Aplicação de redes neurais à economia: demanda por moeda no Brasil

Abstract

Neural network models have been used in the economics basically aiming at obtaining more accurate forecasting. Thus, given the wide generality of this category of models, the possibility of exploring the great analytic power in the estimation of non-linear relations is being wasted. This paper is intended to introduce the main characteristics, involving elements, advantages, difficulties, and restrictions on the use of economic models estimated from neural networks. For such a purpose, the estimation of a money-demand function for Brazil will be used as the first approach. Main results indicate that the deep changes occurred since Cruzado Plan implementation have had strong impacts on the money-demand function, which is shown by marked oscillations in the magnitudes of interest elasticity, income elasticity, and price elasticity of that function.