TY - JOUR
AU - Sovinsky, Michelle
AB - Abstract Every year thousands of firms are engaged in research joint ventures (RJV), where all knowledge gained through research and development (R&D) is shared among members. Most of the empirical literature assumes members are non-cooperative in the product market. But many RJV members are rivals leaving open the possibility that firms may form RJVs to facilitate product market collusion. We examine this by exploiting variation in RJV formation generated by a policy change that affects the collusive benefits but not the research synergies associated with a RJV. We use data on RJVs formed between 1986 and 2001 together with firm-level information from Compustat to estimate a RJV participation equation. After correcting for the endogeneity of R&D and controlling for RJV characteristics and firm attributes, we find the decision to join is impacted by the policy change. We also find the magnitude is significant: the policy change resulted in an average drop in the probability of joining a RJV of |$41\%$| among computer and semiconductor manufacturers, |$34\%$| among telecommunications firms, and |$33\%$| among petroleum refining firms. Our results are consistent with research joint ventures serving a collusive function. “People of the same trade seldom meet together, even for merriment and diversion, but the conversation ends in a conspiracy against the public, or in some contrivance to raise prices.” Adam Smith in Wealth of Nations 1. Introduction Every year thousands of firms are engaged in research joint ventures (RJVs), an agreement in which all knowledge gained through research and development (R&D) is shared among members. RJVs often provide pro-competitive benefits, such as shared risk, increased economies of scale in R&D, asset complementarities, internalized R&D externalities (i.e., overcoming free-rider problems), alleviated financial constraints, and shared cost. However, by construction, RJVs offer firms an opportunity to coordinate. As Martin (1995) notes, “It is conceivable that firms that start to work very closely on R&D projects might start to extend the coordination of their behavior onto other spheres of the life of the firms.” There are numerous ways in which R&D collaborations may lead to collusive product market behavior. For instance, RJV formation could centralize decision-making by combining collaborative efforts with control over competitively significant assets, by imposing collateral restraints that restrict competition among participants, by including member firms’ individual R&D in the collaborative effort, by facilitating the exchange of competitively sensitive information, or by functioning as a vehicle to signal cooperative behavior. Finally, production joint ventures, which involve jointly manufacturing a new or improved product, typically involve agreements on the output level, the price of the joint product, or other competitive variables. Furthermore, it is not uncommon for RJV members to be product market rivals. Examples of direct product market competitors involved in joint RJVs include Xerox and Dupont who formed a RJV to develop copying equipment; Shell and Texaco to refine crude oil; General Motors and Toyota to produce a new type of car; Merck and Johnson & Johnson to develop new over the counter medicines; MCI and Sprint to provide enhanced telecommunications services; Samsung and Sony to develop LCD panels; and SEMATECH, a consortium of leading semiconductor manufacturers established to improve semiconductor manufacturing technology. The possibility that firms may undertake legal RJVs as a means to facilitate illegal product market collusion has generated regulatory scrutiny in a wide variety of industries and RAs.1 Estimating the impact of the returns to collusion on the decision to join a RJV is difficult as there are many legal reasons to join which also result in increased market power of the members. One option is to consider a subset of firms engaged in RJVs and another subset not engaged in RJVs and test whether collusion is higher among the first group. However, such a test would only be able to tell us something about collusive behavior that was detected, but would not inform us about the prevalence of firms that form RJVs with collusive intentions but are not caught. An additional problem is the endogeneity of the choice to join a RJV. In this paper, we propose a test of whether the data are consistent with firms forming RJVs as a way to facilitate collusion in the final goods market. Rather than directly testing for collusion by firms engaged in RJVs, we examine their potentially collusive function through a quasi-experiment. The quasi-experiment examines whether revisions of the antitrust leniency policy in the 1990s, which were enacted to detect collusive behavior, made firms more or less likely to join RJVs. We argue that the policy revision made applying for amnesty easier and more attractive and, hence, reduced the gains from trying to establish a collusive relationship because coconspirators would be more likely to defect and seek amnesty. This change in the value of collusion should change the benefit of joining a RJV only if membership serves some sort of collusive function at the margin. There is empirical evidence that suggests such an investigation is worthwhile. For example, in the next section, we provide evidence of cases where rival firms were in collusive arrangements with RJV members, and one or more of these firms applied for leniency protection. We also examine whether the policy revision differentially impacts firms for whom collusion might be more valuable. To do so, we develop a measure of the RJV’s collusive value to a firm that is considering whether to join a particular RJV. The firm-specific measure of “RJV market power” allows us to obtain a heterogeneous treatment effect of RJV participation. Determining the entire shape of the curve relating the probability of joining a RJV to the market power of the RJV allows us to make a more precise inference on the collusive potential of RJVs. Our test of a RJV’s collusive function is (i) whether the revised leniency policy changed the probability that firms join a RJV and (ii) whether the policy has a differential impact if the RJV market power is larger. Our approach has the advantage that we are able to examine the collusive potential of RJVs without observing costs or prices. One problem in measuring collusive intentions, which plagues the majority of studies of collusion, is defining the relevant product market (Eizenberg and Kovo 2017). To this end, we consider many definitions of the relevant market, ranging from very broad (e.g., three-digit North American Industry Classification System (NAICS)) to very narrow and industry specific (e.g., the market for long distance carriers under the period of telecommunications regulation). We apply our quasi-experiment to three industries with a history of antitrust suits via joint RJV participation: petroleum manufacturing, computer and electronic product manufacturing, and telecommunications.2 We find that the decision to join a RJV is impacted by the policy change and that this impact is significant across market definitions. Specifically, we find that the revised leniency policy reduces the average probability that computer and semiconductor manufacturers join an RJV by |$41\%$| (range of |$21\%$|–|$90\%$|⁠); with a reduction of |$34\%$| (range of |$20\%$|–|$94\%$|⁠) among telecommunications firms, and among firms in petroleum refining the probability decreases by |$33\%$| (range of 24%–|$54\%$|⁠). Our results are consistent with RJVs serving (at least in part) a collusive function. The channels through which R&D cooperation facilitates product market collusion have been examined in a number of theoretical studies. RJVs provide an opportunity for firms to talk openly, exchange information, and coordinate strategies explicitly. The seminal paper of d’Aspremont and Jacquemin (1988) examine collaborative R&D and finds that, in many cases, welfare is reduced if firms collude in output. Greenlee and Cassiman (1999) develop a model in which RJV formation and the decision to collude in the product market are endogenous. They also find that RJVs should not be supported if they involve product market collusion. In addition to explicit collusion, RJVs can enable tacit collusion through creating common assets. Martin (1995) shows that self-enforcing R&D makes it more likely that tacit collusion can be sustained in the product market. In contrast, Levy (2012) finds that limitations on the formation of RJV may not have much of an effect on firms’ ability to collude tacitly, unless alternative forms of technology sharing (such as licensing) are constrained as well. Lambertini, Poddar, and Sasaki (2002) consider horizontal product differentiation and finds that cooperative R&D agreements can destabilize collusion if firms develop the product jointly. Miyagiwa (2009) analyzes the effect of RJVs on consumer welfare in an international context and finds that international RJVs can be welfare enhancing even under tacit collusion. Whereas Cabral (2000) examines R&D collaboration with differentiated probabilities of success through unobservable efforts, and find that product market prices may decrease from the agreements. Further, Cooper and Ross (2009) show that RJVs may induce collusion if they enable firms to signal cooperative behavior. The closest empirical work on this topic is that of Duso, Hendrick Röller, and Seldeslachts (2014). They show that one can examine RJV member market shares to learn about the collusive nature of the venture. Using this motivation, they estimate a market share equation for firms involved in RJVs distinguishing between firms that compete in the same product market (defined by the four-digit Standard Industrial Classification code) from those that do not. They find that product market rivals experience a decline in market shares on average after joining a RJV, which implies that these RJVs are conducive to collusion. Our findings are complementary to theirs, while our approach differs in every respect excepting the data sources. First, we identify the collusive potential of an RJV by using a quasi-experiment that made collusion harder to sustain. Second, we provide motivation for why it is important to define the relevant market carefully in evaluating RJV outcomes, and we define the relevant product market in a variety of ways spanning broad categories to narrow definitions. Third, as we discuss later, it is unlikely that all RJVs survive for the whole sample period, which is important when examining which RJVs are available for the firm to join. We implement a strategy that takes into consideration which RJVs are available for firms to join based on a variety of ways of computing the ending date (which is not observed in any data). Finally, we allow RJVs to have a different impact in terms of collusive value depending on the size of the industry relative to the size of the RJV and the number of firms involved. Both Duso, Hendrick Röller, and Seldeslachts (2014) and this work evolved simultaneously and are nice complements to each other in that we examine a similar issue in different ways. This work is also related to earlier work by Scott (1988) who examined all RJV filings over an 18 month period and found that collaboration may have resulted in less competitive markets. Finally, our results are well in line with the findings of a controlled experiment that examines product market collusion in oligopolistic markets arriving from RJVs (Suetens 2008). The rest of the paper proceeds as follows. In Section 2, we provide background on antitrust investigations of collusion among RJV members, the legal policies surrounding RJV formation, and the impact of the leniency policy. We present the model and estimation technique in Section 3. In Section 4, we discuss the data. We present the results in Section 5. In Section 6, we provide goodness-of-fit and robustness results. In Section 7, we conclude. 2. Motivation and Background 2.1. Leniency Policy The Sherman Act of 1890 makes it illegal for firms to agree to fix prices or engage in other agreements that restrict output or harm consumers. In the United States, antitrust violators face criminal sanctions consisting of fines (for firms and individuals) and jail sentences. The Department of Justice (DOJ) Antitrust Division enacted a leniency policy program in 1978 designed to detect firms engaged in collusive behavior. In 1993, the DOJ substantially revised the program to make it easier and financially more attractive for firms to cooperate. According to a DOJ policy statement, “Leniency means not charging such a firm criminally for the activity being reported.” There were three major revisions: (i) amnesty was made automatic if there was no pre-existing investigation; (ii) amnesty could be granted even if cooperation began after the investigation was underway; and (iii) all directors, officers, and employees of the filing firm are protected from criminal prosecution. There is one important caveat: only the first company to file receives full amnesty. In addition to making it more attractive for corporations to report illegal behavior, in the latter part of 1994, the division implemented the individual leniency policy where individuals would not be criminally charged if they report collusive behavior on their own (not as part of a corporate application). This latter revision makes leniency more appealing as individuals are able to avoid jail time and fines. This also means that collusive parties need to consider the possibility that their behavior is revealed via individual applications further increasing the probability of detection. Accompanying these changes in policy the observed penalties for antitrust violations increased. Prior to 1995, the largest criminal fine was $6 million. In contrast, the average criminal fine was in excess of $6 million after 1996. Total fines imposed in 1997 and 1998 were “virtually identical to the total fines imposed in all of the Division’s prosecutions during the 20 years from 1976 through 1995”. In 1999, total fines imposed exceeded $1.1 billion.3 Since the revisions, cooperation from leniency applications has resulted in numerous convictions and over $4 billion in criminal fines. Theoretical support for the effectiveness of leniency programs is mixed. A significant part of the literature finds that leniency programs can destabilize existing cartels and may deter future cartel formation. Furthermore, an increase in fines may also provide impetus to report collusive behavior (Spagnolo 2004; Bageri, Katsoulakos, and Spagnolo 2013; Bigoni et al. 2015). Although, it is theoretically possible that the policy could have the opposite effect by providing a tool to discourage deviations from collusive agreements (Spagnolo 2000). However, the empirical literature is mostly supportive of the effectiveness of the leniency policy in discouraging collusion (e.g., Miller 2009; De 2010; Zhou 2013; Armoogum, Davies, and Mariuzzo 2017; Bos et al. 2018). Specifically, the literature finds evidence that the revised leniency policy resulted in increased cartel detection rates (Miller 2009), cartel destabilization (De 2010, Bos et al. 2018), and enhanced deterrence more generally (Miller 2009; Zhou 2013; Armoogum, Davies, and Mariuzzo 2017).4 There is also anecdotal evidence that firms reacted to the policy changes by revealing collusive behavior. First, the revision resulted in a surge in amnesty applications. Under the old policy, the Division obtained about one application per year, whereas the revised policy generates more than one application per month. A Deputy Assistant Attorney General of the Division remarked “The early identification of antitrust offences through compliance programs, together with the opportunity to pay zero dollars in fines under the Antitrust Division’s Corporate Amnesty Program, has resulted in a ‘race to the courthouse,’...” Indeed, it is not uncommon for a company to request amnesty a few days after one of its coconspirators has already secured amnesty by filing first.5 In addition to the observed increase in filings, there is documented evidence that the leniency policy led to breaking up international cartels. The most well-known example, made famous in the Hollywood movie,“The Informant,” involved the detection of the international cartel for lysine. In this case, the FBI obtained video recordings of meetings of the cartel members with the help of the whistleblower, Marc Whitacre.6 The cartel had raised lysine prices |$70\%$| within their first six months of cooperation. This case yielded $105 million in criminal fines, which was the largest antitrust penalty at that time (Department of Justice 2003). Some other well-known examples of collusive behavior thwarted via the leniency policy include markets for dynamic random access memory (DRAM) chips, marine hosing (used to funnel oil from tankers to storage facilities), air cargo transportation, graphite electrodes (used in steel making), textiles, construction, food preservatives, chemicals, vitamin sales, fine arts auctions, and USAID construction. Each of theses cases involved multimillion dollar fines and in some cases criminal sentences, whereas the amnesty applicant incurred no fines and received prosecution protection. For instance, in the graphite electrodes investigation, the second company to file paid $32.5 million (10% of annual earnings), the third $110 million (15% of annual earnings), and the fourth $135 million (28% of annual earnings). Mitsubishi was later convicted at trial and was sentenced to pay $134 million (76% of annual earnings). Executives from these companies incurred fines and prison sentences. In the vitamin investigation, F. Hoffmann-La Roche and BASF AG plead guilty and incurred fines of $500 million and $225 million, respectively. Again, executives from these companies served time in prison. In the fine arts auctions case, Sothebys paid $45 million, and the chairman was sentenced to one year in jail and a $7.5 million fine. Finally, in the USAID construction case, firms were ordered to pay fines of $140 million and to pay $10 million in restitution to the US government. An executive for one of the companies received a three year prison sentence. Figure 1 shows the number of new RJV filings across research areas (RAs). The first line denotes the post-corporate leniency policy period and the second line denotes the post-individual leniency policy period. The figure shows a drop in RJV filings that is consistent with the timing of these revisions. The telecommunications RA’s decline starts the earliest and is consistent with the timing of the corporate policy revision. As we will discuss momentarily, the long-distance segment of the telecommunications industry was under close scrutiny until 1996 (during the period of regulation) and, hence, telecom firms may have been more responsive to any policy aimed at deterring collusive behavior. Obviously, there may be other (non-leniency policy related) reasons for firms to reduce their RJV applications. However, this figure suggests that the decline may be due, at least in part, to the changes in policies regarding detection and punishment of collusive behavior via the leniency policy.7 Figure 1. Open in new tabDownload slide Number of new US RJV filings. 2.2. National Cooperative Research and Production Act The National Cooperative Research Act (NCRA), established in 1984, requires all firms interested in securing antitrust protection to file their RJV with the Federal Trade Commission (FTC).8 The NCRA was extended in 1993 to include all firms involved in production joint ventures (and was renamed the National Cooperative Research and Production Act, NCRPA). By filing, if member firms are subjected to criminal or civil action, antitrust authorities are required to apply the (more lenient) rule of reason that determines whether the joint venture improves social welfare rather than the per-se illegality rule.9 If found to fail a rule-of-reason analysis, member firms are granted antitrust protection, which limits their possible antitrust exposure to actual (rather than treble) damages, plus costs and attorneys’ fees with respect to activities identified in the filing.10 In deciding whether to challenge a proposed RJV, the primary consideration of the FTC is whether the venture is likely to give member firms the ability to retard the pace or scope of R&D efforts. In practice, antitrust authorities are unlikely to challenge a RJV when there are at least three independent firms with comparable research capabilities to those of the proposed RJV.11 Furthermore, authorities have indicated they will not challenge RJVs in certain RAs.12 Finally, we should note that the broadening of the NCRPA coincides with the revision of the leniency policy. Note, however, that we would expect to see more RJVs formed due to the NCRPA broadened protection. If the effect of the leniency policy is to reduce RJV applications, the presence of the NCRA revision would strengthen any negative findings. 2.3. Antitrust Cases In many industries, competitors are in RJVs together. The possibility that firms may undertake legal RJVs as a means to facilitate illegal product market collusion has generated regulatory scrutiny in a wide variety of industries and RAs. For example, in the petroleum industry in 1990, antitrust authorities found evidence that six major oil companies, who were involved in RJVs with overlapping membership, were sharing price information through direct contacts among competitors, press releases, and price postings.13 An antitrust lawsuit was filed in 2006 against CITGO Petroleum and Motiva, a RJV between Shell, Texaco, and Saudi Refining, alleging that the defendants conspired with the Organization of the Petroleum Exporting Countries (OPEC) to fix the price of gasoline.14 The following year a group of California gasoline station owners brought a suit against Equilon Enterprises, a RJV between Texaco and Shell, alleging that the RJV violated unfair competition laws and illegally fixed gasoline prices from 1998 to 2001. The suit states that the chairmen of the oil companies met privately for the “purpose of forming and organizing a combination”, that the executives destroyed meeting documents, and that the (now-defunct) RJV violated antitrust laws.15 The suit is similar to a later one which was dismissed by the Supreme Court who ruled that the unified price for the two companies’ brands was not a violation under the rule of reason.16 Texaco had to withdraw from Equilon and Motiva when it merged with Chevron to satisfy federal regulators. In addition, European antitrust authorities required Mobil Corp. to withdraw from a RJV with BP Amoco as a condition for merging with Exxon Corp. There have also been high profile cases in the computer industry. In one such case involving the semiconductor memory market, the DOJ charged four companies (Samsung, Infineon, Hynix, and Elpida) with fixing prices for DRAM. The suit states that company executives discussed the price of DRAM at joint meetings, agreed to fix prices, and exchanged information with competitors. Micron, who was a coconspirator, sought amnesty from prosecution through the DOJ’s leniency policy, and hence was not subject to criminal fines. Samsung, Hynix, Elpida, and Infineon plead guilty and were fined more than $732 million. These companies had been involved in various RJVs including SEMATECH, of which Micron was a member. In another case in 2010, the DOJ claimed that Sony, LG, Samsung, Hitachi, and Toshiba discussed prices for CDs/DVDs and Blu-ray devices during their trade organization meetings. In 2011, Hitachi plead guilty to price fixing and paid $21.1 million dollars in fines. In 2013, Woo Jin Yang, an executive in the joint venture, was sentenced to six months in federal prison for his role in the price fixing scheme.17 Another industry with a history of collusive behavior in which RJVs are commonplace is telecommunications, where nearly 40% of firms are involved in at least one RJV with another direct product market rival. Between 1984 and 1996, telecom firms were not permitted to offer both local and long distance services.18 During this period of regulation, the long distance market consisted of a regulated dominant firm (AT&T), two main competitors (MCI and Sprint), and hundreds of resellers. AT&T was required to provide services to all long distance customers, to file with the Federal Communications Commission (FCC) to add a new service, and to average its rates across consumer markets. MCI and Sprint, despite being unregulated, charged prices a little lower than those of AT&T. Furthermore, almost every new rate decrease proposed by AT&T was challenged under the umbrella of predatory behavior. These observations have led some economists to classify the market for long distance services in the 1990s as collusive with AT&T as the price leader (MacAvoy 1995). It is also notable that from 1984 to 1996, AT&T, MCI, and Sprint were involved in joint RJVs. 3. Econometric Specification In this section, we provide an econometric framework for a firm’s decision to join a particular RJV, which we use to understand the implications of our quasi-experiment on firm RJV joining behavior.19 The model describes the behavior of a firm conditional on the characteristics of the firm, the RJV, and the industry, where we account for the endogeneity of RJV formation. We begin by discussing the motivation behind the variables included in the model specification. Then we formalize the model and present the estimation technique. We conclude with a discussion of how our model is identified. 