Communication in vertical markets: Experimental evidence

https://doi.org/10.1016/j.ijindorg.2016.10.002Get rights and content

Highlights

  • Study experimental vertical markets with upstream monopolist and competing downstream firms.

  • In basic experimental condition with no communication, commitment problem prevents industry from achieving monopoly outcome; allowing public communication among all firms results in fairly complete monopolization.

  • Allowing private communication between upstream and each downstream firm moves industry halfway toward monopoly outcome; allowing public communication among all firms results in fairly complete monopolization.

  • Communication also shifts the distribution of rents between upstream and downstream in a pattern well explained by Nash-in-Nash bargaining.

  • Using third-party coders, unsupervised text mining, among other approaches, we uncover features of the rich chat data that are correlated with market outcomes.

Abstract

An upstream monopolist supplying competing downstream firms may fail to monopolize the market because it is unable to commit not to behave opportunistically. We build on previous experimental studies of this well-known commitment problem by introducing communication. Allowing the upstream firm to chat privately with each downstream firm reduces total offered quantity from near the Cournot level (observed in the absence of communication) halfway toward the monopoly level. Allowing all firms to chat together openly results in complete monopolization. Downstream firms obtain such a bargaining advantage from open communication that all of the gains from monopolizing the market accrue to them. A simple structural model of Nash-in-Nash bargaining fits the pattern of shifting surpluses well. Using third-party coders, unsupervised text mining, among other approaches, we uncover features of the rich chat data that are correlated with market outcomes. We conclude with a discussion of the antitrust implications of open communication in vertical markets.

Introduction

Whether vertical mergers can have anticompetitive effects remains a central question in the largest antitrust cases. For example, in January 2011, the U.S. Department of Justice applied the “most intense scrutiny ever for a planned media merger” before approving the takeover of NBC Universal (an upstream content provider) by Comcast (a downstream cable distributor) subject to a list of conditions (Arango and Stelter, 2011). In April 2015, the European Competition Commission charged Google with the violation of favoring its affiliates over competitors in search displays (Kanter and Scott, 2015).

An influential strand of the theoretical literature (summarized in Rey and Tirole, 2007) connects the anticompetitive effects of vertical restraints to their ability to solve a commitment problem. An upstream monopolist serving downstream competitors might wish to offer contracts restricting output to the joint-profit maximum. It may fail to do so, however, because it has an incentive to behave opportunistically, offering one of the downstream firms a contract increasing their bilateral profits at the expense of all other downstream firms (the same logic extending to the bilateral contract with each downstream firm). In Hart and Tirole (1990), a vertical merger helps to solve this commitment problem by removing its incentive to behave opportunistically in a way that would harm the downstream unit with which it shares profits. While the upstream firm benefits from solving the commitment problem, overall the vertical merger has an anticompetitive effect on the market because prices rise and output falls. Similar anticompetitive effects can arise with vertical restraints aside from mergers including resale price maintenance (O’Brien, Shaffer, 1992, Rey, Vergé, 2004) and non-discrimination clauses (McAfee and Schwartz, 1994).

The commitment problem is a somewhat delicate theoretical proposition. Depending on downstream firms’ beliefs after receiving a deviating secret contract offer—not pinned down in a perfect Bayesian equilibrium—there can be multiple equilibria, with the commitment effect arising in some and not in others (McAfee, Schwartz, 1994, Rey, Vergé, 2004). With symmetric beliefs, downstream firms reject deviating contracts generating negative profits for rivals because they infer that rivals received the same deviating contract. In this way, symmetric beliefs afford the upstream firm the ability to commit to monopolizing the market. With passive beliefs, on the other hand, deviation does not change downstream firms beliefs, increasing their willingness to accept deviating contracts, impairing the upstream firm’s commitment power.

