Research Joint Ventures and Collusion - What can we learn from European Commission Decisions?

Size: px
Start display at page:

Download "Research Joint Ventures and Collusion - What can we learn from European Commission Decisions?"

Transcription

1 Research Jont Ventures and Colluson - What can we learn from European Commsson Decsons? ERASMUS UNIVERSITY ROTTERDAM Erasmus School of Economcs Department of Economcs Supervsor: Dr. Martn Bog Name: Nora Barb Exam number: E-mal address: norabarb@hotmal.t Keywords: Research Jont Ventures, Colluson, Event study Methodology

2 Abstract Ths paper offers an ndrect approach to test for possble antcompettve behavour of frms engaged n European Research Jont Ventures. I analyze the cumulated abnormal returns of the partner frms for the days surroundng the announcement of the RJV. The sample contans frms that can be suspected of colluson and frms that are not suspected of colluson. I assume that Shareholders are aware of ths dfference at the moment of the announcement. I construct a categorcal varable that dstngush between the two groups of frms and usng a smple emprcal specfcaton I test for the effect of ths varable on the abnormal returns of the partcpant frms. At the end of the analyss, however, I fnd not enough evdence to affrm that RJVs facltate colluson behavours. Acknowledgements I would lke to thank my mother and my famly all for ther love and support. I also would lke to thank Dr. M. Bog, Prof J. van Snderen and Prof. R. Zwnkels at the Erasmus School of Economcs for ther thoughtful comments. I would lke to thank Prof. Schnkel at the Amsterdam Center for Law&Economcs for makng avalable the revew of the Commsson Decsons. I would lke to thank K.Felletner for her precous suggestons and the laughter. Specal thanks go to T. Bezemer for sharng wth me the bad and the good moments of the past year. I also would lke to thank all the people that supported me durng the perod of the collecton and the analyss of the data: A. Bonfatt, R. de Luca, M. Talarco, M. Maran, A. Pecora, M. Grosso, N. Daudì and W. Per. A fnal thanks goes to I. Bass and G.dalla Torre, for always beng there. 2

3 Table of contents: Abstract 2 1.Introducton 4 2.European anttrust law European Commsson Decsons 7 3.Lterature revew 8 3.1RJVs and Colluson, ndrect approaches 8 3.2Event studes on allance formaton 12 4.Methodology Sample and Data The Emprcal Specfcaton 18 5.Results and Dscusson 24 6.Conclusons 32 7.References 33 8.Appendx 36 3

4 1.Introducton A Research Jont Venture (RJV) s an entty formed by two or more partcpant frms where knowledge and resources are shared n order to jont cooperate for the research and development of new products and/or new processes. RJV started to became a popular way to share the R&D efforts n the 80 s. Nowadays they are a common nstrument used especally n those sectors for whch R&D requres hgh fxed cost. Moreover, RJV agreements are enhanced by governments for ther postve effects on welfare through the reducton of R&D costs. The US Natonal Cooperatve Research Act (NCRA) of 1984 was ssued to promote R&D cooperaton among US frms. In Europe, there exst two large networks created n order to enhance European R&D allances: the EU Framework Programme for Scence and Technology (EU-FP) and the Eureka Programme 1. These two networks together saw the formaton of over 25,000 R&D allances n the last twenty years. The benefts for the frms that form RJVs are several. Trough ths knd of cooperaton t s possble to avod the duplcaton of R&D costs and take advantage of assets complementartes. Frms learn new sklls and gan access to new nformaton, processes and specalzed resources. RJVs also allow for the nternalzaton of the spllovers and help to overcome possble free-rder problems. Moreover, RJVs gve the partcpants the possblty to share the rsks related to the R&D actvty. Some or all of these benefts are desrables when frms decde to cooperate n R&D (Marìn et al.(2000) and Roller at al. (2007)). Another mportant beneft that frms could take nto consderaton s the possblty to soften market competton through the RJV. It s therefore possble that some RJVs are formed wth ntentons of colluson. Ths paper ams to dscover possble antcompettve behavours of frms partcpatng n European RJVs. I test for ths hypothess usng an ndrect approach. I collect a sub-sample of RJVs for whch the European Commsson adopted a formal decson concernng possble antcompettve behavour 2 and I defne these RJVs as those that can be suspected of colluson. I then collect a random sub-sample of RJVs from a populaton of European RJVs and I defne these RJVs as those not suspected of colluson. I then combne the two sub-samples and analyze the cumulated abnormal 1 For further nformaton please refer to the offcal web pages of these networks: for the EU Framework Program and for the Eureka Program. 2 See next secton for a basc revew of the European anttrust legslaton. 4

