Elsevier

Games and Economic Behavior

Volume 121, May 2020, Pages 368-381
Games and Economic Behavior

Can there be a market for cheap-talk information? An experimental investigation

https://doi.org/10.1016/j.geb.2020.03.002Get rights and content

Abstract

This paper reports on experiments testing the viability of markets for cheap-talk information. We find that the level of trade in these markets is very small and eventually vanishes. Sellers provide low-quality information even when doing so does not increase their monetary payoff. This contributes to the low demand in the market for information. Moreover, we observe the same very low level of activity in the market for information when sellers face no conflict of interest and the noise in the quality of the transmitted information is much lower. Hence, we argue that the collapse of the market for information is a demand phenomenon, and even small uncertainty over the quality of information seems to have a large impact.

Introduction

Information transmission is common in economic life. Transmission occurs even when it is possible to transmit only soft/unverifiable information and when the incentives of senders and receivers do not align. Potential competitors sometimes share information, for example, when recruiters discuss the characteristics of individuals whom they would both like to hire or when corporate raiders discuss potential takeover targets. The drawback of sharing information in this way is that conflicts of interest may make the sources unreliable. However, when information is costly to acquire, market participants can benefit by sharing, perhaps in return for some payment, the information they have acquired – even though the information shared may be noisy – thereby creating a market for information. The lower the rivalry of information or the higher its cost, the more likely soft information is sold and transmitted. The literature on cheap-talk games, starting with Crawford and Sobel (1982), has established a set of conditions for soft strategic information transmission.

This paper reports on a series of experiments testing the viability of a market for information. From a theoretical point of view, the game we designed has an equilibrium where information is indeed exchanged. In addition, previous experimental evidence shows that real subjects often tell the truth, even against their self-interest. This suggests that they derive utility from not breaching a truth-telling norm, and it could strengthen the likelihood of a market for information arising. Thus, both theoretical and experimental reasons support the feasibility of a market for information where the information is unverifiable. In our experiments, however, we find that markets for information are fragile when information is soft. Given the above, this finding is somewhat in contrast with the theoretical predictions and previous experimental evidence. Moreover, we find that when the information exchanged is verifiable, fragility no longer obtains, as the level of activity in the market for information is in line with the theoretical predictions. When information is soft, subjects show a considerably weaker willingness to pay for information, and there is little information exchange. This is due to a lack of trust in information from nonverifiable sources. We find this is true independently of whether there are conflicts of interest between sellers and buyers of information. Hence, the fragility can be attributed to the soft nature of the information.

In our baseline treatment, we use a stylized game in which agents can acquire information by paying a fixed cost. Then, they can sell information to others via nonverifiable reports. That is, a seller of information has no commitment to sending reports with correct information. What makes the game relevant for our purposes is that there is a potential conflict of interest. The seller of information may be able to profit directly from the information he acquired by carrying out trades in the market. In that case, he has an interest in misleading the buyers of information since his payoff is decreasing in the number of informed agents. It is important (and also, we argue, realistic) that this conflict of interest does not always exist. That is, these profitable opportunities arise only with some probability, that is, positive but less than one, and when they do not arise, there is no motive to mislead buyers.

Cabrales and Gottardi (2014), henceforth CG, characterize the ‘informative’ equilibria of such a game (when all agents are self-interested and do not have an intrinsic motive to tell the truth). In these equilibria, sellers of information send truthful reports when they have no opportunities to directly use the information. Otherwise, they send an uninformative message. Markets for information are active in equilibrium, provided the cost of acquiring information is not too high. Other equilibria exist, where the buyer of information expects to obtain uninformative messages. As a result, information is not transmitted.

It is of interest to contrast the findings for the above game with those for an alternative game specification. This was the basis for our main experimental control. In that alternative specification, the sellers of information can send only truthful messages, while the rest of the game remains unchanged. In this situation, when the seller prefers that buyers remain uninformed, he can achieve this outcome only by not selling them a report. The equilibrium of this game exhibits the same payoffs as the informative equilibrium of the game with cheap-talk reports. However, there is no equilibrium with outcomes analogous to the babbling equilibrium. We call the game where reports are cheap-talk messages and can then be false the game with soft information, or Soft game for short. We call the game where reports must be truthful the game with hard information or Hard game.