3.1. Components that Impact RJV Formation The RJV literature points to potential motivations for RJV formation that are not driven by the incentive to collude. These can be categorized by research intensity, firm specific traits, RJV specific traits, as well as economic cycles. In addition, we allow for collusive potential to impact the decision to join an RJV, where we develop a measure of RJV market power. This is best thought of as the collusive value to the firm of joining RJV |$j$|⁠. R&D Intensity. Many papers in the RJV literature show that the expected impact on R&D may be an important motivation for joining a RJV (see Roller, Siebert, and Tombak (2007) and examples therein). For instance, firms may engage in RJVs to take advantage of complementarities among member firms, share R&D-related costs, or overcome free-rider problems. Following the RJV literature, we define |$rd_{{ijt}}$| as the change in R&D intensity of firm |$i$| that would result from joining RJV |$j$| at time |$t.$| It is given by $$\begin{equation} rd_{{ijt}}=\frac{R\&D_{it-1}}{\textit{sales}_{it-1}}-\frac{R\&D_{{ijt}}}{\textit{sales}_{{ijt}}}, \end{equation}$$(1) where |$R\&D_{i}$| represents R&D expenditures and |$\textit{sales}_{i}$| represents gross dollar sales. Firm Characteristics. Firms may have different absorptive capacities, which in turn determine their willingness to form RJVs (Cohen and Levinthal 1989). The absorptive capacity is impacted by factors such as size and past experience with research collaboration (Kogut 1991). We use total assets as a measure of size and as a control for the capital and equipment that a particular firm brings to a RJV. This is consistent with the notion by Irwin and Klenow (1996) that larger firms gain more from RJVs and from R&D knowledge spillovers.20 Much of R&D is funded from retained earnings, and we use free cash flow as a proxy for capital constraints. Firms with a high free cash follow should be more attractive partners in a RJV since they are able to sustain investment without loans or new equity issues (Compte, Jenny, and Rey 2002). RJV Member Characteristics. Baumol (2001) showed that the potential benefits of RJVs may increase with the number of participating firms since technological spillovers increase. The intent to patent is a measure of efficiency with which firms innovate and may proxy for absorptive capacity (see Gugler and Siebert 2007). In addition, the need to standardize may be an incentive to form an RJV to coordinate technology choices. Patent pools represent an important vehicle for standard-setting organizations. This motivates further the need to control for the intent to patent the findings of the RJV if firms substituted R&D coordination via RJVs for standardization purposes with patent pools.21 The theoretical literature suggests that the degree of asymmetries across firms may influence RJV participation (Petit and Towlinski 1999). Previous empirical work (see Hagedoorn, Link, and Vonortas 2000) finds that size asymmetries and the degree of product complementarity between firms influences participation decisions. We include variables designed to capture the attractiveness of a firm to other partners in the RJV, which consist of a measure of firm size relative to the average RJV member (⁠|$\textit {rasset}_{{ijt}}$|⁠) and a measure of capital constraints relative to the average RJV member (⁠|$\textit {rcapcon}_{{ijt}}$|⁠). The measure of firm size relative to the RJV is $$\begin{equation} \textit {rasset}_{{ijt}}=\frac{\textit {assets}_{it-1}-\textit {avgassets}_{jt-1}}{\textit {avgassets}_{jt-1}}, \end{equation}$$(2) where |$\textit {avgassets}_{jt-1}$| are average assets of all members of the RJV in the period previous to RJV |$j$| formation. Relative capital constraints, |$\textit {rcapcon}_{{ijt}},$| are similarly defined, where we use |$\textit {cash}_{\textit {it}}$| as a proxy for capital constraints. Finally, the decision to join a new RJV may be different than the decision to continue. To account for this, we include a variable that captures whether this is the first period firm |$i$| joined RJV |$j$|⁠. State of Economy. Ghosal and Gallo (2001) suggest that antitrust enforcement by the DOJ is countercyclical. R&D investments may also be counter-cyclical; when the economy is weak firms may lack sufficient internal resources to finance long-term R&D projects so they may be more likely to rely on cooperative research arrangements. RJV Market Power. Our measure of the market power of a RJV, |$H_{{ijt}}$|⁠, is motivated by the observation that the larger the joint market shares of the firms engaged in collusive behavior (via the RJV) relative to the other firms in the industry, the higher is the profit to split among members (as the price will be closer to the monopoly price). Hence, the market power of the RJV is a function both of the market shares of the members as well as the overall level of industry concentration. Furthermore, we wish to measure the potential for product market collusion so the market power of the RJV should be relevant only among product market competitors, even though RJV members may be in different industries. RJVs commonly involve a subset of all potential rivals. Hence a cartel formed among RJV members is likely to be partial in the sense that the cartel will involve a subset of all the firms in the industry. If the RJV is formed to facilitate (partial) collusion then the RJV will be the most valuable the larger is its size (Bos and Harrington 2010).22 The intuition is that the cartel price is increasing in capacity. Therefore when a firm joins the cartel the price increases. However, the new member will have lower sales after joining since it will be required to produce below capacity. Each firms output share is proportional to its capacity share, hence the percentage reduction in post-cartel sales is lower for a firm with more capacity. This gives larger firms more incentives to become a member of the cartel. Specifically, suppose firm |$i$| belongs to industry |$k,$| and let |$\Gamma _{{jt}}$| be the subset of firms in industry |$k$| that are engaged in RJV |$j$| at time |$t.$| The collusive value of a partial cartel |$\Gamma _{{jt}}$| (formed via RJV |$j$|⁠) is a function of the total size of the partial cartel: |${\textstyle \sum \nolimits _{r\in \Gamma _{{jt}}}} s_{ {rt}}^{2},$| where |$s_{ {rt}}$| is the market share of firm |$r$| computed as sales of firm |$r$| over total sales in industry |$k$|23 and the probability the cartel is detected. For a prospective cartel member, the antitrust leniency policy revision makes collusion more costly (by increasing the rate of detection). We define the collusive value (i.e., the market power) of the RJV as $$\begin{equation} H_{{ijt}}=\frac{{\textstyle \sum \nolimits _{r\in \Gamma _{{jt}}}} s_{ {rt}}^{2}}{{\textit {HHI}}_{{kt}}} ,\end{equation}$$(3) where |${\textit {HHI}}_{{kt}}$| is the Herfindahl Index for industry |$k$|⁠.24 Why this is a measure of the RJV market power is best understood from the perspective of firm |$i$| who is considering joining RJV |$j.$| When making this decision firm |$i$| may be interested in how much collusive potential joining RJV |$j$| will yield. The number and size of firms in his market is fixed (the denominator) so in assessing the collusive potential of the RJV he will consider his size as well as the size of the other firms in the RJV relative to the overall industry concentration. Notice the larger are the firms in RJV |$j$| the higher is |$H_{{ijt}}$|⁠, which reflects the higher collusive potential of the RJV. If there were only a few large firms in industry |$k$| then the RJV would require fewer members to have substantial market power. A RJV in which most of the large firms in the industry are members has more collusive potential. That is, holding the HHI of the industry fixed, the greater the combined market shares of the participants the greater will be |$H_{{ijt}}$| as consistent with the theory of partial cartels.25 If the RJV consists of all firms in the industry (i.e., is an all-inclusive cartel) then |$H_{{ijt}}=1$|⁠. Our primary measure of the RJV market power is given by equation (3), which is increasing in the fraction of firms in the industry that join the RJV but is non-increasing in the fragmentation of the firms that join conditional on the fraction joining. This is reasonable if we believe that the RJV will be less effective in sustaining collusion relative to the status quo when the members are more fragmented. Alternatively, if it is more difficult to coordinate collusion across many firms, more fragmented firms may have more to gain from joining a RJV if the RJV also acts as a tool to coordinate. To allow for this possibility, we consider an alternative measure of the collusive potential of the RJV that is increasing in both the fraction of firms in industry |$k$| that join the RJV and in their level of fragmentation (which we refer to as the fragmentation measure denoted |$H_{{ijt}}^{\textit {frag}})$|⁠. The fragmentation measure is defined as: industry concentration post-RJV if the RJV acts as a single entity normalized by the pre-RJV industry concentration. To motivate the value to considering both measures suppose there are two industry structures: Market Structure A (MSA) has eight equally-sized firms and Market Structure B (MSB) has four equally-sized firms. If four firms under MSA and two firms under MSB join a RJV, the first measure of the RJV collusive potential (referred to as the primary measure) is identical: |$H_{{ijt}}=1/2$|⁠. The fragmentation measure yields different results: the post-RJV HHI in MSA is |$5/16$| if the RJV acts as a perfectly collusive entity and |$1/8$| under the status quo, yielding |$H_{{ijt}}^{\textit {frag}}=5/2$|⁠. The post-RJV HHI in MSB is |$3/8$| if the RJV acts as a perfectly collusive entity and |$1/4$| under the status quo, yielding |$H_{{ijt}}^{\textit {frag}}=3/2$|⁠. The fragmentation measure indicates the RJV has higher collusive potential under the more fragmented MSA. To take a more extreme example if all the firms in an industry join the same RJV should the RJVs collusive potential be the same or different if there are four or two equally-sized firms in the industry? As this is an empirical question, we consider both the primary and fragmentation measure of the market power of the RJV in estimation. 3.2. Model We develop a model of a firm’s decision to join a particular RJV. The unit of observation is a firm, specific-RJV, time combination. Let |$V_{{ijt}} ^{\ast }$| be the (latent) value to firm |$i=1,\ldots,N$| of engaging in (a new or continuing) RJV |$j$| at time |$t$|⁠: $$\begin{equation} V_{{ijt}}^{\ast }=\alpha _{1}L+\alpha _{2}LH_{{ijt}}+\lambda H_{{ijt}}+\gamma _{1}rd_{{ijt}}+\beta x_{\textit {it}}+\delta z_{{ijt}}+\varepsilon _{{ijt}}.\end{equation}$$(4) If firms enter into a RJV to facilitate collusion, antitrust policy targeted at product market collusion could impact their decision (through an increase in the probability of detection). The |$L$| term is an indicator variable taking on the value of 1 if firm |$i$| enters RJV |$j$| after the leniency policy revision. Some firms may be affected by the corporate leniency policy and/or by the individual leniency policy (that coincides with an observed increase in fines). Therefore, we estimate our model with two definitions of the indicator. The |$L$| either takes on the value of one post-1993 or one post-1995.26 We also conduct robustness checks ex-post with the leniency policy indicator defined over different years. We discuss these tests in Section 6. Furthermore, the potential payoff to collusion in the product market could depend upon the market power of the RJV (the |$H_{{ijt}}$| term). We are primarily interested in the total effect of the leniency policy on RJV formation (determined by the |$\alpha _{1}$| and |$\alpha _{2}$| terms). As we detail in the previous subsection, we include multiple terms to capture potential motivations for RJV formation that are not related directly to the incentive to collude. The |$rd_{{ijt}}$| term represents the expected change in R&D intensity of firm |$i$| after entering RJV |$j$|⁠. Firm-specific terms are captured by |$x_{\textit {it}}$| and include firm size (⁠|$\textit {assets}_{\textit {it}}$|⁠), the number of other RJV’s in which |$i$| is currently engaged and the square, capacity constraints (⁠|$\textit {cash}_{\textit {it}}),$| and industry fixed effects (when we consider definitions of markets with firms from many industries, such as RAs). RJV-specific terms are included in the |$z_{{ijt}}$|⁠. These are the number of members of RJV |$j$|⁠, whether this is the first period firm |$i$| joined RJV |$j,$| whether the intent is to patent the RJV outcome, and measures of firm–RJV asymmetries (⁠|$\textit {rasset}_{{ijt}}$| and |$\textit {rcapcon}_{{ijt}}$|⁠). We include year fixed effects to capture any economic or time-specific variables relevant to RJV formation that are not captured in other variables. The |$\epsilon _{{ijt}}$| is an i.i.d. normally distributed mean zero stochastic term. 3.3. Estimation Firms that join RJVs join on average more than one.27 Hence, firm |$i$| will enter (or continue) RJV |$j$| at time |$t$| if the value to doing so is larger than the value to not doing so. Let |$V_{i0t}^{\ast }$| represent the value to firm |$i$| of not joining a RJV: $$\begin{equation*} V_{i0t}^{\ast }=\gamma _{0}rd_{\textit {it}}+\beta _{0}x_{i0t}+\varepsilon _{i0t}, \end{equation*}$$ where |$rd_{\textit {it}}$| is the average annual intensity of R&D undertaken by firm |$i$| when it is not in a RJV. Hence, firm |$i$| will join RJV |$j$| if |$V_{{ijt}}^{\ast }\ge V_{i0t}^{\ast }$| where |$V_{{ijt}}^{\ast }$| is given in equation (4). Notice that the number of feasible alternatives does not impact the decision to join a particular RJV, although our model allows the number of RJVs a firm is currently engaged in to impact the value to joining a RJV. We do not observe |$V_{{ijt}}^{\ast }$| or |$V_{i0t}^{\ast }$|⁠, instead we observe whether firm |$i$| enters a RJV. Define $$\begin{equation} V_{{ijt}}\equiv \alpha _{1}L+\alpha _{2}LH_{{ijt}}+\gamma (rd_{{ijt}}-rd_{\textit {it}} )+\beta (x_{\textit {it}}-x_{i0t})+\delta z_{{ijt}}.\end{equation}$$(5) Any model of RJV formation must address two issues regarding estimation, both relate to the observation that the value to firm |$i$| of joining RJV |$j$| is a function of |$(rd_{{ijt}}-rd_{\textit {it}}).$| That is, firms consider the expected effect on R&D expenditures when considering whether to form a RJV. However, R&D intensity is influenced by RJV formation. Thus, the first issue to address concerns the endogeneity of R&D. The second issue concerns the effect on R&D from joining a RJV. We can construct |$(rd_{{ijt}}-rd_{\textit {it}})$| from the data when firm |$j$| joins a RJV. However, we do not observe |$rd_{{ijt}}$| if the firm is not engaged in a RJV. We need a consistent estimate of the expected effect of RJV formation on R&D intensity when a RJV is not formed. The endogenous switching model estimation procedure (Lee 1978; Roller, Siebert, and Tombak 2007) allows us to address the endogeneity and missing values issues and to obtain consistent parameter estimates. We discuss the exclusionary restrictions that allow us to identify the parameters of the model in detail in the next subsection. However, there is one more endogeneity concern related to the fact that |$H_{{ijt}}$| is a function of the market shares of member firms and industry concentration and hence may be endogenous. For instance, if establishing a RJV raises barriers to entry it could increase the market power of the involved firms even if they do not collude. We have included the measure of the market power separately (in the |$z_{{ijt}})$| as well as interacted with the leniency policy variable, but it is important to keep this caveat in mind when interpreting the results.28 Estimation is based on the following equation of RJV formation $$\begin{equation} P_{{ijt}}=V_{{ijt}}+\eta _{{ijt}}, \end{equation}$$(6) where |$\eta _{{ijt}}\equiv \varepsilon _{i0t}-\varepsilon _{{ijt}}\, \sim N(0,\sigma _{\eta }^{2})$|⁠.29 We observe |$rd_{{ijt}}$| when firm |$i$| is engaged in RJV |$j$|⁠: $$\begin{equation} rd_{{ijt}}=\lambda _{1}w_{{ijt}}+u_{1it}\;\;\text{if }V_{{ijt}}\ge \eta _{{ijt}}, \end{equation}$$(7) where |$w_{{ijt}}$| includes a constant, the number of members of RJV |$j$|⁠, firm size relative to the average RJV member (⁠|$\textit {rasset}_{{ijt}}$|⁠), and capital constraints relative to the average RJV member (⁠|$\textit {rcapcon}_{{ijt}})$|⁠. Note that the coefficient on the constant term will pick up other effects on R&D of being in RJV such as cost-sharing. If firm |$i$| is not engaged in RJV |$j$| we observe: $$\begin{equation} rd_{\textit {it}}=\lambda _{0}v_{\textit {it}}+u_{0it}\;\;\text{if }V_{{ijt}}<\eta _{{ijt}}, \end{equation}$$(8) where |$v_{\textit {it}}$| includes the assets and capital constraints faced by firm |$i$|⁠. We assume the errors |$(u_{1},u_{0},\eta )\sim N(0,\Omega ).$| To obtain asymptotically efficient estimates, we simultaneously estimate all the parameters of the model by full information maximum likelihood. The parameters of the model are |$\theta =vec\lbrace \alpha _{1},\alpha _{2},\gamma ,\beta ,\delta ,\lambda _{0},\lambda _{1},\Omega \rbrace .$|30 3.4. Identification Our strategy to identify collusive intentions relies on the variation in RJV formation arising from the revisions in the leniency policy. For this to be a reasonable quasi-experiment, the leniency policy should impact collusive behavior but not affect the other motivations to form a RJV. As discussed in Section 2.1, there is sufficient evidence that the revision to the leniency policy has been successful in curbing collusive behavior. Furthermore, there is no evidence that the DOJ changed the leniency policy with an intention to influence RJV formation or R&D investments directly.31 These theoretical arguments provide justification for our exclusionary restriction. In addition, we can test the credibility of our exclusionary restriction by examining whether the institutional adjustment to the leniency policy had an effect on R&D-related activities in the markets we consider.32 We conduct firm-level regressions to examine if either revisions to the corporate or individual leniency policies impacted R&D-related variables including R&D expenditures, R&D intensity, and patents granted.33 If our exclusionary restriction is valid than revisions to the leniency policy should not have a significant impact on R&D related activities. We find results that support our identification strategy across all market definitions and leniency policy variables. That is, revisions to the leniency policies (both corporate and individual) do not have a statistically significant impact (at |$95\%$| confidence) on any of the three R&D related variables in any markets.34 However, we should note that there may be alternative explanations that could yield an impact on RJV formation around the period of the leniency policy revision. For example, computer markets saw the dot-com stock market bubble, which caused excessive speculation of internet-related companies, around 1995. Telecommunications markets changed in 1994 when smartphones were first made available. There also may have been organizational changes made by telecommunications firms during the period of the divestiture of AT&T. These may explain part of the observed trend for telecommunications that we see in the raw data. We cannot control adequately for these or other potential explanations, and so the interpretation of the leniency policy dummy variable should be taken with caution as it may capture these effects. However, to the extent they are specific only to one year they will be controlled for by year fixed effects. We should also note that we do not have to rely on a discrete law change to identify potentially collusive efforts as the effect of the leniency policy revision on RJV formation is allowed to vary with a continuous measure of RJV market power (⁠|$H_{{ijt}}$|⁠). While it is possible that some unknown policy (that has not been controlled for) impacted the propensity to join a RJV at the same time the leniency policy was revised, it is less likely that this hypothetical policy would vary with the RJV market power measure as well. To summarize, the parameters of the model are identified by the leniency policy exclusion restriction that should not impact R&D investments directly (equations (7) and (8)) rather only the decision to enter a RJV. 4. Data Our data cover the period 1986–2001.35 Information on RJVs comes from the CORE database constructed by Link (1996) and includes the name of the RJV, date of filing, general industry classification, and the nature of research to be undertaken. We augment the CORE data with the names of the member firms in each RJV in our time frame, as reported in the Federal Register.36 Firm-level data come from the US Compustat database, which includes industry classification, assets, sales, free cash, and R&D expenditures for over 20,000 publicly traded firms. There are a few data issues to address. First, small firms are underrepresented. They are less likely to file a RJV application with the FTC since they are less likely to be the subject of antitrust investigation, and they are less likely to be in the Compustat database.37 As a result of losing small firms, we observe a few RJVs with only one member, which we drop. If firms add members to the RJV they are required to refile with the FTC; therefore, we observe changes in the composition of RJV membership across years. Unfortunately, firms do not refile when the RJV is terminated. As a result, we observe new RJVs and changes to RJV membership, but not end dates. In practice, many RJVs do not span the period of our data; a RJV formed in 1986 is not likely to be around for new firms to join in 2001. We had to make some assumptions regarding the set of potential RJVs available for each firm to join (i.e., the choice set). We decided to “end” a RJV in the year that we last observe a member join.38 Imposing this restriction, there were 386 RJVs in all industries with an average length of three years.39 Hence, we have approximately 1,200 RJVs in the sample. The firm’s choice set requires some additional explanation. One option would be to assume that every firm in the sample could join every RJV we ever observe in the data. Given that there are over a thousand RJVs in the sample and tens of thousand of firm years, this is computationally infeasible. It also assumes that all firms could contribute to any RJV. To narrow the viable options we assume a firm could join any RJV that was formed or that exists in a given year in which the firm exists. To make the explanation complete, consider an example involving AT&T starting in 1986. AT&T’s choice set in 1986 includes all RJVs in 1986 in which at least one telecommunication firm has joined—there were three such RJVs of which it joined one. In 1987, two new RJVs that included telecommunications firms formed, so AT&T’s choice set in 1987 is four (the two continuing from 1986 which it did not join and the two new RJVs). It joined two of these. No telecommunications firms joined a RJV in 1988, so AT&T choice set in 1988 consisted of two RJVs (the two continuing from 1987 which it did not join) of which it joined one. Hence, the number of RJVs in AT&T’s choice set (and the total number of RJVs joined) in each consecutive year is 3(1), 4(3), and 2(4). AT&T’s choice set continues to evolve over the sample period with new RJVs being created and entering the choice set while others exit either because the firm joins or our ending rule removes the RJV from all the choice sets. When considering the collusive intent of firms it is important to be certain that the level of aggregation is not too broad, so as to include more firms than the relevant antitrust market, nor to narrow, so as to exclude potential rivals.40 This is difficult to address in a sample spanning many industries, therefore, we do not focus on estimates from the entire pooled sample.41 Firms in computer manufacturing, petroleum refining, and telecommunications are involved in numerous RJVs with product market rivals over time, this observation, coupled with a history of antitrust proceedings, motivates us to consider these industries in detail. 