5 returns of the partcpant frms around the day of the announcement of the RJV. I ntroduce a smple emprcal specfcaton that tres to capture the effect of the expected colluson nformaton ncluded n the event RJV announcement lookng at the determnants of the abnormal returns. The focal pont of my emprcal specfcaton s the categorcal varable ECdecson 3 that dstngush between the two sub-samples of RJVs. If the RJVs suspected of colluson were ndeed formed for ths purpose, then ths varable should have a role n explanng the abnormal returns of the shares n the days surroundng the announcement. Ths clam rests on the assumpton that the shareholders knew, at the moment of the announcement, that the RJV had such characterstcs that made t fall nto the scope of the anttrust legslaton. I therefore assume that they were fully aware of the collusve potental of the RJV. The other varables I ncluded n the specfcaton have been used n lterature to explan the determnants of RJV formaton. The nteracton of these varables wth ECdecson should reveal dfferent mpacts on the abnormal returns of the frms n the hypothess a RJV s formed to enhance colluson behavours. There are only few papers n the emprcal lterature that try to assess whether RJVs facltate collusve behavours. Moreover, the exstng lterature s based exclusvely on US data. Oxeley at al. (2009) also offer an event study methodology approach to detect possble colluson behavour through RJV. Instead of analyzng the determnants of the abnormal returns of partcpant frms, they analyze those of rval frms. They fnd that the abnormal returns of rvals move n the same drecton of the abnormal returns of partcpants and argue therefore that RJVs soften competton for the whole sector. Goeree and Helland (2008) try to detect a shft n the probablty for US frms of jonng a RJV after the ntroducton of a revsed lenency program. Duso et al. (2008) nvestgate on the determnants of the stablty of a RJV, argung that stable RJVs are more suspected of colluson. Duso et al. (2009) study nstead the shfts n market share due to RJV partcpaton. They argue that f no ncrease n market share s acheved thorough RJV partcpaton, then the frms mght have formed the RJV to collude. All of these works provde some evdence of antcompettve behavour enhanced by RJV formaton. My approach starts wth the general clam that some RJV are formed wth collusve ntentons. If ths s true, shareholders should be nformed of that, or at least they should 3 See the methodology secton for a detaled explanaton of the ndependent varables. 5

6 suspect t. To fnd out f ndeed some RJVs are formed wth collusve ntentons I therefore analyze the reactons of the shareholders. I observe whether shareholders behave dfferently for frms wth potental for colluson and frms wthout potental for colluson. If they behave dfferently ths mply that they had collusve expectatons from the RJV formaton. If shareholders had collusve expectatons ths most probably mply that at least some RJVs are ndeed formed to enhance colluson. On the other hand, f the collusve potental of a frm add no sgnfcant nformaton for the shareholders, they should behave the same way for both groups of frms. Ths means that all the varables ncluded n my specfcaton should have the same effect for both groups of frms,.e. those wth potental for colluson and those wthout potental for colluson. Ths n turn means that the nteracton terms of these varables wth the varable ECdecson should be all equal to zero. Ths s ndeed what I fnd. However, the varable ECdecson seem to have a postve effect but only when t s not accompaned by nteracton terms. A possble explanaton s that shareholders ndeed attach a postve value to the collusve potental of a frm, but they beleve at the same tme that the RJV wll produce gans besdes the collusve actvty. If ths s the case, the other varables do not have a dfferent mpact whether there s potental for colluson or not,.e. the nteracton terms are not sgnfcant. At the end of my analyss, I can only conclude that shareholders attach a postve value to the collusve potental of frms but I have not enough evdence to assess that RJVs n Europe are ndeed formed wth the ntenton of colluson. The remnder of the paper s as follow. Next secton ncludes a bref ntroducton of the European Anttrust legslaton and a revew of the European Commsson decsons. Secton 3 ncludes the lterature revew. In secton 4 I present the Methodology. Secton 5 dscusses the results. Fnally, secton 5 offers the conclusons. 2. European anttrust law The foundatons of the European anttrust legslaton are ncluded n the artcles 81., 82., 83., 84., 85., and 86. of the Treaty of the European Unon. Artcles 81 and 82 are of major relevance. Art. 81. drectly forbds agreements drect to prevent, restrct or dstort competton. Paragraph 3 of Art. 81 lsts the cases n whch the artcle may be napplcable. Art. 82. refers to the cases of abuse of a domnant poston. Other than n 6

7 the Treaty, anttrust rules are contaned n a seres of regulatons, notces and gudelnes adopted by ether the Commsson or the Councl of the European Unon. Relevant to the R&D jont ventures are the block exemptons to the Artcle 81 of the EC Treaty. The block exemptons are a seres of exceptons for whch the Artcle 81(1) mght not be applcable. They contan a seres of specal cases for whch the European Commsson mght tolerate a certan degree of nfracton of Art 81(1). The block exemptons are dvded nto vertcal agreements, horzontal agreements and lcensng agreements. Two regulatons ncluded n the horzontal agreements, Commsson Regulaton (EC) No 2658/2000 and Commsson Regulaton (EC) No 2659/2000, are of partcular relevance for R&D jont ventures. The latter, n partcular, declares that Art 81(1) mght not applcable n the case of jont research and development, even when these agreements are ndeed restrctve of competton as prescrbed by Art. 81(1) European Commsson Decsons Carree, Günster and Schnkel (2008) propose a revew of all the 538 formal decsons on anttrust publshed on the Offcal Journal of the European Communtes that the European Commsson adopted from the Treaty of Rome n 1957 untl of the 538 Commsson decsons regard the formaton of jont ventures. To notce s that they exclude from ther work all the cases n whch an nvestgaton has been dropped by the Commsson for lack of evdences, and nclude consequently only the cases for whch a formal decson has been adopted. The decsons made by the Commsson can take three forms: negatve clearance, exempton or nfrngement 4. In the case of nfrngement, remedes and sanctons may follow. The paper presents a statstc analyss on the type of decsons, on the sector of the partes, on the economc motvaton of the partes, on the nvestgaton duraton and on the amount of mposed fnes. Regardng the 62 formal decsons on jont ventures 11 receved negatve clearance, 49 receved exempton and only 2 were declared as nfrngements. Is nterestng to notce the dfference wth the horzontal constrants category for whch n 130 out of 219 decsons, the Commsson declared nfrngement. Another nterestng 4 Negatve clearance s declared when, based on the avalable nformaton, the practce object of the nvestgaton does not nfrnge Art 81. Exempton means that the practce under nvestgaton does enter n the scope of Art 81 but that block exemptons apply. Exemptons are only vald for lmted and specfed perods. Fnally, nfrngement s declared when the practce volates Art 81 and no exempton s applcable. 7

