When Optimization Isn't Optimal: Aggregation and Information Contagion

In the information contagion context, agents choose sequentially between two competing products, basing their decisions upon information obtained from a sample of previous adopters. The market shares that each product obtains depend upon the ture difference in performance between the products, but also on the number of previous adopters each agent samples and the way in which agents use the sample information to guide their product choice. We highlight some surprising features of these dependencies. First, it is not socially optimal for agents to be Bayesian optimizers. In fact, we show that a simple rule-of-thumb always leads to an asymptotic market share of 100% for the better product, while Bayesian optimization can result in substantial market share for the inferior product. Second, we show that giving agents access to more information can lead to smaller market share for the superior product. Third, we show that the ability to predict limiting market shares for the two products, even given knowledge of how the products actually perform and how agents process information, depends upon features of agent psychology and their decision rules.