24

CHARLOTTETOWN (ULTRAMAR) HALIFAX (ULTRAMAR)

MONTREAL (ESSO)

OTTAWA (ESSO)

QUEBEC (ESSO)

REGINA (ESSO)

SAINTJOHN (ULTRAMAR)

STJOHN (G.E.O.)

TORONTO (ESSO)

VANCOUVER (ESSO)

WINNIPEG (ESSO)

.3 .2 .1 0 0

.1

.2

.3

Market Share

0

.1

.2

.3

CALGARY (ESSO)

1988

1999 1988

1999 1988

1999 1988

1999

Time

Observed Market Share of the Acquring Firm Average Pre−Merger Market Share of the Acquiring Firm Predicted Combined Post−Merger Market Share

Fig. 1. Pre- and post-merger market share of the acquiring firm in each market.

city and the corresponding market specific Herfindahl–Hirschman Index (HHI).15,16 4.2. Acquiring firms’ market shares and concentration over time Fig. 1 depicts the actual market share of the acquiring firm over time in each market as well as its average pre-merger and predicted post-merger market share. The acquiring firm in each market is shown in parentheses in the title of each subplot. In all 8 markets in Central and Western Canada the acquiring firm is Esso. In 3 of the 4 markets in Atlantic Canada the acquiring firm is Ultramar. In one market in Atlantic Canada the acquiring firm is G.E.O. The average pre-merger market share of a firm is calculated as the average market share of the firm in the 18 months preceding the merger. The predicted post-merger market share of the acquiring firm is calculated as the sum of the acquiring firm’s average pre-merger

15 With respect to Canadian research on gasoline prices, Sen (2003, 2005) estimates the effects of market concentration on retail prices but uses only (post-merger) data from 1991 onwards. Some studies suggest limited variation in retail prices across stations owned by vertically integrated firms within Canadian cities at a point in time or within a short time span (Atkinson, Eckert, & West, 2009). Analyses that have employed weekly, daily, or hourly Canadian data (Eckert, 2002, 2003; Eckert & West, 2004a, 2004b; Noel, 2007a, 2007b), have typically relied on variation within short time-periods and a small number of cities. Houde (2012) studies the effects of commuting patterns on spatial differentiation using station level data (between 1991 and 2006) from Quebec. Carranza, Clark, and Houde (2011) employ station level data from twin cities in Ontario and Quebec to study the effects of price floor regulations on local competition, prices, and productivity. Erutku (2007) studies patterns in Canadian gas prices before long weekends. Erutku and Hildebrand (2010) evaluates collusion and price fixing in rural towns in Quebec. 16 For the secondary specifications that we employ we also obtained monthly data on population and unemployment in each city from Statistics Canada in order to account for demand side shocks. For Charlottetown these variables are measured by their values for the Prince Edward Island. We also experimented with the use of other city and province level covariates such as average household income and government transfers to low income households. Inclusion of these covariates did not alter our results. For these alternative specifications also data on monthly crude oil prices over time were retrieved from the Petroleum Resources Branch of Natural Resources Canada in order to account for input price fluctuations. The data are available at http://nrcan.gc.ca/eneene/sources/pripri/crubru-eng.php. There is very little variation between the Edmonton and Montreal Brent Par at any given time. We use the Edmonton Brent Par for all markets.

market share and the acquired firm’s (Texaco) average pre-merger market share. Fig. 1 shows that in most cities the acquiring firm’s pre-merger market share is between 10% and 20%.17 The notable exception is St. John’s, where the acquiring firm’s average pre-merger market share was 0% as the acquirer, G.E.O., had no prior presence in that market. Accordingly, we treat this market in the empirical analyses as if there was no merger in that market. Comparison of the acquirer’s average pre-merger market shares and the predicted combined post-merger market share in Fig. 1 shows that Texaco’s average pre-merger market share is around 10% in most cities. The notable exception is Saint John, where Texaco’s pre-merger market share is only 2%. Fig. 1 also demonstrates that initially after the merger, the actual market share of each acquiring firm follows the predicted market share closely in all cities. However, over time the actual market shares dip below the predicted values for all cities except Saint John. Fig. 2 depicts the actual HHI, the average pre-merger HHI and the predicted HHI for each city during the sample period. The average pre-merger HHI is calculated as the average HHI for the 18 months preceding the merger. The post-merger predicted HHI is calculated from the average pre-merger market shares by adding Texaco’s average pre-merger market share to the acquiring firm’s average pre-merger market share. Fig. 2 shows that there is considerable variation in concentration across cities and also within each city over time. Similar to what we observed for firm specific market shares (Fig. 1), the merger had a significant – at least a 100 point increase in HHI – impact on concentration in most cities. The two exceptions are again Saint John and St. John’s. Moreover, in 5 cities (Charlottetown, Halifax, Regina, Toronto, Vancouver) out of the 10 remaining cities the post-merger HHI was above 1500, which the Federal Trade Commission and Department of Justice (2010) merger guidelines cite as the threshold for moderately concentrated markets (the Canadian Horizontal Merger Guidelines do not use HHI). While post merger trends right after the merger reveal a close alignment between the actual HHI and predicted post-merger

17 Our agreement with Kent Marketing precludes the reporting of individual firmspecific market shares.

M. Packalen, A. Sen / International Review of Law and Economics 36 (2013) 12–24

CHARLOTTETOWN

HALIFAX

MONTREAL

OTTAWA

QUEBEC

REGINA

SAINTJOHN

STJOHN

TORONTO

VANCOUVER

WINNIPEG

1000 3000 1000 3000 1000

Hirschman−Herfindahl Index

3000

CALGARY

17

1988

1999 1988

1999 1988

1999 1988

1999

Time Observed HHI Average Pre−Merger HHI Predicted Post−Merger HHI

Fig. 2. Pre- and post-merger Hirschman–Herfindahl Index in each market.

