Forecasting in a complex environment: Machine learning sales expectations in a stock flow consistent agent-based simulation model

Abstract

The aim of this paper is to investigate how different degrees of sophistication in agents’ behavioral rules may affect individual and macroeconomic performances. In particular, we analyze the effects of introducing into an agent-based macro model firms that are able to formulate effective sales forecasts by using simple machine learning algorithms. These techniques are able to provide predictions that are unbiased and present a certain degree of accuracy, especially in the case of a genetic algorithm. We observe that machine learning allows firms to increase profits, though this result in a declining wage share and a smaller long-run growth rate. Moreover, the predictive methods are able to formulate expectations that remain unbiased when shocks are not massive, thus providing firms with forecasting capabilities that to a certain extent may be consistent with the Lucas Critique.