4.1. Computer Markets The computer industry is a highly-evolving, rapidly-changing industry. It is characterized both by upstream firms (such as semiconductor producers) selling inputs to PC firms, as well as PC firms selling to final users. The industry consists of several large companies with worldwide sales and a high degree of capital intensity. RJVs started to play a large role in computer markets starting in the late 1980s with the formation of SEMATECH, and they continue to play a large role with over |$10\%$| of all RJV filings registered in computing related markets. Unlike the telecommunications markets (discussed momentarily), the computer industry is unregulated during our sample period and, hence, subject to competitive pressures that have increased the pace of technology (Goettler and Gordon 2011; Lundqvist 2015). Indeed, recently firms in this industry have been convicted of collusive behavior, which was revealed via the leniency policy, making it directly relevant to our study (see discussion in Section 2.3). We consider five relevant market definitions and present descriptive statistics in the top panel of Table 1. A broad definition consists of firms engaged in the computer software research area, “Software RA.” Most RJVs in memory-related industries are associated with the software RA. However, this market definition is likely to be too broad as it contains firms from more than ten three-digit NAICs industries. The other three-digit market definition, “Computer and Electronic Product Manufacturing,” encompasses firms with NAICS classifications that begin with 334. These consist of firms that manufacture computers (such as Dell), computer peripherals, and communications equipment as well as firms that manufacture components for such products (such as Intel). As these firms are not always rivals, indeed Dell is a customer of Intel, the Computer and Electronic Product Manufacturing market is also likely to be too broad a market definition. Table 1. Descriptive statistics. . Compustat . Compustat . Other . Source of firm data: . 3-digit NAICS . 6-digit NAICS . see details below . Level of aggregation: . Mean . Std. Dev. . Mean . Std. Dev. . Mean . Std. Dev. . Computer Markets Compustat; Gartner; Computer/Electronic Computer Manufacture iSuppli Semiconductors HHI 0.04 0.01 0.04 0.04 0.21 0.16 Market share 0.003 0.01 0.005 0.02 0.06 0.14 R&D expenditures 0.23 0.71 0.21 0.71 0.12 0.35 Sales 2.92 9.98 3.47 12.11 1.16 3.18 Assets 3.19 11.46 11.39 60.71 1.30 3.87 Proportion join RJV 0.01 0.10 0.01 0.11 0.01 0.09 RJV HHI 0.12 0.06 0.17 0.14 0.34 0.06 Number of RJVs 246 246 246 Memory/ Software RA Microprocessors HHI 0.05 0.06 0.17 0.11 Market share 0.009 0.03 0.02 0.06 R&D expenditures 0.28 0.87 0.14 0.40 Sales 5.23 16.39 1.15 3.46 Assets 11.44 51.48 1.47 4.52 Proportion join RJV 0.01 0.12 0.01 0.10 RJV HHI 0.16 0.16 0.34 0.06 Number of RJVs 58 246 Telecommunications Markets FCC Long Distance Broadcast Telecom Firms All Years HHI 0.16 0.15 0.22 0.12 Market share 0.05 0.09 0.05 0.11 R&D expenditures 0.55 1.21 0.19 0.53 Sales 8.42 17.04 2.34 6.92 Assets 18.34 36.23 24.54 39.02 Proportion join RJV 0.02 0.14 0.07 0.25 RJV HHI 0.25 0.09 0.34 0.07 Number of RJVs 94 72 Telecom RA Regulated Years HHI 0.23 0.20 0.27 0.13 Market share 0.09 0.20 0.05 0.12 R&D expenditures 0.27 0.84 0.15 0.42 Sales 4.57 15.17 1.97 6.12 Assets 9.51 45.76 21.49 38.12 RJV HHI 0.16 0.16 0.34 0.07 Proportion join RJV 0.02 0.14 0.07 0.26 Number of RJVs 90 53 Petroleum Markets Coal/Crude Extraction Petroleum Refining HHI 0.10 0.03 0.07 0.01 Market share 0.02 0.03 0.03 0.04 R&D expenditures 0.23 0.31 0.33 0.33 Sales 12.60 27.19 30.86 37.29 Assets 13.41 24.78 34.05 32.53 Proportion join RJV 0.09 0.29 0.12 0.33 RJV HHI 0.11 0.12 0.06 0.05 Number of RJVs 140 135 . Compustat . Compustat . Other . Source of firm data: . 3-digit NAICS . 6-digit NAICS . see details below . Level of aggregation: . Mean . Std. Dev. . Mean . Std. Dev. . Mean . Std. Dev. . Computer Markets Compustat; Gartner; Computer/Electronic Computer Manufacture iSuppli Semiconductors HHI 0.04 0.01 0.04 0.04 0.21 0.16 Market share 0.003 0.01 0.005 0.02 0.06 0.14 R&D expenditures 0.23 0.71 0.21 0.71 0.12 0.35 Sales 2.92 9.98 3.47 12.11 1.16 3.18 Assets 3.19 11.46 11.39 60.71 1.30 3.87 Proportion join RJV 0.01 0.10 0.01 0.11 0.01 0.09 RJV HHI 0.12 0.06 0.17 0.14 0.34 0.06 Number of RJVs 246 246 246 Memory/ Software RA Microprocessors HHI 0.05 0.06 0.17 0.11 Market share 0.009 0.03 0.02 0.06 R&D expenditures 0.28 0.87 0.14 0.40 Sales 5.23 16.39 1.15 3.46 Assets 11.44 51.48 1.47 4.52 Proportion join RJV 0.01 0.12 0.01 0.10 RJV HHI 0.16 0.16 0.34 0.06 Number of RJVs 58 246 Telecommunications Markets FCC Long Distance Broadcast Telecom Firms All Years HHI 0.16 0.15 0.22 0.12 Market share 0.05 0.09 0.05 0.11 R&D expenditures 0.55 1.21 0.19 0.53 Sales 8.42 17.04 2.34 6.92 Assets 18.34 36.23 24.54 39.02 Proportion join RJV 0.02 0.14 0.07 0.25 RJV HHI 0.25 0.09 0.34 0.07 Number of RJVs 94 72 Telecom RA Regulated Years HHI 0.23 0.20 0.27 0.13 Market share 0.09 0.20 0.05 0.12 R&D expenditures 0.27 0.84 0.15 0.42 Sales 4.57 15.17 1.97 6.12 Assets 9.51 45.76 21.49 38.12 RJV HHI 0.16 0.16 0.34 0.07 Proportion join RJV 0.02 0.14 0.07 0.26 Number of RJVs 90 53 Petroleum Markets Coal/Crude Extraction Petroleum Refining HHI 0.10 0.03 0.07 0.01 Market share 0.02 0.03 0.03 0.04 R&D expenditures 0.23 0.31 0.33 0.33 Sales 12.60 27.19 30.86 37.29 Assets 13.41 24.78 34.05 32.53 Proportion join RJV 0.09 0.29 0.12 0.33 RJV HHI 0.11 0.12 0.06 0.05 Number of RJVs 140 135 Notes. An observation is a firm-year pair. Sales, assets, and R&D expenditures are in billions of chain weighted $2004. Gartner and iSuppli shares are from published reports. HHI and RJV HHI are calculated at either the three- or six-digit NAICs depending on the market definition. Open in new tab Table 1. Descriptive statistics. . Compustat . Compustat . Other . Source of firm data: . 3-digit NAICS . 6-digit NAICS . see details below . Level of aggregation: . Mean . Std. Dev. . Mean . Std. Dev. . Mean . Std. Dev. . Computer Markets Compustat; Gartner; Computer/Electronic Computer Manufacture iSuppli Semiconductors HHI 0.04 0.01 0.04 0.04 0.21 0.16 Market share 0.003 0.01 0.005 0.02 0.06 0.14 R&D expenditures 0.23 0.71 0.21 0.71 0.12 0.35 Sales 2.92 9.98 3.47 12.11 1.16 3.18 Assets 3.19 11.46 11.39 60.71 1.30 3.87 Proportion join RJV 0.01 0.10 0.01 0.11 0.01 0.09 RJV HHI 0.12 0.06 0.17 0.14 0.34 0.06 Number of RJVs 246 246 246 Memory/ Software RA Microprocessors HHI 0.05 0.06 0.17 0.11 Market share 0.009 0.03 0.02 0.06 R&D expenditures 0.28 0.87 0.14 0.40 Sales 5.23 16.39 1.15 3.46 Assets 11.44 51.48 1.47 4.52 Proportion join RJV 0.01 0.12 0.01 0.10 RJV HHI 0.16 0.16 0.34 0.06 Number of RJVs 58 246 Telecommunications Markets FCC Long Distance Broadcast Telecom Firms All Years HHI 0.16 0.15 0.22 0.12 Market share 0.05 0.09 0.05 0.11 R&D expenditures 0.55 1.21 0.19 0.53 Sales 8.42 17.04 2.34 6.92 Assets 18.34 36.23 24.54 39.02 Proportion join RJV 0.02 0.14 0.07 0.25 RJV HHI 0.25 0.09 0.34 0.07 Number of RJVs 94 72 Telecom RA Regulated Years HHI 0.23 0.20 0.27 0.13 Market share 0.09 0.20 0.05 0.12 R&D expenditures 0.27 0.84 0.15 0.42 Sales 4.57 15.17 1.97 6.12 Assets 9.51 45.76 21.49 38.12 RJV HHI 0.16 0.16 0.34 0.07 Proportion join RJV 0.02 0.14 0.07 0.26 Number of RJVs 90 53 Petroleum Markets Coal/Crude Extraction Petroleum Refining HHI 0.10 0.03 0.07 0.01 Market share 0.02 0.03 0.03 0.04 R&D expenditures 0.23 0.31 0.33 0.33 Sales 12.60 27.19 30.86 37.29 Assets 13.41 24.78 34.05 32.53 Proportion join RJV 0.09 0.29 0.12 0.33 RJV HHI 0.11 0.12 0.06 0.05 Number of RJVs 140 135 . Compustat . Compustat . Other . Source of firm data: . 3-digit NAICS . 6-digit NAICS . see details below . Level of aggregation: . Mean . Std. Dev. . Mean . Std. Dev. . Mean . Std. Dev. . Computer Markets Compustat; Gartner; Computer/Electronic Computer Manufacture iSuppli Semiconductors HHI 0.04 0.01 0.04 0.04 0.21 0.16 Market share 0.003 0.01 0.005 0.02 0.06 0.14 R&D expenditures 0.23 0.71 0.21 0.71 0.12 0.35 Sales 2.92 9.98 3.47 12.11 1.16 3.18 Assets 3.19 11.46 11.39 60.71 1.30 3.87 Proportion join RJV 0.01 0.10 0.01 0.11 0.01 0.09 RJV HHI 0.12 0.06 0.17 0.14 0.34 0.06 Number of RJVs 246 246 246 Memory/ Software RA Microprocessors HHI 0.05 0.06 0.17 0.11 Market share 0.009 0.03 0.02 0.06 R&D expenditures 0.28 0.87 0.14 0.40 Sales 5.23 16.39 1.15 3.46 Assets 11.44 51.48 1.47 4.52 Proportion join RJV 0.01 0.12 0.01 0.10 RJV HHI 0.16 0.16 0.34 0.06 Number of RJVs 58 246 Telecommunications Markets FCC Long Distance Broadcast Telecom Firms All Years HHI 0.16 0.15 0.22 0.12 Market share 0.05 0.09 0.05 0.11 R&D expenditures 0.55 1.21 0.19 0.53 Sales 8.42 17.04 2.34 6.92 Assets 18.34 36.23 24.54 39.02 Proportion join RJV 0.02 0.14 0.07 0.25 RJV HHI 0.25 0.09 0.34 0.07 Number of RJVs 94 72 Telecom RA Regulated Years HHI 0.23 0.20 0.27 0.13 Market share 0.09 0.20 0.05 0.12 R&D expenditures 0.27 0.84 0.15 0.42 Sales 4.57 15.17 1.97 6.12 Assets 9.51 45.76 21.49 38.12 RJV HHI 0.16 0.16 0.34 0.07 Proportion join RJV 0.02 0.14 0.07 0.26 Number of RJVs 90 53 Petroleum Markets Coal/Crude Extraction Petroleum Refining HHI 0.10 0.03 0.07 0.01 Market share 0.02 0.03 0.03 0.04 R&D expenditures 0.23 0.31 0.33 0.33 Sales 12.60 27.19 30.86 37.29 Assets 13.41 24.78 34.05 32.53 Proportion join RJV 0.09 0.29 0.12 0.33 RJV HHI 0.11 0.12 0.06 0.05 Number of RJVs 140 135 Notes. An observation is a firm-year pair. Sales, assets, and R&D expenditures are in billions of chain weighted $2004. Gartner and iSuppli shares are from published reports. HHI and RJV HHI are calculated at either the three- or six-digit NAICs depending on the market definition. Open in new tab A narrow definition comprises establishments that engage in manufacturing or assembling of electronic computers (such as mainframes, personal computers, and servers). The “Computer Manufacturing” definition consists of all six-digit NAICS starting with 33411 and encompasses firms such as Dell, HP, Sun, and Apple. This is a more convincing relevant market, as it does not contain semiconductor manufacturers and hence is more likely to consist of product rivals to the extent that firms selling mainframes compete with PC firms. The second six-digit definition includes firms that are engaged in manufacturing semiconductors and other components for electronic applications. The “Semiconductors” definition consists of all six-digit NAICS starting with 3344 and includes firms such as Intel, AMD, Micron, and Motorola. Semiconductors are used as inputs in computer products, in communications equipment, and in electronics. Semiconductor production, which consists primarily of memory chips and microcomponents, constitutes the largest component of the computer industry. Examining the market at this narrow level is particularly worthwhile given the recent antitrust case against semiconductor/DRAM memory producers who were involved in many joint RJVs. However, one drawback of narrowing the relevant market definition is that the number of observations are fewer and perhaps not sufficient to estimate our model. Furthermore, there are two issues with using the Compustat data at this level. First, Compustat provides total sales data for publicly traded US firms, but does not break sales down at the level of detail we require. For instance, IBM’s microelectronics division was involved in semiconductor sales during the 1990s. However, IBM is one of the world leaders in mainframe computers, thus using the Compustat sales data (which is for all of IBM’s divisions) will lead us to over-estimate the importance of IBM in the semiconductor industry. Second, there are many foreign firms in the semiconductor industry, some of which are major players, such as Samsung, Toshiba, and Infineon. If firms in the United States enter RJVs to facilitate international collusion then the Compustat data will not give us an accurate measure of the RJV market power if it does not take into account sales of international firms. To overcome these problems, we augment the Compustat data with sales data for semiconductor firms published by Gartner Group and iSuppli Corporation.42 These data are provided for the top worldwide semiconductor firms (constituting |$50\%$|–|$70\%$| of worldwide sales) and are limited to sales of semiconductors.43 We use the data on non-US firms to get an accurate measure of the RJV market power as we have information on all worldwide members of the RJV. We use the Gartner/iSuppli sales data from US firms together with Compustat data (on R&D expenditures, etc.) to estimate the model. While much more narrow than the other relevant markets, the semiconductor definition consists of firms that manufacture and sell their chips (such as Intel and Samsung) as well as firms that outsource the manufacturing to other companies (such as Qualcomm). The manufacturing of microprocessors (CPUs) and memory chips (such as DRAM, static random access, and flash memory) accounts for approximately half of the sales of semiconductors. Our final narrow market definition consists of firms that manufacture microprocessors and memory chips. This market encompasses producers of DRAM who were involved in the recent antitrust amnesty case (Zulehner 2003). Furthermore, RJVs play a large role in microprocessor and memory production. Indeed, the largest firms in the market for flash memory products are AMD and Fujitsu, who are the only members of a joint venture in this area (Spansion Inc.). We again use the firm level sales data by Gartner/iSuppli combined with the Compustat data to estimate this market definition. As Table 1 indicates, firms have high R&D intensities (R&D expenditures as a proportion of sales) ranging from |$5\%$| to |$12\%,$| on average. Indeed, semiconductor companies rank highest in R&D intensity: approximately |$13\%$| worldwide, which is higher than R&D intensity in pharmaceutical markets. In our data, firms in the semiconductor industry spend |$10\%$| of their sales on R&D, with firms in memory and microprocessing spending |$12\%$|⁠. Furthermore, as the market definition narrows market concentration grows substantially with industry concentrations at the six-digit level consistent with moderate concentration levels. Across market definitions, the RJV HHI indicates a moderate to high degree of market power for those firms in an RJV ranging from |$12\%$| to |$34\%$|⁠.44 4.2. Telecommunications Markets The telecommunications sector has a history of potentially collusive behavior and RJVs are common among rivals, where 38% of firms involved in at least one RJV with another direct product market competitor. These observations, coupled with an ability to construct a well-defined antitrust market (due to the telecommunications regulatory mandate), makes the telecom industry ideal for our study. We consider four definitions of the relevant telecom antitrust market. At the most aggregate (three-digit NAICS) industry level we consider two potential markets: firms in “Broadcast Telecom” (NAICS 513) and firms involved in telecommunications research, “Telecom RA”, stated as the primary RA in their RJV filing. There are reasons to believe that this level of aggregation may be too broad. For instance, Broadcast Telecom includes wired telecommunications carriers, radio stations, television broadcasters, cable providers, and wireless carriers, which are not always competitors with each other. The Telecom RA also includes firms that are often in very different competitive markets (e.g., firms in publishing as well as chemical manufacturing). Indeed, the descriptive statistics presented in the middle panel of Table 1 indicate the Broadcast Telecom market is less concentrated as given by the three-digit Herfindahl Index (HHI, which is calculated as the sum of squares of the market shares of all firms in the relevant industry). Our two more narrow definitions of the relevant market use data from the FCC’s Report of Common Carriers,45 which permits us to further divide telecom firms into those offering long distance versus local service. Furthermore, the FCC data include all firms in telephony regardless of size. Our final two definitions of the relevant market consist of all firms offering long distance services. Over all years of the data (1986–2001), the market of long distance firms may be too narrow since after 1996 long distance carriers were permitted to offer local services. Therefore, we also consider a subset of the long distance market restricted to the years of regulation. Although the latter is a relatively small sample, this market definition is particularly attractive since, by law, the market includes only these firms and these firms are not permitted to enter other telecom markets. Table 1 indicates that the market for long distance services is more concentrated relative to our other antitrust telecom market definitions. The long distance market was more concentrated during the period of regulation with an HHI suggesting it operated similar to an industry with three equally sized firms. Finally, on average, more firms join a RJV (7%) relative to other telecom antitrust market definitions. Across market definitions, the RJV HHI indicates a moderate to high degree of market power for those firms in an RJV ranging from 16% to 34% with the most concentrated in the long distance markets.46 4.3. Petroleum Markets Our final set of markets involves firms in petroleum related production, where the prevalence of RJVs involving product market rivals is most pronounced. The importance of oil production worldwide and the existence of an international cartel make this industry worthwhile to consider. The petroleum industry is organized into four broad sectors: exploration and production of crude oil and natural gas; transport; refining; and marketing and distribution. Due to data limitations common to studies in this industry, we are only able to examine two market definitions.47 Our first broad three-digit definition contains firms engaged in “Coal and Crude Oil Extraction” who focus on the transformation of crude petroleum and coal into usable products. The dominant process in the transformation is petroleum refining that involves the separation of crude petroleum into components. In addition, this subsector includes establishments that further process refined petroleum and coal products to produce related products such as asphalt coatings and petroleum lubricating oils. Our more narrow definition focuses on firms engaged in petroleum refining (defined at the six-digit NAICS). A few notable aspects of the industry are apparent from the descriptive statistics in the bottom panel of Table 1. First, there are a large number of RJVs active in each relevant market coupled with high probabilities of joining a RJV. Furthermore, across all market definitions, R&D intensity is lower than overall R&D spending as a percentage of sales in other industries. Finally, there are relatively few firms compared to the number of RJVs suggesting that, like telecommunications, there are many RJVs among rival firms. Also similar to the telecommunications industry, the petroleum industry is highly concentrated; however, this is not reflected in our descriptive statistics due to the presence of foreign firms. In addition, the RJV HHI indicates a low to moderate degree of market power of the RJV ranging from |$6\%$| to |$11\%$|⁠. There are a few drawbacks to using the petroleum industry to examine the impact of the leniency program. First, the industry has several large international players, often with substantial government support, that are not publicly traded in the United States and so are not in our data. For example, in 1998, the largest oil producer was the Saudi Arabian Oil Co., and the top five were all state owned firms.48 As a result, our sample (regardless of how the market is defined) will omit important players in the industry, most notably members of OPEC. This important drawback of the data is balanced to some extent, by the fact that the leniency policy was specifically aimed at thwarting cartels that include international firms. So, while we are not able to construct a sample of all the relevant competitors, the behavior of the US-based firms, that we do observe, will still be influenced by the leniency policy even when (or perhaps especially when) they are engaged in international cartels. 