HHIs, for most markets the actual HHI then dips below the predicted post-merger HHI. In some cities the actual HHI eventually surpasses the predicted post-merger HHI.18 In summary, Fig. 1 implies that over time merging firms lost market share in most markets, and Fig. 2 suggests that in most cities the long-run effect of the mergers on concentration was more limited than the short-run effect. Figs. 1 and 2 also show that Saint John and St. John’s are outliers in several respects as (i) the mergers had little or no impact on concentration in these two markets, (ii) Texaco’s pre-merger market in Saint John was very small, and (iii) in the beginning of the sample period, the HHI was much higher for Saint John than for any other city. For these reasons, we also show the regression results for the case when Saint John and St. John’s are excluded from the sample.

For the firm j that acquired Texaco’s assets in market i the variable MARKETSHAREj,i,t denotes the combined market share of the acquiring firm j and Texaco in market i at time t. Thus, for that firm in market i the variable MARKETSHAREj,i,t is calculated as the sum of Texaco’s market share and the acquiring firm’s market share both before and after the merger. For all other firms in market i the variable MARKETSHAREj,i,t simply denotes the market share of the firm

5. Empirical model

long-run merger effects, respectively.20 The market-specific time and firm fixed effects ˛ij and ˛it control for unobservable characteristics that vary across markets but are constant within each market either across time or across firms.21 We estimate model (1) using all observations on vertically integrated firms. The parameters of interest ˇ1 and ˇ2 are identified by the variation within a given market across the vertically integrated firms in whether a given firm acquired Texaco’s assets in that market. Obviously, in each market only one vertically integrated firm acquired Texaco’s assets while the other vertically integrated firms in that market did not acquire Texaco’s assets. If the true merger impact is that a merging firm loses market share, then mergers lead to market share gains for non-merging firms. Consequently, the results obtained using specification (1) will overstate the absolute size of the true merger impact. However, in our analysis the

In our empirical specification, we compare changes in outcomes within a given market for the firm that acquired Texaco’s assets in that market with changes in outcomes for all other vertically integrated firms in the same market.19 With market share as the dependent variable, this specification is formally written as ln MARKETSHARE j,i,t = ˇ1 MERGE 0−2 + ˇ2 MERGE 3+ + ˛ij j,i,t j,i,t + ˛it + uj,i,t .

(1)

18 We also ran some simple market-level OLS regressions with the HHI as the dependent variable, using monthly data from January 1988 to December 1987 in the 12 cities. We created two merger dummy variables to assess the effects of the merger: for a given market-time period combination the short run (long run) dummy variable equals 1 if the merger was completed in that market no more than two years prior (more than two years prior) to the time period. The specification was estimated using city specific unemployment rates and population as controls along with city, month, and year dummy variables. The detailed results are available on request; for the sake of brevity we only note that point estimates of the shortand long-run merger impacts on concentration are 10% and 15%, respectively, and statistically significant. While one cannot place much emphasis on these findings as the identification relies on time-series variation, they suggest that mergers did impact the distribution of market shares across firms also in the long run. 19 We include only vertically integrated firms in set of the control firms so that the set of control firms is comparable with the treatment firms (Esso in some markets, Ultramar in other markets). The vertically integrated firms are Esso, Ultramar, PetroCanada, Shell, Chevron, Co-op (West), Husky, Irving, and Turbo.

j in market i at time t. The dummy variable MERGE 0−2 equals 1 if j,i,t firm j was involved in a merger (acquired Texaco’s assets) in market i no more than two years prior to time t, and is zero otherwise. The dummy variable MERGE 3+ equals 1 if firm j participated in a merger j,i,t (acquired Texaco’s assets) in market i and the merger was completed more than two years prior to time t, and is zero otherwise. and MERGE 3+ thus capture the short- and The variables MERGE 0−2 j,i,t j,i,t

20 Focarelli and Panetta (2003) also use a two-year window to examine short-run effects. This approach is consistent with the Merger Enforcement Guidelines of the Competition Bureau of Canada, which notes a 2-year time period for assessing the competitive effects of a merger. See “Merger Enforcement Guidelines,” Competition Bureau of Canada, September 2004 (available at http://www.competitionbureau.gc.ca/eic/site/cb-bc.nsf/eng/01245.html, last accessed 28 October 2009). 21 As the data are monthly, for each market 12 × T separate dummy variables represent the time fixed effects, where T is the number of years in the sample.

18

M. Packalen, A. Sen / International Review of Law and Economics 36 (2013) 12–24

Table 2.1 Regression results on market share. Dependent variable: ln MARKETSHAREj,i,t Panel A (1988–1994)

(1) All observations on vertically integrated firms* in all 12 markets

(2) Observations with small avg. market share excluded** in all 12 markets

(3) All observations on vertically integrated firms* in 10 markets***

(4) Observations with small avg. market share excluded** in 10 markets***

MERGE 0−2 j,i,t

0.03 [0.03] pwild = 0.365 −0.11 [0.07] pwild = 0.067 Market × Time, Market × Brand

0.01 [0.02] pwild = 0.596 −0.13 [0.07] pwild = 0.029 Market × Time, Market × Brand

0.01 [0.02] pwild = 0.681 −0.14 [0.07] pwild = 0.009 Market × Time, Market × Brand

0.01 [0.02] pwild = 0.629 −0.13 [0.06] pwild = 0.021 Market × Time, Market × Brand

0.96 29.13 pwild = 0.000 5470 Yes****

0.91 21.51 pwild = 0.001 3780 No

0.95 29.96 pwild = 0.000 4714 Yes****

0.88 21.42 pwild = 0.001 3276 No

MERGE 3+ j,i,t

Fixed effects 2

R F-statistic on H0 : ˇ1 = ˇ2 Observations Observation weights

Panel B (1988–1998)

(1)

(2)

(3)

(4)

