Redes neurais artificiais e análise de sensibilidade: uma aplicação à demanda de importancões brasileira

Abstract

This paper presents some non-linear estimates for the Brazilian import demand equation using artificial neural networks. The estimations use quarterly data from 1978 until 1999. Estimations are performed for total and intermediate goods imports and try to measure the sensibility ofthe network output to impulses in the input variables. This sensibility analysis in done using impulses on both each data and clusters of data. The main results show that there exists a non-linear relationship between the variables and that there has been a structural change in 1989 and 1994. Mainly after 1990 the most significant variables are the GDP and the exchange rate, with little relevance for the capacity utilization variable. From 1978 to 1988 all input variables have less significance, but the exchange rate seems to be the most relevant, followed by the capacity utilization, that is important for intermediate goods imports only.