8 analyss s the subdvson by Economc sector. Carree at al. fnd that the majorty of the decsons concern the manufacturng, communcaton and transport ndustry. Dstngushng nstead between the three types of possble decsons, they fnd that nfrngement ncreases over tme wth respect to exempton and negatve clearance. A forthcomng book by Carree, Russo, Schnkel and Guenster focuses on the economcal analyss of the Commsson decsons above mentoned. A chapter s fully dedcated on the jont ventures decsons and t ncludes the lst of the 62 decsons accompaned by the detals of the jont ventures nvolved. The decsons are dvded n three categores: the ones regardng R&D jont ventures, the ones regardng Marketng, Sellng and Producton jont ventures and the ones regardng Strategc and Technologcal Allances. The 34 decsons regardng R&D jont ventures consttute the startng pont of my dataset. 3.Lterature revew 3.1. RJVs and Colluson, ndrect approaches The emprcal lterature on RJVs and Colluson s rather scarce for two man reasons. Frst, colluson s not easy to detect and therefore to prove t wth an emprcal exercse s always necessary to use ndrect ways that mply somewhat subjectve assumptons. Second, data on RJVs are prvate nformaton and consequently not easy to obtan. At ths regard, most of the emprcal lterature on the topc s based on US data as US frms, pror formng a RJV are requested to fll n wth the US Natonal Cooperatve Research Act that render the data collecton easer. Fve recent works attempt to prove ndrectly that colluson s one of the reasons for RJVs formaton, or at least that RJVs facltate tact colluson n the product market. Goeree and Helland (2008) buld up a model to test whether there s a varaton n RJVs formaton after the ntroducton of a new lenency program. They construct two samples, an across ndustry and a telecommuncaton one, ncludng US RJVs over the perod The dea at the base of ther research s that f RJVs are formed wth absolute no ntenton of colluson then the ntroducton of the new polcy should not nfluence the formaton of RJVs. Ths approach s partcularly vald as the new lenency polcy has no nfluence on the formaton of RJVs per sè but t does affect RJVs 8

9 formaton wth collusve ntentons. Goeree and Helland study RJVs formatons n the US n 1993, the year n whch the lenency program has been revsed. The man novelty of the polcy was that amnesty was granted only to the frm that frst flled n wth the Anttrust Dvson of the Department of Justce. They used the year 1993 as a dummy n a regresson model that has as a dependent varable the probablty of formng a RJV. Together wth the dummy year, Goeree and Helland nclude n ther model other varables that accordng to the exstng lterature 5 are lkely to nfluence the probablty of formng an RJV. Ther results suggest that the revson of the lenency polcy leads ndeed to a sgnfcant reducton of the probablty of formng a RJV. Moreover, when restrctng ther test to the sole telecommuncaton ndustry they fnd a reducton of around 25% of the probablty for a frm to jon a RJV n that ndustry. Ther overall fndngs are supportve of the dea of RJVs as a mean of colluson. One of the lmtatons of ther work, however, les n the fact that they do not drectly observe some of the varables ncluded n ther model. The probablty for a frm to jon a RJV s determned by the dfference between the value for the frm of jonng that specfc venture and the value for the frm of not jonng that specfc venture. If the dfference s postve the frm has a hgher probablty of jon. The value for a frm of jonng/not jonng s nfluenced by a number of dfferent varables lke the change n R&D ntensty, the market share and other frm-specfc varables. However t s only possble to observe the value of these varables when the frm actually jons the RJV. Goeree and Helland use expected values of these varables to predct the value for the frm of not jonng that specfc RJV. Gven that they use expected values nstead of observed values, ther results are affected by a certan degree of subjectvty. In my research I face a smlar problem. I do not observe the returns of a frm n the case of no announcement of a RJV; nstead, I only observe the returns of a frm when t does announce a RJV. Lke Goeree and Helland I have to use expectatons nstead of true values and therefore my results, lke thers, are affected by a certan degree of subjectvty. 5 Marín, Sots and Hernán. (2000) fnd that R&D ntensty, ndustry concentraton, frm sze and past experence are all factors that postvely nfluence RJV formaton. Roller, Sebert and Tombak. (2007) fnd smlar results. They extensvely examne ncentve and dsncentves for RJV formaton. Incentves for RJV formatons are cost-sharng opportuntes, reducton of free-rder problems created by R&D spllovers and the fact that frms produce complementary products. They fnd that the factors that have the major mpact on the choce for two frms to partcpate together n a RJV are frm-sze asymmetry, number of partcpants n the RJV, ndustry of the partcpants, the mpact on R&D nvestments and the fact that frms have prevously partcpated n other RJVs. 9

