Money Demand and Seigniorage-maximizing Inflation in Latin America: Approximation, Learning, and Estimation with Neural Networks

Abstract

This paper examines money demand and the seigniorage-maximizing inflation rates in Chile, Colombia, Mexico, and Peru with error-correction models (ECM) and artificial neural network (ANN) methods. The purpose is to approximate more accurately the "true" underlying non-linear functional forms for the long-run equilibrium demand for money, to estimate the learning process in short-run monthly adjustment of money stocks, and to obtain better estimates of the seigniorage-maximizing rates of inflation in a region characterized by macroeconomic instability. Unlike most previous studies, this paper explicitly incorporates parallel-market exchange-rate uncertainty in the short-run demand for money. The ANN model shows that there are various degrees of non-linearity in the long-run demand for money, with Chile ranking highest. However, all of the countries examined showed that a relatively simple ANN model outperformed the ECM for the error-correction or learning process in the short-run demand for money. In one case, Peru, the ANN more than doubled the explanatory power of the linear ECM. The ANN approach also shows that uncertainty plays the dominant role in short-run money demand, and that the seigniorage-maximizing inflation rates are much lower that the predictions of previous studies.