MERGE 0−2 j,i,t

0.03 [0.04] pwild = 0.494 −0.19 [0.10] pwild = 0.018 29.95 pwild = 0.001 8309

0.00 [0.03] pwild = 0.865 −0.23 [0.11] pwild = 0.008 25.48 pwild = 0.002 5760

0.00 [0.02] pwild = 0.971 −0.22 [0.10] pwild = 0.005 31.93 pwild = 0.000 7157

0.00 [0.03] pwild = 0.903 −0.23 [0.10] pwild = 0.003 25.36 pwild = 0.001 4992

MERGE 3+ j,i,t

F-statistic on H0 : ˇ1 = ˇ2 Observations

Notes: Wild cluster-bootstrapped standard errors in brackets (clustered by market). Inference is based on pwild , which is calculated using the wild cluster-bootstrapped distribution of the t-statistic (see Section 6). (*) All observations on all vertically integrated firms in all markets. (**) Same sample as in column (1) except observations on a firm are excluded in markets where the firm’s average market share during the sample period is less than 10%. (***) The two excluded markets are St. John’s and Saint John where the acquirer’s pre-merger market share was small or non-existent. (****) In each market the observations for a firm are weighted by the firm’s average market share in the market.

magnitude of this overstatement is limited because there were still at least 4 vertically integrated firms operating in each market even after the mergers. Thus, in expectation each non-merging vertically integrated firm would be expected to gain at most onethird of the market share that the merging firm loses.22 Caveats associated with such and related general equilibrium concerns are certainly not unique to our difference-in-difference analysis and, while here this caveat implies that the magnitude of the merger impact is slightly overestimated, the sign of the merger effect is consistently estimated. We also estimate model (1) using the proportion of outlets affiliated with a firm as the dependent variable (ln OUTLETSHAREj,i,t ). This outlet-based alternative measure of market share allows us to investigate whether the mergers’ resulted in outlet rationalization.23 6. Results For each analysis we report two sets of estimates: the results when only data up to year 1994 is included (panel A) and the results when the whole sample period is used (panel B). Although in each panel B we only report the estimates of the coefficients of interest,

22 For example, if merging firms always lose 10% points of market share to other vertically integrated firms and post-merger there are 5 vertically integrated firms in each market, a firm loses 10% points of market share in markets where it merged and gains only 10/5−1 = 2.5 % points of market share in markets in which another firm merged. 23 Using ln(OUTLETSHAREj,i,t /POPi,t ) instead as the dependent variable, where POPi,t denotes population in market i at time t would of course yield the same estimates because using specification (1) the estimates are obtained by changes across firms within a given market.

the empirical model in each column of panel B is the same as the empirical model in the corresponding column of panel A. Due to the presence of persistence in the unobserved factors, the estimates are less precise in several cases when the whole sample period is used compared to when data only up to year 1994 is used.24 However, this happens much less frequently than we anticipated. The estimates are ordinary least squares estimates, unless otherwise noted. Standard errors are clustered by market. Inference is based on the p-value denoted pwild , which is calculated using the wild cluster-bootstrapped distribution of the t-statistic. Monte Carlo evidence favors the use of pwild over asymptotic p-values when the number of clusters is small (Cameron, Gelbach, & Miller, 2007). Unlike several other cluster-bootstrap methods, the wild cluster-bootstrap is feasible also when clusters are unbalanced, which is the case in our firm-level analyses. The wild clusterbootstrapped standard errors are presented in brackets. 6.1. Results: market shares Table 2.1 reports the results when the dependent variable is market share as measured from the amount of regular grade gasoline sold. As we employ specification (1), the estimates are identified by comparing changes in the merging firms’ combined market share in a market with the corresponding changes for all other vertically integrated firms in the same market. Because we expect the variation in the error term to be inversely related with the market share of the firm, we weight observations on each firm

24 These persistent unobserved factors include: (i) other mergers such as PetroCanada’s acquisition of G.E.O.’s assets in St John’s in 1996, (ii) the gradual privatization of Petro-Canada that began in the 1990s, and (iii) the outlet rationalization that began in the 1980s.

Table 2.2 Market-specific regression results on market share. Dependent variable: ln MARKETSHAREj,i,t (1) Calgary

(2) Charlottetown

(3) Halifax

(4) Montreal

(5) Ottawa

(6) Quebec

(7) Regina

(8) Toronto

(9) Vancouver

(10) Winnipeg

MERGE 0−2 j,i,t

0.01 [0.02] pwild = 0.547 −0.09 [0.08] pwild = 0.322 Time, Brand

−0.05 [0.08] pwild = 0.651 −0.39 [0.32] pwild = 0.286 Time, Brand

0.06 [0.05] pwild = 0.329 −0.05 [0.06] pwild = 0.489 Time, Brand

−0.00 [0.01] pwild = 0.871 −0.15 [0.11] pwild = 0.307 Time, Brand

0.12 [0.09] pwild = 0.211 0.08 [0.07] pwild = 0.370 Time, Brand

−0.05 [0.09] pwild = 0.612 −0.25 [0.20] pwild = 0.236 Time, Brand

−0.04 [0.05] pwild = 0.539 −0.23 [0.20] pwild = 0.353 Time, Brand

0.00 [0.03] pwild = 0.944 −0.10 [0.10] pwild = 0.493 Time, Brand

0.04 [0.03] pwild = 0.267 −0.07 [0.06] pwild = 0.488 Time, Brand

0.07 [0.07] pwild = 0.307 −0.10 [0.09] pwild = 0.360 Time, Brand

Observations Observation weights

0.94 2.63 pwild = 0.275 504 Yes****

0.93 83.91 pwild = 0.056 420 Yes****

0.95 11.50 pwild = 0.291 420 Yes****

0.98 11.85 pwild = 0.212 504 Yes****

0.98 0.32 pwild = 0.685 430 Yes****

0.90 10.09 pwild = 0.322 504 Yes****

0.90 4.35 pwild = 0.333 504 Yes****

0.97 1.64 pwild = 0.503 420 Yes****

0.99 52.67 pwild = 0.179 504 Yes****

0.97 3.09 pwild = 0.306 504 Yes****

Panel B (1988–1998)

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

(9)

(10)