10 Another approach s proposed by Duso, Pennngs and Seldeslachts (2008). Ther sample ncludes 785 US RJVs over the perod They test the hypothess of RJVs as a mean of colluson through the analyss of the stablty of a RJV. They argue that frms may delay research and development and hence reman n the jont venture longer than necessary wth the sole scope of collude. Stablty s postvely nfluenced by a large number of partcpants, as larger RJVs are found to be more stable, and negatvely nfluenced by the ext or entry of members. Duso et al. use for ther model a dummy stable as a dependent varable verfyng whether varables lke ndustry concentraton, sze of RJV, R&D ntensty and presence of non-proft nsttutons can explan the stablty of a RJV. They fnd that large RJVs n hghly concentrated markets are more stable and therefore more prone to colluson. The lmtaton of these results however les on the fact that the only thng that s explaned and proved n ths work s the postve relaton between the stablty of a RJV and a number of frm and ndustryspecfc varables that nfluence t. The lnk between the stablty of a RJV and ts use as a mean of colluson remans a mere assumpton for whch they do not provde further proof. In my research, on the contrary, I try to establsh a drect lnk between the ntent to collude and the abnormal returns of frms. The lnk s depcted by the dummy ECdecson 6.Ths dummy varable separates my sample n two parts, on one sde I have the frms for whch the European Commsson adopted a formal decson regardng Artcle 81 of the European Treaty and that therefore can be suspected of colluson from the shareholders pont of vew. On the other sde, I have a random draw of frms for whch the European Commsson never adopted a decson on the matter of anttrust. However, my hypotheses have to rely on one assumptons as well: the shareholders of the frst group of frms foresee the possblty of colluson, whch s later nvestgated by the Commsson. In other terms I assume that f a RJV can be restrctve of competton, the shareholders know (or can suspect) t from the moment t s announced. A dfferent approach to detect colluson s tred by Duso, Roller and Seldeslachts (2009) who try to support the hypothess of colluson, ths tme lookng at the market share of the partcpant frms. In ths paper, they use the same sample of 785 US RJVs over the perod used by Duso, Pennngs and Seldeslachts (2008). The man dea underlyng ther model s that a loss of market share due to RJV partcpaton s a suffcent condton to assume ntenton of colluson. They assume two possble 6 The varable s explaned n detal n the Methodology secton. 10

11 explanatons for RJV formaton, one s the realzaton of effcency gans, and the other s the realzaton of beneft through colluson. In the case of effcency gans the partcpatng frms also realze an ncrease n market share. They argue that f no ncrease n market share s acheved through the RJV then the frms must have form t wth the sole ntent of colludng. They analyze the relatonshp between market share and sze of RJVs networks. They fnd that partcpaton n medum sze RJVs networks operatng n the Chemcals, Petroleum, Industral and Commercal machnery, Electronc and Transportaton Equpment ndustres lead to a decrease of a frm market share and t s therefore supportve of colluson behavour.. Gugler and Sebert (2007). They study whether the effcency gans from horzontal mergers and RJVs are sgnfcant enough to justfy antcompettve effects. They fnd evdence of antcompettve effects through mergers, but not through RJVs formaton. Stll ther research s lmted to the semconductor ndustry only. Oxley, Sampson and Slverman (2009) am to detect possble antcompettve behavours n RJVs usng event study methodology. They use a pretty sngular approach that focuses on the abnormal stock returns of rval frms durng the days surroundng the announcement of a RJV. In the event study lterature 7 researchers fnd, n the majorty of the cases, that the announcement of a new partnershp leads to postve abnormal stock returns of the partcpant frms. Ths evdence, however, can be explaned by two dfferent and contrastng hypotheses. Partcpant frms are supposed to gan access to new nformaton, technology and resources and therefore to mprove ther performance through learnng process and effcency gans, otherwse mpossble wthout the R&D cooperaton. Shareholders therefore react postvely to the announcement a new partnershp. Ths frst explanaton s coherent wth the vew that RJVs are procompetton. On the other hand, however, f RJVs were a mean of colluson, and shareholders expected possble antcompettve behavour, they would also react postvely, generatng postve abnormal stock returns. Is therefore dffcult to dstngush whch of the two effects preval when studyng postve abnormal stock returns. To solve the problem, Oxley at al. decde to look at the correlaton between the abnormal returns of the partners and those of rval frms. They argue that f RJVs are formed to enhance nnovaton and progress among the partcpants, then the partners and the rvals abnormal returns should be negatvely correlated. On the other hand, f 7 For the event study lterature see followng secton. 11

12 the RJVs are formed wth the ntent of softenng product market competton, then the abnormal return of the partcpants and those of the rval frms should be postvely correlated. Ths dea follows from the assumpton that all the frms n the ndustry wll beneft from the reduced competton enhanced by the RJV. In ther study, Oxley at al. ndeed fnd postve correlaton between the abnormal stock returns of partcpants and those of the rval frms. They conclude therefore that some RJVs are expected to soften competton. In ther study they analyze 705 allances announcements over the perod n the telecommuncatons and electroncs ndustry and use as a defnton of rval frms those that have the same prmary 4-dgt SIC code as the partcpants. Havng the same 4-dgt SIC code however does not necessarly mply that two frms are drect rvals n the fnal product market. Even though they belong to the same 4-dgt SIC classfcaton, two frms could manufacture and sell products that are not drectly competng wth each other. The problem, however s that n most cases databases only offer the 4-dgt SIC code as a prmary code for frms. For researchers ths s stll the lnk more easly avalable between dfferent frms prmary sectors. Moreover ths analyss use the US market as the fnal market wthout takng nto consderaton possble dfferences among regons. For example two frms wth the same prmary 4- dgt SIC code mght not drect compete n the same geographcal area. The work of Oxeley at al. together wth that of Schut and Frederkslust (2004) 8 represent the startng pont of my analyss Event studes on allance formaton Researchers use event study methodology to examne how the stock market reacts to a specfc event. To do so they look at the abnormal stock returns of a frm n the day of the event. The event s related to the subject under study and t can be lterally anythng that s able to nfluence the behavour of nvestors n the stock market lke a poltcal electon or the announcement of a merger. For the purpose of my research, I lmt my lterature revew to the studes that analyze the stock market reacton to the event announcement of an allance. Fama et al. (1969) gve one of the frst contrbutons to the event study lterature. Ther paper ams to study how the frm market value reacts to the nformaton brought 8 For a descrpton of Schut and Frederkslust s research see followng secton. 12

