Journal of Economic Theory 132 (2007) 274– 290www.elsevier.com/locate/jetThe effects of costless pre-play communication:Experimental evidence from games withPareto-ranked equilibriaAndreas Blumea, Andreas Ortmannb,∗aEconomics Department, University of Pittsburgh, Pittsburgh, PA 15260, USAbCenter for Economic Research and Graduate Education, Charles University, Economics Institute,Academy of Sciences of the Czech Republic, PO Box 882, Politickych veznu, 7, CZ 11121 Prague,Czech RepublicReceived 16 July 2000; final version received 3 March 2005Available online 13 June 2005AbstractCheap talk is shown to facilitate coordination on the unique efficient equilibrium in experimentalorder-statistic games. This result is roughly consistent with theoretical predictions according to whichcheap talk promotes efficient Nash play. The evidence concerning the mechanisms that theory appealsto is mixed: Frequent agreement of messages and actions is consistent with messages being viewedas self-committing. Risk in the underlying game and the absence of self-signaling messages mayexplain why message profiles are not unanimous. Time-varying message profiles can be interpretedas evidence for players trying to negotiate equilibria and/or trying to rely on secret handshakes.© 2005 Elsevier Inc. All rights reserved.JEL classification: C72; C92Keywords: Coordination games; Order-statistic games; Cheap talk; Experiments1. IntroductionWe experimentally investigate the effects of costless pre-play communication, or cheaptalk, in symmetric coordination games of the stag hunt variety [11–15].Van Huyck, Battalio,and Beil (VHBB) [33,34] demonstrate that in so-called Minimum and Median games with∗Corresponding author. Fax: +420 224227143.E-mail address: [email protected] (A. Ortmann).0022-0531/$ - see front matter © 2005 Elsevier Inc. All rights reserved.doi:10.1016/j.jet.2005.03.001A. Blume, A. Ortmann / Journal of Economic Theory 132 (2007) 274– 290 275multiple Pareto-ranked equilibria, the Pareto-efficient equilibria need not be selected, andin fact typically are not selected [7, Chapter 7.4]. Taking as the point of departure a Mini-mum game explored in VHBB [33], Berninghaus and Ehrhart [3] demonstrate the crucialimportance of the frequency of play for these results. Building on a Median game designin VHBB [34], Van Huyck et al. [36] show that a dramatically refined action space allowscohorts of size 7 to creep toward efficiency in some cases. These two papers suggest thatthe consequence of the reduced opportunity cost of experimentation that is the by-productof the increased number of plays in the first case and the refined action space in the secondcase can facilitate, but not guarantee, “tacit coordination”.VHBB [35] show that adding a pre-play auction each round enables experimental partic-ipants to coordinate on the Pareto-efficient equilibrium. Crawford and Broseta [16] providea model of the efficiency-enhancing effects of the pre-play auctions in VHBB [35]. Theypoint out that these auctions are a costly form of pre-play communication, or signaling,that draws on the power of forward induction. Building also on the Median game results inVHBB [34], Cachon and Camerer [6] show that asking participants to pay a fixed price forparticipation also enables participants to coordinate on the Pareto-efficient equilibrium. TheVHBB [35] result, and the Crawford and Broseta [16] explanation of this result, suggestthat the price of the right to play serves as another means of “tacit communication” thatallows participants to eliminate equilibria with payoffs lower than that price, and thus toreduce the strategic uncertainty resulting from the multiplicity of equilibria. The Cachon–Camerer results suggest a role for both forward induction and loss avoidance. Devetag [18]establishes, for critical mass games that are closely related to the games under considerationhere, the importance of the degree of information feedback.In the experiments summarized above the signal is both tacit and costly. Costless sig-naling in coordination games has been investigated by Cooper, De Jong, Forsythe, andRoss (CDFR) [10]. They focus on two-player games with two Pareto-ranked equilibria andstudy one-sided as well as two-sided pre-play communication. They find that cheap talkhas the potential to increase both the incidence of equilibrium play and the frequency of theefficient equilibrium being reached. For a two-by-two game with tension between Paretoefficiency and risk dominance, they find that one-sided communication somewhat increasesthe frequency of the efficient equilibrium being reached whereas two-sided communica-tion guarantees efficiency.1The results in VHBB [33], however, suggest that the resultsof two-player games may be very different from those that involve more than two players,although much of this seems to be driven by the matching protocol used (e.g., [33,9]).It is well known that the addition of a cheap-talk stage to a game does not eliminate anyequilibrium outcomes of the original game. After all, the cheap-talk announcements cansimply be ignored. This argument does not accord well with our intuition for these games.Farrell [20] suggests a rationale for cheap talk’s apparent effects. According to him, if theplayers’ pre-play announcements constitute a Nash equilibrium, this equilibrium becomesa focal point that entices players to follow their announced plans. Subsequent theoretical1Charness [8], Clark et al. [9] and Duffy and Feltovich [19] have recently re-examined similar games withsomewhat different designs. Charness finds a stronger, and Clark et al. a weaker efficiency-enhancing role forpre-play communication. Duffy and Feltovich show that the effectiveness of pre-play communication is a functionof the game. All agree that communication affects play.276 A. Blume, A. Ortmann / Journal of Economic Theory 132 (2007) 274– 290work, extensively discussed in Section 3 below, has provided additional arguments for therole of cheap talk (e.g., evolutionary ones), but also noted possible countervailing forcessuch as risk in the underlying game or the absence of self-signaling messages, i.e. messagesthe sender wants to send if and only if they are true (see [1,21]).In the following we investigate these theoretical considerations by combining the experi-mental frameworks of VHBB [33,34] and CDFR [10]. In particular, we ask whether costlesspre-play communication with a