MONEY TALKS? AN EXPERIMENTAL INVESTIGATION OF CHEAP TALK AND BURNED MONEY

• Published date: 01 November 2015
• Author: Theo Offerman,Randolph Sloof,Thomas de Haan
• Date: 01 November 2015
• DOI: http://doi.org/10.1111/iere.12141

INTERNATIONAL ECONOMIC REVIEW

Vol. 56, No. 4, November 2015

MONEY TALKS? AN EXPERIMENTAL INVESTIGATION OF CHEAP TALK

AND BURNED MONEY∗

BYTHOMAS DE HAAN,THEO OFFERMAN,AND RANDOLPH SLOOF1

Norwegian School of Economics, Norway; University of Amsterdam, The Netherlands;

University of Amsterdam, The Netherlands

We experimentally study the strategic transmission of information in a setting where both cheap talk and money

can be used. Theoretically, many equilibria exist side by side, in which senders use either costless messages, money,

or both. We find that senders prefer to communicate through costless messages. Only when the interest disalignment

between sender and receiver increases does cheap talk tend to break down and high sender types start burning money

to enhance the credibility of their costless messages. A behavioral model assuming that sellers bear a cost of lying fits

the data best.

1. INTRODUCTION

Many strategic interactions contain a phase where the involved parties exchange privately

held information. In the economics literature, two communication channels have been identified

that allow agents to communicate information in a meaningful way. Crawford and Sobel (1982)

showed how in a situation of partial conflict of interests an informed party may employ costless

messages to transmit private information to an uninformed party. In equilibrium, communica-

tion must go via a vague, imprecise language. The conflict of interests shapes the language and

provides a limit to the extent of information transmission. Spence (1973) addressed the question

of how agents can communicate strategically by burning money. In the context of a job market

signaling game in which employers are uninformed about prospective workers’ productivity

type, he showed how high-type job applicants can credibly separate themselves from lower

types by means of obtaining costly (but useless) education. Thus, a sender can credibly signal

information about his type by employing either cheap talk or costly signaling.

In practice, a combination of the two channels is often used though. Take, for instance,

advertising. One acclaimed role of advertising is to provide information to potential buyers

about a product’s attributes (cf. Nelson, 1970, 1974). Some attributes, like price, are easily

verifiable before the product is bought. As there is little room for misrepresentation, such

search attributes can to a large extent be communicated through cheap-talk messages, that is,

the content of the advertisement. Other attributes, like quality, can only be judged upon buying

and actually using the product. These experience attributes are more easily misrepresented,

and the informational value mainly lies in the amount spent: high-quality firms may distinguish

themselves by spending more on—directly uninformative, or dissipative—advertising then low-

quality firms do (Nelson 1970, 1974; Kihlstrom and Riordan, 1984; Milgrom and Roberts, 1986).

As many goods have both search and experience characteristics (and for each attribute, the

distinction is one of degree), costless and costly signaling may be used at the same time. As

another example, a suitor can always blarney his fiancee by simply saying he loves her, but

∗Manuscript received November 2012; revised May 2014.

1Our article benefited significantly from the constructive and helpful comments of three anonymous referees and the

coeditor Charles Horioka. We gratefully acknowledge financial support from the Research Priority Area Behavioral

Economics of the University of Amsterdam. We thank CREED-programmer Jos Theelen for programming the exper-

iment. Please address correspondence to: Randolph Sloof, Roetersstraat 11, 1018 WB Amsterdam, the Netherlands.

Phone: +31 (0)20 525 5241. E-mail: r.sloof@uva.nl.