13 about by the announcement of a stock splt 9. They nclude n ther study 940 stock splt announcements n the US over the perod In the majorty of the cases, a stock splt s followed by the announcement of a dvdend ncrease. As n general drectors of frms are unwllng to decrease dvdends, t s reasonable to assume that they wll ncrease the dvdends f and only f they expect a future and constant postve performance of the frm that wll allow them to mantan the dvdends at the ncreased level. Fama et al. assume therefore that the announcement of a stock splt s a sgnal of future postve performance by the frm. Ther assumpton s confrmed by ther fndngs. They found evdence of postve abnormal returns n the months precedng the announcement of the stock splts and evdence of negatve abnormal returns n the month followng the announcement of the stock splts for those frms for whch the announcements are not followed by an ncrease n dvdends. Hence ther man concluson s that the market s effcent, n the sense that t reacts to the dvdend nformaton ncluded n the announcement of a stock splt. My study bulds on the same dea: the market, n my case, should react to the expected benefts from colluson nformaton ncluded n the announcement of a RJV. McConnell and Nantell (1985) make use of the market model to test smlartes between the market reactons at the announcements of mergers and JVs. Ther am s to separate two possble dfferent causes of wealth gans n mergers: synergy and management dsplacement. To do so they solate the synergy effect trough the analyss of the market reactons to jont ventures, for whch there s no management dsplacement. Ther dea stands on the assumpton that mergers and jont ventures are characterzed by smlar synergy patterns. Ther study ncludes 136 jont ventures announced durng the perod They analyze the abnormal stock returns of partcpant frms for the days around the announcement of the JV. They fnd that lke n mergers, the partcpants of a jont venture gan postve abnormal returns. When testng for sze they fnd that the smaller frms beneft from larger excess rates of return than lager frms. When convertng the excess gans n dollars, though, they fnd that n general the gans are equally dstrbuted. They can conclude that synergy s the prevalent explanaton for wealth gans n mergers. 9 A stock splt s the decson to ncrease the number of outstandng share of a frm by ssung new shares and offerng them to the current shareholders. As the number of shares ncrease, ther prce decrease, gven that the frm mantan the same captalzaton. 13

14 Woolrdg and Snow (1990) study the cumulatve abnormal return for 767 announcements made by 248 companes and reported on the Wall Street Journal n the perod They test three dfferent hypothess regardng strategc decsons adopted by frms managers. The frst one s what they call the Shareholder Value Maxmzaton hypothess. Ths hypothess states that shareholders rewards managers for takng strategc decsons that wll mprove the long-run performance of the frm, lke for example the announcement of a RJV. If ths hypothess s true, then market reactons should be postve. The second hypothess s what the authors call the Insttutonal Investors one. They argue that n US large and nfluental groups of nvestors lke penson and nvestments funds or bank trust departments prefer short-run profts to long-run ones. If ths second hypothess s true, then the market wll react negatvely to the announcement of a long-run strategc decson lke a RJV formaton. Fnally, what they call the Ratonal Expectaton hypothess predcts no reacton by the market as the shareholders expect the managers to take perodc strategc decsons n order to mantan (and not to enhance) the frm compettveness. They take nto consderaton announcements of jont ventures, R&D allances, product and market dversfcaton and captal expendtures. They also test for the sze of the nvestment and the duraton of the project. Ther evdences, n lne wth the rest of the lterature, are supportve of the Shareholder Value Maxmzaton hypothess. They found stronger postve return for longer projects, but not substantal dfferences between small and large nvestments. Wth respect to the dfferent types of announcements, they fnd that JVs and R&D allances report stronger postve returns than product and market dversfcaton and captal expendtures. Koh and Venkatraman (1991) use event study methodology to test the mpact of dfferent JVs formaton strateges on parents market value for the IT sector. Ther sample ncludes 175 JVs and 239 frms over the perod They test for relatedness among the prmary category of the parents and between the category of the parents and the prmary category of the JV. They also test the mpact of market shares asymmetres and more n general the mpact of the sze of the partcpants n an equally owned JV. In lne wth the lterature, they fnd that JV announcements results n postve abnormal returns. To test for ther dfferent hypothess they smply dvde the frms n ther sample nto dfferent categores. Then they analyze the average abnormal returns of the categores. They frst dvde the parents n four categores dependng on the JV role,.e. dependng on weather the JV wll manufacture a totally new product, a 14

15 new product that s related wth the parents busness or whether t wll enter a new geographcal market or t wll enter new sectors n the same geographcal market of the parents. They fnd that dfferent JV roles have a dfferent mpact on the frms abnormal returns. More specfcally ther results ndcate that the parents of RJVs that produce related products or that operated n related markets earn hgher average abnormal returns. Furthermore, they dvde the parents dependng on whether ther actvtes are related to the actvty of the jont venture or not. They also dvde the sample nto related parents and unrelated parents. They arrve to the concluson that both, the relatedness among the actvtes of the parents and between the actvty of the parent and the actvty of the JV, leads to hgher average abnormal returns. Fnally, they dvde ther sample for sze, separatng larger and smaller parents. They fnd that smaller partners report on average hgher gans than larger ones. Das, Sen and Sengupta (1998) use a smlar event study methodology to test dfferent hypothess regardng abnormal share returns of companes partcpatng n 119 strategc allances over the perod Ther focus s on the nature of the allance, dstngushng between technologcal and marketng allances; on the formaton process, ndvduatng whch frm s the frst mover; on the sze and on the relatve dependence of the partners. They fnd that the announcement of a technologcal allance produces larger postve returns than the announcement of a marketng allance. Moreover, they fnd that the proftablty of the partners and ther abnormal share returns are negatvely correlated. Fnally, they fnd evdence that smaller partners enjoy greater abnormal returns than larger partners. Anand and Khanna (2000) use a smlar model to assess whether cumulated experence n managng jont ventures and lcence agreements can explan the abnormal return of the stocks around the day of the announcement of successve agreements. Ther sample ncludes 1976 JV and lcenses by 147 frms over the perod They fnd that learnng process are mportant n the case of jont ventures only and not n the case of lcensng agreements. Moreover, n lne wth prevous emprcal studes, they fnd that these learnng effects are greater for R&D and manufacturng jont ventures than for Marketng or other forms of jont ventures. Reuer and. Koza. (2000) study two dfferent explanatons for jont ventures: the ndgestblty hypothess and asymmetry of nformaton hypothess. Ther dataset ncludes 297 European and US jont ventures, termnated n the perod between 1985 and They dvde ther sample nto four group based on the degree of asymmetry 15