In the absence of a widely accepted refinement of perfect Bayesian equilibrium providing a firm theoretical foundation for selecting one or another equilibrium in this context, Martin et al. (2001) turned to experiments to gauge the significance of the commitment problem. In their baseline treatment in which an upstream monopolist makes secret offers of nonlinear tariffs to two downstream firms, labeled SECRAN, they found that markets were rarely monopolized; industry profits averaged only two thirds of the joint maximum. By contrast, markets were regularly monopolized when either the upstream monopoly was vertically integrated with a downstream firm or when contracts were public. The experiments thus support the view that the commitment problem is genuine.

In this paper, we return to an experimental study of vertical markets with a new focus—on whether allowing firms to communicate can help them solve the commitment problem without resorting to vertical restraints. For the sake of comparison, we start with the same SECRAN treatment as Martin et al. (2001). In addition to this baseline treatment without communication, we run a series of three treatments in which players can communicate whatever messages they want via a messenger-like tool. The communication treatments involve different levels of openness. One allows the upstream firm to engage in private two-way chat with each downstream firm. Another allows all three firms to engage in completely open (three-way) chat. A third is a hybrid of the other two, allowing players the option of using either or both of two- or three-way communication.

Communication is cheap talk in our experiments, so standard results (Crawford and Sobel, 1982) leave open the possibility that adding this form of communication may have no effect on equilibrium. Yet we have a number of good reasons to believe communication might have real effects in our experiments. First, the vertical contracting game involves considerable strategic uncertainty. A downstream firm has to form an out-of-equilibrium belief and other firms have to conjecture what this belief is (or what the distribution of beliefs are in the case of heterogeneous beliefs). Communication could resolve some of this strategic uncertainty. Second, communication could help solve the commitment problem by allowing the upstream firm to make promises. Promises about rival contracts are not legally enforceable in our experiments but could still afford some commitment power if making a bald-faced lie involves a substantial psychological cost. Third, communication has been shown in previous experiments to reduce bargaining frictions (Roth, 1995). On the other hand, communication could conceivably work in the opposite direction, impairing commitment. A conspiracy between the upstream and a downstream firm to deviate to a contract increasing their bilateral profits at the expense of the downstream rival would be easier to hatch if they could communicate privately. Of course, open communication precludes conspiracy, so open communication should either aid commitment or at worst have no effect. When firms are given the option of using either private or open communication, whether or not they are tempted to conspire, undermining commitment, is an interesting empirical question, which can be addressed by the hybrid treatment.

Along with the theoretical motives we just described for studying the effects of communication, we also have practical policy motives. Communication between vertically related firms is presumably the rule rather than the exception in the field,1 the lab adds an important practical element to existing experiments. While a conversation between an upstream and a downstream firm would not violate antitrust law, communication in an open forum involving horizontally along with vertically related firms might raise antitrust concerns. Whether such communication has the potential to restrain competition has so far not been studied.

Our experimental results reveal a remarkably consistent pattern: increasing the openness of communication has a monotonic effect across virtually every market outcome and treatment we study. In the treatment without communication, the same severe commitment problem observed in Martin et al. (2001) occurs: aggregate offered quantity is again much closer to the Cournot than the monopoly level. Two-way communication mitigates but does not solve the commitment problem, cutting the distance between aggregate offered quantity and the monopoly quantity about in half. Three-way communication cuts the remaining distance again in half, resulting in nearly complete monopolization of the market, particularly in the late rounds of play. Results for the hybrid treatment are between the other two, somewhat closer to the treatment with open communication. Further, we find that more open communication leads to more fluid bargaining, captured by an increasing rate of contract acceptance. The increase in acceptance rate, due in part to increasing confidence in the upstream firm’s commitment to monopolize the market, is also due in part to a reduction in the upstream firm’s tariff demands. Overall, the increase in acceptance rates leaves upstream profits essentially unchanged; the increase in industry profit accrues almost entirely to downstream firms.

