Dynamic Spatial Search and Learning in a Fishery

Abstract

This paper measures costs of cognitive biases in learning. Although theoretical literature on learning has been expanding recently, no paper studies how much cost the biases could cause at our best knowledge. We measure the cost by comparing decision-makers’ profits with over- and under-reaction and confirmation bias. An economic agent has a prior belief on the distribution of return to each choice. For each period, he observes his return from this period’s choice and then updates his belief for the next period’s choice. Cost occurs if he alters his choice like the transaction fee of stock trading. He enjoys higher welfare over the infinite horizon if he learns the true data generating process earlier, enjoying higher expected returns and reducing the switching cost. In specific, we simulate our model built on a fisherman’s location choice using confidential survey data of longline halibut fishermen during 2006-2007 in the Gulf of Alaska. We overcome an empirical challenge of no prior belief exists for each choice. This paper contributes to the growing learning literature by showing the effect of biases in monetary metrics and to the literature on empirical dynamic modeling by proposing a method to incorporate non-Bayesian inference from new information.