What Can Be Expected from Mergers after Deregulation?The Case of the Long-distance Bus Industry in France

Thierry Blayac∗ Patrice Bougette†

February 13, 2020

Abstract

This study estimates the competitive effects of horizontal mergers in the Frenchlong-distance bus industry. We examine the two mergers that followed the 2015 Dereg-ulation Act using an exclusive and exhaustive dataset covering eight consecutive quar-ters. We analyze the merger effects by comparing bus links that were affected by themergers with those that were unaffected using difference-in-differences estimations.We find that the two mergers are associated with price increases from 12.08% to 5.3%and frequency decreases of between 24.2% and 29.7%, while observing no effects onload factors. These findings provide evidence of short-run unilateral effects.

Keywords: Long-distance bus industry – Mergers and acquisitions – Deregulated industry– Consolidation – Intramodal competition – Difference-in-differences estimation.

JEL Classification: L11, L41, L43, L92, K23, R40.

∗CEE-M, Univ. Montpellier, CNRS, INRA, Montpellier SupAgro, Montpellier, France. Tel.: +33 (0)434 43 24 81. thierry.blayac@umontpellier.fr.†Corresponding author. Université Côte d’Azur, CNRS, GREDEG, France. Address: 250, rue

Albert Einstein / CS 10269, 06905 Sophia Antipolis Cedex. Tel.: +33 (0)4 93 95 41 05. Email:patrice.bougette@univ-cotedazur.fr.

1 Introduction

France’s long-distance bus industry is a dynamic sector in which consolidation has playedan important role since it was liberalized. Four mergers have occurred since its opening tocompetition in 2015: Flixbus/Megabus and Ouibus/Starshipper in 2016, Blablacar/Ouibusin 2018, and FlixBus/Isilines-Eurolines in 2019. These mergers have strongly affected thestructure of this new market, in which new hopes for mobility have emerged. This studyanalyzes the competitive effects of the 2016 acquisitions on three strategic variables—prices, load factors, and frequencies—for which sufficient hindsight and data are available.This exercise is important because French railways will be fully open to competition by2023, and the Italian experience suggests that this will increase intermodal competition(Gremm, 2018).

The economics literature identifies two main types of impact resulting from a horizontalmerger. On the one hand, a merger can lead to a rationalization of the production system(through efficiency gains or cost synergies). These are the pro-competitive effects of themerger. On the other hand, a merger may increase market power, leading to higherprices, less choice, or reduced innovation and quality (i.e., the anti-competitive effects onconsumers). This tradeoff was analyzed from a theoretical perspective in the pioneeringwork of Williamson (1968) and within an oligopoly framework by Deneckere and Davidson(1985), who focused on price competition, and by Farrell and Shapiro (1990), who focusedon quantity competition.1

We address this tradeoff from an empirical perspective, focusing on how mergers affectprices, load factors, and frequencies in a new industry: France’s intercity bus industry.We examine the 2016 mergers that have created changes in market concentration.2 Weuse original data provided by the French transport regulator Autorité de régulation destransports (ART)3 to show that these two acquisitions affected the competitive dynamicsof the nation’s intercity bus industry. A difference-in-differences (DiD) analysis showsthat the Megabus and Starshipper acquisitions resulted in more concentrated links, whichallowed the other operators to raise their prices and reduce their frequencies, although theentry barriers are low in this type of market. The analysis finds no effects on load factors.In the first quarter after the acquisitions, prices increased by 12.8%, but these increasesdiminished over time. Frequency decreases started one quarter later, ranging from 24.2%to 29.7%.

The rest of the paper is organized as follows. Section 2 briefly reviews the relevantempirical literature. Section 3 provides background on France’s long-distance intercitybus market. Section 4 describes our estimation strategy and data sources as well as themethodology used to construct the merger and comparison markets. Section 5 presentsthe study’s empirical findings. Section 6 concludes the paper.

1For a theoretical survey of the above tradeoff, see Whinston (2006). See also the 2018 special issue ofthis journal edited by Stephen Martin dedicated to the 50th anniversary of the publication of Williamson(1968).

2In other words, the analysis is focused on the links wherein both merging firms were already activebefore the acquisition.

3The transport regulator was renamed “ART” in October 2019. Its former name was “ARAFER”,standing for Autorité de régulation des activités ferroviaires et routières.

2

2 Related Literature

The bus city-pair merger analysis and airline city-pair merger analysis are similar in manyways. This section reviews the recent literature on merger retrospectives, focusing first onairlines and then on the intercity bus industry in Europe.

Peters (2006) analyzes five airline mergers from the 1980s and compares the observedpost-merger price with merger simulations. This retrospective work questions the predic-tive performance of merger simulation and pleads for more flexible models of firm conduct.The results are likely heterogeneous across industries. However, most retrospective studiesfocus on U.S. cases or specific industries (e.g., banking, airlines, hospitals, and petroleum);European mergers remain largely unstudied, due in part to the relatively poor data avail-ability.

Several interesting studies have performed airline merger retrospectives and used thesame methodology employed in this study. For example, Dobson and Piga (2013) analyzethe business model assimilation following two mergers between low-cost European carri-ers.4 Using a DiD approach, they show that, except for the fares advertised only a few daysbefore departure, the acquirer firms maintained most fares at below their pre-acquisitionlevels and increased or stabilized capacities and flight frequencies on the acquired routes.However, these effects were not persistent over time, but were observed only during thefirst year following the merger.

Another airline study of interest, Shen (2017), uses DiD estimation to examine theprice effects of the United-Continental merger. Shen (2017) finds that the fares on non-stop links increased significantly after the merger relative to the competitive rates of theseparate airlines. In direct-route markets, the competitive pre-merger prices of United andContinental Airlines increased by 7.8%, all else being equal. Here, the effect of marketpower outweighed the possible efficiency gains. Interestingly, the effect of consolidationin the airline industry is also linked to airport dominance (e.g., see Bilotkach and Lakew(2014)’s study on US airports).

Of course, a merger may affect market outcomes other than pricing, such as flightfrequencies, firm productivity, cost synergies, or product repositioning. For instance, Chenand Gayle (2019) analyze the effect on service quality of two airline mergers, the Delta-Northwest (DL/NW) (2008) and the Continental-United (CO/UA) (2010) mergers. Theyshow that each merger increased the routing quality of the merging firms’ products—0.45% and 5.28% for the DL/NW and CO/UA merger respectively—in markets where themerging firms did not compete prior to the merger.

Das (2019) focuses on price and quality effects related to the mega-merger betweenAmerican Airlines (AA) and US Airways (US). Using a similar DiD approach, Das (2019)shows that the post-merger price effects are directly related to market size. Indeed, post-merger prices decreased for bigger city-pair markets but increased for the smaller ones.Furthermore, the AA-US merger also had both positive and negative effects on quality(e.g., decrease in cancelled flights, increase in departure and/or arrival delays) but theseeffects are not differentiated according to market size, unlike the price effects. In the samevein, Carlton et al. (2019) carry out a comprehensive investigation of the effect on faresand output of three recent US airline mergers, showing that these mergers have been pro-competitive overall, with significant increases in passenger traffic and frequency, and no

4They analyze two important mergers involving European low-cost carriers (EasyJet’s acquisition ofGo Fly in 2002 and Ryanair’s acquisition of Buzz in 2003).

3

significant unilateral effect on fares.Doi and Ohashi (2019) focus their retrospective work on quality responses to airline

mergers. They estimate a structural model that allows for firms to choose not only pricesbut also product characteristics such as flight frequency. They study the 2002 horizontalmerger between JAL and JAS of Japan. They find that the efficiency gains from the mergerwere significant and that, while the welfare effects of the merger were positive, they variedacross market structures. Using a quasi-natural experiment involving a political choicemade by the Chinese government, Yan et al. (2019) show that the mergers of the early2000s increased the productivity of airline companies.

On-time performance is another variable used to approximate service quality. Princeand Simon (2017) analyze five major airline mergers that have occurred in the US since2000 and study how these mergers affected on-time performance. The authors find that ser-vice quality improved because of efficiency gains from the merger. Nonetheless, efficiencygains may not materialize following a merger or by targeting other strategic variables.Cao et al. (2017) analyze the effect of airline competition on service quality, allowing fornonlinear effects. Using a large database of 5,472 US route-carrier combinations, theyfind that the average length of flight delays and cancellation rates increase along with theconcentration level.