5. Results In this section, we first provide the findings for our control parameters. We then discuss alternative specifications of our main model, followed by results for each industry for different definitions of the relevant antitrust market. As discussed in Section 3, all regressions include a constant, firm assets, firm capacity constraints, number of RJVs that the firm is a member of and its square, number of RJV members, whether the RJV is new to firm |$i$|⁠, relative assets, relative capital constraints, whether intent is to patent R&D outcome, industry fixed effects (for RA markets), and year fixed effects. As a result of including year-effects the parameter estimate for the level effect of the leniency policy dummy (defined as equal to one in all post-leniency years) will be the effect of a dummy variable for the year of the leniency policy effect (relative to the first year). However, the interactions of the RJV market power measures with the leniency policy indicator will take on a non-zero value for all post-leniency years. Our parameter estimates for the control variables are intuitive and consistent with the majority of the RJV formation literature findings. For instance, we find that firms with more assets are significantly more likely to engage in RJVs. The more RJVs a firm is engaged in the more likely they are to join another RJV but there are decreasing returns to joining. Joining a new RJV has a positive and significant impact on the probability of joining. Finally, more relatively capital constrained firms are more likely to join a RJV. Given that the focus of this paper is on the collusive intent underlying RJV formation, we do not report the parameter estimates for the controls across the specifications and samples.49 If the primary motivation for a firm to join a RJV is to foster collusion in the product market, the impact of RJV formation on R&D may be less important. For this reason, we estimate two specifications for all market definitions and samples. The first, the “Without R&D Effects” specification, consists of a model of RJV formation without predicted change in R&D intensity as a regressor. Note that this specification is equivalent to the first stage in a two stage endogenous switching regression. Estimates from this specification will not be consistent if firms consider the predicted change in R&D when making RJV formation decisions. Our second specification, “With R&D Effects”, addresses this by correcting for the endogeneity of R&D and RJV decisions as outlined in Section 3. This is worthwhile to consider because our experiment may affect R&D through the “back door”. For example, because the collusive benefits of the RJV are reduced, the R&D benefits that would have occurred (in the absence of the leniency policy) are not realized in the revised leniency policy environment. Controlling for R&D endogeneity allows us examine the impact of the policy holding R&D intensity constant. Thus, we have estimates of collusive behavior that are not contaminated by the potential “back door” effects of the leniency policy on R&D.50 We report robust standard errors from White’s correction clustered by firm (in parenthesis). The leniency policy revisions applies to all firms so our results could be driven by unobserved trends. For this reason, we estimate additional specifications in which we construct a “ treatment” group of potential colluders in the sense that these are firms that have joined a RJV with other firms in the same final goods industry. Our definition of which firms are in the treatment group depends upon the level of aggregation (i.e., which relevant market) we are considering. Note that firm |$i$| from industry |$k$| would be in the treatment group for a particular RJV |$j,$| if other firms from industry |$k$| are in RJV |$j$|⁠, but not in the treatment group for another RJV |$m,$| if there are no other firms from industry |$k$| in RJV |$m.$| We estimate all specifications in each industry for both leniency policy revision dates (post-corporate and post-individual). We will discuss the details momentarily, but our results showed that firms in the telecommunications industry reacted immediately to the first revisions in the leniency policy (the corporate leniency revisions). While firms in the computer and petroleum refining industries reacted only after the revisions allowed for individuals to obtain amnesty. This finding is perhaps related to the fact that firms in the telecommunications industry were regulated by antitrust authorities until 1996 and, hence, may have been more sensitive to any policy aimed at deterring collusive behavior. For sake of space, we report results for telecommunications with the leniency variable defined as post-corporate revision, and for the other two industries we report the results for post both revisions. We now discuss the estimates from each market in turn. 5.1. Computer Markets We report the estimates for the leniency policy and RJV market power variables for the computer markets in Table 2. We estimated three models for each market definition and R&D effect specification: (i) a model that includes the primary RJV market power measure (RJV HHI); (ii) a model that includes the fragmentation RJV market power measure (RJV HHI Fragmentation); and (iii) a model that is estimated using only firms in the treatment group. Table 2. Estimates for join RJV in computer markets. Level of aggregation: . 3-digit NAICS . 3-digit NAICS . 6-digit NAICS . 6-digit NAICS . 6-digit NAICS . Data source: . Compustat . Compustat . Compustat . Compustat, Gartner/iSuppli Sales . Compustat, Gartner/iSuppli Sales . Relevant market definition: . Computer/Electronic Manufacture . Software RA . Computer Manufacturing . Semiconductors . Memory/Microprocessor . Sample: . All . All . Treatment . All . All . Treatment . All . All . Treatment . All . All . Treatment . All . All . Treatment . . (1) . (2) . (3) . (4) . (5) . (6) . (7) . (8) . (9) . (10) . (11) . (12) . (13) . (14) . (15) . Without R&D effects Post leniency dummy −0.526*** −0.207* −0.539*** −0.611*** −1.031*** −0.546*** −0.625*** −0.700*** −0.548*** −0.524*** −0.353 −0.791*** −0.806*** −0.687** −0.952*** (0.0994) (0.112) (0.101) (0.122) (0.190) (0.125) (0.143) (0.233) (0.156) (0.180) (0.268) (0.231) (0.244) (0.329) (0.299) RJV HHI 0.544|$^{***}$| 0.588|$^{***}$| 0.449|$^{***}$| 0.344|$^{***}$| 0.239|$^{*}$| 0.332|$^{***}$| 0.229 0.514|$^{**}$| 0.409|$^{***}$| 0.800|$^{***}$| (0.0680) (0.0650) (0.126) (0.128) (0.129) (0.123) (0.142) (0.256) (0.147) (0.274) RJV HHI*Post leniency −0.466*** −0.331*** −0.296** −0.399*** −0.153 −0.0909 −0.475** −0.573* −0.383* −0.371 (0.103) (0.103) (0.120) (0.127) (0.188) (0.185) (0.193) (0.295) (0.229) (0.336) RJV HHI fragmentation −0.848*** −0.404*** 0.485*** 0.713*** 0.812*** (0.104) (0.112) (0.0878) (0.130) (0.139) RJV HHI fragmentation|$^{*}$| −0.308** 0.499|$^{***}$| 0.117 0.0659 0.0530 Post leniency (0.125) (0.152) (0.122) (0.136) (0.138) Leniency policy total effect −0.003*** −0.003*** −0.003*** −0.004** −0.011** −0.004* −0.005* −0.005** −0.005* −0.002** −0.002*** −0.015* −0.007*** −0.006*** −0.022*** Wald test statistic 13.39 14.25 13.79 6.058 6.537 5.426 5.226 5.359 4.959 8.888 10.05 4.879 9.256 9.629 10.79 P-value 0.009 0.006 0.006 0.042 0.039 0.053 0.052 0.049 0.053 0.012 0.007 0.087 0.009 0.008 0.005 With R&D effects Post leniency dummy −0.578*** −0.273** −0.570*** −0.674*** −0.940*** −0.632*** −0.614*** −0.761*** −0.569*** −0.550*** −0.423 −0.497** −0.820*** −0.661* −0.928*** (0.106) (0.120) (0.105) (0.146) (0.208) (0.147) (0.158) (0.244) (0.164) (0.193) (0.278) (0.233) (0.255) (0.344) (0.315) RJV HHI 0.558|$^{***}$| 0.602|$^{***}$| 0.464|$^{***}$| 0.379|$^{**}$| 0.307|$^{**}$| 0.336|$^{**}$| 0.307|$^{**}$| 0.858|$^{***}$| 0.447|$^{***}$| 0.883|$^{***}$| (0.0715) (0.0691) (0.153) (0.151) (0.130) (0.133) (0.150) (0.260) (0.140) (0.251) RJV HHI*Post leniency −0.462*** −0.333*** −0.336** −0.416*** −0.157 −0.0938 −0.383* −0.104 −0.305 0.248 (0.110) (0.110) (0.137) (0.144) (0.192) (0.197) (0.207) (0.286) (0.235) (0.342) RJV HHI fragmentation −0.872*** −0.304*** 0.474|$^{***}$| 0.724|$^{***}$| 0.852|$^{***}$| (0.111) (0.115) (0.0926) (0.138) (0.139) RJV HHI fragmentation|$^{*}$| −0.298** 0.365|$^{**}$| 0.137 0.0706 0.001 Post leniency (0.133) (0.153) (0.127) (0.144) (0.141) Leniency policy total effect −0.003*** −0.003*** −0.004*** −0.006* −0.006* −0.008* −0.005* −0.005* −0.005* −0.003** −0.002*** −0.027** −0.009** −0.007** −0.024*** Wald test statistic 13.27 14.02 13.38 5.215 5.586 4.961 4.789 4.975 4.623 6.238 10.28 6.358 8.921 8.588 13.06 P-value 0.009 0.006 0.006 0.062 0.062 0.064 0.060 0.061 0.063 0.044 0.006 0.042 0.012 0.014 0.001 Probability of joining RJV 0.011 0.011 0.015 0.015 0.015 0.025 0.012 0.012 0.017 0.008 0.008 0.045 0.010 0.010 0.038 Number of observations Without R&D effects 415,439 415,439 337,727 254,629 254,629 144,291 93,046 93,046 75,635 91,506 91,506 16,456 52,142 52,142 15,088 With R&D effects 340,343 340,343 276,726 148,821 148,821 88,757 80,048 80,048 65,033 70,145 70,145 13787 43,934 43,934 13,027 Level of aggregation: . 3-digit NAICS . 3-digit NAICS . 6-digit NAICS . 6-digit NAICS . 6-digit NAICS . Data source: . Compustat . Compustat . Compustat . Compustat, Gartner/iSuppli Sales . Compustat, Gartner/iSuppli Sales . Relevant market definition: . Computer/Electronic Manufacture . Software RA . Computer Manufacturing . Semiconductors . Memory/Microprocessor . Sample: . All . All . Treatment . All . All . Treatment . All . All . Treatment . All . All . Treatment . All . All . Treatment . . (1) . (2) . (3) . (4) . (5) . (6) . (7) . (8) . (9) . (10) . (11) . (12) . (13) . (14) . (15) . Without R&D effects Post leniency dummy −0.526*** −0.207* −0.539*** −0.611*** −1.031*** −0.546*** −0.625*** −0.700*** −0.548*** −0.524*** −0.353 −0.791*** −0.806*** −0.687** −0.952*** (0.0994) (0.112) (0.101) (0.122) (0.190) (0.125) (0.143) (0.233) (0.156) (0.180) (0.268) (0.231) (0.244) (0.329) (0.299) RJV HHI 0.544|$^{***}$| 0.588|$^{***}$| 0.449|$^{***}$| 0.344|$^{***}$| 0.239|$^{*}$| 0.332|$^{***}$| 0.229 0.514|$^{**}$| 0.409|$^{***}$| 0.800|$^{***}$| (0.0680) (0.0650) (0.126) (0.128) (0.129) (0.123) (0.142) (0.256) (0.147) (0.274) RJV HHI*Post leniency −0.466*** −0.331*** −0.296** −0.399*** −0.153 −0.0909 −0.475** −0.573* −0.383* −0.371 (0.103) (0.103) (0.120) (0.127) (0.188) (0.185) (0.193) (0.295) (0.229) (0.336) RJV HHI fragmentation −0.848*** −0.404*** 0.485*** 0.713*** 0.812*** (0.104) (0.112) (0.0878) (0.130) (0.139) RJV HHI fragmentation|$^{*}$| −0.308** 0.499|$^{***}$| 0.117 0.0659 0.0530 Post leniency (0.125) (0.152) (0.122) (0.136) (0.138) Leniency policy total effect −0.003*** −0.003*** −0.003*** −0.004** −0.011** −0.004* −0.005* −0.005** −0.005* −0.002** −0.002*** −0.015* −0.007*** −0.006*** −0.022*** Wald test statistic 13.39 14.25 13.79 6.058 6.537 5.426 5.226 5.359 4.959 8.888 10.05 4.879 9.256 9.629 10.79 P-value 0.009 0.006 0.006 0.042 0.039 0.053 0.052 0.049 0.053 0.012 0.007 0.087 0.009 0.008 0.005 With R&D effects Post leniency dummy −0.578*** −0.273** −0.570*** −0.674*** −0.940*** −0.632*** −0.614*** −0.761*** −0.569*** −0.550*** −0.423 −0.497** −0.820*** −0.661* −0.928*** (0.106) (0.120) (0.105) (0.146) (0.208) (0.147) (0.158) (0.244) (0.164) (0.193) (0.278) (0.233) (0.255) (0.344) (0.315) RJV HHI 0.558|$^{***}$| 0.602|$^{***}$| 0.464|$^{***}$| 0.379|$^{**}$| 0.307|$^{**}$| 0.336|$^{**}$| 0.307|$^{**}$| 0.858|$^{***}$| 0.447|$^{***}$| 0.883|$^{***}$| (0.0715) (0.0691) (0.153) (0.151) (0.130) (0.133) (0.150) (0.260) (0.140) (0.251) RJV HHI*Post leniency −0.462*** −0.333*** −0.336** −0.416*** −0.157 −0.0938 −0.383* −0.104 −0.305 0.248 (0.110) (0.110) (0.137) (0.144) (0.192) (0.197) (0.207) (0.286) (0.235) (0.342) RJV HHI fragmentation −0.872*** −0.304*** 0.474|$^{***}$| 0.724|$^{***}$| 0.852|$^{***}$| (0.111) (0.115) (0.0926) (0.138) (0.139) RJV HHI fragmentation|$^{*}$| −0.298** 0.365|$^{**}$| 0.137 0.0706 0.001 Post leniency (0.133) (0.153) (0.127) (0.144) (0.141) Leniency policy total effect −0.003*** −0.003*** −0.004*** −0.006* −0.006* −0.008* −0.005* −0.005* −0.005* −0.003** −0.002*** −0.027** −0.009** −0.007** −0.024*** Wald test statistic 13.27 14.02 13.38 5.215 5.586 4.961 4.789 4.975 4.623 6.238 10.28 6.358 8.921 8.588 13.06 P-value 0.009 0.006 0.006 0.062 0.062 0.064 0.060 0.061 0.063 0.044 0.006 0.042 0.012 0.014 0.001 Probability of joining RJV 0.011 0.011 0.015 0.015 0.015 0.025 0.012 0.012 0.017 0.008 0.008 0.045 0.010 0.010 0.038 Number of observations Without R&D effects 415,439 415,439 337,727 254,629 254,629 144,291 93,046 93,046 75,635 91,506 91,506 16,456 52,142 52,142 15,088 With R&D effects 340,343 340,343 276,726 148,821 148,821 88,757 80,048 80,048 65,033 70,145 70,145 13787 43,934 43,934 13,027 Notes. Robust standard errors clustered by firm are in parentheses. |$^{*}$| indicates significant at the 10% level; |$^{**}$| at 5% level; and |$^{***}$| at 1% level. The total effect is computed at the mean of the independent variables. We include the following controls in all regressions: constant, firm assets, firm capacity constraints, number of RJVs the firm is a member of and its square, whether the RJV is new, number of RJV members, relative assets, relative capital constraints, industry fixed effects (for RAs), and year fixed effects. An observation is a firm–year–RJV combination. Open in new tab Table 2. Estimates for join RJV in computer markets. Level of aggregation: . 3-digit NAICS . 3-digit NAICS . 6-digit NAICS . 6-digit NAICS . 6-digit NAICS . Data source: . Compustat . Compustat . Compustat . Compustat, Gartner/iSuppli Sales . Compustat, Gartner/iSuppli Sales . Relevant market definition: . Computer/Electronic Manufacture . Software RA . Computer Manufacturing . Semiconductors . Memory/Microprocessor . Sample: . All . All . Treatment . All . All . Treatment . All . All . Treatment . All . All . Treatment . All . All . Treatment . . (1) . (2) . (3) . (4) . (5) . (6) . (7) . (8) . (9) . (10) . (11) . (12) . (13) . (14) . (15) . Without R&D effects Post leniency dummy −0.526*** −0.207* −0.539*** −0.611*** −1.031*** −0.546*** −0.625*** −0.700*** −0.548*** −0.524*** −0.353 −0.791*** −0.806*** −0.687** −0.952*** (0.0994) (0.112) (0.101) (0.122) (0.190) (0.125) (0.143) (0.233) (0.156) (0.180) (0.268) (0.231) (0.244) (0.329) (0.299) RJV HHI 0.544|$^{***}$| 0.588|$^{***}$| 0.449|$^{***}$| 0.344|$^{***}$| 0.239|$^{*}$| 0.332|$^{***}$| 0.229 0.514|$^{**}$| 0.409|$^{***}$| 0.800|$^{***}$| (0.0680) (0.0650) (0.126) (0.128) (0.129) (0.123) (0.142) (0.256) (0.147) (0.274) RJV HHI*Post leniency −0.466*** −0.331*** −0.296** −0.399*** −0.153 −0.0909 −0.475** −0.573* −0.383* −0.371 (0.103) (0.103) (0.120) (0.127) (0.188) (0.185) (0.193) (0.295) (0.229) (0.336) RJV HHI fragmentation −0.848*** −0.404*** 0.485*** 0.713*** 0.812*** (0.104) (0.112) (0.0878) (0.130) (0.139) RJV HHI fragmentation|$^{*}$| −0.308** 0.499|$^{***}$| 0.117 0.0659 0.0530 Post leniency (0.125) (0.152) (0.122) (0.136) (0.138) Leniency policy total effect −0.003*** −0.003*** −0.003*** −0.004** −0.011** −0.004* −0.005* −0.005** −0.005* −0.002** −0.002*** −0.015* −0.007*** −0.006*** −0.022*** Wald test statistic 13.39 14.25 13.79 6.058 6.537 5.426 5.226 5.359 4.959 8.888 10.05 4.879 9.256 9.629 10.79 P-value 0.009 0.006 0.006 0.042 0.039 0.053 0.052 0.049 0.053 0.012 0.007 0.087 0.009 0.008 0.005 With R&D effects Post leniency dummy −0.578*** −0.273** −0.570*** −0.674*** −0.940*** −0.632*** −0.614*** −0.761*** −0.569*** −0.550*** −0.423 −0.497** −0.820*** −0.661* −0.928*** (0.106) (0.120) (0.105) (0.146) (0.208) (0.147) (0.158) (0.244) (0.164) (0.193) (0.278) (0.233) (0.255) (0.344) (0.315) RJV HHI 0.558|$^{***}$| 0.602|$^{***}$| 0.464|$^{***}$| 0.379|$^{**}$| 0.307|$^{**}$| 0.336|$^{**}$| 0.307|$^{**}$| 0.858|$^{***}$| 0.447|$^{***}$| 0.883|$^{***}$| (0.0715) (0.0691) (0.153) (0.151) (0.130) (0.133) (0.150) (0.260) (0.140) (0.251) RJV HHI*Post leniency −0.462*** −0.333*** −0.336** −0.416*** −0.157 −0.0938 −0.383* −0.104 −0.305 0.248 (0.110) (0.110) (0.137) (0.144) (0.192) (0.197) (0.207) (0.286) (0.235) (0.342) RJV HHI fragmentation −0.872*** −0.304*** 0.474|$^{***}$| 0.724|$^{***}$| 0.852|$^{***}$| (0.111) (0.115) (0.0926) (0.138) (0.139) RJV HHI fragmentation|$^{*}$| −0.298** 0.365|$^{**}$| 0.137 0.0706 0.001 Post leniency (0.133) (0.153) (0.127) (0.144) (0.141) Leniency policy total effect −0.003*** −0.003*** −0.004*** −0.006* −0.006* −0.008* −0.005* −0.005* −0.005* −0.003** −0.002*** −0.027** −0.009** −0.007** −0.024*** Wald test statistic 13.27 14.02 13.38 5.215 5.586 4.961 4.789 4.975 4.623 6.238 10.28 6.358 8.921 8.588 13.06 P-value 0.009 0.006 0.006 0.062 0.062 0.064 0.060 0.061 0.063 0.044 0.006 0.042 0.012 0.014 0.001 Probability of joining RJV 0.011 0.011 0.015 0.015 0.015 0.025 0.012 0.012 0.017 0.008 0.008 0.045 0.010 0.010 0.038 Number of observations Without R&D effects 415,439 415,439 337,727 254,629 254,629 144,291 93,046 93,046 75,635 91,506 91,506 16,456 52,142 52,142 15,088 With R&D effects 340,343 340,343 276,726 148,821 148,821 88,757 80,048 80,048 65,033 70,145 70,145 13787 43,934 43,934 13,027 Level of aggregation: . 3-digit NAICS . 3-digit NAICS . 6-digit NAICS . 6-digit NAICS . 6-digit NAICS . Data source: . Compustat . Compustat . Compustat . Compustat, Gartner/iSuppli Sales . Compustat, Gartner/iSuppli Sales . Relevant market definition: . Computer/Electronic Manufacture . Software RA . Computer Manufacturing . Semiconductors . Memory/Microprocessor . Sample: . All . All . Treatment . All . All . Treatment . All . All . Treatment . All . All . Treatment . All . All . Treatment . . (1) . (2) . (3) . (4) . (5) . (6) . (7) . (8) . (9) . (10) . (11) . (12) . (13) . (14) . (15) . Without R&D effects Post leniency dummy −0.526*** −0.207* −0.539*** −0.611*** −1.031*** −0.546*** −0.625*** −0.700*** −0.548*** −0.524*** −0.353 −0.791*** −0.806*** −0.687** −0.952*** (0.0994) (0.112) (0.101) (0.122) (0.190) (0.125) (0.143) (0.233) (0.156) (0.180) (0.268) (0.231) (0.244) (0.329) (0.299) RJV HHI 0.544|$^{***}$| 0.588|$^{***}$| 0.449|$^{***}$| 0.344|$^{***}$| 0.239|$^{*}$| 0.332|$^{***}$| 0.229 0.514|$^{**}$| 0.409|$^{***}$| 0.800|$^{***}$| (0.0680) (0.0650) (0.126) (0.128) (0.129) (0.123) (0.142) (0.256) (0.147) (0.274) RJV HHI*Post leniency −0.466*** −0.331*** −0.296** −0.399*** −0.153 −0.0909 −0.475** −0.573* −0.383* −0.371 (0.103) (0.103) (0.120) (0.127) (0.188) (0.185) (0.193) (0.295) (0.229) (0.336) RJV HHI fragmentation −0.848*** −0.404*** 0.485*** 0.713*** 0.812*** (0.104) (0.112) (0.0878) (0.130) (0.139) RJV HHI fragmentation|$^{*}$| −0.308** 0.499|$^{***}$| 0.117 0.0659 0.0530 Post leniency (0.125) (0.152) (0.122) (0.136) (0.138) Leniency policy total effect −0.003*** −0.003*** −0.003*** −0.004** −0.011** −0.004* −0.005* −0.005** −0.005* −0.002** −0.002*** −0.015* −0.007*** −0.006*** −0.022*** Wald test statistic 13.39 14.25 13.79 6.058 6.537 5.426 5.226 5.359 4.959 8.888 10.05 4.879 9.256 9.629 10.79 P-value 0.009 0.006 0.006 0.042 0.039 0.053 0.052 0.049 0.053 0.012 0.007 0.087 0.009 0.008 0.005 With R&D effects Post leniency dummy −0.578*** −0.273** −0.570*** −0.674*** −0.940*** −0.632*** −0.614*** −0.761*** −0.569*** −0.550*** −0.423 −0.497** −0.820*** −0.661* −0.928*** (0.106) (0.120) (0.105) (0.146) (0.208) (0.147) (0.158) (0.244) (0.164) (0.193) (0.278) (0.233) (0.255) (0.344) (0.315) RJV HHI 0.558|$^{***}$| 0.602|$^{***}$| 0.464|$^{***}$| 0.379|$^{**}$| 0.307|$^{**}$| 0.336|$^{**}$| 0.307|$^{**}$| 0.858|$^{***}$| 0.447|$^{***}$| 0.883|$^{***}$| (0.0715) (0.0691) (0.153) (0.151) (0.130) (0.133) (0.150) (0.260) (0.140) (0.251) RJV HHI*Post leniency −0.462*** −0.333*** −0.336** −0.416*** −0.157 −0.0938 −0.383* −0.104 −0.305 0.248 (0.110) (0.110) (0.137) (0.144) (0.192) (0.197) (0.207) (0.286) (0.235) (0.342) RJV HHI fragmentation −0.872*** −0.304*** 0.474|$^{***}$| 0.724|$^{***}$| 0.852|$^{***}$| (0.111) (0.115) (0.0926) (0.138) (0.139) RJV HHI fragmentation|$^{*}$| −0.298** 0.365|$^{**}$| 0.137 0.0706 0.001 Post leniency (0.133) (0.153) (0.127) (0.144) (0.141) Leniency policy total effect −0.003*** −0.003*** −0.004*** −0.006* −0.006* −0.008* −0.005* −0.005* −0.005* −0.003** −0.002*** −0.027** −0.009** −0.007** −0.024*** Wald test statistic 13.27 14.02 13.38 5.215 5.586 4.961 4.789 4.975 4.623 6.238 10.28 6.358 8.921 8.588 13.06 P-value 0.009 0.006 0.006 0.062 0.062 0.064 0.060 0.061 0.063 0.044 0.006 0.042 0.012 0.014 0.001 Probability of joining RJV 0.011 0.011 0.015 0.015 0.015 0.025 0.012 0.012 0.017 0.008 0.008 0.045 0.010 0.010 0.038 Number of observations Without R&D effects 415,439 415,439 337,727 254,629 254,629 144,291 93,046 93,046 75,635 91,506 91,506 16,456 52,142 52,142 15,088 With R&D effects 340,343 340,343 276,726 148,821 148,821 88,757 80,048 80,048 65,033 70,145 70,145 13787 43,934 43,934 13,027 Notes. Robust standard errors clustered by firm are in parentheses. |$^{*}$| indicates significant at the 10% level; |$^{**}$| at 5% level; and |$^{***}$| at 1% level. The total effect is computed at the mean of the independent variables. We include the following controls in all regressions: constant, firm assets, firm capacity constraints, number of RJVs the firm is a member of and its square, whether the RJV is new, number of RJV members, relative assets, relative capital constraints, industry fixed effects (for RAs), and year fixed effects. An observation is a firm–year–RJV combination. Open in new tab The results show a significant and negative impact of the individual leniency policy (post leniency dummy) for all five market definitions and all specifications. In addition, the results show that firms are more likely to join RJVs with higher market power (RJV HHI) across all specifications. The interaction of these two variables (RJV HHI*Leniency) shows that the higher the market power of the RJV, the more an impact the leniency policy has on the decision to join. We examine this in more detail below in Figure 2. However, the role of the RJV as a means to coordinate collusion across many fragmented firms (RJV HHI fragmentation) does not seem to be supported by that data. The estimates are not consistent in sign across specifications, nor are the interactions with the post-leniency dummy variable. Figure 2. Open in new tabDownload slide Leniency policy effects on probability join RJV in computer markets. For policy purposes, the overall effect is of particular interest since it will tell us the average impact on RJV formation. The predicted total effect of the revised leniency policy on the probability of joining a RJV is given in the final rows of the upper and lower panels. The total effect is the difference in the predicted probability of joining a RJV under a revised policy versus the probability of joining under no revision, evaluated at the mean of the regressors. Indeed, the results indicate the total effect is identical or very close across the R&D effects within specifications and is significant and negative across market definitions. We now discuss the total effect of the leniency policy in more detail for each market definition. The most broad definition of the relevant market is computer and electronic product manufacturing, which consists of firms that manufacture computers, computer peripherals, and components. The estimates (columns (1)–(3)) show that the predicted total effect of the leniency policy is around |$-0.003,$| which implies the revision resulted in a |$28\%$| reduction in the |$1\%$| observed probability of joining a RJV. The effect is of a similar magnitude among firms that enter RJVs with other rivals. As we narrow the market definition to the Software RA (columns (4)–(6)), we again find significant negative effects across all specifications. However, the effects are significant at lower levels larger in magnitude. Again, within specifications, the results are of similar magnitudes whether or not we control for endogenous R&D. Both measures of RJV market power show negative and significant impacts on the probability of joining. The previous market definitions included firms that manufacture computers together with those that manufacture inputs for computers. We consider the subsets individually in the next three market definitions. The first narrow definition, computer manufacturing consists of firms that manufacture or assemble mainframes, personal computer, servers, and so forth (estimates are in columns (7)-(9)). The revised leniency policy again has a significant and negative effect of the a similar magnitude across specifications, where the total effect of the policy is to reduce the probability of joining by |$30\%$|–|$40\%.