Modelos baseados em agentes em economia e finanças: estimação por máxima verossimilhança simulada e dinâmica evolutiva de mercados heterogêneos

Abstract

This work aims to develop and analyze models in economics and finance featuring heterogeneous agents with bounded rationality. In the discrete-time framework, agents are allowed to adapt their beliefs at each time step by selecting from a finite set of predictors, each representing a distinct forecasting strategy. Each predictor is a function of past prices and provides an estimate for the next-period price. Furthermore, each predictor is associated with a publicly evaluated performance metric. To conduct a more detailed analysis, we first consider the cobweb model with linear supply and demand functions, alongside a set of two predictors: one representing rational behavior and the other naive behavior. We show that increasing the agents’ rationality, modeled through the intensity of choice among predictors, drives the price dynamics toward chaos, characterized by the emergence of a strange attractor. Subsequently, we consider a heterogeneous agent model for asset pricing. Using the Simulated Nonparametric Maximum Likelihood Estimator based on kernel methods, we calibrate the model to empirical data from different asset classes. The estimations reveal that the analyzed markets exhibit strong heterogeneity and low agent adaptability. Moreover, the model is able to replicate empirical statistical features that standard models typically fail to reproduce. We thus conclude that global dynamic complexity is an intrinsic feature of an equilibrium concept based on fully rational behavior.