That different communication treatments led to dramatically different divisions of surplus between upstream and downstream firms initially surprised us as we had not designed the treatments to look for such effects. In Section 5, we propose a simple bargaining model providing a straightforward explanation. In the absence of communication, the upstream firm makes take-it-or-leave-it offers; opening a communication channel affords participating subjects an opportunity to bargain. We assume bargaining outcomes are given by the widely used “Nash-in-Nash” solution concept proposed by Horn and Wolinsky (1988), recently given non-cooperative foundations by Collard-Wexler et al. (2016). According to this solution concept, each bargain maximizes the Nash product assuming that other bargains occurring simultaneously are efficiently consummated. Our bargaining model delivers the same pattern of surplus division observed in the experiments: opening a two-way communication channel in the model causes the upstream firm to lose bargaining power and moving to three-way communication reduces upstream surplus yet further.

Section 6 delves into the content of communication to uncover correlations between content features and market outcomes. To deal with the difficulty in quantifying the rich content data, we take several analytical approaches: counting messages, employing third-party coders, and using text-mining methods to extract keywords. The communication stage appears to function like a bargaining process, with discussions successfully converging to a contract that is the one that ends up being offered. When the upstream firm is successful at committing to the monopoly outcome, his or her messages tend to mention deals given to all both downstream firms and market prices. Commitment sometimes breaks down when a subject tries to strike an exclusive deal to sell the entire industry quantity, inevitably leading to oversupply as exclusion proves unenforceable.

From a policy perspective, our results imply that some forms of communication can effectively function as an anticompetitive vertical restraint. In particular, allowing an upstream firm to discuss contracts with several downstream firms in a “smoke-filled room” (or simply to exchange public pronouncements) has the potential to substantially restrict output. On the other hand, if firms already have such forums for open communication, vertical mergers and restraints themselves may not raise further antitrust concerns.

Regarding its relationship to the literature, our paper is the first experimental study of communication in a vertically related market. Our paper is closest to the one on which we build, Martin et al. (2001), which provides an experimental test of the theories of anticompetitive vertical restraints (vertical mergers, public contracts) put forth by the papers mentioned earlier (Hart, Tirole, 1990, O’Brien, Shaffer, 1992, McAfee, Schwartz, 1994, Rey, Vergé, 2004, Rey, Tirole, 2007); (see Avenel, 2012 and Caprice and Rey, 2015 for more recent developments). Other experiments in vertically related markets include Mason and Phillips' (2000) study of equilibrium when the upstream input is demanded by a Cournot duopoly in one market and perfectly competitive firms in another. Durham (2000) and Badasyan et al. (2009) analyze whether vertical merger mitigates the double-marginalization problem. Normann (2011) investigates whether vertical merger has an anticompetitive “raising rivals’ cost” effect in a bilateral duopoly. None of these papers studies communication, the focus of the present paper.

Also related is the experimental literature on exclusive dealing (Landeo, Spier, 2009, Smith, 2011, Boone, Müller, Suetens, 2014). As in our setting, the vertical contract exerts an externality on other downstream firms. The nature of the externality is different: rather than secretly oversupplying a rival, an initial exclusive contract diverts demand that would otherwise prompt a more efficient upstream firm to enter, which then would supply other downstream firms at lower prices. Landeo and Spier (2009) and Smith (2011) show that communication between downstream firms reduces entry-deterring exclusion.

Our paper contributes to a large literature on cheap talk in experimental games. Theory suggests that potential gains from cheap talk are greatest in games of common rather than conflicting interests (Farrell and Rabin, 1996). Consistent with theory, experiments find large gains from cheap talk in coordination games (see Crawford, 1998 for a survey).2 However, cheap talk also increases the rate of cooperation in dilemma games (Dawes, McTavish, Shaklee, 1977, Isaac, Ramey, Williams, 1984, Balliet, 2010) in which neoclassical theory would suggest agreements to cooperate should be worthless. Our result that communication aids monopolization has a similar flavor, although decision making is more complex in our setting: final output is the result of a negotiation between upstream and downstream firms rather than being one firm’s unilateral choice.3,4

Within the literature on cheap talk in experimental games, ours is closest to studies of the effect of cheap talk on bargaining. Adding a round of face-to-face communication before offers are made results in near perfect rates of agreement (Roth, 1995). Typed messages—the sort of communication also used in our experiments—does not improve efficiency as much but still improves upon no communication (Brosig, Ockenfels, Weimann, 2003, Andersson, Galizzi, Hoppe, Kranz, van der Wiel, Wengström, 2010, Zultan, 2012). Ours is the first to study how cheap talk between vertically related players affects bargaining with externalities. In this setting, the openness of communication becomes an important treatment variable. We find that private communication improves efficiency somewhat and open communication still more, reaching 92% agreement rates.

