‘Homo economicus’ as an intuitive statistician (2): Bayesian diagnostic learning

Abstract

The bounded rationality programme views the economy as a society of intuitive statisticians. The key for the success of this programme is the existence of a ‘tight enough’ theory of statistical inference. We have so far shown that there is no entirely data-driven algorithm that receives a finite sample of data and yields the model that best approximates the process generating the data. Learning an interpretable model of a choice situation requires starting with a parametric probability model. To analyse the programme further, we now examine the possibility of a ‘tight enough’ theory of learning within the general framework of the Bayesian theory, which is primarily a theory of parametric inference.