Learning with Limited Memory: Bayesianism vs Heuristics; Tai-Wei Hu (University of Bristol)

Abstract

Bayesian analysis is considered the optimal way of processing information. However, it often leads to problems for decision-makers with constrained cognitive capacity. Modelling such constrained capacity by finite automata, we answer two questions in the context of Wald's (1947) sequential analysis, namely in what environments is optimal Bayesian analysis possible even with constraints; also, when it is not possible what simplifications in the analysis enable us to obtain a satisfactory outcome. We identify two features of the simplified analysis: information stickiness (ignoring information) and rule stickiness (ignoring small differences in the environment).

Click here to view the paper.