16 nformaton between the partners. They fnd stronger postve abnormal returns for the group wth the largest degree of nformaton asymmetry (when the prmary 3-dgts SIC codes of both partners and that of the JV all dffers). They conclude n favour of the asymmetry of nformaton hypothess. A paper by Schut and Frederkslust (2004) bulds on the model developed by Fama at al. and study the shareholders wealth effects of 233 RJVs announcements n the Netherlands over the perod Schut and Frederkslust am at establshng the strategy factors that have an mpact on the stock prces of the partcpants companes. They dvde these factors nto three categores: Strategc Content, Strategc Context and Strategc Control. The frst group ncludes varables that dscern between the motves of a RJV: market, technology, and effcency. The varable dversfcaton also belongs to the Strategc Content. Ths varable measures the relatedness between the prmary sector of the RJV and the prmary sector of the partcpants. To the Strategc Context belong the varables Indvdualsm, Related Sze, and Partner Relatedness. In the group Strategc Control Schut and Frederkslust nclude the dummes Mnorty, Majorty and Equalty, wth the ntent to capture the effects of the ownershp structure. They fnd that the 75% of the RJV announcements ncluded n ther research produce postve abnormal returns. All the varables they ncluded are sgnfcant and have the expected sgn. In my study I ncluded many of the factors ndvduated by Schut and Frederkslust n ths research. 4. Methodology 4.1. Sample and Data My dataset s consttuted by two dstnct sub-samples. The frst sub-sample contans data on 27 RJVs for whch the European Commsson adopted a formal decson relatve to Artcle 81 of the Treaty of the European Unon. The second sub-sample represents a random draw of equal sze from a populaton of European RJVs. Obtan the true populaton of RJVs formed n Europe for the perod under study s very dffcult. However, a good approxmaton can be obtaned through the database Securtes Data Company, Jont Ventures and Allances (SDC). I therefore retreved 16

17 from SDC the 3729 RJVs formed n the EU 27 countres from 1984 untl today. As the frst sub-sample only covers the perod , I lmted the SDC sample untl the year Moreover, not all the observatons n SDC nclude data regardng the ownershp structure of the RJVs. For ths reason, I had to restrct further the sample to 728 RJVs, ncludng only those observatons for whch the nformaton on ownershp was avalable. Usng the program Evews I extracted a random sample of 27 observatons, matchng the sze of the frst sub-sample. To establsh the date of the announcement of the RJVs I used, for the frst subsample, the reports of the European Commsson publshed n the Offcal Journal of the European Communtes together wth the database LexsNexs News. I was not able to determne the dates for 7 RJVs and I therefore excluded them from the sample. As the dates reported n SDC are not always accurate 10, I followed the same procedure for the second sub-sample. The process leadl to the modfcaton of 5 dates. Other observatons have been excluded durng the curse of the analyss because of the pecular behavour of the asset prces under study. The fnal dataset s a combnaton of the two sub-samples and consst of 70 frmobservatons related to 42 RJVs over the perod For lack of data, ths frst dataset does not allow to control for the sze of the frms. To collect frm-specfc data I used the database Thomson One Banker; ths source, however, dd not contan data on the number of employees, on the total assets and on the sales for some of the frms ncluded n my sample. I therefore restrcted t to 535 frm-observatons complete of sze data. The model s tested on both, the 70 frm-observatons sample and the 55 frmobservatons sample. RJV specfc data lke naton, name and number of partcpant frms, ownershp structure and deal descrpton come from the reports of the European Commsson for the frst sub-sample and from SDC for the second sub-sample. SDC also contans data on the prmary 4-dgt SIC code of the partcpant frms and of the venture tself. The prmary 4-dgt SIC codes of the frms n the frst sub-sample come nstead form Thomson One Banker. No nformaton was provded regardng the prmary 4-dgt SIC code of the RJVs n the reports of the European Commsson nor elsewhere. However, followng the accurate descrpton of the deal that the Commsson made for every sngle case, I was able to attrbute a rather accurate 4-dgt SIC code to every venture n 10 Other researchers fnd naccuracy n the dates reported by SDC. See for example Anand and Khanna (2000); and McGahan and Vllalonga. (2005). 17

