Explaining Violations of Bayesian Inference

Empirical studies show that agents often violate Bayes' rule in updating probability expectations. This paper deals with errors in combining observations with prior knowledge. Such errors neccessarily occur when agents have limited information-processing capacities. It is shown that rational control of errors could lead to systematic deviations from Bayesian inference which are consistent with empirical behaviour. It is also shown that simple adaptive updating rules are more robust against errors and have a better MSE-performance than the Bayesian rule.