MERGE 0−2 j,i,t

0.01 [0.02] pwild = 0.539 −0.16 [0.15] pwild = 0.272 2.46 pwild = 0.260 768

−0.05 [0.08] pwild = 0.599 −0.54 [0.44] pwild = 0.232 27.71 pwild = 0.241 640

0.06 [0.05] pwild = 0.382 −0.17 [0.14] pwild = 0.213 37.02 pwild = 0.054 640

−0.00 [0.01] pwild = 0.905 −0.19 [0.16] pwild = 0.223 5.49 pwild = 0.220 768

0.12 [0.08] pwild = 0.203 0.03 [0.05] pwild = 0.557 2.18 pwild = 0.469 650

−0.05 [0.09] pwild = 0.584 −0.30 [0.24] pwild = 0.229 14.70 pwild = 0.225 768

−0.04 [0.05] pwild = 0.505 −0.30 [0.28] pwild = 0.312 2.16 pwild = 0.297 768

0.00 [0.03] pwild = 0.894 −0.09 [0.10] pwild = 0.349 1.37 pwild = 0.456 619

0.04 [0.03] pwild = 0.262 −0.15 [0.12] pwild = 0.349 54.72 pwild = 0.093 768

0.07 [0.07] pwild = 0.245 −0.20 [0.17] pwild = 0.283 5.51 pwild = 0.279 768

MERGE 3+ j,i,t

Fixed effects 2

R F-statistic on H0 : ˇ1 = ˇ2

MERGE 3+ j,i,t

F-statistic on H0 : ˇ1 = ˇ2 Observations

M. Packalen, A. Sen / International Review of Law and Economics 36 (2013) 12–24

Panel A (1988–1994)

Notes: Wild cluster-bootstrapped standard errors in brackets (clustered by brand). Inference is based on pwild , which is calculated using the wild cluster-bootstrapped distribution of the t-statistic (see Section 6). Regressions in this table are market-specific versions of the regression reported column 1 of Table 2.1. Hence, observations in each column consist of the observations on all vertically integrated firms in that market, and in each market the observations for each firm are weighted by the firm’s average market share in that market.

19

20

M. Packalen, A. Sen / International Review of Law and Economics 36 (2013) 12–24

Table 3.1 Regression results on outlets. Dependent variable: ln OUTLETSHAREj,i,t Panel A (1988–1994)

(1) All observations on vertically integrated firms* in all 12 markets

(2) Observations with small avg. market share excluded** in all 12 markets

(3) All observations on vertically integrated firms* in 10 markets***

(4) Observations with small avg. market share excluded** in 10 markets***

MERGE 0−2 j,i,t

0.01 [0.01] pwild = 0.415 −0.04 [0.05] pwild = 0.521 Market × Time, Market × Brand

−0.00 [0.02] pwild = 0.816 −0.04 [0.05] pwild = 0.489 Market × Time, Market × Brand

0.00 [0.01] pwild = 0.747 −0.05 [0.05] pwild = 0.394 Market × Time, Market × Brand

−0.00 [0.02] pwild = 0.809 −0.04 [0.05] pwild = 0.461 Market × Time, Market × Brand

0.97 1.03 pwild = 0.354 5473 Yes****

0.92 0.50 pwild = 0.533 3780 No

0.96 1.04 pwild = 0.371 4717 Yes****

0.94 0.50 pwild = 0.550 3276 No

MERGE 3+ j,i,t

Fixed effects 2

R F-statistic on H0 : ˇ1 = ˇ2 Observations Observation weights

Panel B (1988–1998)

(1)

(2)

(3)

(4)

MERGE 0−2 j,i,t

0.02 [0.02] pwild = 0.282 −0.04 [0.06] pwild = 0.467 1.07 pwild = 0.330 8312

0.00 [0.02] pwild = 0.988 −0.05 [0.06] pwild = 0.406 0.72 pwild = 0.402 5760

0.01 [0.01] pwild = 0.463 −0.05 [0.05] pwild = 0.409 1.01 pwild = 0.331 7160

−0.00 [0.02] pwild = 0.986 −0.05 [0.06] pwild = 0.416 0.72 pwild = 0.381 4992

MERGE 3+ j,i,t

F-statistic on H0 : ˇ1 = ˇ2 Observations

Notes: Same as notes to Table 2.1 (only change is the dependent variable)

in each market by the firm’s average market share in that market over the sample period.25 The weighted least squares estimates are reported in the odd-numbered columns. In the even-numbered columns we report the unweighted least squares results when observations on each firm are excluded in markets where the firm’s average market share is less than 10% during the sample period. Columns 1 and 2 report results when data on all 12 markets is used. Columns 3 and 4 report results when the two markets (St. John’s and Saint John) where the acquiring firm’s pre-merger market share was small or non-existent (see the discussion in Section 4.2) are excluded.26 Point estimates of the short-run effect on the merging firms’ combined market share range from 0% to 3% and are statistically insignificant. In contrast, the estimates indicate that in the long run the mergers had a large and statistically significant negative impact on the merging firms’ combined market share. Point estimates of the long-run effect are between −11% and −23%. With one exception (column 1 of Panel A) this long-run impact is statistically significant at the 5% level. In all cases the hypothesis that the short

25 For a firm with a small market share in a given market we expect more of the variation in the (logarithm of the) firm’s market share to be due to unobserved factors compared to a firm with a larger market share in that market. 26 As there was no merger in the St. John’s market (see Section 4.2), the inclusion of data for St. John’s in columns 1 and 2 influences neither the estimated coefficients nor their significance (the latter feature is due to bootstrapping). The St. John’s market is included in the analyses because in previous versions of this paper we also employed across-market comparisons. We also estimated two alternative versions of model (1) using data on all 12 markets. These alternative versions do not include the full set of fixed effects. Therefore, these analyses also rely on time-series variation for identification. In the first alternative we use year and month fixed effects as well as a market-specific time trend in place of time fixed effects. In the second alternative we use only month fixed effects in place of time fixed effects. These results (omitted here) were very similar to the corresponding results discussed above for the difference-in-difference specification (1).