$| Among semiconductor firms (such as Intel, AMD, Micron, and Motorola) (columns (10)–(12)), the total effect of the leniency policy is again negative and significant across specifications (a reduction in the probability of joining of |$23\%$|–|$59\%$|⁠). A further classification of semiconductors into its relevant six-digit components, microprocessors and memory (the last three columns) yield a much larger significant impact of the revision: among this subset, the total effect reduced RJV formation by |$57\%$|–|$90\%$|⁠. In summary, we find evidence of collusive behavior among computer manufacturers, but the behavior is more pronounced among memory and microprocessor producers, which is supported by evidence on collusive cases reported via the amnesty policy in the market for DRAM memory. Again, our results for the computer markets are consistent in sign and significance across all relevant markets. Across all specifications and market definitions, the leniency policy revision resulted in an average (median) drop in the significant probability of joining a RJV of |$41\%(34\%).$| It is also interesting to note that the estimated total effect is the same across specifications for both measures of the RJV market power (except for the software RA), even though the coefficient estimates for the RJV HHI measures differ. Figure 2 illustrates the total effect of the revision on the probability of joining a RJV for all values of RJV market power. Again, we present the results for one broadly defined three-digit market, computer and electronic manufacturing (dark line), and one narrowly defined market, memory and microprocessors (light line), both for the With-R&D-Effects specification. The figure reveals the same pattern as with telecommunications, suggesting that the higher the market power of the RJV, the more an impact the leniency policy has on the decision to join a RJV in computer markets. Again, as RJV market power increases, the probability of joining a RJV increases when there is no leniency policy (both solid lines). When there is a revised leniency policy, the probability of joining a RJV is lower (both dashed lines) and is impacted very little by the market power of the RJV. 5.2. Telecommunications Markets The top (lower) panel of Table 3 presents the estimates from the first-stage probit “Without R&D Effects” (endogenous switching regression “With R&D Effects”) specifications under differing aggregation levels for firms in the telecommunications market. The results show a negative impact of the individual leniency policy across (post leniency dummy) for all four market definitions and all specifications, and the impact is significant for the majority of the specifications. In addition, the results show that firms are more likely to join RJVs with higher market power (RJV HHI) across all specifications. The interaction of these two variables (RJV HHI*Leniency) shows that the benefit of joining an RJV with higher market power post leniency policy is declining on average when it is significant, except for the long-distance markets where the likelihood of joining an RJV post leniency policy is increasing in the market power of the RJV. This latter finding is not intuitive and we examine this in more detail momentarily when we discuss Figure 3. As with computer firms, the role of the RJV as a means to coordinate collusion across many fragmented firms (RJV HHI fragmentation) does not seem to be supported by that data. The estimates are not consistent in sign across specifications, nor are the interactions with the post-leniency dummy variable. Recall we posit that if firms enter RJVs to collude then the impact on R&D may be less important. If our conjecture is correct then, when the leniency policy has a significant effect on the decision to join, the results should be similar across R&D specifications (if the “back door” impact on R&D does not matter). We now discuss the total effect of the leniency policy in more detail for each market definition. Figure 3. Open in new tabDownload slide Leniency policy effects on probability join RJV in Telecom. Table 3. Estimates for join a RJV in telecom markets. Level of aggregation: . 3-digit NAICS . 3-digit NAICS . Long distance carriers . Relevant market definition: . Telecom RA . Broadcast telecom . All years . Regulated years . Sample: . All . All . Treatment . All . All . Treatment . All . All . Treatment . All . All . Treatment . No ATT . . (1) . (2) . (3) . (4) . (5) . (6) . (7) . (8) . (9) . (10) . (11) . (12) . (13) . Without R&D effects Post leniency dummy −0.334*** −0.290* −0.338*** −0.459* −0.383 −0.647** −0.627*** −0.726*** −0.684*** −0.845*** −0.805* −0.936*** −0.927*** (0.0845) (0.169) (0.0933) (0.277) (0.273) (0.326) (0.132) (0.259) (0.145) (0.166) (0.466) (0.175) (0.172) RJV HHI 0.404|$^{***}$| 0.511|$^{***}$| 0.995|$^{*}$| 0.615 0.777|$^{***}$| 0.709|$^{***}$| 0.773|$^{***}$| 0.756|$^{***}$| 0.746|$^{***}$| (0.133) (0.136) (0.600) (0.540) (0.121) (0.131) (0.142) (0.148) (0.154) RJV HHI*Post leniency −0.175 −0.329** −0.272 −0.0951 0.257 0.217 0.326 0.704|$^{*}$| 0.563 (0.139) (0.143) (0.492) (0.593) (0.166) (0.184) (0.394) (0.367) (0.401) RJV HHI fragmentation −0.0759 0.695|$^{***}$| −1.317*** −1.388*** (0.164) (0.153) (0.179) (0.219) RJV HHI fragmentation|$^{*}$| −0.0888 −0.277** 0.0807 −0.0811 Post leniency (0.147) (0.111) (0.232) (0.480) Leniency policy total effect −0.011** −0.011* −0.018** −0.0161 −0.0298 −0.0321 −0.016** −0.019** −0.022** −0.014** −0.015** −0.019** −0.014** Wald test statistic 8.858 9.450 9.170 3.213 3.904 3.464 7.077 7.572 7.011 7.936 8.171 7.062 7.309 P-value 0.0173 0.052 0.0115 0.174 0.157 0.148 0.0288 0.0223 0.0304 0.0231 0.0224 0.0340 0.0302 With R&D effects Post leniency dummy −0.284*** −0.357 −0.318*** −0.421 −0.454 −0.613 −0.677*** −0.640** −0.734*** −0.927*** −0.616 −1.066*** See (0.104) (0.259) (0.117) (0.432) (0.461) (1.836) (0.138) (0.275) (0.152) (0.172) (0.472) (0.193) Note RJV HHI 0.814|$^{***}$| 0.893|$^{***}$| 1.905 1.603 0.750|$^{***}$| 0.668|$^{***}$| 0.746|$^{***}$| 0.680|$^{***}$| (0.179) (0.176) (1.798) (12.48) (0.127) (0.136) (0.154) (0.155) RJV HHI*Post leniency −0.568*** −0.716*** −0.334 −0.275 0.355|$^{**}$| 0.342|$^{*}$| 0.563 1.076|$^{***}$| (0.162) (0.173) (0.817) (4.222) (0.172) (0.192) (0.401) (0.389) RJV HHI fragmentation −0.226 1.005|$^{***}$| −1.268*** −1.356*** (0.275) (0.248) (0.186) (0.235) RJV HHI fragmentation|$^{*}$| −0.0560 −0.458*** −0.0507 −0.337 Post leniency (0.247) (0.125) (0.242) (0.475) Leniency policy total effect −0.015** −0.013 −0.025** −0.017 −0.054 −0.052 −0.016** −0.020** −0.023** −0.014** −0.016** −0.019* Wald test statistic 5.924 6.993 6.639 0.915 1.514 0.0775 6.775 7.184 6.678 7.309 7.502 6.765 P-value 0.0379 0.051 0.0299 0.486 0.341 0.881 0.0343 0.0267 0.0372 0.0302 0.0280 0.0511 Probability of joining RJV 0.016 0.016 0.024 0.021 0.021 0.038 0.058 0.058 0.069 0.058 0.058 0.058 0.069 Number of observations Without R&D effects 52,159 52,159 35,966 11,648 11,648 6,410 1,627 1,627 1,397 844 844 744 257 With R&D effects 37,899 37,899 27,006 2,389 2,389 1,346 334 334 295 257 257 226 Level of aggregation: . 3-digit NAICS . 3-digit NAICS . Long distance carriers . Relevant market definition: . Telecom RA . Broadcast telecom . All years . Regulated years . Sample: . All . All . Treatment . All . All . Treatment . All . All . Treatment . All . All . Treatment . No ATT . . (1) . (2) . (3) . (4) . (5) . (6) . (7) . (8) . (9) . (10) . (11) . (12) . (13) . Without R&D effects Post leniency dummy −0.334*** −0.290* −0.338*** −0.459* −0.383 −0.647** −0.627*** −0.726*** −0.684*** −0.845*** −0.805* −0.936*** −0.927*** (0.0845) (0.169) (0.0933) (0.277) (0.273) (0.326) (0.132) (0.259) (0.145) (0.166) (0.466) (0.175) (0.172) RJV HHI 0.404|$^{***}$| 0.511|$^{***}$| 0.995|$^{*}$| 0.615 0.777|$^{***}$| 0.709|$^{***}$| 0.773|$^{***}$| 0.756|$^{***}$| 0.746|$^{***}$| (0.133) (0.136) (0.600) (0.540) (0.121) (0.131) (0.142) (0.148) (0.154) RJV HHI*Post leniency −0.175 −0.329** −0.272 −0.0951 0.257 0.217 0.326 0.704|$^{*}$| 0.563 (0.139) (0.143) (0.492) (0.593) (0.166) (0.184) (0.394) (0.367) (0.401) RJV HHI fragmentation −0.0759 0.695|$^{***}$| −1.317*** −1.388*** (0.164) (0.153) (0.179) (0.219) RJV HHI fragmentation|$^{*}$| −0.0888 −0.277** 0.0807 −0.0811 Post leniency (0.147) (0.111) (0.232) (0.480) Leniency policy total effect −0.011** −0.011* −0.018** −0.0161 −0.0298 −0.0321 −0.016** −0.019** −0.022** −0.014** −0.015** −0.019** −0.014** Wald test statistic 8.858 9.450 9.170 3.213 3.904 3.464 7.077 7.572 7.011 7.936 8.171 7.062 7.309 P-value 0.0173 0.052 0.0115 0.174 0.157 0.148 0.0288 0.0223 0.0304 0.0231 0.0224 0.0340 0.0302 With R&D effects Post leniency dummy −0.284*** −0.357 −0.318*** −0.421 −0.454 −0.613 −0.677*** −0.640** −0.734*** −0.927*** −0.616 −1.066*** See (0.104) (0.259) (0.117) (0.432) (0.461) (1.836) (0.138) (0.275) (0.152) (0.172) (0.472) (0.193) Note RJV HHI 0.814|$^{***}$| 0.893|$^{***}$| 1.905 1.603 0.750|$^{***}$| 0.668|$^{***}$| 0.746|$^{***}$| 0.680|$^{***}$| (0.179) (0.176) (1.798) (12.48) (0.127) (0.136) (0.154) (0.155) RJV HHI*Post leniency −0.568*** −0.716*** −0.334 −0.275 0.355|$^{**}$| 0.342|$^{*}$| 0.563 1.076|$^{***}$| (0.162) (0.173) (0.817) (4.222) (0.172) (0.192) (0.401) (0.389) RJV HHI fragmentation −0.226 1.005|$^{***}$| −1.268*** −1.356*** (0.275) (0.248) (0.186) (0.235) RJV HHI fragmentation|$^{*}$| −0.0560 −0.458*** −0.0507 −0.337 Post leniency (0.247) (0.125) (0.242) (0.475) Leniency policy total effect −0.015** −0.013 −0.025** −0.017 −0.054 −0.052 −0.016** −0.020** −0.023** −0.014** −0.016** −0.019* Wald test statistic 5.924 6.993 6.639 0.915 1.514 0.0775 6.775 7.184 6.678 7.309 7.502 6.765 P-value 0.0379 0.051 0.0299 0.486 0.341 0.881 0.0343 0.0267 0.0372 0.0302 0.0280 0.0511 Probability of joining RJV 0.016 0.016 0.024 0.021 0.021 0.038 0.058 0.058 0.069 0.058 0.058 0.058 0.069 Number of observations Without R&D effects 52,159 52,159 35,966 11,648 11,648 6,410 1,627 1,627 1,397 844 844 744 257 With R&D effects 37,899 37,899 27,006 2,389 2,389 1,346 334 334 295 257 257 226 Notes. Robust standard errors clustered by firm are in parentheses. |$^{*}$| indicates significant at the 10% level; |$^{**}$| at 5% level; and |$^{***}$| at 1% level. The total effect is computed at the mean of the independent variables. An observation is a a firm–year–RJV combination. We include the following controls in all regressions: constant, firm assets, firm capacity constraints, number of RJVs the firm is a member of and its square, number of RJV members, relative assets, whether the RJV is new, relative capital constraints, industry fixed effects (for RA markets), and year fixed effects. There were not enough observations to run specification (13) with R&D effects. Open in new tab Table 3. Estimates for join a RJV in telecom markets. Level of aggregation: . 3-digit NAICS . 3-digit NAICS . Long distance carriers . Relevant market definition: . Telecom RA . Broadcast telecom . All years . Regulated years . Sample: . All . All . Treatment . All . All . Treatment . All . All . Treatment . All . All . Treatment . No ATT . . (1) . (2) . (3) . (4) . (5) . (6) . (7) . (8) . (9) . (10) . (11) . (12) . (13) . Without R&D effects Post leniency dummy −0.334*** −0.290* −0.338*** −0.459* −0.383 −0.647** −0.627*** −0.726*** −0.684*** −0.845*** −0.805* −0.936*** −0.927*** (0.0845) (0.169) (0.0933) (0.277) (0.273) (0.326) (0.132) (0.259) (0.145) (0.166) (0.466) (0.175) (0.172) RJV HHI 0.404|$^{***}$| 0.511|$^{***}$| 0.995|$^{*}$| 0.615 0.777|$^{***}$| 0.709|$^{***}$| 0.773|$^{***}$| 0.756|$^{***}$| 0.746|$^{***}$| (0.133) (0.136) (0.600) (0.540) (0.121) (0.131) (0.142) (0.148) (0.154) RJV HHI*Post leniency −0.175 −0.329** −0.272 −0.0951 0.257 0.217 0.326 0.704|$^{*}$| 0.563 (0.139) (0.143) (0.492) (0.593) (0.166) (0.184) (0.394) (0.367) (0.401) RJV HHI fragmentation −0.0759 0.695|$^{***}$| −1.317*** −1.388*** (0.164) (0.153) (0.179) (0.219) RJV HHI fragmentation|$^{*}$| −0.0888 −0.277** 0.0807 −0.0811 Post leniency (0.147) (0.111) (0.232) (0.480) Leniency policy total effect −0.011** −0.011* −0.018** −0.0161 −0.0298 −0.0321 −0.016** −0.019** −0.022** −0.014** −0.015** −0.019** −0.014** Wald test statistic 8.858 9.450 9.170 3.213 3.904 3.464 7.077 7.572 7.011 7.936 8.171 7.062 7.309 P-value 0.0173 0.052 0.0115 0.174 0.157 0.148 0.0288 0.0223 0.0304 0.0231 0.0224 0.0340 0.0302 With R&D effects Post leniency dummy −0.284*** −0.357 −0.318*** −0.421 −0.454 −0.613 −0.677*** −0.640** −0.734*** −0.927*** −0.616 −1.066*** See (0.104) (0.259) (0.117) (0.432) (0.461) (1.836) (0.138) (0.275) (0.152) (0.172) (0.472) (0.193) Note RJV HHI 0.814|$^{***}$| 0.893|$^{***}$| 1.905 1.603 0.750|$^{***}$| 0.668|$^{***}$| 0.746|$^{***}$| 0.680|$^{***}$| (0.179) (0.176) (1.798) (12.48) (0.127) (0.136) (0.154) (0.155) RJV HHI*Post leniency −0.568*** −0.716*** −0.334 −0.275 0.355|$^{**}$| 0.342|$^{*}$| 0.563 1.076|$^{***}$| (0.162) (0.173) (0.817) (4.222) (0.172) (0.192) (0.401) (0.389) RJV HHI fragmentation −0.226 1.005|$^{***}$| −1.268*** −1.356*** (0.275) (0.248) (0.186) (0.235) RJV HHI fragmentation|$^{*}$| −0.0560 −0.458*** −0.0507 −0.337 Post leniency (0.247) (0.125) (0.242) (0.475) Leniency policy total effect −0.015** −0.013 −0.025** −0.017 −0.054 −0.052 −0.016** −0.020** −0.023** −0.014** −0.016** −0.019* Wald test statistic 5.924 6.993 6.639 0.915 1.514 0.0775 6.775 7.184 6.678 7.309 7.502 6.765 P-value 0.0379 0.051 0.0299 0.486 0.341 0.881 0.0343 0.0267 0.0372 0.0302 0.0280 0.0511 Probability of joining RJV 0.016 0.016 0.024 0.021 0.021 0.038 0.058 0.058 0.069 0.058 0.058 0.058 0.069 Number of observations Without R&D effects 52,159 52,159 35,966 11,648 11,648 6,410 1,627 1,627 1,397 844 844 744 257 With R&D effects 37,899 37,899 27,006 2,389 2,389 1,346 334 334 295 257 257 226 Level of aggregation: . 3-digit NAICS . 3-digit NAICS . Long distance carriers . Relevant market definition: . Telecom RA . Broadcast telecom . All years . Regulated years . Sample: . All . All . Treatment . All . All . Treatment . All . All . Treatment . All . All . Treatment . No ATT . . (1) . (2) . (3) . (4) . (5) . (6) . (7) . (8) . (9) . (10) . (11) . (12) . (13) . Without R&D effects Post leniency dummy −0.334*** −0.290* −0.338*** −0.459* −0.383 −0.647** −0.627*** −0.726*** −0.684*** −0.845*** −0.805* −0.936*** −0.927*** (0.0845) (0.169) (0.0933) (0.277) (0.273) (0.326) (0.132) (0.259) (0.145) (0.166) (0.466) (0.175) (0.172) RJV HHI 0.404|$^{***}$| 0.511|$^{***}$| 0.995|$^{*}$| 0.615 0.777|$^{***}$| 0.709|$^{***}$| 0.773|$^{***}$| 0.756|$^{***}$| 0.746|$^{***}$| (0.133) (0.136) (0.600) (0.540) (0.121) (0.131) (0.142) (0.148) (0.154) RJV HHI*Post leniency −0.175 −0.329** −0.272 −0.0951 0.257 0.217 0.326 0.704|$^{*}$| 0.563 (0.139) (0.143) (0.492) (0.593) (0.166) (0.184) (0.394) (0.367) (0.401) RJV HHI fragmentation −0.0759 0.695|$^{***}$| −1.317*** −1.388*** (0.164) (0.153) (0.179) (0.219) RJV HHI fragmentation|$^{*}$| −0.0888 −0.277** 0.0807 −0.0811 Post leniency (0.147) (0.111) (0.232) (0.480) Leniency policy total effect −0.011** −0.011* −0.018** −0.0161 −0.0298 −0.0321 −0.016** −0.019** −0.022** −0.014** −0.015** −0.019** −0.014** Wald test statistic 8.858 9.450 9.170 3.213 3.904 3.464 7.077 7.572 7.011 7.936 8.171 7.062 7.309 P-value 0.0173 0.052 0.0115 0.174 0.157 0.148 0.0288 0.0223 0.0304 0.0231 0.0224 0.0340 0.0302 With R&D effects Post leniency dummy −0.284*** −0.357 −0.318*** −0.421 −0.454 −0.613 −0.677*** −0.640** −0.734*** −0.927*** −0.616 −1.066*** See (0.104) (0.259) (0.117) (0.432) (0.461) (1.836) (0.138) (0.275) (0.152) (0.172) (0.472) (0.193) Note RJV HHI 0.814|$^{***}$| 0.893|$^{***}$| 1.905 1.603 0.750|$^{***}$| 0.668|$^{***}$| 0.746|$^{***}$| 0.680|$^{***}$| (0.179) (0.176) (1.798) (12.48) (0.127) (0.136) (0.154) (0.155) RJV HHI*Post leniency −0.568*** −0.716*** −0.334 −0.275 0.355|$^{**}$| 0.342|$^{*}$| 0.563 1.076|$^{***}$| (0.162) (0.173) (0.817) (4.222) (0.172) (0.192) (0.401) (0.389) RJV HHI fragmentation −0.226 1.005|$^{***}$| −1.268*** −1.356*** (0.275) (0.248) (0.186) (0.235) RJV HHI fragmentation|$^{*}$| −0.0560 −0.458*** −0.0507 −0.337 Post leniency (0.247) (0.125) (0.242) (0.475) Leniency policy total effect −0.015** −0.013 −0.025** −0.017 −0.054 −0.052 −0.016** −0.020** −0.023** −0.014** −0.016** −0.019* Wald test statistic 5.924 6.993 6.639 0.915 1.514 0.0775 6.775 7.184 6.678 7.309 7.502 6.765 P-value 0.0379 0.051 0.0299 0.486 0.341 0.881 0.0343 0.0267 0.0372 0.0302 0.0280 0.0511 Probability of joining RJV 0.016 0.016 0.024 0.021 0.021 0.038 0.058 0.058 0.069 0.058 0.058 0.058 0.069 Number of observations Without R&D effects 52,159 52,159 35,966 11,648 11,648 6,410 1,627 1,627 1,397 844 844 744 257 With R&D effects 37,899 37,899 27,006 2,389 2,389 1,346 334 334 295 257 257 226 Notes. Robust standard errors clustered by firm are in parentheses. |$^{*}$| indicates significant at the 10% level; |$^{**}$| at 5% level; and |$^{***}$| at 1% level. The total effect is computed at the mean of the independent variables. An observation is a a firm–year–RJV combination. We include the following controls in all regressions: constant, firm assets, firm capacity constraints, number of RJVs the firm is a member of and its square, number of RJV members, relative assets, whether the RJV is new, relative capital constraints, industry fixed effects (for RA markets), and year fixed effects. There were not enough observations to run specification (13) with R&D effects. Open in new tab Columns (1)–(3) give the estimates for firms that join a RJV in the Telecom RA.51 The coefficient estimates for the level and/or RJV HHI interactions with the leniency policy dummy are negative and usually significant. The negative coefficients indicate that firms are less likely to join a RJV after the revision and that the effect is more pronounced as the RJV market power (RJV HHI) increases. The first specification yields a predicted total effect of about |$-0.01,$| which implies a |$69\%$| reduction in the |$1.6\%$| probability of joining a Telecom RA venture.52 As column (3) indicates, the effect is more pronounced among firms that enter RJVs with other rivals where the total effect is a reduction between |$75\%$| and |$100\%$| from the |$2.4\%$| probability of joining (i.e., the treatment group). The results are consistent whether or not we control for the endogenous nature of RJV formation (top or bottom panels). However, the fragmentation measure of RJV market power (column (2)) does not yield as significant of an effect of the leniency policy. Among firms in Broadcast Telecom (in columns (4)–(6)), the total effect is significant only at the |$80\%$| level and only for the Without R&D Effect specification. Note also that these markets are not ideal in that it they involve firms that are not in the same antitrust market over a period of the data (due to federal regulations). We now turn to the narrowest of our definitions of long-distance firms, and the ones for which we have additional FCC data. The revision has a highly negative and significant effect on RJV formation across almost all specifications (columns (7)–(13)), where the probability of entering a RJV decreases by about |$28\%$| (⁠|$33\%$| among firms in the treatment group). The market over all years may be too narrow post-1996 as it does not include local providers. During the regulated years (columns (10)–(13)), we find the leniency policy significantly lowers the probability of joining (for all specifications). The |$5.8\%$| observed probability of joining is reduced by |$24\%$| to |$33\%.$| In column (13), we control for if AT&T is a member of the RJV. We anticipate that RJVs formed without AT&T would have the most collusive potential since the firms that needed to coordinate were the non-AT&T firms. We do not have enough data to estimate this specification for the with R&D effects specification, but the without R&D effects specification indicates that the leniency policy revision had a significant negative impact on the probability of joining a non-AT&T RJV (a reduction of |$20\%).$| Our results are consistent in sign, significance, and magnitude across all but one relevant market (which is insignificant). They show the corporate leniency policy revision resulted in a drop in the probability of joining a RJV, where the mean (median) drop across significant definitions is |$34\%$| (⁠|$28\%)$|⁠. The effect is more pronounced among firms that join only with product market rivals (the treatment group), where the mean (median) reduction is |$45\%$| (⁠|$33\%).$| The total effect under the RJV fragmentation market power measure is not significant or significant at a lower level than under other definitions. This is not surprising as there are only two large firms (MCI and Sprint) that need to coordinate and many small re-sellers. Hence, there is less of a need for a coordination device to facilitate collusion. The total effect, while informative, is calculated at the mean of the regressors including RJV market power. Figure 3 illustrates the total effect of the leniency policy revision on the probability of joining a RJV for all values of RJV market power (⁠|$H_{{ijt}}).$| We present the results for one broadly defined market, the Telecom RA, and one narrowly defined market, long distance providers during the period of regulation.53 The light gray lines show the total effect for the Telecom RA and the dark black lines for long distance carriers during the regulated period (both for the With-R&D-Effects specification). The figure reveals that the higher the market power of the RJV, the more an impact the leniency policy has on the decision to join a RJV in the Telecom RA. The figure reiterates the previous results namely the probability of joining a RJV is lower after the leniency policy is implemented. Furthermore, it shows that as RJV market power increases, the probability of joining a RJV increases when there is no leniency policy (both solid lines). When there is a revised leniency policy the probability of joining a RJV is lower (both dashed lines). The figures for the other market definitions with significant total effects are similar to the Telecom RA figure. These results suggest that the higher the market power of the RJV the more collusive potential it has, which results in a differential effect of the leniency policy on the probability of joining a RJV. Notably, even though firms in the computer industry are distinct from those in the telecommunications industry, Figures 2 and 3 show a similar pattern. The probability that a telecom firm enters an RJV prior to the leniency policy is increasing in the market power of the RJV, which is similar to the findings in other industries. However, unlike other markets, firms in long distance join more RJVs the higher is the RJV market power even after the leniency policy is in effect (although they join with an overall lower probability). This finding is likely an artifact of the composition of this market. Recall that during this time the long distance market consisted of a regulated dominant firm (AT&T) and two main competitors (MCI and Sprint)—combined these firms held more than 80% of the market share. These three firms were in RJVs together and it seems that they continued to enter RJVs together even after the leniency policy although they joined fewer than prior to the policy. 5.3. Petroleum Markets Table 4 presents the estimates for the leniency policy and RJV market power variables for the petroleum markets, where the total effect of the increase in leniency policy is negative (when it is significant). Columns (1)–(3) present the results for the coal and crude extraction firms. The total effect of the leniency policy is significant at the |$1\%$| level for the without R&D effects specifications, with a |$24\%$|–|$35\%$| reduction in the |$1\%$| observed probability of joining a RJV. The impact is of the same magnitude when we control for R&D endogeneity among firms that enter RJVs with other rivals (column (3)) significant at the |$10\%$| level. The results for the narrower market definition involving firms engaged in petroleum refining are in columns (4)–(6), where the leniency policy has a negative effect, although it is significant at the |$1\%$| level only in one specification. Table 4. Estimates for join a RJV in petroleum markets. . 3-digit NAICS . 