In terms of merger control efficiency, although the two acquisitions we analyze werenot notified to the competition authority (below the notification thresholds), retrospectivestudies can help those responsible for public policy to enhance the decision making ofcompetition authorities. For instance, Friberg and Romahn (2015) analyze the effectof asset divestitures on prices and welfare following the Carlsberg-Pripps merger in theSwedish beer market. They show that the sale of assets by the merging firms had adisciplining effect on the price increases associated with the merger.

A strand of the financial literature looks at mergers in deregulated industries. Forinstance, Becker-Blease et al. (2008) conclude that mergers and acquisitions following thederegulation of the US electric power sector did not create long-term value for investorsand that any efficiencies or synergies that may have been associated with deregulationwere captured by other industry stakeholders. In the EU, liberalization was followed bya wave of consolidation (Domanico, 2007). One of the risks in this process is the creationof barriers by large incumbents designed to maintain their position and prevent the entryof more efficient market operators. Ovtchinnikov (2013) finds that mergers followingderegulation represent an exit from poorly performing industries. This result suggeststhat regulatory changes affect the industry’s merger dynamics. Mergers in deregulatedindustries can serve an important contractionary role.

Finally, a stream of research focuses on the effects of deregulation in the intercitybus industry. Aarhaug and Fearnley (2016) observe that deregulation in Norway led tosignificant growth in passenger numbers and the emergence of a new form of competitionbased on innovation and digital technologies.5 New entry has been important in Italyand Germany (Dürr and Hüschelrath, 2017; Beria et al., 2018), although deregulationresulted in strong consolidation in Germany (Dürr et al., 2016; de Haas and Schäfer, 2017).6

5Empirical evidence has been found of the effect of liberalization in the rail industry. The UK, Norway,and Sweden were among the first to open their rail industries to competition in Europe. See Nilsson et al.(2013) for the Swedish case.

6The operator Flixbus acquired MeinFernBus (2015) and Postbus (2016). Both acquisitions were belowthe notification threshold so that they were not reviewed by the German competition authority.

4

Deutsche Bahn exited the intercity bus market in 2016.7 Average fares tend to decrease (seeDürr et al. (2016) for Germany, supported by the increased intermodal rail competitionfound by Gremm (2018)), but some prices have increased (see the notable British caseanalyzed in White (1990)). Finally, Fageda and Sansano (2018) show that intermodalcompetition matters in Europe. They analyze the determinants of prices and frequenciesdepending on competition type. Using data collected from a sample of links in six largeEuropean countries, they find that intramodal competition is based on frequencies, whileintermodal competition is based on prices.

3 Industry Background

The Macron Law of August 6, 2015 marked the start of a deregulation process in linewith the general trend towards liberalization in the transport services industry and, morebroadly, the network industries in the EU and the US. In France’s intercity bus services, theFrench government aimed to increase the mobility of the more price-sensitive populationand offer alternative transport modes.

Until 2011, non-agreed regular interregional coach transport services were almost non-existent in France. Few data on this market were available during that period, as regularinterurban transport by bus was not statistically distinguished from urban public transportby bus. Bus transport is less developed in France than elsewhere because of the successivegovernments’ decisions in the 1970s to favor rail over bus. This choice proved useful sincethe rail policy was a great national success, as seen in the TGV speed records in the 1970s,the TGV Atlantic line in 1989 designed to run at 300 km/h in commercial service, thenEurostar in 1994, and Thalys and TGV Duplex two years later. However, this preventedthe emergence of alternative modes of low-cost transportation.

From 2011 to 2015, long-distance bus travel in France was restricted by the cabotageregulation of international lines: Journeys could not involve more than one region, andthe number of domestic passengers could not exceed 50% on journeys between two pointsin the region.8

The Macron Law removed these cabotage constraints and opened the long-distancemarket completely.9 This led to the offer of new links that would have been impossibleunder cabotage. The transport regulator ART had full responsibility for monitoring thederegulation process, and the Macron Law endowed it with the responsibility for datacollection and monitoring for the new market (see below).

By late 2015, five bus operators had emerged following the Macron Law. Isilines-Eurolines was a subsidiary of the Transdev group10 and was a leader in the Europeaninternational bus transport market. The group was particularly active in the Netherlands,Belgium, the Czech Republic, and Spain. It operated a bus network covering 25 countries.The second operator, Ouibus (formerly IDBus), was launched by the French rail incum-

7There are similarities in the effects of such liberalization between Germany and Italy; the differencesare caused by industrial, geographical, and historical factors (see Grimaldi et al. (2017)).

8See Figure 1 for the key events in the deregulation of the French intercity bus industry. Blayac andBougette (2017) provide further details and identify some of the pricing strategies used.

9By “long-distance services,” we mean links of over 100 km. Below this threshold, the opening of anew service is allowed by the transport regulator only if a local transport organization can demonstratethe non-economic sustainability of a competitive public mode of transportation on that route.

10Transdev was owned by the Caisse des Dépôts group and Véolia. In October 2018, Rethmann—aGerman industrial group—acquired Véolia’s stake in Transdev (33%).

5

Figure 1: Key steps in the deregulation of the French intercity bus industry

bent operator SNCF—Société Nationale des Chemins de Fer—which can be considereda multimodal operator.11 The third operator, the German Flixbus uses a platform busi-ness model with rapid entry and multiple stops. It does not own any buses but connectscustomers via independent and local bus operators.12 The fourth and fifth operators wereMegabus, owned by the UK Stagecoach group, and the French Starshipper consisting ofa network of independent bus operators based mainly in South West France. It is worthmentioning that, besides these five major operators, the market also includes other localindependent operators (see Figure 4 in the appendix).

Competition on France’s bus links has been relatively heterogeneous since liberaliza-tion. For instance, in early 2019, the 10 busiest links accounted for about 30% of totaldemand.13 On some representative links with particularly fierce intermodal competition,bus operators have enhanced their position by offering lower tariffs (Blayac and Bougette,2017). Bus operators seem to have used an aggressive pricing strategy, with a gradualincrease in fares over time, to generate demand in this new market (see Figure 2 for theevolution of average revenue from 2015 to 2019). In addition, entries have occurred on var-ious scales on many of the links studied. This pattern can be expected in most deregulatedmarkets.

This dynamic new entry phase tends to be followed by a consolidation phase andmergers between operators (Dürr et al., 2016, e.g.). The long-distance bus markets inFrance and Germany generally adhere to this classic pattern. In France, the first wave of

11In November 2018, the leading EU carpooling operator Blablacar announced its intention to acquireOuibus. Blablacar entered the intercity bus market via a new fundraising campaign (“BlaBlaCar rachèteOuibus à la SNCF,” Les Échos, Nov. 12, 2018).

12More precisely, FlixMobility GmbH is the parent company of mobility platforms FlixBus and Flix-Train. Capitalization includes equity firms such as General Atlantic, Silver Lake, TCV, Permira, and HVHoltzbrinck Ventures. For further information on the group’s strategy, see “How Flixbus conquered theEuropean coach market”, The Economist, May 10, 2018.

13Ibid.

6

Figure 2: Average revenue per passenger/100km and per bus (evolution t/t-1)(ARAFER, 2019b)

acquisitions occurred three quarters after the Macron Law. On July 1, 2016, FlixBus tookover Megabus’s retail business in continental Europe (i.e., Germany, Italy, France, Spain,and Benelux) and its cross-border services to London. On July 24, 2016, Starshipperbecame part of the Ouibus franchise, which complemented Ouibus’s connections. Thesetwo mergers did not have to be approved by the French competition agency because theywere below the notification threshold. Both are indicative of the optimization of businessstrategies. The effects of such strategies are the focus of this study. The study assessesthe competitive effects of these two mergers on the whole intercity bus market. Theacquisitions of Ouibus by Blablacar in 2018 and of Isilines-Eurolines by Flixbus in 2019are too recent to be included.14

4 Data and Empirical Strategy

This section describes the database provided by the ART (4.1). We also explain how wecreated variables from the original data set (4.2). Finally, we briefly describe the empiricalstrategy used to assess the effects of the bus operator mergers (4.3).

4.1 Database and Useful Data Set

The data used in this study were provided by ART. Under the Macron Law, bus operatorsare obliged to provide ART with information on all commercial links. The bus operatorsprovide two quarterly spreadsheets: one for the areas served, the modalities of access tothe services, the use of the services delivered, and the main economic and social results;and another for the quality of their services. They are also required to provide annualfinancial and economic information.15

14The integration of the Eurolines-Isilines brands by FlixBus has been effective since May 1, 2019 whilethat of Ouibus into BlaBlaBus has been effective since July 1, 2019 (ARAFER, 2019a).

