This website uses cookies.

By clicking the "Accept" button or continuing to browse our site, you agree to first-party and session-only cookies being stored on your device to enhance site navigation and analyze site performance and traffic. For more information on our use of cookies, please see our Privacy Policy.

  1. Home
  2. ASSA Annual Meeting
  3. 2021
  4. Preliminary Program
  5. Economics of Information
« Back to Results

Economics of Information

Paper Session

Sunday, Jan. 3, 2021 12:15 PM - 2:15 PM (EST)

Hosted By: American Economic Association
  • Chair: Ethan Struby, Carleton College

Information Processing: Contracts versus Communication

Andreas Blume
,
University of Arizona
Inga Deimen
,
University of Arizona
Sean Inoue
,
Louisiana State University

Abstract

We consider the trade-off between imperfect control and communication in organizations.
A principal anticipates receiving private information and hires an agent to take an action for her. The principal has the ability to contractually tie the agent's action to the state, but this control is incomplete. States not covered by a contract induce a communication game. Close alignment of interests favors communicating and, thus, ceding authority to the agent, and vice versa. Contracting increases the number of actions that can be induced through communication. Optimal contracts that do not cover all states both substitute for and facilitate communication.

Finding the Wise and the Wisdom in a Crowd: Simultaneously Estimating the Underlying Qualities of Raters and Items from a Series of Reviews

Nicolas Carayol
,
University of Bordeaux
Matthew O. Jackson
,
Stanford University

Abstract

Consumers and businesses rely on others' ratings of items when making choices. However, individual reviewers vary in their accuracy, and some are biased -- either systematically over- or under-rating items relative to others' tastes, or even deliberately distorting a rating. We provide a new technique that processes ratings by a group of reviewers over a set of items and simultaneously evaluates the individual reviewers' accuracies and biases, and provides unbiased and efficient estimates of the items' true qualities. Monte Carlo simulations show that our technique generates significant improvements over an average of ratings in recovering qualities even with small data sets, and that this improvement increases as the number of items increases. Revisiting the famous 1976 wine tasting that compared Californian and Bordeaux wines, we find substantial variation in reviewers' accuracies and a ranking that differs from the original one based on average ratings. In addition, we apply our methodology to more than forty-five thousand ratings of Bordeaux wines. Our estimated wine qualities significantly predict prices when controlling for prominent experts' ratings and numerous fixed effects. We also find that the elasticity of a wine price in an expert's ratings increases with that expert's accuracy.

Why Echo Chambers Are Useful

Ole Jann
,
CERGE-EI
Christoph Schottmüller
,
University of Cologne

Abstract

Why do people appear to forgo information by sorting into “echo chambers”? We model a society in which information is dispersed and preferences are polarized. Segregation into small, homogeneous groups can then be individually rational and Pareto-efficient, as it maximizes the amount of communication that takes place. We examine the optimal communication structure and give sufficient conditions for when it is attainable through endogenous group formation. A major problem is that people tend to segregate inefficiently little. Using data from Twitter, we show several behavioral patterns that are consistent with the results of our model.

Reputational Cheap Talk versus Reputational Delegation

Xin Gao
,
University of Cambridge

Abstract

Should a principal keep control or delegate the decision making to an
agent who has reputational concerns? Consider a two-period repeated
game. In each period, the uninformed principal first decides whether to
delegate to the informed agent who is either aligned or biased. If delegating,
the agent takes an action. If not, the agent sends a cheap talk
message to the principal who takes an action. In the second period, the
principal prefers communication. The first-period's authority allocation
depends on a prior cut-off: Delegation dominates communication only
if the prior about the agent being aligned is above this cut-off.

Expectation Formation under Uninformative Signals

Pascal Kieren
,
University of Mannheim
Martin Weber
,
University of Mannheim

Abstract

Can individuals distinguish informative from uninformative signals when forming beliefs? The neoclassical theory of probabilistic beliefs assumes that people update their prior beliefs according to Bayes’ Theorem whenever new (informative) information arrives. In reality however, not every signal contains new information, either because the signal is simply uninformative, or because the information is ambiguous.
In this article, we study whether individuals can distinguish objective pieces of new information from pieces which do not contain information. In a modified bookbag-and-poker-chip experiment (Grether, 1980) with 600 participants, subjects repeatedly observe new signals to learn about an underlying state of the world. In this environment, we exogenously manipulate both the informational content of signals (informative vs. uninformative) and the valence of signals (high vs. low payoffs). Our design allows us to test (i) whether individuals can distinguish informative from uninformative signals, and (ii) whether the belief updating process is affected by the valence of uninformative signals.
We find that individuals update their priors even after observing signals which contain no information about the underlying state of the world. Importantly, the direction in which they update depends on the valence of the signal. Subjects who observe positive uninformative signals increase their prior beliefs, while subjects who observe negative uninformative signals decrease their priors. Whereas a Bayesian forecaster would not update his prior beliefs after observing an uninformative signal, individuals tend to adjust their priors with about 34% of the strength as if the observed signal would contain information. This bias becomes even more pronounced when the valence of the uninformative signals is at odds with subjects’ prior beliefs and the more uninformative signals individuals observe.
Our results provide novel insights why individuals form and entertain false beliefs in environments where potentially new information is easily accessible but costly to verify (e.g. online media).

Increasing Returns and the Efficient Organization of Information

Michael Mandler
,
Royal Holloway University of London

Abstract

Increasing returns are inherent in the organization of information. If information can be stored in devices that attain k states, each value of k defines a different type of information "factory" with the number N of k-state devices indicating the scale of the factory. The number of facts that can be stored is then an exponential function of N.
As in conventional industries, increasing returns implies that efficient factories should operate at a large scale. A small k is therefore the efficient way to store a given number of facts since N can then be large. The optimal k is normally 2, even when the marginal cost of increases in k converges to 0. This result explains the prevalence of transistors in digital information storage, a puzzle for the engineering literature since it contends that 3-state devices are optimal.
Increasing returns also characterize the sequential acquisition of information. If we ask an agent a tree of questions then the paths of answers specify the information the agent can reveal. The number of paths will be an exponential function of the number of questions: increasing returns again prevails. Consequently, to minimize the time agents spend answering questions, the number of answers to questions should be small and the scale (number of questions) should be large.
Although the information storage and information acquisition problems are formally similar, most questions for the latter problem should have at least 3 answers, in contrast to the optimal number of 2 for states in storage devices. The difference stems from an additional cost of processing answers to questions that counterintuitively raises the efficient number of answers per question: agents typically fail to deduce that if all but one of the answers are rejected then the remaining answer must be accepted.
JEL Classifications
  • D8 - Information, Knowledge, and Uncertainty