6-digit NAICS . Level of aggregation: . Coal manufacture and crude extraction . Petroleum refining . Relevant market definition: . All . All . Treatment . All . All . Treatment . Sample: . (1) . (2) . (3) . (4) . (5) . (6) . Without R&D effects Post leniency policy dummy −1.494*** 1.496 −1.054*** −0.952* 0.244 −0.826 (0.335) (1.170) (0.404) (0.535) (1.123) (0.556) RJV HHI −0.28 1.297 6.102|$^{***}$| 6.312|$^{***}$| (0.649) (0.984) (2.127) (2.190) RJV HHI*Post leniency −0.0447 −1.628 −3.043 −3.632 (2.958) (2.881) (3.693) (3.601) RJV HHI fragmentation 2.060|$^{***}$| 2.275|$^{***}$| (0.562) (0.560) RJV HHI fragmentation*Post leniency −2.797*** −1.960** (1.025) (0.946) Total effect of leniency policy −0.034*** −0.029*** −0.029*** −0.040* −0.065*** −0.043 Wald test statistic 31.614 56.840 14.443 4.654 6.010 2.550 P-value 0.000 0.000 0.001 0.091 0.039 0.232 With R&D effects Post leniency policy dummy −1.449 0.145 −1.041* −0.889 0.349 −0.802 (1.409) (158.60) (0.574) (0.892) (1.459) (1.374) RJV HHI −0.202 1.315 6.331|$^{***}$| 6.434** (1.293) (0.954) (2.250) (2.542) RJV HHI*Post leniency policy 0.0641 −1.585 −3.029 −3.544 (5.607) (3.224) (4.297) (4.794) RJV HHI fragmentation 0.077 2.308|$^{***}$| (72.24) (0.619) RJV HHI fragmentation*Post leniency −0.174 −2.043* (211.90) (1.049) Total effect of leniency policy −0.032 0.042 −0.029* −0.038* 0.012 −0.039 Wald test statistic 1.164 0.037 5.250 5.807 3.838 4.276 P-value 0.424 0.982 0.072 0.055 0.147 0.118 Probability of joining RJV 0.095 0.095 0.123 0.121 0.121 0.152 Number of observations Without R&D effects 15,727 15,727 11,893 7,105 7,105 5,987 With R&D effects 5,142 5,142 4,232 3,766 3,766 3,204 . 3-digit NAICS . 6-digit NAICS . Level of aggregation: . Coal manufacture and crude extraction . Petroleum refining . Relevant market definition: . All . All . Treatment . All . All . Treatment . Sample: . (1) . (2) . (3) . (4) . (5) . (6) . Without R&D effects Post leniency policy dummy −1.494*** 1.496 −1.054*** −0.952* 0.244 −0.826 (0.335) (1.170) (0.404) (0.535) (1.123) (0.556) RJV HHI −0.28 1.297 6.102|$^{***}$| 6.312|$^{***}$| (0.649) (0.984) (2.127) (2.190) RJV HHI*Post leniency −0.0447 −1.628 −3.043 −3.632 (2.958) (2.881) (3.693) (3.601) RJV HHI fragmentation 2.060|$^{***}$| 2.275|$^{***}$| (0.562) (0.560) RJV HHI fragmentation*Post leniency −2.797*** −1.960** (1.025) (0.946) Total effect of leniency policy −0.034*** −0.029*** −0.029*** −0.040* −0.065*** −0.043 Wald test statistic 31.614 56.840 14.443 4.654 6.010 2.550 P-value 0.000 0.000 0.001 0.091 0.039 0.232 With R&D effects Post leniency policy dummy −1.449 0.145 −1.041* −0.889 0.349 −0.802 (1.409) (158.60) (0.574) (0.892) (1.459) (1.374) RJV HHI −0.202 1.315 6.331|$^{***}$| 6.434** (1.293) (0.954) (2.250) (2.542) RJV HHI*Post leniency policy 0.0641 −1.585 −3.029 −3.544 (5.607) (3.224) (4.297) (4.794) RJV HHI fragmentation 0.077 2.308|$^{***}$| (72.24) (0.619) RJV HHI fragmentation*Post leniency −0.174 −2.043* (211.90) (1.049) Total effect of leniency policy −0.032 0.042 −0.029* −0.038* 0.012 −0.039 Wald test statistic 1.164 0.037 5.250 5.807 3.838 4.276 P-value 0.424 0.982 0.072 0.055 0.147 0.118 Probability of joining RJV 0.095 0.095 0.123 0.121 0.121 0.152 Number of observations Without R&D effects 15,727 15,727 11,893 7,105 7,105 5,987 With R&D effects 5,142 5,142 4,232 3,766 3,766 3,204 Notes. Robust standard errors clustered by firm are in parentheses. |$^{*}$| indicates significant at the 10% level; |$^{**}$| at 5% and |$^{***}$| at 1%. The total effect is computed at the mean of the independent variables. We include the following controls in all regressions: constant, firm assets, firm capacity constraints, number of RJVs the firm is a member of and its square, number of RJV members, relative assets, relative capital constraints, industry fixed effects (for RAs), whether the RJV is new, and year fixed effects. An observation is a firm-year-RJV combination. Open in new tab Table 4. Estimates for join a RJV in petroleum markets. . 3-digit NAICS . 6-digit NAICS . Level of aggregation: . Coal manufacture and crude extraction . Petroleum refining . Relevant market definition: . All . All . Treatment . All . All . Treatment . Sample: . (1) . (2) . (3) . (4) . (5) . (6) . Without R&D effects Post leniency policy dummy −1.494*** 1.496 −1.054*** −0.952* 0.244 −0.826 (0.335) (1.170) (0.404) (0.535) (1.123) (0.556) RJV HHI −0.28 1.297 6.102|$^{***}$| 6.312|$^{***}$| (0.649) (0.984) (2.127) (2.190) RJV HHI*Post leniency −0.0447 −1.628 −3.043 −3.632 (2.958) (2.881) (3.693) (3.601) RJV HHI fragmentation 2.060|$^{***}$| 2.275|$^{***}$| (0.562) (0.560) RJV HHI fragmentation*Post leniency −2.797*** −1.960** (1.025) (0.946) Total effect of leniency policy −0.034*** −0.029*** −0.029*** −0.040* −0.065*** −0.043 Wald test statistic 31.614 56.840 14.443 4.654 6.010 2.550 P-value 0.000 0.000 0.001 0.091 0.039 0.232 With R&D effects Post leniency policy dummy −1.449 0.145 −1.041* −0.889 0.349 −0.802 (1.409) (158.60) (0.574) (0.892) (1.459) (1.374) RJV HHI −0.202 1.315 6.331|$^{***}$| 6.434** (1.293) (0.954) (2.250) (2.542) RJV HHI*Post leniency policy 0.0641 −1.585 −3.029 −3.544 (5.607) (3.224) (4.297) (4.794) RJV HHI fragmentation 0.077 2.308|$^{***}$| (72.24) (0.619) RJV HHI fragmentation*Post leniency −0.174 −2.043* (211.90) (1.049) Total effect of leniency policy −0.032 0.042 −0.029* −0.038* 0.012 −0.039 Wald test statistic 1.164 0.037 5.250 5.807 3.838 4.276 P-value 0.424 0.982 0.072 0.055 0.147 0.118 Probability of joining RJV 0.095 0.095 0.123 0.121 0.121 0.152 Number of observations Without R&D effects 15,727 15,727 11,893 7,105 7,105 5,987 With R&D effects 5,142 5,142 4,232 3,766 3,766 3,204 . 3-digit NAICS . 6-digit NAICS . Level of aggregation: . Coal manufacture and crude extraction . Petroleum refining . Relevant market definition: . All . All . Treatment . All . All . Treatment . Sample: . (1) . (2) . (3) . (4) . (5) . (6) . Without R&D effects Post leniency policy dummy −1.494*** 1.496 −1.054*** −0.952* 0.244 −0.826 (0.335) (1.170) (0.404) (0.535) (1.123) (0.556) RJV HHI −0.28 1.297 6.102|$^{***}$| 6.312|$^{***}$| (0.649) (0.984) (2.127) (2.190) RJV HHI*Post leniency −0.0447 −1.628 −3.043 −3.632 (2.958) (2.881) (3.693) (3.601) RJV HHI fragmentation 2.060|$^{***}$| 2.275|$^{***}$| (0.562) (0.560) RJV HHI fragmentation*Post leniency −2.797*** −1.960** (1.025) (0.946) Total effect of leniency policy −0.034*** −0.029*** −0.029*** −0.040* −0.065*** −0.043 Wald test statistic 31.614 56.840 14.443 4.654 6.010 2.550 P-value 0.000 0.000 0.001 0.091 0.039 0.232 With R&D effects Post leniency policy dummy −1.449 0.145 −1.041* −0.889 0.349 −0.802 (1.409) (158.60) (0.574) (0.892) (1.459) (1.374) RJV HHI −0.202 1.315 6.331|$^{***}$| 6.434** (1.293) (0.954) (2.250) (2.542) RJV HHI*Post leniency policy 0.0641 −1.585 −3.029 −3.544 (5.607) (3.224) (4.297) (4.794) RJV HHI fragmentation 0.077 2.308|$^{***}$| (72.24) (0.619) RJV HHI fragmentation*Post leniency −0.174 −2.043* (211.90) (1.049) Total effect of leniency policy −0.032 0.042 −0.029* −0.038* 0.012 −0.039 Wald test statistic 1.164 0.037 5.250 5.807 3.838 4.276 P-value 0.424 0.982 0.072 0.055 0.147 0.118 Probability of joining RJV 0.095 0.095 0.123 0.121 0.121 0.152 Number of observations Without R&D effects 15,727 15,727 11,893 7,105 7,105 5,987 With R&D effects 5,142 5,142 4,232 3,766 3,766 3,204 Notes. Robust standard errors clustered by firm are in parentheses. |$^{*}$| indicates significant at the 10% level; |$^{**}$| at 5% and |$^{***}$| at 1%. The total effect is computed at the mean of the independent variables. We include the following controls in all regressions: constant, firm assets, firm capacity constraints, number of RJVs the firm is a member of and its square, number of RJV members, relative assets, relative capital constraints, industry fixed effects (for RAs), whether the RJV is new, and year fixed effects. An observation is a firm-year-RJV combination. Open in new tab On average, the leniency policy revision resulted in a drop in the significant probability of joining a RJV of |$33\%.$| While, our results are consistent in sign and magnitude across markets, the leniency effects are not significant at the |$1\%$| level in the majority of the specifications. This may be attributable to the data limitations that we mentioned earlier, namely, we do not have data on the major market players in the petroleum industries and so our market definition does not capture all relevant competitors. Nonetheless, the results are suggestive, although not as robust as those for the other two markets. Figure 4 illustrates the total effect for all values of RJV market power in our two markets coal/crude extraction (light line and left axis) and petroleum refining (dark line and right axis). As the other figures, we find that the higher the market power of the RJV, the more an impact the leniency policy has on the decision to join a RJV for firms in petroleum refining. When there is a revised leniency policy, the probability of joining a RJV is lower (both dashed lines) and is not impacted by the market power of the RJV. Figure 4. Open in new tabDownload slide Leniency policy effects on probability join RJV in petroleum markets. 5.4. Policy Implications The economic damage caused by cartels is significant (Hyytinen, Steen, and Toivanen 2018, 2019). Connor (2003) estimates that modern international cartels resulted in |$28\%$| higher prices. The graphite electrode cartel, for instance, caused more than a |$60\%$| price increase. Small increases also have the potential to cause substantial harm. For example, the lysine cartel resulted in |$17\%$| higher prices: an overcharge of more than |$\$75$| million in the United States and |$\$200$| million worldwide (Connor 2003). Secondly, cartels may cause dynamic inefficiencies if firms have fewer incentives to innovate to improve their market position.54 Finally, cartels may generate ‘X-inefficiencies” in that efficient operation is not as necessary because collusive profits are sufficient to compensate for higher costs (Leibenstein 1966). To the extent that firms in alliances and trade associations (i.e., non-research focused ventures) are registered, and hence protected, under the NCRPA, the welfare implications of collusion are clear. Alliances and trade associations are not engaged in R&D and hence realize no efficiency gains to offset the welfare losses due to collusion. Therefore, welfare is unambiguously lower under collusion. Alliances are prevalent in many industries. For example, in the airline industry, a number of antitrust concerns over code-sharing has raised collusive concerns. A study by Oum and Park (1997) found that the 30 largest airlines were involved in over 300 various types of alliances in 1996 alone. In the early 1980s, antitrust authorities voiced concern over the cartel-inducing properties of alliances formed in the movie industry. Specifically, major movie companies created a RJV where members would provide movies exclusively to a pay network for a limited time before making them available to other networks.55 On the other hand, there are many potential benefits to R&D collaboration as discussed in Section 1. Whether overall welfare is reduced as a result of collusion among RJV members depends on the magnitude of the welfare loss due to product market collusion relative to the welfare gain due to R&D collaboration. The welfare implications depend both on the nature of competition in the industry, the characteristics of the RJV, and the extent to which RJVs help to overcome inefficiencies associated with R&D (such as high levels of spillovers (see Irwin and Klenow 1994)). Notice that antitrust authorities are faced with a similar dilemma when considering whether to approve a proposed merger. Mergers can generate efficiencies (e.g., by decreasing costs) but lead to increased market concentration that may overcompensate for the welfare gains. In deciding whether to permit a merger, antitrust authorities consider each case individually while applying a “rule of thumb” based on industry concentration.56 Our results suggest a parallel approach be taken with a subset of RJV applications. First, the conventional wisdom that cartels are most easily established in concentrated industries leads to the suggestion that RJVs formed in concentrated industries should be more closely examined. Second, every partial cartel leads to prices above the competitive level. However, the overcharge, and thus the damage caused, increases in the size of the cartel relative to the market. Therefore, the combined market share of the cartel relative to industry concentration (i.e., the RJV market power index) might provide an indication of the welfare implications of collusion. A RJV “rule-of-thumb” could be formulated based on the market power of the RJV. 6. Specification and Robustness Checks We conducted a number of goodness-of-fit, specification, and robustness checks, which we detail below.57 Overall our results are robust to alternative specifications, controls for potential serial correlation, and the inclusion of unobserved firm and RJV effects. Table 5 presents tests of the fit of the model. We find that the empirical model has strong explanatory power before the policy change, but that it fits well over the whole time period only when we include the policy dummy and interaction terms. The relevant pre-leniency sample is prior to the corporate leniency policy revision for the telecom markets and prior to the individual leniency policy for the computer and petroleum markets. The restricted (full) model is the model without (with) the leniency policy indicators and interactions. The first column presents the pseudo R-squared measure of fit of the restricted model over the pre-leniency period. The results of the second column show that the restricted model does not fit as well in the post-leniency period for any market definition. The final two columns report the results of a likelihood ratio test of the null hypothesis that the restricted model fits as well as the full model over the entire sample period. As the P-value shows we can reject the null hypothesis for most markets at the 1% level. We can reject the null at the 5% level for Telecom RA and the long distance firms under the period of regulation; and at the 10% level for computer manufacturing. There is only one market where the restricted model does not give a worse fit than the full model, that for long distance firms over all years (in the telecom industry). Table 5. Goodness-of-fit results. . . Restricted model . Likelihood ratio test: . . . Psuedo R-squared . Null hypothesis restricted model . . . Pre-leniency . Post-leniency . Fits as good as full model . . . Sample . Sample . Test statistic . P-value . Telecom RA 0.217 0.146 9.17 0.010 Broadcast 0.329 0.240 17.83 0.000 Long distance all years 0.445 0.375 1.65 0.439 Long distance regulated 0.445 0.398 7.46 0.024 Computer Computer/electronic 0.305 0.232 32.22 0.000 Software RA 0.308 0.234 24.31 0.000 Computer manufacturing 0.289 0.212 5.51 0.064 Semiconductors 0.283 0.197 9.60 0.008 Memory/microproc 0.303 0.228 10.78 0.005 Petroleum Coal/crude extraction 0.486 0.445 29.56 0.000 Petroleum refining 0.444 0.351 13.84 0.001 . . Restricted model . Likelihood ratio test: . . . Psuedo R-squared . Null hypothesis restricted model . . . Pre-leniency . Post-leniency . Fits as good as full model . . . Sample . Sample . Test statistic . P-value . Telecom RA 0.217 0.146 9.17 0.010 Broadcast 0.329 0.240 17.83 0.000 Long distance all years 0.445 0.375 1.65 0.439 Long distance regulated 0.445 0.398 7.46 0.024 Computer Computer/electronic 0.305 0.232 32.22 0.000 Software RA 0.308 0.234 24.31 0.000 Computer manufacturing 0.289 0.212 5.51 0.064 Semiconductors 0.283 0.197 9.60 0.008 Memory/microproc 0.303 0.228 10.78 0.005 Petroleum Coal/crude extraction 0.486 0.445 29.56 0.000 Petroleum refining 0.444 0.351 13.84 0.001 Notes. Results are reported for the “Without R&D Effects” specifications using the primary measure of the RJV measure of market power. All regressions include same controls as in main specification. The likelihood ratio test statistic assumes no heteroskedasticity in standard errors. The Wald statistic for the significance of the leniency policy variables is presented in the estimates tables. The restricted (full) model is the model without (with) the leniency policy indicators. The relevant pre-leniency sample is prior to revision for telecom and prior to increase in fines for computer and petroleum. Open in new tab Table 5. Goodness-of-fit results. . . Restricted model . Likelihood ratio test: . . . Psuedo R-squared . Null hypothesis restricted model . . . Pre-leniency . Post-leniency . Fits as good as full model . . . Sample . Sample . Test statistic . P-value . Telecom RA 0.217 0.146 9.17 0.010 Broadcast 0.329 0.240 17.83 0.000 Long distance all years 0.445 0.375 1.65 0.439 Long distance regulated 0.445 0.398 7.46 0.024 Computer Computer/electronic 0.305 0.232 32.22 0.000 Software RA 0.308 0.234 24.31 0.000 Computer manufacturing 0.289 0.212 5.51 0.064 Semiconductors 0.283 0.197 9.60 0.008 Memory/microproc 0.303 0.228 10.78 0.005 Petroleum Coal/crude extraction 0.486 0.445 29.56 0.000 Petroleum refining 0.444 0.351 13.84 0.001 . . Restricted model . Likelihood ratio test: . . . Psuedo R-squared . Null hypothesis restricted model . . . Pre-leniency . Post-leniency . Fits as good as full model . . . Sample . Sample . Test statistic . P-value . Telecom RA 0.217 0.146 9.17 0.010 Broadcast 0.329 0.240 17.83 0.000 Long distance all years 0.445 0.375 1.65 0.439 Long distance regulated 0.445 0.398 7.46 0.024 Computer Computer/electronic 0.305 0.232 32.22 0.000 Software RA 0.308 0.234 24.31 0.000 Computer manufacturing 0.289 0.212 5.51 0.064 Semiconductors 0.283 0.197 9.60 0.008 Memory/microproc 0.303 0.228 10.78 0.005 Petroleum Coal/crude extraction 0.486 0.445 29.56 0.000 Petroleum refining 0.444 0.351 13.84 0.001 Notes. Results are reported for the “Without R&D Effects” specifications using the primary measure of the RJV measure of market power. All regressions include same controls as in main specification. The likelihood ratio test statistic assumes no heteroskedasticity in standard errors. The Wald statistic for the significance of the leniency policy variables is presented in the estimates tables. The restricted (full) model is the model without (with) the leniency policy indicators. The relevant pre-leniency sample is prior to revision for telecom and prior to increase in fines for computer and petroleum. Open in new tab Our first robustness check considers that, in our descriptive framework including controls for observable industry, RJV, and firm characteristics may not be sufficient as there may be unobserved firm- or RJV-specific factors that affect the value of entering a RJV. To determine if this impacts our results, we estimate a number of fixed effects logit models of the decision to enter a RJV.58 We present the results for the “Without R&D” specification for the primary measure of market power (i.e., the counterpart to the results reported in the top panel first column of each market definition in Tables 2–4). Table 6 includes the total effect of the leniency policy across market definitions for a logit regression without fixed effects (for comparison), with RJV fixed effects, and with firm fixed effects. The total effect of the revised leniency policy does not change when firm and RJV fixed effects are included, where the effect is significant and negative across almost all specifications. Furthermore, for most specifications, the magnitude of the total effect does not change significantly, although in some instances the total effect is higher when fixed effects are included. These results indicate that our findings are robust to inclusion of firm and RJV fixed effects. Table 6. Fixed effect logit results. Industry . Market definition . Included . Total effect of . Wald . P-value . . . fixed effects . leniency policy . statistic . . Telecom Research area None −0.001** 8.784 0.012 RJV −0.009*** 38.020 0.000 Firm −0.024* 5.488 0.064 Broadcast None −0.011*** 13.420 0.001 RJV −0.019** 9.233 0.010 Firm −0.039 3.088 0.213 Long dstance all years None −0.038 1.852 0.396 RJV −0.593 4.171 0.124 Firm −0.076 0.722 0.697 Long distance regulated None −0.001*** 16.060 0.000 RJV −0.370*** 144.700 0.000 Firm −0.032*** 36.980 0.000 Computer Computer/electronic None −0.001*** 26.790 0.000 RJV −0.034*** 36.360 0.000 Firm −0.017** 19.520 0.000 Software RA None −0.001*** 23.480 0.000 RJV −0.080*** 95.790 0.000 Firm −0.167*** 16.690 0.000 Computer manufacturing None −0.003** 9.225 0.010 RJV −0.547*** 391.000 0.000 Firm −0.040*** 17.050 0.000 Semiconductors None −0.001*** 12.330 0.002 RJV 0.000 0.112 0.946 Firm −0.005*** 10.300 0.006 Memory/microprocessors None −0.002*** 9.751 0.008 RJV −0.016 2.287 0.319 Firm −0.009** 8.416 0.015 Petroleum Coal/crude extraction None −0.023*** 27.130 0.000 RJV 0.473 2.014 0.156 Firm −0.011* 5.719 0.057 Petroleum refining None −0.028** 7.976 0.019 RJV −0.144*** 9815.000 0.000 Firm −0.084*** 14.740 0.001 Industry . Market definition . Included . Total effect of . Wald . P-value . . . fixed effects . leniency policy . statistic . . Telecom Research area None −0.001** 8.784 0.012 RJV −0.009*** 38.020 0.000 Firm −0.024* 5.488 0.064 Broadcast None −0.011*** 13.420 0.001 RJV −0.019** 9.233 0.010 Firm −0.039 3.088 0.213 Long dstance all years None −0.038 1.852 0.396 RJV −0.593 4.171 0.124 Firm −0.076 0.722 0.697 Long distance regulated None −0.001*** 16.060 0.000 RJV −0.370*** 144.700 0.000 Firm −0.032*** 36.980 0.000 Computer Computer/electronic None −0.001*** 26.790 0.000 RJV −0.034*** 36.360 0.000 Firm −0.017** 19.520 0.000 Software RA None −0.001*** 23.480 0.000 RJV −0.080*** 95.790 0.000 Firm −0.167*** 16.690 0.000 Computer manufacturing None −0.003** 9.225 0.010 RJV −0.547*** 391.000 0.000 Firm −0.040*** 17.050 0.000 Semiconductors None −0.001*** 12.330 0.002 RJV 0.000 0.112 0.946 Firm −0.005*** 10.300 0.006 Memory/microprocessors None −0.002*** 9.751 0.008 RJV −0.016 2.287 0.319 Firm −0.009** 8.416 0.015 Petroleum Coal/crude extraction None −0.023*** 27.130 0.000 RJV 0.473 2.014 0.156 Firm −0.011* 5.719 0.057 Petroleum refining None −0.028** 7.976 0.019 RJV −0.144*** 9815.000 0.000 Firm −0.084*** 14.740 0.001 Notes. Results are reported for the “Without R&D Effects” specifications using the primary measure of the RJV measure of market power. All regressions include same controls as in main specification including year dummies (and industry dummies for RAs). Open in new tab Table 6. Fixed effect logit results. Industry . Market definition . Included . Total effect of . Wald . P-value . . . fixed effects . leniency policy . statistic . . Telecom Research area None −0.001** 8.784 0.012 RJV −0.009*** 38.020 0.000 Firm −0.024* 5.488 0.064 Broadcast None −0.011*** 13.420 0.001 RJV −0.019** 9.233 0.010 Firm −0.039 3.088 0.213 Long dstance all years None −0.038 1.852 0.396 RJV −0.593 4.171 0.124 Firm −0.076 0.722 0.697 Long distance regulated None −0.001*** 16.060 0.000 RJV −0.370*** 144.700 0.000 Firm −0.032*** 36.980 0.000 Computer Computer/electronic None −0.001*** 26.790 0.000 RJV −0.034*** 36.360 0.000 Firm −0.017** 19.520 0.000 Software RA None −0.001*** 23.480 0.000 RJV −0.080*** 95.790 0.000 Firm −0.167*** 16.690 0.000 Computer manufacturing None −0.003** 9.225 0.010 RJV −0.547*** 391.000 0.000 Firm −0.040*** 17.050 0.000 Semiconductors None −0.001*** 12.330 0.002 RJV 0.000 0.112 0.946 Firm −0.005*** 10.300 0.006 Memory/microprocessors None −0.002*** 9.751 0.008 RJV −0.016 2.287 0.319 Firm −0.009** 8.416 0.015 Petroleum Coal/crude extraction None −0.023*** 27.130 0.000 RJV 0.473 2.014 0.156 Firm −0.011* 5.719 0.057 Petroleum refining None −0.028** 7.976 0.019 RJV −0.144*** 9815.000 0.000 Firm −0.084*** 14.740 0.001 Industry . Market definition . Included . Total effect