15For further details on the ART’s data-collection process, see www.arafer.fr/transmission-de-donnees-par-les-autocaristes.

7

We use only quarterly information. The database is exhaustive and the data periodcovers eight consecutive quarters from the creation of this new market in France; August2015 (Q4-2015) to the third quarter of 2017 (Q3-2017). The market opened in August2015 (Q3-2015). However, Q3-2015 data are not used as observations since the periodcovered only half of the quarter.

The database provided by ART includes information on each link for each bus oper-ator and each quarter. The ART database consists of an Excel spreadsheet with 23,411observations and 14 variables. The exhaustive list of variables comprises the name of thebus operator (commercial brand), the quarter, the link number, the city of origin, the des-tination city,16 the INSEE code of the city of origin,17 the INSEE code of the destinationcity, a dummy variable indicating whether the link is actually marketed, the number ofpassengers carried, the number of seats on the bus, the length of the trip, the number oftrips made, and the average revenue per passenger/100 km.18 Based on the informationprovided, we create new variables useful for the econometric modeling (see Section 4.2).

In the first step, we restrict the data set to observations fully available for the 14variables: the new database then contains 17,303 observations. Then, we reduce thedataset to only those links operated throughout the period under consideration (from Q4-2015 to Q3-2017). The database then contains 4,904 observations. However, this level ofdisaggregation is not relevant for our research question since we do not aim to analyzethe merger effects by operator or link but rather the global merger effects on the wholeset of links. Consequently, in the final step, we aggregate the data. As mentioned, theobservations refer to a given link between an origin and a destination, operated by a givenbus brand, and in a given quarter. These observations are used at the aggregate levelbased on links not bus operators. Overall, at the aggregate level, the database includesonly 257 links19 operated from Q4-2015 to Q3-2017, these are used to study the globalmerger effects. As mentioned, three quarters of observations are available in the pre-merger period (Q4-2015, Q1-2016, and Q2-2016) and four quarters of observations areavailable post-merger (Q4-2016, Q1-2017, Q2-2017, and Q3-2017); the quarter Q3-2016 inwhich the two mergers occurred, was neutralized. Ultimately, 257 links are observed for7 quarters. Finally, econometric models are estimated on the basis of 1,799 observations(257 × 7).

4.2 Description of Variables Used in the Econometric Modeling

This section first describe the aggregation process from individual observations to a linkand then provides a short description of the variables used in the econometric analysis(see Section 5). Descriptive statistics are provided in Tables 1 and 2. The variables usedin the econometric modeling are described in the appendix (see Table 6). We start withthe dependent variables: price, load factor, and frequency.

16Due to data-collection constraints, a link between a city of origin A and a destination city B is non-oriented in the study. In other words, one cannot distinguish between the numbers of passengers travelingfrom A to B or from B to A. The figure expresses the aggregation of both.

17INSEE is the French National Institute of Statistics and Economic Studies.18Some bus operators implement yield management systems and use price discrimination techniques

but these behaviors cannot be captured by the quarterly data-collection process.19Though the reduction in the number of observations and links may seem drastic, the 257 links studied

represent more than 90% of the passengers transported during the three quarters before the mergers.

8

4.2.1 Construction of Dependent Variables

For a given quarter, the average price on a given link i is computed as a weighted meanof the average revenue per passengers/100 km of each bus operator on this link for thisquarter. The weight factor is the number of passengers carried by the corresponding busoperator on this link for this quarter divided by the total number of passengers on thegiven link and in the given quarter. Formally, we obtain

Pricei,t =∑

j

wi,j,t × Pricei,j,t (1)

where subscript i denotes the link, j the bus operator, and t the quarter under consid-eration. wi,j,t represents the weight of the bus operator j for the link i and quarter t.Therefore, we have

wi,j,t =Passengersi,j,t∑j Passengersi,j,t

(2)

Table 1 provides some descriptive statistics for the price. The econometric regressionsuse the logarithm of the variable. For all the links under study, the average price is about4.38 e/100km per passenger. This average price is strongly increasing before and after themergers (from 3.54 e/100km to 4.83 e/100km). One of our aims is to determine whetherthese price increases are due to the mergers between operators or to the aggressive pricingstrategies used by operators when opening this new market (Blayac and Bougette, 2017).This average bus price is lower than that of carpooling or train services in France, equalto 5.94 e/100km and 13.42 e/100km (respectively) in the same period.

The methodology used for the two other dependent variables (load factor and fre-quency) is similar, but the weight factor changes. We use the number of seats× kmsupplied by the bus operators for a given link and a given quarter as the weighting factor.The load factor variable is not directly provided in the ART database but, for a given linkand a given quarter, we are able to compute the total number of supplied seats for eachbus operator, which is equal to the product of the number of trips made (frequency) bythe number of seats on the bus (bus size). The load factor for bus operator j on link iand for quarter t is then obtained by dividing the number of passengers carried by busoperator j by the total number of seats supplied by j. Formally, we obtain:

Loadi,j,t =Passengersi,j,t

Frequencyi,j,t × Bus-Sizei,j,t(3)

We then compute an average load factor for a given link i and a given quarter t bythe weighted average. The weight factor is the number of supplied seats per kilometer ofa given bus operator j divided by the total number of supplied seats per kilometer of thelink i for quarter t. Formally, we obtain

Loadi,t =∑

j

wi,j,t × Loadi,j,t (4)

9

where subscript i denotes the link, j the bus operator, and t the quarter under considera-tion. wi,j,t represents the weight of bus operator j for link i and quarter t. Therefore, weobtain

wi,j,t =Frequencyi,j,t × Bus-Sizei,j,t × Distancei,j,t∑j Frequencyi,j,t × Bus-Sizei,j,t × Distancei,j,t

(5)

In the econometric modeling, we use the logarithm of this variable. As Table 1 shows,the average load factor is low, at about 13.39%, with a relatively modest evolution beforeand after the merger: this is far from the load factor required to reach the breakeven point(Crozet and Guihéry, 2018). This average load factor is heterogeneous within the links,as illustrated by the magnitude of the standard deviation.

The number of trips supplied by bus operator j on link i for quarter t is directly pro-vided in the ART database; it is denoted “Frequency”. To compute the average frequencyon link i for quarter t, we use a weighted average with the same weight factor as that usedfor the load factor (see equation 5). We thus obtain

Frequencyi,t =∑

j

wi,j,t × Frequencyi,j,t (6)

where subscript i denotes the link, j the bus operator, and t the quarter under considera-tion. wi,j,t represents the weight of bus operator j for link i in quarter t.

In the econometric modeling, we use the logarithm of this variable. For the 257 linksunder study, the frequency supplied is about 1,554 trips per quarter on average, with asignificant reduction before and after the mergers (from 1,698 to 1,457). It will be inter-esting to analyze whether this reduction is due to the mergers and a supply rationalizationor to the other factors affecting the links.

4.2.2 Details of Explanatory Variables

We now discuss the explanatory variables that will be used in the econometric modeling.In addition to the classical variables used in the DiD method (e.g., time fixed effects,treatment, and interaction terms),20 we also introduce several variables useful for capturingthe specificity of each link.

We use the distance between the city of origin and the destination city. In our sample,the average distance of the links is about 357 km which is rather short. We also use adummy variable reflecting the radial characteristic of the link,21 specifically to account forbeginnings or endings in Paris, a radial characteristic seen in about 25% of the links underconsideration.

The links’ sociodemographic characteristics likely play a major role in the pricing andfrequency strategies of bus operators. Therefore, we include in our analysis the population,the unemployment rate, and the reference tax income.22 Specifically, as a link is defined asthe relation between a city of origin and a destination city, we use the geometric mean ofthe previous variables. Table 1 shows that the trips provided by the bus operators joinedmedium-sizes cities, of about 260,000 inhabitants. The city-pairs are characterized byan unemployment rate of about 9% and an annual reference tax income of about 19,000

20More information about these variables is provided in Table 6 in the appendix.21In contrast to transversal links.22For more details about these variables, see Table 6. Data are for 2016.

