Um metodo para aplicação de redes neurais na estimativa de elasticidades de funções econômicas

Abstract

This article studies the efficiency and limitations of artificial neural networks when approximating functions. The aim is to evaluate the possibility of estimating partial derivatives of economy functions from these networks. Methodological aspects are observed in order to define the topology and the parameters of these neural networks to be able to control the results significance. Knowing the partial derivatives of economy functions is possible to know theirs elasticity, allowing research aboutthe weight of each exogenous variable in the composition of the endogenous variable. Some analyses and discernments of economic phenomenon can be taken with this information. At last, we present a hypothetical experiment about the applicability of function approximation with neural networks in economy.