of . Wald . P-value . . . fixed effects . leniency policy . statistic . . Telecom Research area None −0.001** 8.784 0.012 RJV −0.009*** 38.020 0.000 Firm −0.024* 5.488 0.064 Broadcast None −0.011*** 13.420 0.001 RJV −0.019** 9.233 0.010 Firm −0.039 3.088 0.213 Long dstance all years None −0.038 1.852 0.396 RJV −0.593 4.171 0.124 Firm −0.076 0.722 0.697 Long distance regulated None −0.001*** 16.060 0.000 RJV −0.370*** 144.700 0.000 Firm −0.032*** 36.980 0.000 Computer Computer/electronic None −0.001*** 26.790 0.000 RJV −0.034*** 36.360 0.000 Firm −0.017** 19.520 0.000 Software RA None −0.001*** 23.480 0.000 RJV −0.080*** 95.790 0.000 Firm −0.167*** 16.690 0.000 Computer manufacturing None −0.003** 9.225 0.010 RJV −0.547*** 391.000 0.000 Firm −0.040*** 17.050 0.000 Semiconductors None −0.001*** 12.330 0.002 RJV 0.000 0.112 0.946 Firm −0.005*** 10.300 0.006 Memory/microprocessors None −0.002*** 9.751 0.008 RJV −0.016 2.287 0.319 Firm −0.009** 8.416 0.015 Petroleum Coal/crude extraction None −0.023*** 27.130 0.000 RJV 0.473 2.014 0.156 Firm −0.011* 5.719 0.057 Petroleum refining None −0.028** 7.976 0.019 RJV −0.144*** 9815.000 0.000 Firm −0.084*** 14.740 0.001 Notes. Results are reported for the “Without R&D Effects” specifications using the primary measure of the RJV measure of market power. All regressions include same controls as in main specification including year dummies (and industry dummies for RAs). Open in new tab Second, we conducted a “placebo” test of whether we would conclude that the leniency policy revisions had a significant negative effect on RJV formation if we incorrectly assigned the year of the policy change. For all market definitions, we assigned placebo leniency years for the three years surrounding the correct leniency policy revision. For the telecommunications industry, there was no effect of the placebo policy during any year or market definition with the exception of the broadly defined three-digit telecommunications RA. The placebo policy change had a very small effect (on average a |$2\%$| drop in RJV formation over the placebo years) for the telecom RA, although the effect was statistically significant. This result suggests the telecom RA is not an appropriate market definition, but it is one of many and is the least preferred for other reasons as mentioned in previous sections. For the computer industry, there was no negative effect of the placebo policy for any years or market definitions.59 For the petroleum industry, there was no effect of the incorrect leniency policy change for any market definitions or years. Third, the estimates presented in the results section address potential serial correlation in the errors by clustering. An alternative way to limit the effects of potential serial correlation is to run the regressions in a tighter window around the leniency policy. We estimated the regressions using only data from 1991 to 1997. The results from this robustness check do not change in sign or significance, although the total effect of the leniency policy revision is smaller in magnitude for the three-digit computer market definitions and the computer manufacturing market definition. Not surprisingly, given the restricted sample, the significance values are lower. The results suggest that the negative impact of the leniency policy on RJV formation is not an artifact of the sample period. Finally, to examine if our results are sensitive to limiting the set of RJVs which firms can enter (by considering specific markets), we estimated our model using a pooled sample of firms across all industries. Constructing a sample that consists of all possible firm–year–RJV combinations in all industries yields a dataset of unmanageable size. Therefore, we restricted our analysis to firms with complete Compustat data that joined a RJV (in a given year) and a random sample of firms that did not join (in a given year).60 Our pooled sample consists of 1,651 firms yielding 13,399 firm years and 133,654 firm-year-RJV observations. Firms in the pooled sample undertake more R&D and have more assets, free cash, and sales than an average Compustat firm. Market concentration at the six-digit NAICs is higher (0.280) than at the three-digit level (0.082). Table 7 presents the results for the pooled sample across all three-digit industries for both R&D effects specifications. Columns (1) and (4) include the post leniency policy revision and its interaction with the RJV market power as regressors, for the without R&D and with R&D effects, respectively. The results suggest that, holding industry, firm, and RJV characteristics constant, the revision of the leniency policy resulted in a significant reduction of the probability a firm joins a RJV in the set of available RJVs to the firm regardless of industry. The total effect of the leniency policy is to reduce the probability of joining by |$1.5\%$| to |$4\%$|⁠, although the reduction is larger for RJVs among rivals (columns (3) and (6)) where it is |$2\%$| to |$5.5\%.$| The total effects are significantly negative at the |$1\%$| level. When the leniency policy indicator is interacted with the fragmentation measure of RJV market power, in columns (2) and (5), the results indicate the leniency policy level effect is significant and negative, but the interaction is significant and positive resulting in an overall positive effect of the leniency policy. The impact is not significant at the |$1\%$| level and the magnitude is also small (⁠|$1\%$|–|$2\%$|⁠). One potential problem in interpreting the across industries results is that the RJV measure of market power differs across industries and is of a similar magnitude within industries. Hence, while industry fixed effects are included as control variables, it may still be that the RJV market power effects are reflecting differences across industries in RJV market power. Furthermore, differences across industries (such as the degree of product homogeneity and many dispersed customers) may influence the feasibility or the optimal structure of the partial cartel. These points further motivate the necessity to examine firm behaviors within more narrowly defined markets. However, the exercise is useful as it presents an overall picture which is consistent with our industry-specific results in that the effect on average is negative and significant, which suggest our industry-specific findings are not due to a limiting of the potential choice set of firms. Table 7. Estimates for join a RJV for pooled three-digit industries. . Without R&D effects . With R&D effects . Specification: . All . All . Treatment . All . All . Treatment . Sample: . (1) . (2) . (3) . (4) . (5) . (6) . Post leniency policy dummy −0.178*** −0.247** −0.171*** −0.113* −0.195* −0.108* (0.0645) (0.126) (0.0655) (0.0580) (0.115) (0.0591) RJV HHI 0.717|$^{***}$| 0.655|$^{***}$| 0.645|$^{***}$| 0.582|$^{***}$| (0.0726) (0.0759) (0.0711) (0.0719) RJV HHI*Post leniency 0.331|$^{***}$| 0.292|$^{***}$| 0.301|$^{***}$| 0.266|$^{***}$| (0.0774) (0.0781) (0.0716) (0.0704) RJV HHI fragmentation −0.716*** −0.629*** (0.0872) (0.0795) RJV HHI fragmentation*Post leniency 0.296|$^{**}$| 0.276|$^{**}$| (0.128) (0.114) Total effect of leniency policy −0.0008*** 0.0002|$^{*}$| −0.0011*** −0.0003*** 0.0004|$^{**}$| −0.0004*** Wald test statistic 19.57 5.450 15.54 17.84 6.960 14.50 P-value 0.0001 0.0655 0.000422 0.0001 0.0308 0.0007 Observations 222,059 222,059 146,458 166,916 166,916 112,747 . Without R&D effects . With R&D effects . Specification: . All . All . Treatment . All . All . Treatment . Sample: . (1) . (2) . (3) . (4) . (5) . (6) . Post leniency policy dummy −0.178*** −0.247** −0.171*** −0.113* −0.195* −0.108* (0.0645) (0.126) (0.0655) (0.0580) (0.115) (0.0591) RJV HHI 0.717|$^{***}$| 0.655|$^{***}$| 0.645|$^{***}$| 0.582|$^{***}$| (0.0726) (0.0759) (0.0711) (0.0719) RJV HHI*Post leniency 0.331|$^{***}$| 0.292|$^{***}$| 0.301|$^{***}$| 0.266|$^{***}$| (0.0774) (0.0781) (0.0716) (0.0704) RJV HHI fragmentation −0.716*** −0.629*** (0.0872) (0.0795) RJV HHI fragmentation*Post leniency 0.296|$^{**}$| 0.276|$^{**}$| (0.128) (0.114) Total effect of leniency policy −0.0008*** 0.0002|$^{*}$| −0.0011*** −0.0003*** 0.0004|$^{**}$| −0.0004*** Wald test statistic 19.57 5.450 15.54 17.84 6.960 14.50 P-value 0.0001 0.0655 0.000422 0.0001 0.0308 0.0007 Observations 222,059 222,059 146,458 166,916 166,916 112,747 Notes. Robust standard errors clustered by firm are in parenthesis. An observation is a firm–RJV–year combination. All specifications include the usual controls plus industry dummies. Total effect is computed at the mean of the regressors. |$^{*}$| indicates significant at 10% level; |$^{**}$| at 5% level; and |$^{***}$| at 1% level. Open in new tab Table 7. Estimates for join a RJV for pooled three-digit industries. . Without R&D effects . With R&D effects . Specification: . All . All . Treatment . All . All . Treatment . Sample: . (1) . (2) . (3) . (4) . (5) . (6) . Post leniency policy dummy −0.178*** −0.247** −0.171*** −0.113* −0.195* −0.108* (0.0645) (0.126) (0.0655) (0.0580) (0.115) (0.0591) RJV HHI 0.717|$^{***}$| 0.655|$^{***}$| 0.645|$^{***}$| 0.582|$^{***}$| (0.0726) (0.0759) (0.0711) (0.0719) RJV HHI*Post leniency 0.331|$^{***}$| 0.292|$^{***}$| 0.301|$^{***}$| 0.266|$^{***}$| (0.0774) (0.0781) (0.0716) (0.0704) RJV HHI fragmentation −0.716*** −0.629*** (0.0872) (0.0795) RJV HHI fragmentation*Post leniency 0.296|$^{**}$| 0.276|$^{**}$| (0.128) (0.114) Total effect of leniency policy −0.0008*** 0.0002|$^{*}$| −0.0011*** −0.0003*** 0.0004|$^{**}$| −0.0004*** Wald test statistic 19.57 5.450 15.54 17.84 6.960 14.50 P-value 0.0001 0.0655 0.000422 0.0001 0.0308 0.0007 Observations 222,059 222,059 146,458 166,916 166,916 112,747 . Without R&D effects . With R&D effects . Specification: . All . All . Treatment . All . All . Treatment . Sample: . (1) . (2) . (3) . (4) . (5) . (6) . Post leniency policy dummy −0.178*** −0.247** −0.171*** −0.113* −0.195* −0.108* (0.0645) (0.126) (0.0655) (0.0580) (0.115) (0.0591) RJV HHI 0.717|$^{***}$| 0.655|$^{***}$| 0.645|$^{***}$| 0.582|$^{***}$| (0.0726) (0.0759) (0.0711) (0.0719) RJV HHI*Post leniency 0.331|$^{***}$| 0.292|$^{***}$| 0.301|$^{***}$| 0.266|$^{***}$| (0.0774) (0.0781) (0.0716) (0.0704) RJV HHI fragmentation −0.716*** −0.629*** (0.0872) (0.0795) RJV HHI fragmentation*Post leniency 0.296|$^{**}$| 0.276|$^{**}$| (0.128) (0.114) Total effect of leniency policy −0.0008*** 0.0002|$^{*}$| −0.0011*** −0.0003*** 0.0004|$^{**}$| −0.0004*** Wald test statistic 19.57 5.450 15.54 17.84 6.960 14.50 P-value 0.0001 0.0655 0.000422 0.0001 0.0308 0.0007 Observations 222,059 222,059 146,458 166,916 166,916 112,747 Notes. Robust standard errors clustered by firm are in parenthesis. An observation is a firm–RJV–year combination. All specifications include the usual controls plus industry dummies. Total effect is computed at the mean of the regressors. |$^{*}$| indicates significant at 10% level; |$^{**}$| at 5% level; and |$^{***}$| at 1% level. Open in new tab 7. Conclusion This paper empirically examines an important and relatively unexplored issue: Do RJVs serve as a collusive device? RJVs allow for easy communication among partners, and members are granted antitrust protection. It is possible that permitting firms to legally collude in R&D may facilitate illegal collusion in the final goods market. If this is the case, firms may undertake RJVs for anticompetitive reasons with possible negative social welfare repercussions. To separately identify the intention to collude from other (legal) reasons to form a RJV, we take advantage of a shift in antitrust policy which made product market collusion more difficult to sustain. We exploit the variation in RJV formation generated by a revision to the leniency policy that effects the collusive benefits of a RJV but not the research synergies associated with that venture. We examine RJVs in three industries with a history of antitrust litigation that are characterized by RJV membership involving product market rivals: telecommunications, computer manufacturing, and petroleum refining. We find that the leniency policy revision has a significant negative effect on the probability of joining a RJV in telecommunications and computer industries across market definitions. The results are robust to a variety of modifications and specifications. Our findings are consistent with collusive behavior on the part of telecommunications firms (particularly over the years of telecom regulation) and computer manufacturers. We also find support for collusive behavior among petroleum refiners. Our results indicate the revised leniency policy reduces the average probability that computer and semiconductor manufacturers join an RJV by |$41\%$| (range of |$21\%$|–|$90\%);$| with a reduction of |$34\%$| (range of |$20\%$|–|$94\%)$| among telecommunications firms, and among firms in petroleum refining the probability decreases by |$33\%$| (range of |$24\%$|–|$54\%).$| Furthermore, our findings show that the higher the market power of the RJV, the more collusive potential it has, which generates a differential effect of the leniency policy on the probability of joining a RJV. To the extent that antitrust authorities wish to detect and prohibit collusion brought about through RJV formation, our results suggest they should be more concerned when RJVs have a high joint market share relative to industry concentration. Notes The editor in charge of this paper was Imran Rasul. Acknowledgements This research has benefited from tremendous input from Eric Helland; comments at Carlos III Madrid, Cergy Pontoise, Claremont McKenna, DOJ Antitrust Division, EARIE meetings, Hebrew University, Laussane, Mannheim, Melbourne, Monash, UC Davis, Zurich, American Law and Economics meetings, CEPR Applied Micro meetings, IOS meetings, and SEA meetings; and discussions with John Asker, Linda Cohen, Jonah Gelbach, Jacob Goeree, Phil Haile, Hugo Hopenhayn, Patricia Langohr, Josh Rosett, Ralph Siebert, Janet Smith, Guofu Tan, and Otto Toivanen. I gratefully acknowledge support from the Financial Economics Institute and the Lowe Institute of Political Economy at Claremont McKenna College, the European Research Council Grant #725081 FORENSICS, and the Deutsche Forschungsgemeinschaft (DFG, German Research Foundation) through CRC TR 224 Project A02. I thank Katy Femia and Elissa Gysi for research assistance and Albert Link for providing the CORE data. I thank the editor and three referees for many valuable comments. The author is affiliated at CEPR. Footnotes 1. For an extensive discussion see Brodley (1990), Jorde and Teece (1990), and Shapiro and Willig (1990). 2. For example, see Bourreau, Sun, and Verboven (2018) and Genaokos et al. (2018) for studies of collusion in telecommunications; Asker, Collard-Wexler, and De Loecker (2019), Byrne and de Roos (2019), and Clark and Houde (2013) for studies on market power in the petroleum industry; and Zulehner (2003) for an examination of the semiconductor industry. 3. See Kobayashi (2001), Spratling (1999), and Verboven and van Dijk (2009). 4. Brenner (2009) examined the impact of the 1996 EU leniency policy on cartel deterrance and found no effect. However, the studies that examined later revisions (in 2002 and 2006) found the revised policy did impact cartel formation (De 2010; Zhou 2013). Marvao and Spagnolo (2018) provide a survey of the empirical literature. 5. Antitrust Division, US DOJ, Annual Report FY 2001. 6. A link to these videos can be found at https://www.justice.gov/atr/speech/caught-act-inside-international-cartel. 7. Figure 1 shows an increase in filing up until 1993. One of the reasons was a change in management structure in the 1980s, which saw RJV alliances as an acceptance of a firm’s technological limitations (Hemphill 1997). A survey of 4,182 technological alliances (compiled by the Maastricht Economic Research Institute of Innovation and Technology) found that the most commonly cited reasons for RJVs were to gain access to a market, to exploit complimentary technologies, to reduce the time taken for innovation, and R&D sharing (The Economist 1993:16). Second, up until 1993, federal antitrust agencies had not challenged any joint production venture that did not also involve joint marketing (H.R. No. 103-94:184). Finally, this system was further strengthened in the 1990s by a series of programs actively promoting government-industry-university partnerships and efforts to “channel” private sector R&D activity in technological areas with potentially widespread economic returns (Antitrust & Trade Regulation Report 1993, 688:1). 8. According to the NCRA, a RJV is “any group of activities, including attempting to make, making or performing a contract, by two or more persons for the purposes of (a) theoretical analysis, experimentation, or systematic study of phenomena or observable facts, (b) the development or testing of basic engineering techniques, (c) the extension of investigative finding or theory of a scientific or technical nature into practical application for experimental and demonstration purposes..., (d) the collection, exchange, and analysis of research information, or (e) any combination of the [above].” 9. If a behavior is per se illegal then authorities need only prove the behavior exists, there is no allowable defense for the accused parties. Under the rule of reason authorities are required to examine the inherent effect and the intent of the practice. 10. Prevailing defendents are entitled to recover costs and attorneys’ fees if an action is found to be “frivolous, unreasonable, without foundation, or in bad faith.” See 15 USC section 4304(a)(2)(2000). 11. Competitor Collaboration Guidelines, Section 4.3. “Absent extraordinary circumstances, the Agencies do not challenge a competitor collaboration on the basis of effects on competition in an innovation market where three or more independently controlled research efforts in addition to those of the collaboration possess the required specialized assets or characteristics and the incentive to engage in R&D that is a close substitute for the R&D activity of the collaboration. In determining whether independently controlled R&D efforts are close substitutes, the Agencies consider, among other things, the nature, scope, and magnitude of the R&D efforts; their access to financial support; their access to intellectual property, skilled personnel, or other specialized assets; their timing; and their ability, either acting alone or through others, to successfully commercialize innovations.” www.ftc.gov/os/2000/04/ftcdojguidelines.pdf. 12. For example, authorities will permit modifications to RJVs involving pharmaceutical firms engaged in cardiovascular research; those formed by the four US manufacturers of centrifugal pumps (used by electrical utilities) that focus on improving pump performance; or RJVs formed to conduct R&D relating to computer aided design and manufacturing. See US DOJ Business Review Letter to American Heart Association March 20, 1998; the Pump Research and Dev. Comm., 1985; and to the Computer Aided Mfg. Int’l Inc. 1985, respectively. 13. See Coordinated Proceedings in Petroleum Products Antitrust Litigation, 906 F2d 432 (9th Cir. 1990) and Petroleum Products Antitrust Litigation, 906 F.2d 432 (9th Cir. 1990). The firms were Texaco, Inc., Union Oil Co. of California, Atlantic Richfield Co., Exxon Corp., Mobil Oil Corp., and Shell Oil. 14. On January 9, 2009, the case was dismissed due to lack of subject matter jurisdiction. See Refined Petroleum Products Antitrust Litigation (MDL No. 1886) (2006). 15. The lawsuit hinges on a marketing deal that allowed former rivals to collude on prices starting in 1998, when Shell and Texaco formed Equilon Enterprises and Motiva Enterprises. Equilon and Motiva began operating when inflation-adjusted crude oil prices hit their lowest levels post-1930 yet wholesale prices were higher by 20–40 cents a gallon. Franchises typically sign long-term contracts with oil suppliers, making it difficult to switch to another brand or an independent supplier. 16. Texaco Inc. v. Dagher, 547 U.S. 1; 2006. 17. http://www.law360.com/articles/441336/dell-accuses-toshiba-sony-of-fixing-prices-for-disk-drives (Accessed September 2, 2016). 18. In 1984, AT&T relinquished its hold on the local market when the DOJ ordered AT&T to divest its local telephony business. These companies became the Regional Bell Operating companies or RBOCs. Local operators were not permitted to offer long distance services until the Telecommunications Act of 1996. 19. The decision to enter into an RJV may depend upon the decisions of rival firms (Greenlee and Cassiman 1999; and Yi and Shin 2000). We do not estimate a structural model of firms’ decisions because we would need to specify the game played among competing firms in R&D choices, RJV formation, and product market decisions. This game is best specified in a dynamic setting. Estimation would need to address the simultaneity of R&D decisions, RJV formation, and product market decisions, which would require assumptions on the nature of equilibrium and a means to choose among multiple equilibria. Second, addressing the nature of product competition would require estimates from competitive and collusive models of product market behavior. We could compare actual to predicted markups under both models (Nevo 2000), but this requires cost data (not easy to obtain and often proprietary). Finally, the telecom industry was regulated. So the model would have to address strategic behavior in a regulated industry. The model presented in this section captures the collusive intent of firms absent the additional structure and data requirements needed to estimate a structural model. 20. Hernan, Marin, and Siotis (2003) consider the decision to join a RJV in the European Union. They find that sectoral R&D intensity, industry concentration, firm size, and past RJV participation positively influence the probability of forming a RJV. 21. We thank an anonymous referee for this point. 22. Partial cartels have been observed in many industries. For example, a cartel in carbonless paper production had combined market shares of about |$85\%$| (Levenstein and Suslow 2006); a cartel among shipping firms in the North Atlantic constituted |$75\%$| of the market (Escrihuela-Villar 2003); and, famously, petroleum manufacturing firms in the United States and Russia are excluded from the OPEC cartel. There is a growing theoretical literature that examines partial cartels. For example, Bos and Harrington (2010) consider partial cartels among firms in dynamic differentiated products industries. They, and other papers in the theoretical literature, assume that a cartel member’s demand is proportional to the pre-cartel size of the firm. This allocation rule is motivated by cases as cited in Bos and Harrington (2010), these include the Norwegian cement cartel, and several German cartels during the early 1900s. 23. As the collusive value is increasing in the sum of the market shares of the colluding firms, it is also increasing in the sum of the squared market shares of the colluding firms. 24. Where the industry changes depending upon the relevant market we consider. We discuss this in more detail in Section 5. 25. Notice that we cannot use the measure of RJV market power to compare across industries. That is, holding fixed the participants and their market shares, the greater the HHI of the industry the lower is |$H_{{ijt}}$|⁠. 26. We also note that the data identify a structural break in RJV formation that occurred in 1993 for the telecommunications RJVs and in 1995 for computer and petroleum RJVs. We also find this in our estimation results which we discuss in more detail in Section 5. 27. We provide evidence for specific industries in the following section. In addition, across all industries in our data, the average number of RJVs firms join is 1.721 (with a standard deviation of 4.748) and the mean number of RJVs joined among joiners is 3.292 (with a standard deviation of 6.162). 