Section snippets

Market model

Consider a simplified version of the model due to Rey and Tirole (2007).5

Experimental design

We build on the experimental design of Martin et al. (2001). We will maintain their baseline treatment—which they called SECRAN because it involves secret contracts with randomly re-matched players—as our baseline treatment with no communication here. We will then introduce treatments allowing for different forms of communication.

The market, shown in Fig. 1, involves three subjects, one playing the role of the upstream firm (called a producer in the experiment) and two playing the role of

Results

To streamline the discussion of our results, we will confine the initial discussion to the distinct treatments No Chat, Two Chat, and Three Chat. Once the relationship between Two Chat and Three Chat is understood, we can study which one the hybrid treatment Choose Chat is closer to.

The top part of Table 1 can be interpreted as summary statistics for the main experimental variables. It regresses these variables (X, Ti, ai, ...) on an exhaustive set of treatment indicators, suppressing the

Rationalizing effects on surplus division

We designed our experiments to test the hypothesis that communication can help vertically related firms monopolize a market by solving a commitment problem. The results, as seen, bear this hypothesis out. We found another set of results for which we did not have a priori hypotheses—results related to the division of surplus between upstream and downstream firms varied across communication treatments—which were strong, systematic and beg explanation. It is worth recapitulating these results.

Analysis of chat content

In this section we draw further insights about the effect of communication by analyzing the content of the chat itself. The rich content data does not lend itself to easy quantification (Kimbrough et al., 2008), so in this section we take a series of approaches to do so: counting messages, employing third-party coders, and mining the text for keywords. Unlike the results reported to this point, the results in this section should be interpreted as associations, not causal relationships.

Conclusion

In this paper, we introduce communication to a strategically complex vertical market. One upstream and two downstream firms can jointly earn monopoly rents but they may well fail to do so due to a commitment problem (Hart, Tirole, 1990, Rey, Tirole, 2007). The relevance of this commitment problem in turn depends on technical modeling assumptions: the (possibly heterogeneous) beliefs players maintain may suggest different equilibria in which the market may or may not be monopolized. In addition

Acknowledgments

The authors thank John de Figueiredo, Maura Doyle, Gautam Gowrisankaran, Ali Hortaçsu, Robin Lee, Francisco Martinez, Paul Novosad, Patrick Rey and conference participants at CISS Turunç, DFG-ANR Rennes, EARIE Milan, ESA Jerusalem, IIOC Boston, LEW London, VfS Berlin (IO Committee) and VfS Münster for helpful discussions. We are grateful to the editor and referees for extensive, insightful comments, which substantially improved the paper. We are grateful to Deutsche Forschungsgemeinschaft

References (52)

  • O. Andersson et al.

    Do antitrust laws facilitate collusion? Experimental evidence on costly communication in duopolies

    Scandinavian Journal of Economics

    (2007)
  • T. Arango et al.

    Comcast receives approval for NBC universal merger

    New York Times

    (2011)
  • E. Avenel

    Upstream capacity constraint and the preservation of monopoly power in private bilateral contracting

    Journal of Industrial Economics

    (2012)
  • N. Badasyan et al.

    Vertical integration of successive monopolists: a classroom experiment

    Perspectives on Economic Education Research

    (2009)
  • D. Balliet

    Communication and cooperation in social dilemmas: a meta-analytic review

    Journal of Conflict Resolution

    (2010)
  • M. Bigoni et al.

    Fines, leniency and rewards in antitrust

    RAND Journal of Economics

    (2012)
  • J. Boone et al.

