Skip to main content

Informational Cascade vs Herd Behavior


This academic paper talks about how the authors conduct an experimental study to distinguish informational cascade from herd behavior. The two concepts are used interchangeably in some contexts, but the authors distinguish them buy their subtle differences. An informational cascade describes when an infinite sequence of individuals ignore their private information when making a decision, whereas herd behavior occurs when an infinite sequence of individuals make an identical decision, not necessarily ignoring their private information. An implication of the difference between the two is that since informational cascade is purely imitative and hence is uninformative, social learning ceases in such a situation, whereas individuals in a herd may still provide information.


The authors conduct a lab experiment in the effort of distinguishing informational cascade from herd behavior. In the experiment setting, a group of individuals is given a private signal of a number between -10 and 10, based on a uniform distribution. The decision to make is whether to perform action A or action B where A is profitable if the sum of all the signals is positive, and B is profitable if the sum of all the signals is negative. Before one knows her own signal, she is given the history of actions of previous individuals, and is asked to choose a cut-off such that action A will be chosen if her private signal is greater than the cut-off, and B will be chosen otherwise. Only after she has reported her cut-off is she informed of her private signal and allowed to make her decision. An informational cascade is observed when starting with some individual, all people choose either 10 or -10 as their cut-off, whereas herd behavior is characterized by starting with some individual, all take the same action.


The result shows that roughly 36% of the rounds a herd behavior involving at least 5 subjects is observed. Although by theory information cascade should occur less frequently than herd behavior, it is also observed in 34.7% of the rounds.


The content relates to our class in that it studies the concept of informational cascade we covered in class using an experimental methodology, similar to the herding experiment we have in the textbook. Additionally, Bayesian rules and techniques from econometrics, such as GLS and MLE estimation, are also used to provide quantitative justification for their results.



Leave a Reply

Blogging Calendar

November 2017
« Oct   Dec »