Fragility of Information Cascades: An Experimental Study using Elicited Beliefs

This paper examines the occurrence and fragility of information cascades in laboratory experiments. One group of low informed subjects make predictions in sequence. In a matched pairs design, another set of high informed subjects observe the decisions of the first group and make predictions. According to the theory of information cascades (Bikhchandani, Hirshleifer, and Welch, 1992), if initial decisions coincide, an information cascade should occur: it is rational for subsequent players with low quality information to follow the observed pattern regardless of their private information. However, an information cascade should be fragile: it is always rational for subsequent players with high quality information to follow their private information. In line with existing experiments on information cascades, we find some evidence that low informed subjects follow the herd when it is rational, and this herding behavior occurs more frequently if there is a pronounced imbalance. The main finding of this paper is that information cascades are not fragile. We find strong evidence that highly informed subjects follow the herd regardless of their private information. In accordance with those observations we show, by explicitly eliciting subjects' beliefs about the state, that beliefs are not constant in the number of previous decisions that coincide, whether or not an information cascade already occurred. Subjects' behavior can be understood with a statistical model that allows for the possibility of errors in earlier decisions.