    Naked exclusion in the lab: the case of sequential contracting

    Journal of Industrial Economics

    (2014)
  • J. Brosig et al.

    The effect of communication media on cooperation

    German Economic Review

    (2003)
  • S. Caprice et al.

    Buyer power from joint listing decisions

    Economic Journal

    (2015)
  • G. Charness et al.

    Promises and partnership

    Econometrica

    (2006)
  • A. Collard-Wexler et al.

    ‘Nash-in-Nash’ Bargaining: A Microfoundation for Applied Work

    (2016)
  • D.J. Cooper et al.

    Communication, renegotiation, and the scope for collusion

    American Economic Journal: Microeconomics

    (2009)
  • V. Crawford

    A survey of experiments on communication via cheap talk

    Journal of Economic Theory

    (1998)
  • R.M. Dawes et al.

    Behavior, communication, and assumptions about other people’s behavior in a commons dilemma situation

    Journal of Personality and Social Psychology

    (1977)
  • Y. Durham

    An experimental examination of double marginalization and vertical relationships

    Journal of Economic Behavior and Organization

    (2000)
  • J. Farrell et al.

    Cheap talk

    Journal of Economic Perspectives

    (1996)
  • Cited by (17)

    • Let's lock them in: Collusion under consumer switching costs

      2023, International Journal of Industrial Organization
    • How communication makes the difference between a cartel and tacit collusion: A machine learning approach

      2023, European Economic Review
      Citation Excerpt :

      To the best of our knowledge, ours is the first study using LDA to understand how communication affects behavior in experimental markets.6 The relative rank differential statistic due to Huerta (2008), which we use to analyze the communication content in different market settings, is also employed in Moellers et al. (2017), Odenkirchen (2018), and Fourberg (2018). The finding that there is a connection between communication and price levels is consistent with previous studies showing that firms jointly set higher prices in treatments with unrestricted communication than in treatments without communication (see Isaac et al., 1984; Davis and Holt, 1998; Apesteguia et al., 2007; Cooper and Kühn, 2014; Dijkstra et al., 2021).7

    • Is natural language processing the cheap charlie of analyzing cheap talk? A horse race between classifiers on experimental communication data

      2022, Journal of Behavioral and Experimental Economics
      Citation Excerpt :

      On the other hand, their approach might not be as easy to replicate nor as efficient when it comes to larger datasets. This might be why many experimenters use fewer evaluators than Houser and Xiao (2011), as for example Ismayilov and Potters (2016) and Moellers et al. (2017). Houser and Xiao (2011) recruited 25 evaluators per treatment.

    • Collusion and bargaining in asymmetric Cournot duopoly—An experiment

      2019, European Economic Review
      Citation Excerpt :

      Moreover, with earned roles and when firms can talk, they often collude by producing equal amounts—a strategy unknown in the existing analyses of asymmetric Cournot oligopoly. We further employ coding analysis (Houser and Xiao, 2011) to investigate the nature of the agreements reached in the treatments with communication and text-mining analysis (Moellers et al., 2017) to identify the language suitable for successful collusion. For collusion (tacit or explicit) to be a subgame-perfect Nash equilibrium, we need a repeated game.

    • The value of entrant manufacturers: A study of competition and risk for donor-funded procurement of essential medicines

      2019, European Journal of Operational Research
      Citation Excerpt :

      Horn and Wolinsky (1988) study a setting that is closely related to ours, where one firm engages multiple other firms in bilateral negotiations. The authors introduce, as one of the first, a bilateral Nash bargaining model (also referred to as Nash-in-Nash (Collard-Wexler, Gowrisankaran, & Lee, forthcoming in 2017; Moellers, Normann, & Snyder, 2017) or Nash-Nash bargaining (Feng & Lu, 2013)), where each negotiation between two firms is modeled by a Nash bargaining game, and disagreement points describe individual utilities in the case of disagreement. They argue that this approach appropriately describes a negotiation’s outcome with alternating offers which can oftentimes be observed in real-world settings (Horn and Wolinsky, 1988, p. 411).

    View all citing articles on Scopus
    View full text