10

Table 1: Sample Descriptive Statistics

Whole Sample Pre–Merger Post–Merger

Variables Mean Mean Mean(SD) (SD) (SD)

Price (e/100km) 4.38 3.54 4.83(61.03) (39.10) (47.89)

Load Factor (%) 13.39 12.13 14.23(39,257) (38,365) (39,659)

Frequency 1,554 1,698 1,457(4.82E+6) (4.93E+6) (4.71E+6)

Distance (km) 357.20 347.56 364.44(200.50) (192.73) (205.95)

Passengers (per route) 31,818 33,065 31,136(2.07E+6) (2.01E+6) (2.11E+6)

HHI Passenger 0.49 0.46 0.51(11.78) (12.18) (11.31)

HHI Frequency 0.51 0.51 0.51(746) (800) (703)

HHI Capacity 0.51 0.50 0.51(747) (803) (702)

Seat × Kilometer Supplied 29.3E+6 29.0E+6 29.4E+6

(10.17E+10) (96.3E+9) (10.55E+10)

O–D Geometric Mean (Year 2016) Mean Minimum Maximum(SD)

Population 260,203 25,745 1,378,835(196,025)

Unemployment Rate (%) 8.93 5.93 13.44(1.16)

Reference Tax Income (e/Year/Household) 19,130 11,885 31,303(3,836)

Radial (From/To Paris) Frequency PercentYes 64 24.90%No 193 75.10%

11

e per household. These three sociodemographic characteristics have a high degree ofheterogeneity.

Table 1 also provides information about several variables that are not directly used inthe econometric analysis but that are useful for building weight factors or other variables,such as Passenger, HHI Passenger, HHI Frequency, HHI Capacity, and Seat×Kilometersupplied. The quarterly number of passengers on a bus link is approximately 32,000 (thusless than 360 passengers per day in both directions; the potential for traffic developmentis therefore significant). At the same time, the number of seats per kilometer suppliedis high, which may explain the weakness of the load factor. With the data in the ARTdatabase, we compute three HHI (Passenger, Frequency, and Capacity) which are about0.5, illustrating the strong concentration on each link of our sample.

Let us take a more in-depth view of this long-distance bus market for the quarterimmediately preceding the merger operations (i.e., Q2–2016) as well as for the quarterimmediately following (i.e., Q4–2016). As Table 2 shows, the overall market structuredoes not significantly change after the mergers: 27% of the links are monopoly, 33% areduopoly, and 40% are oligopoly. Nevertheless, the concentration seems to be increasing,especially in frequency and capacity. While the number of bus operators does not changefor 56% of the links considered, it decreases for 32% of them, which seems quite logical inmergers between operators. Surprisingly, however, we observe an increase in the numberof bus operators for 12% of the links. Focusing on the variation of the HHI passengers,mergers lead to an increase in the market concentration for 31% of the links, a decreasefor 28% of the links and stability for 41% of the links.

These elements are sufficiently interesting to be introduced into the econometric anal-ysis —not in their current form, however, but via dummy variables. Thus, concerning theevolution of the market structure, we create nine dummy variables to reflect the evolutionof the number of bus operators on each link between the quarter immediately before themerger (Q2–2016) and the quarter immediately after it (Q4–2016). We apply the sametreatment to the HHI evolution: We create a dummy variable equals to 1 if the concen-tration on the link increases (∆HHI > +10%), 0 if the concentration remains stable(−10% ≤ ∆HHI ≤ +10%), and -1 if the concentration decreases (∆HHI < −10%).Finally, the phenomenon wherein bus operators enter a link is captured by the creation ofa final dummy variable equal to 1 for links on which we have seen at least one operatorenter and 0 otherwise, always for the quarters immediately before and after the merger.

4.3 Assessing Various Effects of Mergers

Two methods can be used to assess the effects for mergers between operators: one basedon merger simulations and the estimation of a structural model and the other based onthe DiD approach. We chose the DiD approach as the most appropriate technique fortwo main reasons (Ormosi et al., 2015). First, the DiD approach does not require anassumption about the type of competition within the market. Second, the lack of costcharacteristics data would have led to less relevant merger simulations results.23

The DiD approach requires the identification of a control group associated with thetreatment group. The simplest way to constitute the two groups is to use a naturalexperiment.

23For a discussion on which approach to use for retrospective merger studies, see the EC report byOrmosi et al. (2015). Both methods may be applied to the same case. See also Peters (2006).

12

Table 2: Pre and Post–Mergers Market Structures

Variables Pre–Merger (Q2–2106) Post–Merger (Q4–2016)Number of links 257

Market Structure:Monopoly 67 (26.07%) 71 (27.63%)Duopoly 88 (34.24%) 85 (33.07%)Oligopoly 102 (39.69%) 101 (39.30%)

Herfindahl-Hirschmann Index:Passenger 0.67 0.68Frequency 0.59 0.64Capacity 0.60 0.64

∆ Bus Operators:From 5 to 3 5 (1.95%)From 4 to 3 39 (15.18%)From 3 to 2 17 (6.61%)From 3 to 1 1 (0.39%)From 2 to 1 19 (7.39%)No changes 144 (56.03%)From 1 to 2 15 (5.84%)From 1 to 3 1 (0.39%)From 2 to 3 16 (6.22%)

∆ HHI Passenger:∆HHI > +10% 80 (31.13%)−10% ≤ ∆HHI ≤ +10% 105 (40.86%)∆HHI < −10% 72 (28.01%)

13

4.3.1 Natural Experiment Approach

This method requires the identification of the links where the operators under study wereactive before the acquisitions. We use data from Q4-2015 to Q2-2016 and retain only thelinks for which information is available across all three quarters.

The treatment group comprises links for which the merging parties are present si-multaneously. In other words, in the treatment group, we find links that were operatedsimultaneously by, at least, Ouibus and Starshipper, or by Flixbus and Megabus, or by allfour operators (Ouibus, Starshipper, Flixbus, and Megabus). These are the links on whichoverlapping occurred. In the control group, we find links where operators other than thefour were present (mainly Isilines-Eurolines) or links where only one of the two mergingentities was active (Ouibus or Starshipper, Flixbus or Megabus).

Finally, to define the two groups, we ensure that the selected links—in the control aswell as the treatment group—are still marketed after the merger and for each of the fourquarters observed (Q4-2016 to Q3-2017).

The quarter in which the merger took place (Q3-2016) is neutralized in our study.This procedure leads to 168 links in the control group and 89 in the treatment group (seethe online appendix for the composition of the two groups). The two groups accountedfor more than 90% of the passengers transported during the three quarters preceding themerger between the operators.

4.3.2 Testing DiD Assumptions

The validity of the DiD method relies on the assumption of a parallel trend betweenthe treatment and control groups before the merger (Q4-2015 to Q2-2016). We checkthe validity of this assumption via graphical analysis. For each group, we compute theaverage price (per 100 km), the average load factor, and the average frequency for eachquarter during the period analyzed. We use a weighted approach to compute the averageprice, load factor, and frequency. For the former, the weight factor is the total numberof passengers carried on each link; while for the latter two, it is the number of seats× kmsupplied.

Thus, for a given group and quarter, the price of a link contributes to the average pricein proportion to the number of passengers carried, while the load factor and the frequencyof a link contribute to the average load factor and the average frequency (respectively) inproportion to the number of seats× km supplied. The computational results are providedin Table 3 and are depicted graphically in Figure 3 below.

Figure 3 shows that this assumption does not seem to be unrealistic, although theevolution between the first and second quarters (i.e., Q4-2015 and Q1-2016) regardingaverage fares raises some questions. One possible explanation is that the operator Flixbus(the market leader in Germany) entered the market in Q4-2015 but not necessarily at thebeginning of the period.24

In addition to the graphical analysis, the parallel trend assumption can be tested basedon a comparison of the evolution rate of the three variables under consideration between

24We tried to account for this by introducing a correction factor in terms of frequency and passengerscarried for the links operated by Flixbus. However, this method was not sufficiently convincing to beretained: It leads to strategic reactions from the rivals, especially in terms of prices being ignored. Itwould be legitimate to assume that considering the strategic response of other suppliers would probablyresult in a fall in the average price in the treatment group for Q4-2015.

