Self-sacri?cial Behavior in Information Cascades

Abstract

In sequential decision making, information cascades occur when agents base their decisions on the actions of other agents, ignoring their own observations. This can cause rational agents to perform sub-optimally at group level. The problem seems to stem from a disbalance in on one hand providing new information for the group (independent behavior), and other the other hand making use of this publicly available information (dependent behavior). Carried out in extremes, the fi?rst type of behavior would ignore potentially useful information, while the latter would lead to possibly incorrect herding-behavior. In this thesis, I investigate how in-group agents should ideally balance out the use of publicly accessible information, so that they can maximize their result as a group. A solution is provided for a speci?c example of sequential decision making, known as the urn problem. The main ?finding is that a perfect balance indeed exists, and that the information provided by early actors contributes the most to the optimal group result.