1385

C

(2015) by the Economics Department of the University of Pennsylvania and the Osaka University Institute of Social

and Economic Research Association

1386 DE HAAN,OFFERMAN,AND SLOOF

unless he is morally inclined not to lie, this is a cheap-talk message. The bachelor, however, can

also communicate his commitment through offering personalized gifts that are either costly to

find or costly to purchase. This may explain why many men buy (sometimes very expensive)

gifts for their wives/girlfriends (cf. Camerer, 1988). 2

The availability of two types of communication raises the question of exactly how people mix

between costless and costly signals in transmitting private information. The main purpose of this

article is to explore this question in an experiment. In particular, we investigate how the (simul-

taneous) use of and the interplay between cheap-talk and costly signaling varies with the level

of interest alignment between sender and receiver. Intuitively, one would a priori expect that

senders gain credibility when they support their fine talk with conspicuous expenses on costly

signals, especially when interests become less aligned and cheap talk theoretically loses (much

of) its informative value. One might even conjecture that the availability of more informative,

costly signals renders cheap talk by and large meaningless. At the same time, however, it seems

well conceivable that senders prefer to avoid the use of the costly communication channel in

order to save on (potentially high) signaling costs.

Our experimental setup corresponds to the one explored theoretically by Austen-Smith and

Banks (2000). They augment the canonical model for strategic cheap-talk communication of

Crawford and Sobel (1982) with the possibility that the sender may use costly signals as well. In

particular, besides a cheap-talk message, senders may impose costs on themselves by publicly

burning money.3Although this costly signal is in itself a pure social waste, it provides a very

precise measure of how much a sender is prepared to spend to get his true type recognized.

Austen-Smith and Banks show that the set of equilibria dramatically increases when costly

signals can be used. All original equilibria of the Crawford-Sobel (“cheap talk only”) setup are

preserved, but by using the costly channel, new, more informative equilibria exist side by side.

In fact, there exists a continuum of semipooling equilibria, ranging from a complete pooling

equilibrium to a fully separating one. These equilibria differ profoundly in the use of and the

interaction between money and costless messages to signal information.4

In the presence of multiple equilibria, the exact interplay between money and words becomes

an important empirical question. We address this question head on in the controlled environ-

ment of the lab. In our experiment, we implement the standard uniform-quadratic setting of

the strategic communication game, in which the sender’s type is uniformly distributed and the

players’ preferences are represented by quadratic loss functions. (This setting has been the

working horse for most applications of the Crawford and Sobel, 1982, model.) The sender’s

2As expensive gifts may have a resale value, the true signaling value may lie in the purchase price minus the resale

value. Personalizing the gift through, e.g., an inscription lowers the resale value and thus increases the signaling value.

3In contrast to Spence’s original formulation, in Austen-Smith and Banks (2000), the costs of signaling are inde-

pendent of the state. The aforementioned example of advertising provides a real-world case that arguably fits this

assumption. Indeed, most models on directly uninformative, dissipative advertising assume the costs of advertisements

to be state-independent. For industrial organization applications, see, e.g., Kihlstrom and Riordan (1984), Milgrom and

Roberts (1986), Hertzendorf (1993), Bagwell and Ramey (1994), and Bagwell (2007) for an overview. See, e.g., Prat

(2002a, 2002b) and Potters and Van Winden (1992) for applications in political science, such as political campaigning

and lobbying. Within an organizational context, Gibbons et al. (2013, p. 380) discuss the Austen-Smith and Banks

(2000) model, interpreting money burning in reduced form as “...fighting through red tape and bureaucratic sclerosis.”

4Bagwell and Ramey (1993) theoretically explore the interplay between cheap talk and burning money in a more

specific model of informative advertising. In their model, advertising may provide information that helps buyers with

heterogeneous tastes to match with their preferred products. One group of buyers prefers high-quality, high-priced

goods, whereas another group of buyers has a taste for low-priced, low-quality goods. Optimal coordination requires

that firms identify themselves. Firms may potentially do so through the content of their advertising, but these cheap-talk

claims need not be truthful: Firms may pretend to be of a different type to trick consumers of the claimed type to visit

their store. If the share of consumer types is about equal, firms do not benefit from misrepresenting their type, and

the content of the advertisement is credible. If not, high market share firms must turn to costly advertising to signal

their quality. Under free entry, high market shares follow from high fixed costs. Bagwell and Ramey (1993) therefore

conclude that cheap talk can communicate quality when fixed costs are largely independent of quality, whereas if fixed

costs vary significantly with quality, the higher fixed-cost quality levels need to use dissipative advertising to signal

quality.