14

Table 3: Weighted Average – Price, Load Factor, and Frequency

Price (e/100km) Load Factor (%) Frequency (per quarter)Period Control Treatment Control Treatment Control Treatment

Q4-2015 3.27 3.26 3.31 8.63 452 1231Q1-2016 3.66 3.22 3.68 8.86 770 2336Q2-2016 3.95 3.65 5.53 12.21 654 1903Q3-2016 / / / / / /Q4-2016 4.57 4.90 6.13 12.59 662 2131Q1-2017 4.35 4.58 5.63 13.48 542 1393Q2-2017 4.66 4.86 6.60 15.84 545 1458Q3-2017 4.99 4.99 7.50 15.39 689 1644

Weight Factor Passenger Supplied Seats/km Supplied Seats/km

3,00

3,50

4,00

4,50

5,00

Q4-2015 Q1-2016 Q2-2016 Q3-2016 Q4-2106 Q1-2017 Q2-2017 Q3-2017

Control Group Treatment Group

(a) Bus average prices

2,00%

4,00%

6,00%

8,00%

10,00%

12,00%

14,00%

16,00%

Q4-2015 Q1-2016 Q2-2016 Q3-2016 Q4-2106 Q1-2017 Q2-2017 Q3-2017

Control Group Treatment Group

(b) Bus average load factors

0

500

1000

1500

2000

2500

Q4-2015 Q1-2016 Q2-2016 Q3-2016 Q4-2106 Q1-2017 Q2-2017 Q3-2017

Control Group Treatment Group

(c) Bus average frequencies

Figure 3: Pre- and post-merger evolution of three dependent variables

15

the two groups. For each group, we compute the relative price evolution, the relative loadfactor evolution, and the relative frequency evolution between the quarter preceding themerger (Q2-2016) and the first quarter of the study period (Q4-2015). Since the groupsare sufficiently large, we use a Z-test to compare the average price evolution, the averageload factor evolution, and the average frequency evolution in each group. Table 4 presentsthe results.

Table 4: Test of the common trend assumption25

Price Load Factor FrequencyControl Treatment Control Treatment Control Treatment

Pre-merger Evolution Rate 0.1600 0.0936 -0.1773 -0.1417 0.0603 0.0612Standard Deviation 0.2570 0.6093 0.2123 0.1707 0.3441 0.1980Sample Size 168 89 168 89 168 89Null Hypothesis H0 Same evolution between groups (Q4-2015 to Q2-2016)Z–Test 0.98 -1.46 -0.03Decision Rule Accept H0 Accept H0 Accept H0

The Z-test’s decision rule does not allow us to reject the common trend assumptionat the 5%-level for any of the variables under investigation. Therefore, we use the DiDmethod to assess how the carrier mergers impacted the long-distance bus market in France(i.e., in a natural experiment).

5 Results

This section presents the econometric modeling (5.1) and the results of the estimationprocess (5.2). Finally, we provide economic interpretations and policy implications of theresults (5.3).

5.1 Econometric Modeling

Following the recommendation of Mariuzzo and Ormosi (2019), we use a temporal dis-aggregation model that is more detailed than the classical DiD model with two groupsand two periods. Mariuzzo and Ormosi (2019) underline the importance of consider-ing post-merger dynamics.26 This method consists in introducing a dummy variable foreach quarter and a another dummy variable for each of the interaction terms betweenthe treatment group and the quarters post-merger. Formally, for the three dependent

25In our econometric modeling, we use variables transformed into logarithms. The tests thereforeconcern the evolution of the various quantities taken in logarithms.

26Specifically, Mariuzzo and Ormosi (2019) show that using pooled models can lead to erroneous con-clusions about post-merger effects. They recommend using unpooled models for a period of at least twoyears after the merger.

16

variables under study (price, load, and frequency), we estimate the following equation:

ln(Yi,t) =7∑

t=1β1,t ·Qt + β2 · TREAT +

7∑t=4

β3,t · Inter-Qt + β4 · KM

+β5 · PARIS + β6 · POP-GM + β7 · UR-GM + β8 · INC-GM+β9 · 5-TO-3 + β10 · 4-TO-3 + β11 · 3-TO-2 + β12 · 3-TO-1 (7)

+β13 · 2-TO-1 + β14 · DMY-HHI + β15 · ENTRY-POST + �i,t

where the subscript i refers to an observation—specifically a link between a city of originand a destination city. Subscript t defines the quarters. Quarters 1 to 3 reflect the pre-merger period, while quarters 4 to 7 correspond to the post–merger period.

To avoid endogeneity problems, we adopt the methodology of Carlton et al. (2019) anduse weighted regressions. For the price regression, we use the total number of passengerscarried for each link; for the load factor and frequency regressions, we use the total numberof seats× km supplied for each link. Thus, as in Carlton et al. (2019), the weights do notvary across quarters for a given link, avoiding correlation with the error term. Thus, forthe price regression, the weight of link i, prwi, is equal to

prwi =∑7

t=1∑5

j=1 Passengersi,j,t∑257i=1

∑7t=1

∑5j=1 Passengersi,j,t

(8)

For the load factor and frequency regressions, the weight of link i, lfrwi, is equal to

lfrwi =∑7

t=1∑5

j=1 Supplied-Seat-Kilometeri,j,t∑257i=1

∑7t=1

∑5j=1 Supplied-Seat-Kilometeri,j,t

(9)

The semi-logarithmic form of equation (7) allows us to directly measure in percentagesthe estimated coefficients associated with the interaction terms, such as the impact ofmergers on our three dependent variables.

5.2 Econometric Results of Natural Experiment

In this section, we estimate the model defined in equation (7) for the three dependentvariables under study (price, load factor, and frequency). The results of the econometricestimations are provided in Table 5, where we distinguish between variables directly asso-ciated with the DiD method (first part of the table) and those introduced to capture thespecificity of each link (second part of the table).

Concerning the price regression, the overall statistical validity of the model is good,as evidenced by the F-test. At the individual level (t-test), all the DiD coefficients arestatistically significant, at the 5% level at least. For the other variables introduced in theregression, some of the coefficients are not statistically significant. This is particularlytrue of the coefficients associated with variables describing the market structure evolution(3-TO-1, 2-TO-1, and ENTRY-POST). The explanatory power of the model is high, withan adjusted R–squared of about 0.99.

Furthermore, the econometric results allow us to conclude that the mergers have asignificant effect on the average price per passenger for 100 km. Given the positive andvery significant coefficient for each interaction term, the results show that prices rose for

17

the treatment group as a result of the mergers between Flixbus and Megabus and betweenOuibus and Starshipper. However, the price effect diminishes over time (from 12.8% inQ4-2016 to 5.3% in Q3-2017). The price increase due to the mergers is therefore onlytransitory, supporting the recommendation of Mariuzzo and Ormosi (2019) to focus onthe post-merger dynamics. These results are consistent with the findings in the mergerretrospective literature, where several studies show that price effects are high immediatelyafter the merger and then diminish quickly (Focarelli and Panetta, 2003; Dobson and Piga,2013).

Among the other factors influencing the price on a given link, prominent roles areplayed by the variable PARIS (positive correlation) and by the geometric mean of theunemployment rate (negative correlation). Thus, ceteris paribus, links from or to Pariswill have higher prices, while links between two towns with high unemployment rates willhave lower prices. This means that bus operators take into account both market valueand customer value in their pricing strategies. This appears rational and in line with theMacron Law’s objective of giving back purchasing power to the French consumers.

In the load factor regression, the F-test allows us to confirm the global validity ofthe model. The goodness of fit is high, since the adjusted R–squared is about 0.93.Nevertheless, none of the estimated coefficients associated with the interaction terms isstatistically significant at the individual level (t-test). This means that the bus operatormergers had no differentiated effects (between the treatment and control groups) on theload factor.

Among the variables used to characterize each link, the dummy variables introducedin the model to capture the market structure evolution for the quarter immediately afterthe merger explain, ceteris paribus, the evolution of the load factor. Thus, we can assessthe positive impact of the variable 5-TO-3 on the load factors: The links for which compe-tition between operators was strong before the mergers and where the four merging firmsoperated simultaneously will have, ceteris paribus, a higher load factor than the otherlinks. The same result can be established for the links that saw their duopolistic situationturn into a monopoly immediately after the merger (2-TO-1 ).

Concerning the frequency regression, the overall statistical validity is satisfied (F-test)and the explanatory power of the model is very high (Adj-R2 = 0.9927). The interactionterms coefficients are statistically significant at the 1% level apart for the first quarterafter the merger (Inter-Q4−2016). Given the negative signs and values between 0.242 and0.298 (in absolute value), this means that the bus operator merger reduced the suppliedfrequencies from 24.2% to 29.7% for the treatment group. However, this decrease didnot occur immediately after the merger: The rationalization of the frequencies suppliedbetween the merging firms took some time. Apart from the variables specific to the DiDmethod, the econometric results show that distance (KM ) plays a negative role in thenumber of frequencies offered on a given link. This result appears logical and can beexplained by French legislation on driving times.27

5.3 Economic and Policy Implications

This study paves the way for further analysis of the effects of mergers on other character-istics, such as operators’ costs. A merger can allow a bus operator to achieve the critical

27Bus drivers must take a break of not more than 45 minutes every four and a half hours on the road,and must not work more than 10 hours per day.