and long-run effects on market share are the same can be rejected, as is indicated by the F-statistics on the null hypothesis ˇ1 = ˇ2 . In Table 2.2 are shown the market-specific estimates for the firm-level specification (1). For each market the short- and longrun merger effects on market shares are estimated by comparing changes for the merging firm in that market with changes for all non-merging vertically integrated firms in the same market. The results are market-specific versions of the results shown earlier in column 1 of Table 2.1. Coefficient estimates in both panels of Table 2.2 again indicate a clear (although now in most cases a statistically insignificant) difference between the short and long-run effects. None of the short-run estimates are below −0.05 in either panel A or panel B. In contrast, 9 out of 10 long-run estimates in panel A (panel B) are below −0.05 (below −0.09). In addition we note that, perhaps due to the small number of cities, comparison of these market-specific results and the level of concentration in each market (as measured by the HHIs shown in Fig. 2), does not reveal a strong relationship between concentration and merger effects on market shares.27 A potential concern is that our results might be confounded by other unobserved events that impacted the acquiring firms (or the industry in general) and also coincided with the timing of the acquisitions. One such possible factor was the rationalization of retail outlets that occurred during the 1980s and 1990s (see e.g. Sen & Townley, 2010). However, this strategy was adapted by all vertically integrated firms and hence should not bias our estimates

27 Mergers that result from antitrust authority ordered divestitures might have different efficiency impacts than other mergers. As the market-specific results in Table 2.1 suggest, and as formal tests not reported here have confirmed, for the mergers analyzed here there is no differential merger impact in markets where the mergers were the result of antitrust authority ordered divestitures (Charlottetown and Halifax in Atlantic Canada) compared to markets where the mergers were planned (the 8 cities in Central and Western Canada). Our results thus suggest that neither type of merger led to merger-specific efficiencies.

Table 3.2 Market-specific regression results on outlets. Dependent variable: ln OUTLETSHAREj,i,t . (1) Calgary

(2) Charlottetown

(3) Halifax

(4) Montreal

(5) Ottawa

(6) Quebec

(7) Regina

(8) Toronto

(9) Vancouver

(10) Winnipeg

MERGE 0−2 j,i,t

0.03 [0.02] pwild = 0.308 −0.08 [0.08] pwild = 0.434 Time, Brand

−0.10 [0.14] pwild = 0.483 −0.15 [0.17] pwild = 0.421 Time, Brand

−0.01 [0.02] pwild = 0.545 0.02 [0.06] pwild = 0.843 Time, Brand

−0.02 [0.01] pwild = 0.213 −0.01 [0.04] pwild = 0.860 Time, Brand

0.04 [0.04] pwild = 0.349 0.14 [0.12] pwild = 0.463 Time, Brand

0.02 [0.08] pwild = 0.890 −0.07 [0.09] pwild = 0.475 Time, Brand

0.02 [0.05] pwild = 0.629 −0.38 [0.26] pwild = 0.052 Time, Brand

−0.03 [0.02] pwild = 0.240 −0.02 [0.01] pwild = 0.313 Time, Brand

−0.01 [0.01] pwild = 0.137 −0.01 [0.02] pwild = 0.801 Time, Brand

0.10 [0.10] pwild = 0.405 0.07 [0.11] pwild = 0.544 Time, Brand

Observations Observation weights

0.98 3.19 pwild = 0.281 504 Yes****

0.88 0.87 pwild = 0.428 420 Yes****

0.91 0.13 pwild = 0.764 420 Yes****

0.98 0.08 pwild = 0.754 504 Yes****

0.97 2.98 pwild = 0.405 433 Yes****

0.94 5.73 pwild = 0.247 504 Yes****

0.96 29.63 pwild = 0.041 504 Yes****

1.00 1.71 pwild = 0.297 420 Yes****

0.99 0.05 pwild = 0.798 504 Yes****

0.97 0.14 pwild = 0.738 504 Yes****

Panel B (1988–1998)

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

(9)

(10)

MERGE 0−2 j,i,t

0.03 [0.03] pwild = 0.347 −0.13 [0.11] pwild = 0.334 7.16 pwild = 0.253 768

−0.14 [0.14] pwild = 0.504 −0.13 [0.15] pwild = 0.327 0.35 pwild = 0.588 640

−0.01 [0.02] pwild = 0.487 0.10 [0.08] pwild = 0.362 2.65 pwild = 0.242 640

−0.02 [0.01] pwild = 0.210 −0.04 [0.09] pwild = 0.690 0.03 pwild = 0.754 768

0.02 [0.04] pwild = 0.324 0.18 [0.15] pwild = 0.200 6.91 pwild = 0.197 653

0.02 [0.08] pwild = 0.871 −0.12 [0.11] pwild = 0.307 3.79 pwild = 0.258 768

0.02 [0.05] pwild = 0.657 −0.42 [0.30] pwild = 0.045 23.96 pwild = 0.080 768

−0.03 [0.02] pwild = 0.201 0.05 [0.05] pwild = 0.524 5.50 pwild = 0.397 619

−0.01 [0.01] pwild = 0.122 −0.05 [0.05] pwild = 0.388 0.82 pwild = 0.523 768

0.10 [0.10] pwild = 0.369 0.01 [0.10] pwild = 0.830 1.14 pwild = 0.504 768

MERGE 3+ j,i,t

Fixed effects 2

R F-statistic on H0 : ˇ1 = ˇ2

MERGE 3+ j,i,t

F-statistic on H0 : ˇ1 = ˇ2 Observations

M. Packalen, A. Sen / International Review of Law and Economics 36 (2013) 12–24

Panel A (1988–1994)

Notes: Same as notes to Table 2.2 (only change is the dependent variable).

