Comments on the paper

Abstract

The paper by Sylvain Barde presents the explorations into the powers of a novel technique (to economics) called Maximum Entropy (MaxEnt hereafter). The methodology was introduced to economics by Foley (1994), but to the present day its potential is largely untapped. MaxEnt allows predicting solutions to agent-based models analytically. The previous use of methodology has been in image reconstruction, where predictions are made about the original image based on the noisy signal at hand. The approach has a great potential on reducing computational time required to run full-fledged agent-based models that are very often NP-difficult. A particularly intriguing feature of the methodology is that time is implicitly embedded in it. This might not be important in image reconstruction but it is very important in economics as it allows to predict not only the time invariant/equilibrium solution to the model but also to describe the transitional path to it. In previous paper (Barde 2012) the sufficient conditions for the applicability of the methodology have been derived. The same paper has applied the MaxEnt methodology to Schelling’s (1969, 1971) model of segregation. It has been demonstrated that MaxEnt is powerful with respect to the models with fixed proportion of distinct populations. In current paper the methodology is applied to two models with recruitment. These are the models of ant behavior by Kirman (1993) and that of language competition by Abrams and Strogatz (2003). The distinction with respect to the previous application is that recruitment allows the proportion between the (competing) populations to vary. The properties of the two models discussed are well known. In light of this, the performance of the methodology is tested on different time horizons. Using rigorous computational approach it is demonstrated