18

Table 5: Econometric results of the mergers’ effects

LPRICE LLOAD LFREQ

Variables Coefficients [SD] Coefficients [SD] Coefficients [SD]

Q4−2015 1.145∗∗∗ [0.071] −6.511∗∗∗ [0.310] 3.918∗∗∗ [0.256]Q1−2016 1.156∗∗∗ [0.071] −6.509∗∗∗ [0.310] 4.782∗∗∗ [0.256]Q2−2016 1.266∗∗∗ [0.071] −6.152∗∗∗ [0.310] 4.652∗∗∗ [0.256]Q4−2016 1.419∗∗∗ [0.072] −6.069∗∗∗ [0.312] 4.717∗∗∗ [0.258]Q1−2017 1.372∗∗∗ [0.072] −6.122∗∗∗ [0.313] 4.513∗∗∗ [0.259]Q2−2017 1.448∗∗∗ [0.072] −6.015∗∗∗ [0.313] 4.538∗∗∗ [0.259]Q3−2017 1.516∗∗∗ [0.072] −5.875∗∗∗ [0.313] 4.755∗∗∗ [0.259]TREAT −0.043∗∗∗ [0.016] 0.119∗∗ [0.059] 0.429∗∗∗ [0.057]Inter-Q4−2016 0.128∗∗∗ [0.021] −0.150 [0.103] −0.096 [0.085]Inter-Q1−2017 0.109∗∗∗ [0.021] 0.007 [0.103] −0.275∗∗∗ [0.085]Inter-Q2−2017 0.097∗∗∗ [0.021] 0.029 [0.103] −0.242∗∗∗ [0.085]Inter-Q3−2017 0.053∗∗ [0.021] −0.101 [0.103] −0.297∗∗∗ [0.085]

KM −2.69E−4∗∗∗ [2.35E−5] −7.60E−4∗∗∗ [1.02E−4] −1.62E−3∗∗∗ [8.62E−5]PARIS 0.030∗∗ [1.20E−2] −0.501∗∗∗ [0.062] −0.282∗∗∗ [0.051]POP-GM −1.15E−7∗∗∗ [2.45E−8] 1.68E−6∗∗∗ [1.10E−7] 1.10E−6∗∗∗ [9.45E−8]UR-GM −0.014∗∗∗ [4.90E−3] 0.042∗ [0.023] 0.058∗∗∗ [0.019]INC-GM 4.52E−6∗∗ [2.01E−6] 1.43E−4∗∗∗ [8.67E−6] 8.77E−5∗∗∗ [7.12E−6]5-TO-3 0.044∗∗ [0.018] 0.436∗∗∗ [0.089] 0.209∗∗ [0.088]4-TO-3 −0.027∗ [0.015] / 0.259∗∗∗[0.057]3-TO-2 0.112∗∗∗ [0.019] −0.190∗∗ [0.088] −0.417∗∗∗[0.075]3-TO-1 / / −1.531∗∗∗[0.524]2-TO-1 / 0.360∗∗∗ [0.109] −1.003∗∗∗[0.086]DMY-HHI −0.015∗∗∗ [0.005] −0.072∗∗∗ [0.026] /

Obs. 1,799 1,799 1,799F-test 10,524.90 1,165.19 11,189.80(p− value) < 0.0001 < 0.0001 < 0.0001Adj.R2 0.9919 0.9315 0.9927

Period: Q4-2015/Q3-2017; 3 quarters pre-merger and 4 quarters post-merger; Q3-2016 isdropped. Significance level: ***=1%, **=5%, *=10%

19

size required for profitability. However, even after merger, the load factors may not besufficient to ensure continued profitability (Crozet and Guihéry, 2018).

Tables 7–8 recap the descriptive statistics of the main variables for Q2-2016 (pre-merger) and Q2-2017 (post-merger). We choose these two quarters, which are one yearapart, to avoid the seasonality problems that might hinder direct comparison. On average,the number of operators decreased for the links included in the treatment group, whilethe control was not affected by the two mergers (see Tables 7 and8).

All HHIs are increasing at the treatment and control group levels, making the marketmore concentrated following the two mergers. These increases are stronger in the treatmentgroup (between 14% and 24.39% depending on the type of HHI) than in the control group(between 1.32% and 4.35% depending on the type of HHI). Analyzing the descriptivestatistics for the control and treatment groups shows that the groups are affected differentlyby the two mergers. These results are in line with those obtained regarding the numberof operators.

Tables 7 and 8 show that the load factors increased by 23.76% in the treatment groupand by 16.56% in the control group. Since the interaction coefficients in the DiD regressionare not significant (see Table 5), these increases were not due to the two mergers studied.The econometric estimation demonstrates that there is no differentiated effect occurredbetween the treatment and control groups. Thus, the load factors would have risen evenin the absence of the two mergers.

In the treatment group, the frequency decreases significantly (-28.71%), and (as men-tioned), the load factors increase significantly. This indicates rationalization and marketmaturity phenomena. Frequencies also decrease in the control group (-19.92%), and loadfactors increase but to a degree lower than the decrease in frequencies. As the interactionterms in the DiD regressions on frequencies are highly significant (see Table 5 for the lastthree quarters under study), this shows that the two mergers directly affected frequenciesdownwards, with a one quarter lag.

The previous analytical results are corroborated by the data in Figure 3 which is richin information. First, it confirms the net price effect due to the mergers obtained by theDiD estimates. Second, regarding the load factors, a roughly parallel development occurswithin the treatment and control groups, although the load factors’ increase is greaterin the treatment group than in the control group. The DiD shows that the two mergersproduced no significant effects. Third, we see that frequencies decrease to a much greaterdegree in the treatment group than in the control group; this is due to the mergers, as theinteraction terms are clearly significant (see Table 5).

These findings indicate that, in 2016, the market had yet to mature and was searchingfor the critical size at which it would become profitable. The latest acquisitions sincethen have confirmed this trend (see the timeline in Figure 1). In early 2020, a duopolybetween Blablabus and Flixbus emerged.28 The strength of intermodal competition is thena major challenge for the further development of the intercity bus market. For example, inGermany, intermodal competition has been insufficient to counterbalance Flixbus’s marketpower. Flixbus also invested in rail transport after the German rail market was opened tocompetition.29 Similarly, FlixTrain hopes to launch five French inter-city links in January

28Despite its leading position in 2016, the Isilines-Eurolines group lacked the critical mass required tocontinue this consumer business. For this reason, Transdev decided to exit this activity to refocus thegroup’s activities and prepare for the French rail sector’s opening up to competition.

29O’Brien, Chris. “FlixBus company launches FlixTrain to reinvent rail travel.” Venturebeat, March 6,

20

1, 2021.30 Further, the Blablabus brand represents a challenge to Flixbus’s developmentstrategy. Thus, the new brand appears to pose tough competitor for FlixBus by using thesame business model.31 It is also a fairly flexible market, with no major sunk costs.

Another example, a Norwegian case, shows that a high degree of intermodal compe-tition may affect the future of the bus market (Aarhaug et al., 2018). The success ofNorway’s bus sector liberalization was largely due to the offer of links in markets under-served by other transport modes. These markets have benefited from better alternativetransport services. Buses have become less attractive due to the emergence of competi-tion among low-cost carriers via the offer of (i.e., competition from) rail solutions, and aslowdown has been observed in the number of passengers transported by bus in Norwayin recent years. How the French long-distance bus market will develop is probably tooearly to determine: this will be difficult to predict until it competes directly with railtransport.32

6 Conclusion

The effect of the two 2016 French intercity bus industry acquisitions triggered questionsabout how a newly deregulated market can lead to a monopoly, as occurred in Germany.The mergers were aimed at achieving critical size and profitability.33 Using a DiD approachwith original data provided by the regulator, we find evidence of unilateral effects in termsof prices and frequencies. However, no effects on load factors have been demonstrated. Inthe first quarter post-acquisition, prices increased by 12.8%, but these increases diminishedover time. For instance, the price increase for the fourth quarter post-merger amounts to5.3%. Frequency decreases started one quarter later, and ranged from 24.2% to 29.7%.

This new market and data availability provide opportunities for future studies. Forinstance, it would be interesting to measure how acquisitions impact other transportationmodes, especially rail and carpooling. According to Fageda and Sansano (2018), pricecompetition is important for intermodal competition; Blablacar’s acquisition of Ouibusprovides an opportunity to study the strength of such intermodal competition. Finally,comparisons with the German market would be fruitful. The German market has evolvedrapidly from a situation in which new entry was important to one in which Flixbus hasmarket power. Among the research questions relevant to the French case, those that needto be investigated to complement this work are the following: Will bus operators becomeprofitable in the near future? How will the rail liberalization achieved in 2023 affect the

2018. https://venturebeat.com/2018/03/06/flixbus-company-launches-flixtrain-to-reinvent-rail-travel (ac-cessed January 6, 2020). German railway deregulation started in 1994. Flixbus uses a business modelsimilar to its bus services model by partnering with local train operators.