21

22

M. Packalen, A. Sen / International Review of Law and Economics 36 (2013) 12–24

given that our identification strategy is to compare in each market the changes for the merging firm to the changes for other vertically integrated firms. A related potential concern is that in each market the merger might have led the merged firm to focus more on wholesale activities at the expense of retail activities. Unfortunately, data on firm-specific wholesale activities in these Canadian gasoline markets does not appear to be available for the pre- and post-merger periods. Hence, we cannot address this concern directly. To address this issue indirectly, we have examined whether in a given market the merger firm increased the combined share of the merged firm and all independent retailers. If merger-efficiencies allowed the merged firm to increase its sales to independents more than it decreased its own retail sales, the combined market share of the merged firm and all independent firms should have increased post-merger. The results show that the coefficient on the long-run merger variable remains negative (for specifications corresponding to columns 1 and 2 of Panel B in Table 2.1), indicating that the data do not support the hypothesis that the merger led to a significant increase in the merged firm’s sales to independent retailers. We also examined whether these results are driven by the consent order required additional divestitures in Central and Western Canada (see Footnote 7) instead of post-merger competition. These additional divestitures were to be completed by February 1991. Accordingly, we re-estimated model (1) with three MR 1991−JUN 1991 0−FEB 1991 merger dummy variables MERGE j,i,t , MERGE j,i,t 0−FEB and MERGE 2+ . In each market the variable MERGE j,i,t j,i,t

1991

MR 1991−JUN 1991

(MERGE j,i,t ) is set to equal 1 between the merger date and February 1991 (between March 1991 and June 1991) for the firm that merged in that market and is zero otherwise. The variis defined as before. If our finding that the merger able MERGE 2+ j,i,t eventually led to a decline in the merging firms’ combined market share is driven by the consent order required additional divestitures, most of the decline in the merging firms’ combined market share should be reflected already in the coefficient on the variable MR 1991−JUN 1991 MERGE j,i,t as this variable represents the period subsequent to the period when these additional divestitures were to be completed. Using the data on the 8 markets in Central and Western Canada for the years 1988–1998 and the observations on all vertically integrated firms with each observation weighted by the firm’s average market share in that market during the sample period (as in columns 1 and 3 of Table 2.1), point estimates of the coefficients MR 1991−JUN 1991 0−FEB 1991 on the variables MERGE j,i,t , MERGE j,i,t and MERGE 2+ are 0.04, −0.03 and −0.17, respectively.28 The low j,i,t estimated values (0.04 and −0.03) of the coefficients on the variMR 1991−JUN 1991 0−FEB 1991 ables MERGE j,i,t and MERGE j,i,t , and the large difference (−0.03 vs. −0.17) between the estimated values of the MR 1991−JUN 1991 and MERGE 2+ , coefficients on the variables MERGE j,i,t j,i,t indicate that most of the market share decline takes place neither during the period when the additional divestitures were to be completed (by February 1991) nor during the subsequent period (from March 1991 to June 1991). This finding implies that our main finding above – that the mergers eventually led to a decline in the merging firms’ combined market share – is not driven by the

28 The difference between the estimates of the coefficients on the variMR 1991−JUN 1991 and MERGE 2+ is statistically significant (F = 33.69, ables MERGE j,i,t j,i,t

pwild = 0.007). When only the data for the years 1988–1994 is used point estiMR 1991−JUN 1991 0−FEB 1991 , MERGE j,i,t mates of the coefficients on the variables MERGE j,i,t and MERGE 2+ are 0.04, −0.03 and −0.11, respectively, and the difference between j,i,t MR 1991−JUN 1991

the estimates of the coefficients on the variables MERGE j,i,t is again statistically significant (F = 13.62, pwild = 0.018)

and MERGE 2+ j,i,t

consent order required additional divestitures in Central and Western Canada.29 The main empirical result obtained in this section – that the merging firms did not regain market share in the long-run – thus appears relatively robust. When this result is combined with the discussion of the predictions of standard merger theories in Section 2.1, the implication is that the mergers examined here did not result in merger-specific efficiencies. Before further elaborating on this implication in the concluding section, we next examine outlet rationalization as one potential explanation for the long-run decrease in the merging firms’ combined market share. 6.2. Results: outlets Results shown in Tables 3.1 and 3.2 replicate the analyses shown in Tables 2.1 and 2.2 except that now the dependent variable is the share of outlets as opposed to volume-based market share.30 Results in Table 3.1 show that on average across markets the mergers did not have a significant short or long-run impact on outlet rationalization. Point estimates of the long-run impact range from −4% to −5% and there is no evidence of a difference between short and long-run effects. The conclusion that the mergers generally did not have a significant impact on outlet rationalization is supported by results from the market-specific analyses shown in Table 3.2. With one exception (Ottawa), the estimates in Table 3.2 suggest that the magnitude of the short-run merger effect on outlets was negligible. Moreover, comparison of estimates of the long-run merger effects on outlets (in Table 3.2) and on market shares (Table 2.2) indicate that the mergers had a more limited impact on outlets than on market shares. Only 4 out of 10 long-run market-specific estimates of the mergers effects on outlets in panel A (panel B) of Table 3.1 are below −0.05 (below −0.09) whereas, as discussed above, 9 out of 10 of the long-run market-specific estimates of the mergers’ market share effects (in Table 2.2) were below −0.05 (below −0.09). Thus it seems that at least some of the merger effect on market shares is due to factors other than outlet rationalization.

7. Conclusion Recent studies have pointed to the considerable lack of retrospective empirical research on consummated mergers and the need for such studies to inform policy makers on the optimal degree of regulation (see e.g. Ashenfelter et al., 2009). Moreover, the merger retrospectives that exist have focused mostly on short-run price effects. Short-run analyses, however, yield an incomplete picture of merger effects because the gestation lag in absorbing mergerinduced efficiencies may be considerable. The focus on price effects is also limiting. Price effects do not necessarily reveal whether merger-specific efficiencies are present, and analysis of non-price effects is also more amenable to identification strategies that are based on within-market as opposed to across-market comparisons.

29 This conclusion is also supported by the results from market-specific regressions (see the discussion above on Table 2.2) which indicate that the merging firms’ combined market share declined also in the two markets in Atlantic Canada (Charlottetown and Halifax) whereas the consent order required additional divestitures concerned only Central and Western Canada. Further evidence to support this conclusion that our results on market shares are not driven by the consent order required additional divestitures is our finding below that the mergers did not have a strong impact on outlet rationalization (see Section 6.2). 30 We do not report the results for sales because for the specification (1) the results would be the same as the results are when the dependent variable is market share as in both cases the effects are identified from relative changes in comparison to other firms in the same market.

