Shachar Kariv

Bayesian learning in social networks

Abstract We extend the standard model of social learning in two ways. First, we introduce a social network and assume that agents can only observe the actions of agents to whom they are connected by this network. Secondly, we allow agents to choose a different action at each date. If the network satisfies a connectedness assumption, the initial diversity resulting from diverse private information is eventually replaced by uniformity of actions, though not necessarily of beliefs, in finite time with probability one. We look at particular networks to illustrate the impact of network architecture on speed of convergence and the optimality of absorbing states. Convergence is remarkably rapid, so that asymptotic results are a good approximation even in the medium run.

Distinguishing Informational Cascades from Herd Behavior in the Laboratory

This paper reports an experimental test of how individuals learn from the behavior of others. By using techniques only available in the laboratory, we elicit subjects beliefs. This allows us to distinguish informational cascades from herd behavior. By adding a setup with continuous signal and discrete action, we enrich the ball-andurn observational learning experiments paradigm of Lisa R. Anderson and Charles Holt (1997). We attempt to understand subjects behavior by estimating a model that allows for the possibility of errors in earlier decisions.

Estimating ambiguity aversion in a portfolio choice experiment

We report a laboratory experiment that enables us to estimate parametric models of ambiguity aversion at the level of the individual subject. We use two main specifications, a “kinked” specification that nests Maxmin Expected Utility, Choquet Expected Utility, α-Maxmin Expected Utility, and Contraction Expected Utility and a “smooth” specification that nests the various theories referred to collectively as Recursive Expected Utility. Our subjects solved a series of portfolio-choice problems. The assets are Arrow securities corresponding to three states of nature, where the probability of one state is known and the remaining two are ambiguous. The sample exhibits considerable heterogeneity in preferences, as captured by parameter estimates. Nonetheless, there exists a strong tendency to equalize the demands for the securities that pay off in the ambiguous states, a feature more easily accommodated by the kinked specification than by the smooth specification. We also find that a large number of subjects are well described by the ambiguity-neutral Subjective Expected Utility model.

Observational learning under imperfect information

Abstract We explore Bayes-rational sequential decision making in a game with pure information externalities, where each decision maker observes only her predecessors binary action. Under perfect information the martingale property of the stochastic learning process is used to establish convergence of beliefs and actions. Under imperfect information, in contrast, beliefs and actions cycle forever. However, despite the stochastic instability, over time the private information is ignored and decision makers become increasingly likely to imitate their predecessors. Consequently, we observe longer and longer periods of uniform behavior, punctuated by increasingly rare switches.