30“FlixTrain seeks to enter French passenger market,”, Railway Gazette, June 18, 2019.https://www.railwaygazette.com/news/single-view/view/flixtrain-seeks-to-enter-french-passenger-market.html, accessed December 30, 2019.

31See “BlaBlaBus: Fresh competition for Flixbus takes to the road in Germany,” The Local.de(https://www.thelocal.de/20190626/germanys-flixbus-faces-competition-with-launch-of-new-long-distance-bus-operator, accessed December 30, 2019).

32However, road and rail transport show significant differences in rolling stock capacity and servicefrequencies. Bus transport may replace rail transport on certain links but may not be able to entirelyreplace “heavy” rail transport.

33Yves Lefranc-Morin, General Manager of FlixBus France, said “We are still too much in this marketto make it profitable. It is a volume business, you have to reach a significant size to make our costsprofitable,’ (“Mauvais karma pour les cars Macron,” Libération, September 6, 2018).

21

intercity bus market? Would it have been better to open the rail market to competitionbefore liberalizing the bus industry?

Acknowledgements

This article has benefited from Editor Lawrence J. White’s guidance and the construc-tive suggestions of two anonymous reviewers from this journal. The authors thank theFrench transport regulator Autorité de régulation des transports (ART) for the data ac-cess and sector expertise provided, which contributed greatly to the quality of this study.We are grateful to Aude Le Lannier, Nicolas Quinones-Gil, Anthony Martin, and AnneYvrande-Billon. We also thank Dimitri Dubois, Emmanuelle Lavaine, Xavier Fageda,Javier Asensio, and participants at the CREM research seminar in Caen (May 2018), thefirst Rencontres Francophones Transport Mobilité in Lyon (June 2018), the ITEA confer-ence in Hong-Kong (June 2018), and the 15th World Conference on Transport Research(WCTR) in Mumbai (May 2019), the 34th Annual Congress of the EEA in Manchester(August 2019), and the 46th EARIE Annual Conference in Barcelona (August 2019), forthe insightful comments they offered. The usual disclaimer applies.

Funding

This research did not receive any specific grant from funding agencies in the public, com-mercial, or not-for-profit sectors. The authors declare that they have no conflict of interest.

Appendix

Figure 4: The number of bus operators in the French long-distance bus industry(2015–2018) – The Club des cinq vs. local independent operators (Source: ARAFER

(2019b)

22

Table 6: Short description of variables used in the econometric modeling

Name (Notation) Description

Dependent VariablesPrice (LPRICE) This is the average revenue per passenger for 100 km. For each quarter

and each link, we have the average revenue per passenger for 100 km foreach bus operator. We compute a weighted average for the link. Theweight factor is the number of passengers carried by the correspondingbus operator on this link for this quarter divided by the total number ofpassengers of the given link and quarter. In the regression, we take thelogarithm of this variable.

Load Factor (LLOAD) This is the average load factor on a given link for a given quarter. For eachquarter and each link, we first compute for each bus operator a load factorwhich is equal to the total number of passengers carried by this operatordivided by the capacity of this operator (F requency×BusSize). We thencompute an average load factor for a given link and a given quarter via theweighted average. The weight factor is the number of supplied seats perkilometer of a given bus operator divided by the total number of suppliedseats per kilometer of the link for this quarter. In the regression, we takethe logarithm of this variable.

Frequency (LFREQ) This is the average frequency on a given link for a given quarter. Foreach quarter and each link, we have directly in the ART database, thefrequency provided by each bus operator. We compute a weight averagefrequency for a given link and a given quarter. The weighted factor is thesame as for the load factor. In the regression, we take the logarithm ofthis variable.

Explanatory VariablesTime Fixed Effect (Q4−2015 toQ3−2017)

Dummy variable associated with each quarter. Therefore, we obtain sevendummy variables: Q4−2015 to Q3−2017. The quarter in which the mergersbetween operators occur is neutralized (i.e., Q3−2016).

Treatment (TREAT) Dummy variable equals 1 if the links belong to the treatment group and0 otherwise.

Interaction (Inter-Q4−2016 toInter-Q3−2017)

These interaction terms are defined as the product of the treatment vari-able by the time fixed effect, only for quarters posterior to the merg-ers. Therefore, we have four interactions terms (i.e., Inter-Q4−2016 toInter-Q3−2017)

Distance (KM) This is the distance in kilometer between the Origin and Destinationcities.

Radial Trip (PARIS) Dummy variable equals to 1 if the link beginning or ending in Paris(French capital).

Population (POP–GM) We use the population of the city of origin and destination to computethe geometric mean of the population on each link. To this end, we usedata from year 2016.

Unemployment (UR–GM) We use the unemployment rate of the city of origin and destination tocompute the geometric mean of the unemployment rate on each link. Tothis end, we use data from year 2016.

Reference Tax Income (INC–GM)

Each year, the French tax authorities compute the “Reference Tax In-come” of each household. The latter is an indicator of the standard ofliving of a tax household. This provides the average Reference Tax In-come of households living in a given city. We then compute the geometricmean of the Average Reference Tax income on each link. To this end, weuse data from year 2016.

Market Structure (X–TO–Y) We define a dummy variable for the evolution of every market structurepossible in the data set. To this end, we focus on the quarter immediatelybefore the merger (Q2–2016) and the one immediately after it (Q4–2016).For instance, the dummy variable 5–TO–3 equals 1 if there were 5 busoperators on the link before the merger, and 3 after it, and 0 otherwise.Therefore, we create the following 9 dummy variables: 5-TO-3, 4-TO-3,3-TO-2, 3-TO-1, 2-TO-1, No changes, 1-TO-2, 1-TO-3, and 2-TO-3.

HHI Evolution (DMY–HHI) From the ∆HHIP assengers variable (see Table 2), we build a dummyvariable, called DMY-HHI, which is equal to 1 if the concentration on agiven link increased (> +10%) between the pre-merger quarter and thepost-merger quarter (i.e. Q2–2016 and Q4–2016), if the concentrationremained stable (between -10% and +10%), and -1 if the concentrationdecreased (< −10%).

Entry (ENTRY-POST) We build a dummy variable equal to 1 if an entry has occurred on the linkbetween the quarter before the merger (Q2–2016) and the quarter afterthe merger (Q4–2016).

23

Tabl

e7:

Nat

ural

expe

rimen

t–

Des

crip

tive

stat

istic

sfr

omth

etr

eatm

ent

grou

p

Pre

-mer

ger

(Q2-

2016

)P

ost-

mer

ger

(Q2-

2017

)G

row

thra

te

Varia

ble

Mea

nM

in.

Max

.M

ean

Min

.M

ax.

Mea

nM

in.

Max

.

No.

oper

ator

s3.

271

52.

491

3-2

3.85

0-4

0.00

Freq

uenc

ies

1408

.74

72.8

5801

.910

04.2

478

4928

-28.

717.

14-1

5.06

No.

ofpa

ssen

gers

1150

8.4

2411

4727

1049

5.97

1189

990

-8.8

0-5

4.17

-21.

56P

rice

3.95

2.26

8.07

53.

937.

6726

.58

73.8

9-4

.96

Load

Fact

or10

.52

0.08

54.2

213

.02

0.21

56.4

23.7

616

2.50

4.02

HH

IPa

ss0.

50.

261

0.57

0.34

114

.00

30.7

70

HH

IFr

eq0.

410.

211

0.51

0.34

124

.39

61.9

00

HH

IC

apa

0.41

0.21

10.

510.

341

24.3

961

.90

0Su

pply

Seat

Km

19,8

59,3

90.0

944

6,19

1.2

131,

710,

123

15,7

24,1

04.1

349

5,22

0.2

115,

445,

200

-20.

8210

.99

-12.

35So

urce

:A

utho

rs’

com

puta

tion

s

24

Tabl

e8:

Nat

ural

expe

rimen

t–

Des

crip

tive

stat

istic

sfr

omth

eco

ntro

lgro

up

Pre

-mer

ger

(Q2-

2016

)P

ost-

mer

ger

(Q2-

2017

)G

row

thra

te

Varia

ble

Mea

nM

in.

Max

.M

ean

Min

.M

ax.