M. Packalen, A. Sen / International Review of Law and Economics 36 (2013) 12–24

Our contribution is to propose a market share based approach to complement existing merger analyses. Studying mergers’ market share effects is important because standard merger theory predicts that the market share of the merged firm is less than the merging firms’ combined pre-merger market share and that the presence of merger induced efficiencies determines whether the merged firm regains market share in the long run. A comparison of a merger’s short- and long-run market share impacts will thus demonstrate whether merger-specific efficiencies are present. Using market shares to determine the presence of merger-induced efficiencies is advantageous as it is often difficult to obtain reliable direct measurements of a merger’s cost impacts. In our application we examine the mergers that were created by the divestiture of Texaco’s Canadian assets to provide an empirical analysis of the short and long-run merger effects on market shares. Unlike most other studies we are able to exploit panel level data on market shares for the universe of vertically integrated firms in each market. This enables us to use city-specific difference-in-difference specifications to examine the merger effects. Our findings come with caveats (many of which apply to most difference-in-difference analyses) and may not generalize to other settings, which is why we place more emphasis on the methodological innovation in our analysis. An important aspect of the proposed approach is the reliance on within-market comparisons as an alternative to analyses that are based on across-market comparisons. Variety in identification approaches is desirable because within- and across-market comparisons are susceptible to different sets of unobserved confounding factors. For example, one downside of a market share based approach is that magnitude but not the sign of the estimates is biased because the outcome variable is market share (see the penultimate paragraph of Section 5). At the same time, general equilibrium effects introduce bias also to price based approaches and attempts to avoid such bias by selecting geographically distant control markets increases other biases that arise due to control and treatment groups being too different. Hence, both the existing and our approach are far from perfect, which is why we believe that exploring new approaches – as we have done here – is highly beneficial. Our point estimates of the long-run merger effect on market shares range between −11% and −23%. We also find that the long-run merger effect is statistically significantly larger in magnitude than the short-run merger effect, which itself is statistically insignificant. Our market-specific estimates indicate that the merger impacts on outlets were more modest than the merger impacts on market shares, implying that at least some of the merger effect on market shares is due to factors other than the elimination of under performing assets. Our finding that the merging firms do not regain market share supports the related findings in Mueller (1985) and runs counter to the predictions of standard merger theory, which focuses on potential efficiencies created by mergers and predicts that the merging firms will gain market share in the long run. Our results on market shares and outlets suggest that dynamic merger impacts arose neither through these potential efficiencies created by mergers nor through the elimination of under performing assets, underscoring the potential importance of refinements of standard merger theories. Instead, dynamic merger effects may arise, for example, through a merger’s impact on the investment incentives of all firms (see e.g. Katz & Shelanski, 2007; Vasconcelos, 2006). Our analysis and Mueller (1985) both found that the merging firms did not retain their customers in the long run, which raises questions about the motives to merge and how firms benefit from mergers ex-post. In the literature on gasoline mergers, the analysis and discussion on the likely sources of potential merger-specific efficiencies appears especially scarce. By contrast, the specific sources of potential efficiencies from outlet rationalization have

23

been examined at length. This is somewhat surprising, given the central role that merger-specific efficiencies have in merger theory, and given the potentially diverging implications of efficiencies derived from variable vs. fixed cost savings (Rubinobitz, 2008). Hence, extending merger analyses to cover variables that would shed direct light on sources of potential or actual merger-specific efficiencies, on the motives to merge, and on mergers’ ex-post benefits for the merging firms in the long run is an especially worthy direction for future work. Acknowledgement We thank Wai Hong Choi for research assistance and an anonymous reviewer for comments. References Ashenfelter, O. C., & Hosken, D. (2008). The effect of mergers on consumer prices: Evidence from five selected case studies. NBER Working Paper No. 13859. Ashenfelter, O. C., Hosken, D., & Weinberg, M. (2009). Generating evidence to guide merger enforcement. CEPS Working Paper No. 183. Atkinson, B., Eckert, A., & West, D. S. (2009). Price matching and the domino effect in a retail gasoline market. Economic Inquiry, 47(3), 568–588. Borenstein, S. (1990). Airline mergers, airport dominance, and market power. American Economic Review, 80(2), 400–404. Cameron, A. C., Gelbach, J. B., & Miller, D. (2007). Bootstrap-based improvements for inference with clustered errors. NBER Technical Working Paper No. 344. Carranza, J. E., Clark, R., & Houde, J. -F. (2011). Price controls and market structure: Evidence from gasoline retail markets, Mimeo. Chouinard, H. M., & Perloff, J. M. (2007). Gasoline price differences: Taxes, pollution regulations, mergers, market power, and market conditions. The B.E., Journal of Economic Analysis and Policy, 7(1). Article 8 Coloma, G. (2002). The effect of Repsol-YPF merger on the Argentine gasoline market. Review of Industrial Organization, 21(4), 399–418. Compte, O., Jenny, F., & Rey, P. (2002). Capacity constraints, mergers and collusion. European Economic Review, 46(1), 1–29. Conference Board of Canada. (2001). The Final Fifteen Feet of Hose: The Canadian Gasoline Industry in the Year 2000. Ottawa: The Conference Board. Dafny, L. S. (2009). Estimation and identification of merger effects: An application to hospital mergers. Journal of Law & Economics, 52(3), 523–550. Demsetz, H. (1973). Industry structure, market rivalry, and public policy. Journal of Law & Economics, 16(1), 1–9. Dranove, D., & Lindrooth, R. (2003). Hospital consolidation and costs: Another look at the evidence. Journal of Health Economics, 22(6), 983–997. Eckert, A. (2002). Retail price cycles and the response asymmetry. Canadian Journal of Economics, 35(1), 52–77. Eckert, A. (2003). Retail price cycles and the presence of small firms. International Journal of Industrial Organization, 21(2), 151–170. Eckert, A., & West, D. S. (2004a). A tale of two cities: Price uniformity and price volatility in gasoline retailing. Annals of Regional Science, 38(1), 25–46. Eckert, A., & West, D. S. (2004b). Retail gasoline prices cycles across spatially dispersed gasoline stations. Journal of Law & Economics, 47(1), 245–273. Erutku, C. (2007). Les Prix Eleves de L’essence Avant Les Longues Fins de Semaine: Mythe ou Realite? Canadian Public Policy, 33, 85–92. Erutku, C., & Hildebrand, V. (2010). Conspiracy at the pump. Journal of Law & Economics, 53, 223–237. Farrell, J., & Shapiro, C. (1990). Horizontal mergers: An equilibrium analysis. American Economic Review, 80(1), 107–126. Farrell, J., & Shapiro, C. (2001). Scale economies and synergies in horizontal merger analysis. Antitrust Law Journal, 68, 685–710. Federal Trade Commission and Department of Justice, 2010, Horizontal Merger Guidelines. issued August 19, 2010. Fisher, F. M. (1987). Horizontal mergers: Triage and treatment. Journal of Economic Perspectives, 1, 23–40. Focarelli, D., & Panetta, F. (2003). Are mergers beneficial to consumers? Evidence from the market for bank deposits. American Economic Review, 93(4), 1152–1172. Hastings, J. (2004). Vertical relationships and competition in retail gasoline markets: Empirical evidence from contract changes in Southern California. American Economic Review, 94(1), 317–328. Hastings, J., & Gilbert, R. (2005). Market power, vertical integration and the wholesale price of gasoline. Journal of Industrial Economics, 53(4), 469–492. Houde, J.-F. (2012). Spatial differentiation and vertical mergers in retail markets for gasoline. American Economic Review, 102(5), 2147–2182. Katz, M. L., & Shelanski, H. A. (2007). Mergers and innovation. Antitrust Law Journal, 74(1), 1–85. Kim, E. H., & Singal, V. (1993). Mergers and market power: Evidence from the airline industry. American Economic Review, 83(3), 549–569. Kuhn, K.-U., & Motta, M. (1999). The Economics of Joint Dominance. Mimeo. McAfee, P., & Williams, M. (1992). Horizontal mergers and antitrust policy. Journal of Industrial Economics, 40(2), 181–187.

