Do Artificial Neural Networks Provide Better Forecasts? Evidence from Latin American Stock Indexes

ABSTRACT Forecasting is a key activity for academics and investors in the fields of finance and economics. This paper explores the usefulness of the non-linear artificial neural network (ANN) for forecasting Latin American stock indexes. Our goal is to estimate and compare the forecast accuracy of the ANN with three traditional models: random walk, ARMA, and GARCH. Our results provide strong support for the ANN as a potentially useful device for predicting Latin American stock indexes. ANN forecasts are more accurate than those of more traditional methods, and the results are robust using the Diebold and Mariano test and encompassing regressions. RESUMEN. Trazar previsiones es, ciertamente, una de las actividades claves de los academicos e inversores en el campo de las finanzas y la economia. Este estudio explora la utilidad de una red artificial neural no lineal (ANN) para pronosticar los indices bursatiles latinoamericanos. Nuestra meta consiste en estimar y comparar la precision de la prevision de una ...