Mea

nM

in.

Max

.

No.

oper

ator

s1.

831

31.

831

30

00

Freq

uenc

ies

563.

4572

.82,

503.

445

1.21

342,

572

-19.

92-5

3.30

2.74

No.

ofpa

ssen

gers

2,05

2.8

2116

,452

2,17

2.54

422

,484

5.83

-80.

9536

.66

Pric

e3.

972.

486.

825.

083.

4614

.46

27.9

639

.52

112.

02Lo

adFa

ctor

6.28

0.24

29.7

77.

320.

2430

.82

16.5

60

3.53

HH

IPa

ss0.

760.

341

0.77

0.34

11.

320

0H

HI

Freq

0.69

0.34

10.

720.

331

4.35

-2.9

40

HH

IC

apa

0.7

0.34

10.

720.

341

2.86

00

Supp

lySe

atK

m9,

677,

955.

2840

3,60

3.2

44,7

95,0

11.3

8,85

6,13

0.79

329,

188

41,5

56,3

26.8

7-8

.49

-18.

44-7

.23

Sour

ce:

Aut

hors

’co

mpu

tati

ons

25

References

Aarhaug J, Fearnley N (2016) Deregulation of the Norwegian Long Distance Express CoachMarket. Transport Policy 46:1–6

Aarhaug J, Farstad E, Fearnley N, Halse AH (2018) Express Coaches: An Up-Hill Battleafter Liberalization? Research in Transportation Economics 72:82–91

ARAFER (2019a) Marché librement organisé des services interurbains spar autocar –Bilan du 2e trimestre 2019. Tech. rep., Autorité de régulation des activités ferroviaireset routières, URL https://www.arafer.fr/wp-content/uploads/2019/10/arafer_bilan-marche-autocars-2019-t2.pdf

ARAFER (2019b) Rapport annuel 2018 sur le marché du transport par autocaret les gares routières. Tech. rep., Autorité de régulation des activités ferroviaireset routières, URL https://www.arafer.fr/wp-content/uploads/2019/07/rapport_annuel_trv_2018.pdf

Becker-Blease JR, Goldberg LG, Kaen FR (2008) Mergers and Acquisitions as a Responseto the Deregulation of the Electric Power Industry: Value Creation or Value Destruc-tion? Journal of Regulatory Economics 33(1):21–53

Beria P, Nistri D, Laurino A (2018) Intercity Coach Liberalisation in Italy: Fares Deter-minants in an Evolving Market. Research in Transportation Economics 69:260–269

Bilotkach V, Lakew PA (2014) On Sources of Market Power in the Airline Industry: PanelData Evidence from the US Airports. Transportation Research Part A 59:288–305

Blayac T, Bougette P (2017) Should I Go by Bus? The Liberalization of the Long-DistanceBus Industry in France. Transport Policy 56:50–62

Cao KH, Krier B, Liu CM, McNamara B, Sharpe J (2017) The Nonlinear Effects ofMarket Structure on Service Quality: Evidence from the U.S. Airline Industry. Reviewof Industrial Organization 51(1):43–73

Carlton DW, Israel M, MacSwain I, Orlov E (2019) Are Legacy Airline Mergers Pro- orAnti-Competitive? Evidence from Recent U.S. Airline Mergers. International Journalof Industrial Organization 62:58–95

Chen Y, Gayle PG (2019) Mergers and Product Quality: Evidence from the Airline In-dustry. International Journal of Industrial Organization 62:96–135

Crozet Y, Guihéry L (2018) Deregulation of Long Distance Coach Services in France.Research in Transportation Economics 69:284–289

Das S (2019) Effect of Merger on Market Price and Product Quality: American and USAirways. Review of Industrial Organization 55(3):339–374

Deneckere RJ, Davidson C (1985) Incentives to Form Coalitions with Bertrand Competi-tion. Rand Journal of Economics 16(4):473–486

Dobson PW, Piga CA (2013) The Impact of Mergers on Fares Structure: Evidence fromEuropean Low-Cost Airlines. Economic Inquiry 51(2):1196–1217

Doi N, Ohashi H (2019) Market Structure and Product Quality: A Study of the 2002Japanese Airline Merger. International Journal of Industrial Organization 62:158–193

Domanico F (2007) Concentration in the European Electricity Industry: The InternalMarket as Solution? Energy Policy 35(10):5064–5076

Dürr NS, Hüschelrath K (2017) Patterns of Entry and Exit in the Deregulated GermanInterurban Bus Industry. Transport Policy 59:196–208

Dürr NS, Heim S, Hüschelrath K (2016) Deregulation, Competition, and Consolidation:

26

https://www.arafer.fr/wp-content/uploads/2019/10/arafer_bilan-marche-autocars-2019-t2.pdfhttps://www.arafer.fr/wp-content/uploads/2019/10/arafer_bilan-marche-autocars-2019-t2.pdfhttps://www.arafer.fr/wp-content/uploads/2019/07/rapport_annuel_trv_2018.pdfhttps://www.arafer.fr/wp-content/uploads/2019/07/rapport_annuel_trv_2018.pdf

The Case of the German Interurban Bus Industry. Journal of Transport Economics andPolicy 50(2):164–188

Fageda X, Sansano S (2018) Factors Influencing Prices and Frequencies in the Interur-ban Bus Market: Evidence from Europe. Transportation Research Part A: Policy andPractice 111:266–276

Farrell J, Shapiro CD (1990) Horizontal Mergers: An Equilibrium Analysis. AmericanEconomic Review 80(1):107–126

Focarelli D, Panetta F (2003) Are Mergers Beneficial to Consumers? Evidence from theMarket for Bank Deposits. American Economic Review 93(4):1152–1172

Friberg R, Romahn A (2015) Divestiture Requirements as a Tool for Competition Policy:A Case from the Swedish Beer Market. International Journal of Industrial Organization42:1–18

Gremm C (2018) The Effect of Intermodal Competition on the Pricing Behaviour of a Rail-way Company: Evidence from the German Case. Research in Transportation Economics72(October 2017):49–64

Grimaldi R, Augustin K, Beria P (2017) Intercity coach liberalisation. The cases of Ger-many and Italy. Transportation Research Procedia 25:474–490

de Haas S, Schäfer JT (2017) Consolidations in the German Interurban Bus Industry:Effects on Prices and Quantities. MAGKS Papers on Economics No 31-2017

Mariuzzo F, Ormosi P (2019) Post-Merger Price Dynamics Matters, So Why do MergerRetrospectives Ignore it? Review of Industrial Organization 55(3):403–429

Nilsson JE, Pyddoke R, Hulten S, Alexandersson G (2013) Developments in Transport Pol-icy: The Liberalisation of Railway Passenger Transport in Sweden. Journal of TransportEconomics and Policy 47:307–312

Ormosi P, Mariuzzo F, Havell R (2015) A Review of Merger Decisions in the EU: Whatcan we Learn from Ex-post Evaluations? Tech. rep., European Union, Luxembourg

Ovtchinnikov AV (2013) Merger Waves Following Industry Deregulation. Journal of Cor-porate Finance 21:51–76

Peters C (2006) Evaluating the Performance of Merger Simulation: Evidence from theU.S. Airline Industry. The Journal of Law and Economics 49(2):627–649

Prince JT, Simon DH (2017) The Impact of Mergers on Quality Provision: Evidence fromthe Airline Industry. The Journal of Industrial Economics 65(2):336–362

Shen Y (2017) Market Competition and Market Price: Evidence from United/ContinentalAirline Merger. Economics of Transportation 10:1–7

Whinston MD (2006) Lectures on Antitrust Economics. The MIT Press, Cambridge, Mas-sachusetts

White PR (1990) Bus Deregulation: A Welfare Balance Sheet. Journal of Transport Eco-nomics and Policy 24(3):311–332

Williamson OE (1968) Economies as an Antitrust Defense: The Welfare Tradeoffs. Amer-ican Economic Review 58:13–36

Yan J, Fu X, Oum TH, Wang K (2019) Airline Horizontal Mergers and Productivity:Empirical Evidence from a Quasi-natural Experiment in China. International Journalof Industrial Organization 62:358–376

27

IntroductionRelated LiteratureIndustry BackgroundData and Empirical StrategyDatabase and Useful Data SetDescription of Variables Used in the Econometric ModelingConstruction of Dependent VariablesDetails of Explanatory Variables

Assessing Various Effects of MergersNatural Experiment ApproachTesting DiD Assumptions

ResultsEconometric ModelingEconometric Results of Natural ExperimentEconomic and Policy Implications

Conclusion