24

M. Packalen, A. Sen / International Review of Law and Economics 36 (2013) 12–24

Mark, M. (2002). Journal pricing and mergers: A portfolio approach. American Economic Review, 92(1), 259–269. Mueller, D. C. (1985). Mergers and market share. Review of Economics and Statistics, 67(2), 259–267. Nevo, A. (2000). Mergers with differentiated products: The case of the ready-to-eat cereal industry. RAND Journal of Economics, 31(3), 395–421. Noel, M. D. (2007a). Edgeworth price cycles: Evidence from the Toronto retail gasoline market. Journal of Industrial Economics, 55(1), 69–92. Noel, M. D. (2007b). Edgeworth price cycles, cost-based pricing and sticky pricing in retail gasoline markets. Review of Economics and Statistics, 89(2), 324–334. Perry, M. K., Perry, M. K., & Porter, R. H. (1985). Oligopoly and the incentive for horizontal merger. American Economic Review, 75, 219–227. Peters, C. (2006). Evaluating the performance of merger simulation: Evidence from the U.S. airline industry. Journal of Law & Economics, 49, 627–649. Pinske, J., & Slade, M. (2001). Mergers, brand competition, and the price of a pint. European Economic Review, 48, 617–643. Prager, R., & Hannan, T. (1998). Do substantial horizontal mergers generate significant price effects? Evidence from the banking industry. Journal of Industrial Economics, 46(4), 433–452. Rubinobitz, R. (2008). The role of fixed cost savings in merger analysis. Journal of Competition Law and Economics, 5(2), 233–247. Salant, S. W., Switzer, S., & Reynolds, R. J. (1983). Losses from horizontal merger: The effects of an exogenous change in industry structure on Cournot–Nash equilibrium. Quarterly Journal of Economics, 98(2), 185–199. Sen, A. (2003). Higher prices at the gas pump: International crude oil price fluctuations or local market concentration? An empirical investigation. Energy Economics, 25(3), 269–288.

Sen, A. (2005). Does increasing the market share of smaller firms result in lower prices? Review of Industrial Organization, 26(4), 371–389. Sen, A., & Townley, P. G. C. (2010). Estimating the impacts of outlet rationalization on retail prices, industry concentration, and sales: Empirical evidence from canadian gasoline markets. Journal of Economics & Management Strategy, 19(3), 605–633. Simpson, J., & Taylor, C. (2008). Michigan gasoline pricing and the Marathon–Ashland and ultramar diamond shamrock transaction. Journal of Law & Economics, 51(1), 135–152. Taylor, C., & Hosken, D. (2007). The economic effects of the Marathon–Ashland joint venture: The importance of industry supply shocks and vertical market structure. Journal of Industrial Economics, 55(3), 419–451. Taylor, C., Kreisle, N., & Zimmermann, P. R. (2010). Vertical relationships and competition in retail gasoline markets: Comment. American Economic Review, 100(3), 1269–1276. Vasconcelos, H. (2005). Tacit collusion, cost asymmetries, and mergers. RAND Journal of Economics, 36(1), 39–62. Vasconcelos, H. (2006). Endogenous mergers in endogenous sunk cost industries. International Journal of Industrial Organization, 24(2), 227–250. Vita, M., & Sacher, S. (2001). The competitive effects of not-for-profit hospital mergers: A case study. Journal of Industrial Economics, 49(1), 63–84. Williamson, O. E. (1968). Economies as an antitrust defense: The welfare tradeoffs. American Economic Review, 58, 18–35. Winston, C., Mahesri, V., & Dennis, S. (2009). Long run effects of mergers: The case of U.S. Western Railroads. Manuscript. White, L. J. (1987). Antitrust and merger policy: A review and critique. Journal of Economic Perspectives, 1, 13–22.