We perform experiments for both the Hard and the Soft games. The only difference between them lies in the quality of information in the reports sold. As argued above, the theoretical prediction is that the information transmitted is the same in the two cases. The random assignment to the treatment then allows us to attribute the cause of differences in outcomes to this feature. We believe that this is a novel experimental design. Indeed, how much information will be sold in the market and the role played by the hard vs. soft nature of the information transmitted is an open and relevant empirical question. In contrast, the previous theoretical and empirical literature on communication games tended to emphasize the alignment of preferences between senders and receivers, which we will show is less important in our case.

Our experimental findings allow us to conclude that the market for information in the Soft game does not work well. The quality of information (with respect to the quality predicted by the informative equilibrium in the theoretical analysis mentioned above) is rather poor. The market is extremely small and fragile.

We find that there are far fewer purchases of reports from informed players in the Soft game and, in the last rounds of the experiment on the game, these purchases decline further and practically disappear. This is in contrast with what we get for the Hard game, where we observe that the size of the market for information is much larger and, importantly, does not decrease as the game is repeated.

We should point out that the games briefly described above are static. Hence, reputational concerns do not affect the choice of sellers about the informational content of the reports they send. In the experiment, the games are repeated a few times. Thus, in principle, the sellers of information may benefit from creating a reputation for honesty in the Soft game. This makes the nonviability of information markets in the laboratory even more striking. The experimental results provide a negative reply to our motivating question. Despite being “theoretically” possible, markets for information do not develop with Soft information.

Having provided an answer to our main question, we also investigate the mechanism leading to this finding. In line with previous literature, we observe many truthful messages from sellers of information, even when sending a truthful message reduces the gains they can earn from their superior information. This favors the emergence of a well-functioning information market. A counteracting effect, however, operates in the opposite direction. Sellers of information who cannot profit from it often either lie or send uninformative messages.

The fact that some uninterested sellers are not sending informative messages is a novel finding, to the best of our knowledge. To understand this behavior, it is important to note the following. When a seller of information cannot profit directly from it and sends truthful reports, a receiver may benefit from it. Hence, his expected payoff could be higher than that of the sender. Thus, a possible explanation for the behavior we observe is that the seller is envious or nonprosocial. In that case, he may prefer to lie and thereby lower the payoff gained by the buyer of information. Of course, alternative explanations are possible. For example, a babbling equilibrium could prevail in the message part of the game: sellers could randomize their reports or send the same report regardless of the information they have. This would also lead to the collapse of the market for information.

To investigate the determinants of this behavior, we measure the social preferences of subjects playing the games. However, there is a low number of purchases of information, and thus messages sent, in the Soft game. This makes it quite hard to test the hypothesis that the low quality of information arises from envy motives or a lack of prosociality. In Cabrales et al. (2016), we analyze a sender-receiver game closely related to the message component of the Soft game. In the experimental evidence we obtain for that game, we find that social preferences can explain the anomalous, nontruthful behavior of senders. However, we cannot unambiguously conclude from this finding that the same is true in the situation considered here. The two games still exhibit some significant differences. We leave this question for future research.

The lack of truthfulness in the reports could contribute to the low participation in the market for information. However, we argue that this is not the only reason for the low level of activity in this market. This emerges from the experimental findings for a variant of the baseline treatment, which we call Uninterested. In that treatment, the sellers of information cannot profit from using the information directly.1 Hence, a conflict of interest between sellers and buyers of information never arises. Because of this, the seller has no direct benefit from misinforming the buyers of information. We find that the quality of information is indeed significantly higher than in the Soft treatment. Up to 92% of the reports are then truthful (instead of approximately only half of the reports in the Soft treatment). However, information sales are still rather low, very far below those in the most informative equilibrium and significantly below those in the Hard treatment. The results obtained for the Uninterested treatment are thus important. They suggest that the certainty of having accurate information compared to information whose quality is uncertain but still very likely to be accurate has very different implications for agents' willingness to purchase information. This is an important finding. Past theoretical literature on information transmission, starting with Crawford and Sobel (1982), emphasizes that the alignment of interests between the sender and receiver is the key for the emergence of communication. We find that the certainty about the accuracy of information is more important.

