Authors:
Takahiro Obata
and
Setsuya Kurahashi
Affiliation:
Graduate School of Business Sciences, Humanities and Social Sciences, University of Tsukuba, Tokyo 112-0012, Japan
Keyword(s):
Macroeconomic Agent-Based Models, Parameter Estimation, Evolutionary Computations, Real-Coded Genetic Algorithms.
Abstract:
This study reports the estimation of model parameters for a macroeconomic agent-based model (ABM) using evolutionary computation methods. In an ABM, the parameter settings of the model are important in terms of verifying the validity of its outputs, because the parameter settings are closely related to these outputs, and determining whether the set parameters are appropriate. Conventionally, model parameters are qualitatively set by researchers based on values confirmed from empirical studies in related fields. However, in recent years, attempts to quantitatively determine model parameters using metaheuristic methods and Bayesian estimation-based methods have become widespread. In this study, we attempted to estimate time-varying parameters using a real-coded genetic algorithm, a type of evolutionary computation method, based on an inverse simulation method, which has not been used in macroeconomic ABM parameter estimation. The analysis confirmed that parameter estimation works well
when the economic conditions to be assimilated are simple, whereas it is difficult when economic conditions change in a short time, such as before and after economic shocks.
(More)