28. Note some of these variables may be endogenous, but our primary focus is on the impact of RJV formation. For this we need to control for a number of variables, but we are not arguing that our estimates provide a causal effect of these variables on RJV formation. However, we conducted robustness checks without |$H_{{ijt}}$| as a regressor and found that both the signs and signficance of the leniency policy regressor was unchanged in our markets. The impact on the probability of joining an RJV was lower for petroleum refinining but not significantly different for the firms in telecommunications and computers. 29. The parameters of |$V_{{ijt}}$| are identified up to the factor |$\sigma _{n},$| hence we normalize |$\sigma _{n}=1.$| 30. See Maddala (1983, pp. 223–224). The model could be estimated in stages. First, consistent estimates of the predicted probabilities (⁠|$\widehat{P}_{{ijt}})$| come from a reduced form probit regression obtained by substituting equations (7) and (8) into (5). To control for the endogeneity of R&D, equations (7) and (8) are corrected by including control variables (constructed using the inverse Mill’s ratio and the predicted probit probabilities |$\widehat{P}_{{ijt}}).$| Least squares yields consistent estimates of the corrected R&D equations. The predicted values from the corrected R&D equations are used to construct the predicted difference in R&D intensity, |$(\widehat{rd}_{{ijt}}-\widehat{rd} _{\textit {it}})$|⁠, from joining a RJV for all firm–RJV combinations. The probit selection equation in (6) could be estimated after including the predicted R&D difference as a regressor, which Lee (1978) shows yields consistent estimates of the parameters. However, to obtain asymptotically efficient estimates all parameters of the model should be estimated simultaneously. 31. The revision appears to have been motivated by the desire to thwart international cartels. See www.usdoj.gov/atr/public/speeches/206611.htm. It is possible that firms may have anticipated the policy change. We conduct placebo tests in Section 6 in which we vary the date of the policy change. The results suggest that firms reacted to the actual revision dates. 32. We thank an anonymous referee for this idea. 33. We obtained patent data from the NBER U.S. Patent Citations Data File. These contain information on almost 3 million US patents starting in 1963 and programs that compute patent stock, which are matched to firm-identifiers in Compustat. 34. For both leniency policy revision dates (post-corporate and post-individual), we ran sets of regressions representing different market definitions: five with firms from all markets (with different fixed effects) and others that parallel the markets in Table 1. In all regressions, we control for firm assets, sales, free cash, industry classification fixed effects, RA fixed effects, and year fixed effects. Regression results are available upon request. 35. Link and Bauer (1989) document that cooperative research efforts were occuring informally before the NCRA was implemented in 1984. It is likely that RJV applications in 1985 may capture a portion of the pre-1985 stock. For this reason we include all RJVs starting in 1986. 36. See http://www.gpoaccess.gov/fr/index.html. 37. The Compustat data do not contain information on non-publically traded firms or non-profit firms. 38. Our results are robust to changes in our end date assumption. 39. For more on RJVs filed under NCRA see Link (1996), who provides an overview; Majewski and Williamson (2002), who examine contract details of NCRA applicants; and Berg, Duncan, and Friedman (1982). We also note that an RJV may span more than one industry if the member firms are engaged in R&D in more than one industry. 40. See Pittman and Werden (1990) for a discussion of the divergence between industry classifications and antitrust markets. 41. However, we do conduct robustness checks with the entire sample (see Section 6). 42. Prior to 2000, the sales data are from Gartner Dataquest Press Releases (www.gartner.com) and post-2000 data are from iSuppli Corporation (www.isuppli.com). Sales data are released in March. 43. The press releases report the sales for the top 20 firms (approximately 70% of total semiconductor sales) in all years except 1997, 1998, and 1999 when only the top ten firm sales are reported (approximately 50% of total semiconductor sales). 44. In the United States, if a market has an HHI of 25% or higher then it is considered to be highly concentrated. https://www.justice.gov/atr/horizontal-merger-guidelines-0. 45. See www.fcc.gov/Bureaus/Common_Carrier/Reports/. 46. In the United States, if a market has an HHI of 25% or higher then it is considered to be highly concentrated. https://www.justice.gov/atr/horizontal-merger-guidelines-0. 47. Unfortunately, due to the presence of the OPEC cartel, it is difficult to find accurate data on sales of worldwide petroleum producers. The capacity data that are available are not representative of sales due to the fact that firms often do not operate up to capacity. 48. Source: Energy Information Admin http://www.eia.doe.gov/. 49. Parameter estimates are available upon request. 50. We do not present the parameter estimates for the R&D equations due to space considerations. For most specifications, we find that the more RJVs a firm is a member of and the more capital constrained is a firm the lower is its R&D intensity. In many specifications, correcting for endogenous R&D is necessary (i.e., the parameter estimates for the inverse mills correction terms are significant). 51. Firms in the telecom research area span many industries so the regressions for all specifications include dummy varibles for three-digit NAICS. 52. Given the method used to construct our sample, we may be underestimating this probability. If no firm joins a RJV then we remove it from the choice set of all firms. Thus, if RJVs are systematically exiting the sample due to the leniency policy, we would underestimate the impact of the revision on RJV formation. 53. Recall, if all firms in industry |$k$| are in RJV |$j$| then the RJV Herfindahl would be the highest possible (⁠|$H_{{ijt}}=1)$| indicating the RJV has very high market power in that industry. If there were only a few large firms in industry |$k$| then the RJV would require fewer members to have substantial market power. 54. However, the conclusion could be to the contrary if due to higher profits the colluding firms have more means to innovate. Moreover, a firm may benefit more from its innovation when it is not under competitive pressure. 55. United States v. Columbia Pictures Indus 507 F. Supp. 412 n. 47 (S.D.N.Y 1980). 56. According to the US Merger guidelines, mergers are generally not challenged when the HHI is smaller than 1,000, when the HHI is between 1,000 and 1,800 and the merger will increase the HHI by less than 100 points or when the HHI is larger than 1,800 and the merger will increase the HHI by less than 50 points. All other mergers might be challenged. 57. Due to space constraints, we discuss some results for which we do not present parameter estimates. To avoid many repetitions we note here that details for all results are available upon request. 58. Due to the “incidental parameters problem,” a fixed effects probit regression will not give consistent estimates of the parameters. The logit does not suffer from this problem. See Greene (2000) for a discussion. 59. A couple of years generated a positive effect for some broadly defined market definitions via the RJV-market-power and placebo-leniency interaction parameter. 60. As our selection criteria is whether a firm joined a RJV, there may be concern about sample selection bias. However, selection bias is mitigated due to the panel aspect of our data. That is, since the data are a firm–RJV–year panel, a firm will potentially have a number of years in which it does not join a RJV, and a number of years during which it could potentially join (i.e., other members of its industry have joined) but it does not. Hence, this firm is included among joiners in some years and non-joiners in others. In many ways the estimation strategy is to estimate a panel of the probability of joining but to include those firms that never join to allow for systematic differences between the two groups. References Armoogum Khemla Davies S. W. Mariuzzo Franco ( 2017 ). “Cartel Enforcement and Deterrence Over the Life of a Competition Authority.” SSRN Working Paper No. 2993149 . OpenURL Placeholder Text WorldCat Asker John Collard-Wexler Allan De Loecker Jan ( 2019 ). “(Mis)Allocation, Market Power and Global Oil Extraction.” American Economic Review , 109 ( 4 ), 1568 – 1615 . Google Scholar Crossref Search ADS WorldCat Bageri Vasiliki Katsoulakos Yannis Spagnolo Giancarlo ( 2013 ). “The Distortive Effects of Antitrust Fines Based on Revenue.” Economic Journal , 123 , F545 – F557 . Google Scholar Crossref Search ADS WorldCat Baumol William ( 2001 ). “When is Interfirm Coordination Beneficial? The Case of Innovation.” International Journal of Industrial Organization , 19 , 727 – 737 . Google Scholar Crossref Search ADS WorldCat Berg Sanford Duncan Jerome Friedman Philip ( 1982 ). Joint Venture Strategies and Corporate Innovation . Olgeschlagen Gunn & Hain , Cambridge . Google Scholar Google Preview OpenURL Placeholder Text WorldCat COPAC Bigoni Maria Fridolfsson Sven-Olof Le Coq Chloé Spagnolo Giancarlo ( 2015 ). “Trust, Leniency, and Deterrence.” The Journal of Law, Economics, and Organization , 31 , 663 – 689 . Google Scholar Crossref Search ADS WorldCat Bos Iwan Harrington Joseph ( 2010 ). “Endogenous Cartel Formation with Heterogeneous Firms.” RAND Journal of Economics , 41 , 92 – 117 . Google Scholar Crossref Search ADS WorldCat Bos Iwan Davies Stephen Harrington Joseph Ormosi Peter ( 2018 ). “Does Enforcement Deter Cartels? A Tale of Two Tails.” International Journal of Industrial Organization , 59 , 372 – 405 . Google Scholar Crossref Search ADS WorldCat Bourreau Marc Sun Yutec Verboven Frank ( 2018 ). “Market Entry, Fighting Brands and Tacit Collusion: the Case of the French Mobile Telecommunications Market.” SSRN Working Paper . OpenURL Placeholder Text WorldCat Brenner Steffen ( 2009 ). “An Empirical Study of the European Corporate Leniency Program.” International Journal of Industrial Organization , 27 , 639 – 645 . Google Scholar Crossref Search ADS WorldCat Brodley Joseph ( 1990 ). “Antitrust Law and Innovation Cooperation.” Journal of Economic Perspectives , 4 , 97 – 112 . Google Scholar Crossref Search ADS WorldCat Byrne David de Roos Nicolas ( 2019 ). “Learning to Coordinate: A Study in Retail Gasoline.” American Economic Review , 109 ( 2 ), 591 – 619 . Google Scholar Crossref Search ADS WorldCat Cabral Luis ( 2000 ). “R&D Cooperation and Product Market Competition.” International Journal of Industrial Organization , 18 , 1033 – 1047 . Google Scholar Crossref Search ADS WorldCat Clark Robert Houde Jean-François ( 2013 ). “Collusion with Asymmetric Retailers: Evidence from a Gasoline Price-Fixing Case.” American Economic Journal: Microeconomics , 5 , 97 – 123 . Google Scholar OpenURL Placeholder Text WorldCat Cohen Wesley Levinthal Daniel ( 1989 ). “Innovation and Learning: The Two Faces of R&D.” Economic Journal , 99 , 569 – 596 . Google Scholar Crossref Search ADS WorldCat Compte Olivier Jenny Frédéric Rey Patrick ( 2002 ). “Capacity Constraints, Mergers and Collusion.” European Economic Review , 46 , 1 – 29 . Google Scholar Crossref Search ADS WorldCat Connor John ( 2003 ). “Private International Cartels: Effectiveness, Welfare, and Anticartel Enforcement.” Working Paper No. 03-12 , Dept. of Agricultural Economics, Purdue University . OpenURL Placeholder Text WorldCat Cooper Russell Ross Thomas ( 2009 ). “Sustaining Cooperation with Joint Ventures.” Journal of Law, Economics, and Organization , 25 , 31 – 54 . Google Scholar Crossref Search ADS WorldCat d’Aspremont Claude Jacquemin Alexis ( 1988 ). “Cooperative and Noncooperative R&D in Duopoly with Spillovers.” American Economic Review , 78 ( 5 ), 1133 – 1137 . Google Scholar OpenURL Placeholder Text WorldCat De Oindrila ( 2010 ). “Analysis of Cartel Duration: Evidence from EC Prosecuted Cartels.” International Journal of the Economics of Business , 17 , 33 – 65 . Google Scholar Crossref Search ADS WorldCat Department of Justice ( 1993 ). “Corporate Leniency Policy.” https://www.justice.gov/atr/file/810281/download (Accessed October 23, 2018) . OpenURL Placeholder Text WorldCat Department of Justice ( 2003 ). “The Modern Leniency Program After Ten Years: A Summary Overview of the Antitrust Division’s Criminal Enforcement Program.” https://www.justice.gov/atr/speech/modern-leniency-program-after-ten-years-summary-overview-antitrust-divisions-criminal (Accessed November 29, 2020) . OpenURL Placeholder Text WorldCat Duso Tomaso Röller Lars Hendrick Seldeslachts Jo ( 2014 ). “Collusion through Joint R&D: An Empirical Assessment.” Review of Economics and Statistics , 96 , 349 – 370 . Google Scholar Crossref Search ADS WorldCat Eizenberg Alon Assaf Kovo ( 2017 ). “Inferring Market Definitions and Competition Groups from Empirically-Estimated Demand Systems: A Practitioner’s Guide.” Working Paper , Hebrew University of Jerusalem . OpenURL Placeholder Text WorldCat Escrihuela-Villar Marc ( 2003 ). “Partial Cartels.” Doctoral Thesis , University of Alicante . OpenURL Placeholder Text WorldCat Ghosal Vivek Gallo Joseph ( 2001 ). “The Cyclical Behavior of the Department of Justice’s Antitrust Enforcement Activity.” International Journal of Industrial Organization , 19 , 27 – 54 . Google Scholar Crossref Search ADS WorldCat Goettler Ronald Gordon Brett ( 2011 ). “Does AMD Spur Intel to Innovate More?” Journal of Political Economy , 119 , 1141 – 1200 . Google Scholar Crossref Search ADS WorldCat Greene William ( 2000 ). Econometric Analysis , 4th ed. Prentice Hall, Engelwood Cliffs . Google Scholar Google Preview OpenURL Placeholder Text WorldCat COPAC Greenlee Patrick Cassiman Bruno ( 1999 ). “Product Market Objectives and the Formation of Research Joint Ventures.” Managerial and Decision Economics , 20 , 115 – 130 . Google Scholar Crossref Search ADS WorldCat Gugler Klaus Siebert Robert ( 2007 ). “Market Power versus Efficiency Effects of Mergers and Research Joint Ventures: Evidence from the Semiconductor Industry.” Review of Economics and Statistics , 89 , 645 – 659 . Google Scholar Crossref Search ADS WorldCat Hagedoorn John Link Albert Vonortas Nicoloas ( 2000 ). “Research Partnerships.” Research Policy , 29 , 567 – 586 . Google Scholar Crossref Search ADS WorldCat Hemphill Thomas ( 1997 ). “U.S. Technology Policy, Intraindustry Joint Ventures, and the National Cooperative Research and Production Act of 1993.” Business Economics , 32 , 48 – 54 . Google Scholar OpenURL Placeholder Text WorldCat Hernan Roberto Marin Pedro Siotis Georges ( 2003 ). “An Empirical Analysis of the Determinants of Research Joint Venture Formation.” Journal of Industrial Economics , 51 , 75 – 89 . Google Scholar Crossref Search ADS WorldCat Hyytinen Ari Steen Frode Toivanen Otto ( 2018 ). “Cartels Uncovered.” American Economic Journal: Microeconomics , 10 , 190 – 222 . Google Scholar OpenURL Placeholder Text WorldCat Hyytinen Ari Steen Frode Toivanen Otto ( 2019 ). “Anatomy of Cartel Contracts.” Economic Journal , 129 , 2155 – 2191 . Google Scholar Crossref Search ADS WorldCat Irwin Douglas Klenow Peter ( 1994 ). “Learning-by-Doing Spillovers in the Semiconductor Industry.” Journal of Political Economy , 102 , 1200 – 1227 . Google Scholar Crossref Search ADS WorldCat Irwin Douglas Klenow Peter ( 1996 ). “High-tech R&D Subsidies: Estimating the Effects of Sematech.” Journal of International Economics , 40 , 323 – 344 . Google Scholar Crossref Search ADS WorldCat Jorde Thomas Teece David ( 1990 ). “Innovation and Cooperation: Implications for Competition and Antitrust.” Journal of Economic Perspectives , 4 , 75 – 96 . Google Scholar Crossref Search ADS WorldCat Kobayashi Bruce ( 2001 ). “Antitrust, Agency, and Amnesty: An Economic Analysis of the Criminal Enforcement of the Antitrust Laws Against Corporations.” George Washington Law Review 69 , 715 – 744 . Google Scholar OpenURL Placeholder Text WorldCat Kogut Bruce ( 1991 ). “Joint Ventures and the Option to Expand and Acquire.” Management Science , 37 , 19 – 33 . Google Scholar Crossref Search ADS WorldCat Lambertini Luca Poddar Sougata Sasaki Dan ( 2002 ). “Research Joint Ventures, Product Differentiation, and Price Collusion.” International Journal of Industrial Organization , 20 , 829 – 854 . Google Scholar Crossref Search ADS WorldCat Lee Lung-Fei ( 1978 ). “Unionism and Wage Rates: A Simultaneous Equations Model with Qualitative and Limited Dependent Variables.” International Economic Review , 19 , 415 – 433 . Google Scholar Crossref Search ADS WorldCat Leibenstein Harvey ( 1966 ). “Allocative Efficiency vs. X-Efficiency.” American Economic Review , 56 ( 3 ), 392 – 415 . Google Scholar OpenURL Placeholder Text WorldCat Levinstein Margaret Suslow Valerie ( 2006 ). “What Determines Cartel Success?” Journal of Economic Literature , 44 , 43 – 95 . Google Scholar Crossref Search ADS WorldCat Levy Nadav ( 2012 ). “Technology Sharing and Tacit Collusion.” International Journal of Industrial Organization , 30 , 204 – 216 . Google Scholar Crossref Search ADS WorldCat Link Albert ( 1996 ). “Research Joint Ventures: Patterns from Federal Register Filings.” Review of Industrial Organization , 11 , 617 – 628 . Google Scholar Crossref Search ADS WorldCat Link Albert Bauer Laura ( 1989 ). Cooperative Research in US Manufacturing . Lexington Books . Google Scholar Google Preview OpenURL Placeholder Text WorldCat COPAC Lundqvist Björn ( 2015 ). Joint Research and Development under US Antitrust and EU Competition Law . Edward Elgar Publishers , Cheltenham, UK . Google Scholar Crossref Search ADS Google Preview WorldCat COPAC MacAvoy Paul ( 1995 ). “Tacit Collusion under Regulation in the Pricing of Interstate Long-Distance Telephone Services.” Journal of Economics and Management Strategy , 4 , 147 – 185 . Google Scholar Crossref Search ADS WorldCat Maddala Gangadharrao Soundalyarao ( 1983 ). Limited Dependent and Qualitatitve Variables in Econometrics . Cambridge University Press , Cambridge . Google Scholar Crossref Search ADS Google Preview WorldCat COPAC Majewski Suzanne Williamson Dean ( 2002 ). “Endogenous Spillovers, Strategic Blocking, and the Design of Contracts in Collaborative R&D: Evidence from the NCRA Filings of R&D Joint Ventures.” Economic Analysis Group Working Paper , Antitrust Division, US Department of Justice . OpenURL Placeholder Text WorldCat Martin Steve ( 1995 ). “R&D Joint Ventures and Tacit Product Market Collusion.” European Journal of Political Economy , 11 , 733 – 741 . Google Scholar Crossref Search ADS WorldCat Marvão Catarina Spagnolo Giancarlo ( 2018 ). “What Do We Know about the Effectiveness of Leniency Policies? A Survey of the Empirical and Experimental Evidence.” In Handbook of Game Theory and Industrial Organization , Volume II , edited by Corchon Luis C. Marini Marco A.. Edward Elgar Publishing . Google Scholar Crossref Search ADS Google Preview WorldCat COPAC Miller Nathan ( 2009 ). “Strategic Leniency and Cartel Enforcement.” American Economic Review , 99 ( 3 ), 750 – 768 . Google Scholar Crossref Search ADS WorldCat Miyagiwa Kaz ( 2009 ). “Collusion and Research Joint Ventures.” Journal of Industrial Economics , 57 , 768 – 784 . Google Scholar Crossref Search ADS WorldCat Nevo Aviv ( 2000 ). “Mergers with Differentiated Products: The Case of the Ready-to-Eat Cereal Industry.” Rand Journal of Economics , 31 , 395 – 421 . Google Scholar Crossref Search ADS WorldCat Oum Hoon Tae Park Jong-Hun ( 1997 ). “Airline Alliances: Current Status, Policy Issues, and Future Directions.” Journal of Air Transport Management , 3 , 133 – 144 . Google Scholar Crossref Search ADS WorldCat Petit Maria Towlinski Boleslaw ( 1999 ). “R&D Cooperation or Competition.” European Economic Review , 43 , 185 – 208 . Google Scholar Crossref Search ADS WorldCat Pittman Russell Werden Gregory ( 1990 ). “The Divergence of SIC Industries from Antitrust Markets: Indications from Justice Department Merger Cases.” Economics Letters , 33 , 283 – 286 . Google Scholar Crossref Search ADS WorldCat Roller Lars-Hendrik Siebert Ralph Tombak Mihkel ( 2007 ). “Why Firms Form (or Do Not Form) RJVs.” Economic Journal , 117 , 1122 – 1144 . Google Scholar Crossref Search ADS WorldCat Scott John ( 1988 ). “Diversification versus Cooperation in R&D Investment.” Managerial Decision Economics , 9 , 173 – 186 . Google Scholar Crossref Search ADS WorldCat Shapiro Carl Willig Robert ( 1990 ). “On the Antitrust Treatment of Production Joint Ventures.” Journal of Economic Perspectives , 4 , 113 – 130 . Google Scholar Crossref Search ADS WorldCat Smith Adam ( 1776 ). An Inquiry into the Nature and Causes of the Wealth of Nations . W. Strahan and T. Cadell Publishers , London . Google Scholar Google Preview OpenURL Placeholder Text WorldCat COPAC Spagnolo Giancarlo ( 2000 ). “Self Defeating Antitrust Laws: How Leniency Programs Solve Bertrand’s Paradox and Enforce Collusion in Auctions.” Working paper . OpenURL Placeholder Text WorldCat Spagnolo Giancarlo ( 2004 ). “Divide et Impera: Optimal Leniency Programs.” CEPR Discussion Paper No. 4840 . OpenURL Placeholder Text WorldCat Spratling Gary ( 1999 ). “Negotiating the Waters of International Cartel Prosecutions.” Address to the Thirteenth Annual National Institute on White Collar Crime. https://www.justice.gov/atr/speech/negotiating-waters-international-cartel-prosecutions-antitrust-division-policies-relating (Accessed October 16, 2021) . OpenURL Placeholder Text WorldCat Suetens Sigrid ( 2008 ). “Does R&D Cooperation Facilitate Price Collusion? An Experiment.” Journal of Economic Behavior and Organization , 66 , 822 – 836 . Google Scholar Crossref Search ADS WorldCat Verboven Frank van Dijk Theon ( 2009 ). “Cartel Damages Claims and the Passing-on Defense.” Journal of Industrial Economics , 57 , 457 – 491 . Google Scholar Crossref Search ADS WorldCat Yi Sang-Seung Shin Hyukseung ( 2000 ). “Endogenous Formation of Research Coalitions with Spillovers.” International Journal of Industrial Organization , 18 , 229 – 256 . Google Scholar Crossref Search ADS WorldCat Zhou Jun ( 2013 ). Evaluating Leniency with Missing Information on Undetected Cartels: Exploring Time-Varying Policy Impacts on Cartel Duration . SSRN Working Paper No. 1798424 . OpenURL Placeholder Text WorldCat Zulehner Christine ( 2003 ). “Testing Dynamic Oligopolistic Interaction: Evidence from the Semiconductor Industry.” The International Journal of Industrial Organization , 21 , 1527 – 1556 . Google Scholar Crossref Search ADS WorldCat © The Author(s) 2021. Published by Oxford University Press on behalf of European Economic Association. This is an Open Access article distributed under the terms of the Creative Commons Attribution Non-Commercial-NoDerivs licence (https://creativecommons.org/licenses/by-nc-nd/4.0/), which permits non-commercial reproduction and distribution of the work, in any medium, provided the original work is not altered or transformed in any way, and that the work is properly cited. For commercial re-use, please contact journals.permissions@oup.com © The Author(s) 2021. Published by Oxford University Press on behalf of European Economic Association.
TI - Do Research Joint Ventures Serve a Collusive Function?
JF - Journal of the European Economic Association
DO - 10.1093/jeea/jvab041
DA - 2022-02-16
UR - https://www.deepdyve.com/lp/oxford-university-press/do-research-joint-ventures-serve-a-collusive-function-0xCnNoKl5h
SP - 430
EP - 475
VL - 20
IS - 1
DP - DeepDyve
ER -