First, we should mention the seminal work of Crawford and Sobel (1982) on strategic information transmission. They study how the alignment of preferences between sender and receiver affects information transmission (Sobel, 2013 reviews the vast theoretical literature that followed).2 With respect to that paper (and the subsequent literature), we consider a different and richer game structure. This yields some novel results. In particular, the amount of information available to agents is endogenously determined. We also allow payments to be required for the receipt of messages. Crucially, the alignment of interests between senders and receivers is not commonly known. It depends on whether the sender can or cannot directly profit from the information acquired, and this is only privately known by the sender.

The experimental literature on information transmission has concentrated primarily on analyzing sender-receiver games à la Crawford and Sobel (1982). A first series of papers (e.g., Dickhaut et al., 1995; Blume et al., 1998, 2001; and Kawagoe and Takizawa, 1999) demonstrates that when the interests of the sender and receiver are perfectly aligned (the underlying game is one of common interest), play tends to converge to informative/separating equilibria, although other equilibria (babbling/pooling) exist. A more recent strand of the literature (see Sánchez-Pagés and Vorsatz, 2007; Kawagoe and Takizawa, 2005; Cai and Wang, 2006; and Wang et al., 2010) finds more evidence of truth-telling than the most informative equilibrium in Crawford-Sobel would predict in games in which interests are not perfectly aligned, an outcome that can be explained by a truth-telling norm.3 While in our experiments we also find some evidence of aversion to lying, we also observe a substantial amount of deception/misinformation even when lying does not increase the senders' payoff but reduces that of the receivers.

The paper is organized as follows. The Soft and the Hard games and their equilibria are presented in Section 2, while Section 3 describes the experimental design and the results. Finally, some robustness checks of our findings are briefly presented in Section 4, together with some concluding remarks. The experimental instructions, together with the description of the findings for the Uninterested treatments and additional results, are collected in the Appendix, available online.

Section snippets

The game and equilibria

The main objective of this paper is to assess and compare the viability of information markets in the presence of soft and of hard information. To this effect, we consider two simple variants of the model proposed in CG.

There is one object for sale. The object can be of one of 2 possible varieties, assumed to be equally likely ex ante. Let υ{1,2} be the true variety of the object. There are 3 potential buyers. Each buyer i{1,2,3} has utility 200 for one variety θi and 100 for the other

Design of the experiment

At the beginning of the experiment, subjects are divided into groups of three individuals. The subjects in any given group interact for 20 iterations of the game, and this feature is common information. Additionally, within each group of three subjects, each individual is randomly assigned a player position (1, 2 or 3) that remains fixed throughout the experiment.

We consider two main treatments (which we label Soft and Hard), where we implement the Soft and Hard games described in Section 2.

Discussion

We have established with the Hard and Soft treatments that the nature of the information that is transmitted makes a significant difference for the viability of a market for information. We have also shown that the source of this outcome is a steep decrease in the willingness to pay for information in the Soft relative to the High treatment. We now deepen the analysis by showing that the reason for this outcome has more to do with certainty vs. uncertainty over the quality of information

Conclusion

To sum up, in this paper, we experimentally study the viability of markets for information where the information transmitted is soft, that is, transmitted via cheap-talk reports. In this environment, the game played by agents features equilibria with and without information transmission, and hence, an empirical assessment of the viability of information transmission seems necessary. Furthermore, previous results in the experimental literature on cheap-talk games suggest that subjects in the lab

References (23)

  • G. Charness et al.

    Experimental games on networks: underpinnings of behavior and equilibrium selection

    Econometrica

    (2014)
  • Cited by (6)

    We are grateful to Vince Crawford, the editor, as well as an associate editor and three anonymous referees for very helpful comments. We also thank Andreas Blume, Marco Casari, Matthew Embrey, Peter Eso, Antonio Guarino, Navin Kartik, Meg Meyer, Marco Ottaviani, Santiago Sánchez-Pagés, Joel Sobel, Matthias Sutter, Marc Vorsatz, and seminar participants at the London Behavioral and Experimental Group, University of East Anglia, Università Bocconi, EUI, QMUL, Oxford University, Sussex University, and WZB Berlin for comments. Miguel A. Meléndez-Jiménez acknowledges financial support from the Junta de Andalucíaıa-FEDER through project UMA18-FEDERJA-243 and from the Spanish Ministry of Science, Innovation and Universities through Project RTI2018-097620-B-I00. All errors are our own.

    View full text