18 the frst sub-sample. Ths attrbuton however, remans subjectve. The same apply for the classfcaton of the RJVs nto two categores dependng on whether the actvty of the RJV s only lmted to R&D or whether the partcpant frms also decde to manufacture and/or market the products together. To classfy the RJVs nto the two groups I pad careful attenton to the descrptons of the deals ncluded n the reports of the Commsson and n SDC. Frm specfc data lke number of employees, total assets and sales come from Thomson One Banker. The stock market data all come from DataStream. I collected daly share prces for each frm ncluded n the sample for a perod of 170 days pror the announcement of the RJV. For the same perods, I collected the daly prce of the Morgan Stanley Captal Internatonal Index (MSCI) World that I used as a benchmark. For the free rsk rate, I collected the daly yeld on 3 month US Treasury Bll The Emprcal Specfcaton The emprcal specfcaton I use for my research tres to capture the effect of the expected colluson nformaton ncluded n the event RJV announcement usng event study methodology. If shareholders react postvely to the announcement of a RJV because they expect some beneft from a possble antcompettve behavour, then all of the ncluded varables should have a role n explanng the abnormal return of the shares n the days surroundng the announcement. The varable ECdecson s meant to capture the shareholders expectatons of colluson. The varable dvdes the sample n two groups of frms: those that have been nvestgated by the European Commsson n relaton to Art 81 of the Treaty of the European Unon and those for whch a formal decson has not been adopted. I defne the frst group of frms as those that can be suspected of colluson and the second group of frms as those not suspected of colluson. My hypothess rests on the assumpton that the same facts and crcumstances regardng the state of competton between the undertakngs that brought the European Commsson to adopt a formal decson were known to the shareholders at the moment of the announcement. All of these cases have been brought to the attenton of the European Commsson by notfcaton of the partes themselves, as requred by Councl regulaton 17, effectve untl May Untl that date the undertakngs had the oblgaton to notfy to the 18

19 Commsson any arrangements that could, even only potentally, fall wthn the scope of Artcle 81. The varable ECdecson try then to capture n a backward lookng fashon, the shareholders expectatons of a possble antcompettve behavour gven the fact that they knew, at the moment of the announcement, that the RJV had such characterstcs that made t fall nto the scope of Art 81. The other varables I ncluded n the specfcaton have been used prevously n lterature ether to try to detect colluson trough RJV formaton or ether to try to depct the determnants of RJV formaton. The varable OtherActvty try to capture to whch extent the frms nvolved n the RJV are workng together. The role of a RJV can be confned to sole R&D actvty but t can go further and nclude the manufacturng phase, the marketng phase as well as the supplyng phase. Each of these steps requres a hgher degree of nvolvement of the frms and the dsclosure of frm-senstve nformaton. The more frms work together sharng technology, processes, best practces, supply channels, etc. the better ther coordnaton become. The ratonale of ths varable les on the assumpton that f the partes set up the RJV wth the ntent of softenng the competton, a jont collaboraton n the manufacture and/or market and/or supply phases can only mprove ther coordnaton. Therefore, f the RJV are ndeed a mean of colluson the varable OtherActvty s expected to appear wth postve sgn. In the emprcal lterature 11 however, R&D and Technologcal agreements n general are consdered more beneft enhancng than marketng ones. If the motves underlyng the RJV are not those of antcompettve behavour, then one should expect a negatve sgn nstead. To capture the dfferent expected mpact of ths varable on the abnormal returns of the shares dependng whether the reasons underlyng RJV formaton are those of colluson or not, I nclude an nteracton term between the varable ECdecson and OtherActvty. The combnaton of these two varables s therefore excepted to carry a postve sgn. 11 For example, Das, Sen and Sengupta (1998) argue that Technologcal allances are characterstc of emergng markets, whle marketng agreements occur most often n mature market. The stock returns n case of a marketng announcement are lower smply because shareholders expect benefts to last for a shorter tme. Moreover, they argue, marketng allances can be a sgnal of weakness. Koh and Venkatraman (1991) also fnd stronger returns n case of Technologcal agreements wth respect to Marketng, Lcensng and Supply ones. Oxley, Sampson and Slverman. (2009) argue for the two dfferent and contrastng effects of JV that combne R&D and other actvtes, although, n lne wth the lterature, they fnd a negatve mpact for Market allances. 19

20 The varable Blateral dvdes the RJV n the sample nto two categores: those formed by two members and those formed by more than two members. Ths varable s meant to capture the stablty and therefore the easness of coordnaton n a RJV. When RJVs are formed by more than two members, eluson of competton s more dffcult to acheve as all the members have to effectvely coordnate to acheve colluson. The stablty of the venture n general decrease wth the number of partcpants (Duso et al. 2008) and effectve coordnaton and cooperaton toward colluson becomes more dffcult. On the other hand, the hgher the number of partcpants, the larger the benefts due to costs sharng. To control for the dfferent nfluence that ths varable should have for frm wth ntentons of colluson and frms wthout ntentons of colluson I construct an nteracton term for the varables Bltlateral and ECdecson. The varable s therefore expected to appear wth a postve sgn. The varable Partcpants s defned as the number of partcpants n a RJV and t s subject to the same consderatons of the varable Blateral. The hgher the number of partcpants, the more dffcult the coordnaton. Even for ths varable, I construct an nteracton term wth ECdecson n order to capture ts dfferent mpact. The hgher the value of ths nteracton varable the lower the expected mpact on the share returns of the partcpant frms. A negatve sgn s therefore expected. The varable Horzontal dstngushes between agreements n whch two or more partcpants operate n the same sector and agreements n whch none of the partcpants operates n the same sector. The ratonale for ths varable s easy to understand: the more related are the partners sectors, the easer the coordnaton toward market colluson. 12 More mportant, the jont market share reached trough the jont venture ncreases the market power of the partcpant frms. It s nterestng to verfy whether the nteracton of the varables Horzontal and ECdecson produces the postve effect that s expected. The nteracton term s therefore expected to carry a postve sgn. Another factor of nterest for the shareholders s the magntude of the benefts the frm wll acheve through colluson. The bgger the market share of the frm and that of ts partners, the hgher ther gans. Unfortunately, I was not n possesson of the data regardng the market shares of the partcpants. I therefore use the sze of the frms, measured by annual sales, total assets and number of employees as a proxy for market 12 Oxley et al. (2009) draw a smlar concluson argung that horzontal agreements n concentrated sectors facltate market coordnaton. 20