MONEY TALKS? 1387

bliss point regarding the receiver’s action is bunits above the bliss point of the receiver, with

bias parameter b>0 representing the level of interest disalignment. Within this setting, we

investigate how subjects use the costless and the costly communication channel to signal their

private information and which of these two channels is most prominent. We focus, in particular,

on the impact of different levels of interest disalignment on the mixture of communication

methods employed. In the experiment, we investigate our conjecture that successful communi-

cation through cheap talk gets harder as the bias bgets larger. This may induce a shift toward

communication through burning money when preferences become less aligned.

We obtain the following main experimental findings: Senders appear to have a strong pref-

erence for costless messages. They predominantly choose to communicate through cheap talk,

and cheap talk appears more informative than the most informative Crawford and Sobel equi-

librium predicts. Only when the level of interest disalignment increases and costless messages

turn out to be insufficiently useful for high types to separate themselves from low(er) types do

sender subjects start to burn increasing amounts of money to get their exaggerated cheap-talk

messages across. The amounts of money involved, however, are well below the levels required

by the costly signaling equilibria (as derived in Austen-Smith and Banks, 2000). Nevertheless,

the induced reaction of receivers indicates that senders actually do gain credibility by backing

up their costless messages with burned money.

Our finding that cheap talk is more informative than standard theory predicts is in line with

previous experimental findings. Studying settings that allow for costless communication only,

Dickhaut et al. (1995), Cai and Wang (2006), S´

anchez-Pag´

es and Vorsatz (2007), and Wang et al.

(2010) find that senders consistently overcommunicate as compared to the most informative

equilibrium in cheap-talk games.5Our results indicate that this finding is not an artifact of

imposed limitations on communication, but continues to hold when subjects are allowed to use

alternative, costly communication channels as well.

A plausible explanation of why senders’ cheap-talk messages are more informative than

the standard model predicts is that senders are lying averse. In a recent contribution, Kartik

(2009) analyzes the original setting of Crawford and Sobel under the assumption that senders

bear a cost of lying. He shows that partial separation by means of messages may then occur

in equilibrium, in line with the experimental findings on cheap-talk games. The Kartik model

is readily extended to the situation where senders may burn money as well (see our Section

2). In that case, even full separation is possible, with low types separating through costless

messages and high types by means of burned money. In our empirical analysis, we verify which

equilibrium model fits our data best. We find that Kartik’s original lying cost model—which does

not include the use of burned money—does so. However, there are some clear signs that our

data have not (yet) reached equilibrium. In particular, the equilibrium models do not explain

the actual use and effectiveness of moderate costly signals when preferences are less aligned.

Our article contributes to a small experimental literature comparing the effectiveness of words

and actions in various applications. Duffy and Feltovich (2002, 2006) compare the effectiveness

of cheap talk and observations of previous round actions of others in three different 2×2 games

(prisoners’ dilemma, stag-hunt, and chicken). They find that, when studied in isolation, either

source of information makes coordination and cooperation more likely. Interestingly, having

both information sources present lowers the likelihood of good outcomes, mainly because the

two information channels may lead to opposing interpretations (i.e., point to different expected

actions of the opponent).6Celen et al. (2010) investigate the role of both predecessor’s advice

about how to play and observation of their actions in a standard social learning game. They

observe that advice is more effective despite the fact that the two forms of communication

are theoretically equally informative. A key difference with these earlier papers is that in our

5In contrast, Brandts and Charness (2003) investigate how senders’ communication about intended play in a complete

information game affects receivers’ reciprocal responses. They find that punishment by the receiver is much more likely

if, for the receiver, unfavorable outcome was reached after a deceptive message.

6In a trust game, Bracht and Feltovich (2009) compare cheap talk and observation of the trustees’ past decision and

find that observation leads to much more cooperation, whereas cheap talk has little effect.