Discussion paper Series 1 / Volkswirtschaftliches Forschungszentrum der Deutschen Bundesbank 2005,34
The literature applying information-theoretic ideas to economics has so far considered only Gaussian uncertainty. Ex post Gaussian uncertainty can be justified as optimal when the associated optimization problem is linear-quadratic, but the literature has often assumed Gaussian uncertainty even where it cannot be justified as optimal. This paper considers a simple two-period optimal saving problem with a Shannon capacity constraint and non-quadratic utility. It derives an optimal ex post probability density for wealth in two leading cases (log and linear utility) and lays out a general approach for handling other cases numerically. It displays and discusses numerical solutions for other utility functions, and considers the feasibility of extending this paper?s approaches to general non-LQ dynamic programming problems. The introduction of the paper discusses approaches that have been taken in the existing literature to applying Shannon capacity to economic modeling, making criticisms and suggesting promising directions for further progress.