21 share. Yet, for the purpose of my research, studyng the relatve sze of a frm wth respect to ts partners other than the sze of the frm tself would be of nterest. Agan, I was not n possesson of the sze data for all the partners and therefore I could not construct a varable for the relatve sze. Even for the varable sze there exst dfferent and competng hypothess 13 n lterature. To assess the dfferent effects of Sze for the two groups of frms I construct an nteracton term wth the varable ECdecson. The term s expected to carry postve sgn, as the sze of a frm can be drectly lnked to the magntude of ts gans n case of collusve behavour. The larger the sze, the hgher the gans. The ownershp structure of the jont venture says whether the frm has a majorty, a mnorty or an equal nterest n the venture. When the control s equally dvded between the partcpants t s possble to reach a hgher coordnaton. A dsparty of control n fact, could not result n the same degree of coordnaton as the frm wth the lower partcpaton mght not be eager to engage n a collusve behavour when t has to submt to the other frm s decsons. At ths regard, t s nterestng to analyze the nteracton effect of the varable Equalty wth the varable Sze. If a large frm engages n an equally owned RJV ths could be a sgnal for the other partcpants and for the shareholders that t s favourable to a frutful coordnaton. As argued before, ths effect would be better captured from the relatve sze of the frm s partners. Agan, I was not n possesson of ths data. When the ownershp structure allows for a frm to have a majorty nterest, effectve coordnaton mght not be acheved therefore rendng colluson more dffcult. If ths s the case, the varable Majorty should appear wth a negatve sgn. However, n case a majorty nterest s present, and f n spte of the coordnaton problems the frms ndeed acheve effectve colluson, the gans wll be most probably hgher for the frm that owns the majorty nterest. In ths case the reacton of the shareholders should be postve. Whch one of the effects wll preval remans therefore uncertan. Even for the ownershp structure I nclude the nteracton term. The coeffcent s expected to carry a postve sgn for the nteracton ECdecson*Equalty, whle the sgn of the nteracton ECdecson*Majorty remans uncertan. 13 Koh and Venkatraman (1991) fnd that smaller frms earn postve abnormal returns, whle the returns of the larger partners are nsgnfcant. Roller et al. (2007) argue a smlar hypothess. Marìn at al. (2000) argue that the absolute sze of a frm n terms of number of employees and total assets, also capture the capablty of the frm to cope wth possble large fxed costs related wth the RJV. The sze of a frm can be a sgnal for the shareholders that the frm s ndeed able to sustan the costs related wth the new venture. Goeree at al. (2008) nfer that frm wth larger assets are more lkely to engage n RJV. 21

22 The varable Dversfcaton depcts the degree of relatedness between the prmary sector of the partcpant frms and the prmary sector of the RJV 14. The hgher the value of ths varable, the greater the dstance between the actvtes of the RJV and those of ts parent frms. If frms form RJVs to enhance antcompettve behavour, the gans wll be hgher f the RJV s able to soften the competton n the prmary market of the frm. It follows that ths varable should appear wth a negatve sgn. To dstngush ths effect for those frms more suspected of colluson I nclude the nteracton term ECdecson*Dversfcaton, whch s expected to carry a negatve sgn. The model s specfed as follow: CAR = α + β0ecdecson + β1otheractvty + β2blateral + β3partcpants + β4 Horzontal + β + 5Sze + β 6Equalty + β 7Majorty + β8dversfcaton β 9 ECdecson * OtherActvty + β ECdecson + ECdecson * Partcpan ts 10 * Blteral + β10 Equalty * Sales β11 + β ECdecson + Ecdecson * Equalty 12 * Horzontal + β13 ECdecson * Sze β15 + β * + 16 ECdecson * Majorty + β17ecdecson Dversfcaton ε (1) The dependent varable The dependent varable represents the cumulate abnormal returns of a frm s share prce over the days surroundng the announcement of a RJV. The abnormal returns have been calculated usng the smple verson of the Captal Asset-Prcng Model 15 (CAPM) as follows: E( er t ) = α + β ( emr ) + ε (2) t t where: 14 Balakrshnan and Koza (1993) constructed ths varable to depct the relatedness of to partner frms. Schut and Frederkslust (2004) ncluded t n ther research as measure of the relatedness between the actvtes of the partcpant frms and those of the RJV. 15 See e.g. Levy and Post. (2005). 22

23 E ( ert ) represents the expected excess return over the free rsk rate of the asset at tme t emr t represents the excess market return over the free rsk rate at tme t α s the constant term of asset β represents the systematc rsk of asset ε t s the error term Followng the lterature n event study methodology 16 regresson (2) s estmated for a perod that goes from 50 untl 170 days pror the announcement of the RJV. The estmated excess return s then subtracted from the effectve excess return to obtan the abnormal return of share at tme t: AR t = er E er ) (3) t ( t Abnormal returns are calculated for the day of the announcement.e. day 0 and for the precedng and the followng day.e. day -1 and day +1. I then calculated two cumulated abnormal returns, one ncludng the three days and one ncludng only day 0 and day +1: CAR 1 = AR 1 + AR 0 + AR + 1 (4) CAR 2 = AR0 + AR + 1 (5) 16 McConnel and Nantell (1985) estmate ther model for the perod that goes from 180 untl 61 days before the announcement; Das et al (1998) use a larger span, from 200 untl 10 days before the announcement; Reuer and Koza. (2000) estmate ther model from 250 untl 50 days before the announcement; Schut and Frederkslust (2004) for the perod startng 200 untl 51 days before the announcement, fnally Oxeley et al. (2009) use the span from 170 untl 21 days before the announcement. 23