Forecasting the Colombian Exchange Market Index (IGBC) using Neural Networks

Abstract

In this article, the daily Colombian exchange market index (IGBC) is forecasted using linear models, artificial neural networks and adaptive neuro-fuzzy inference systems with the aim of evaluate the accuracy of the forecasts when nonlinear models are used. In addition, we evaluate the explanatory power of other international market indexes, oil prices and exchange rates. Our findings are the following: first, an autoregressive neural network better captures the behavior of the IGBC in comparison with linear and adaptive neuro-fuzzy models; second, the preferred explanatory variables are able to explain complex properties as heteroskedasticity and non-normality of the residuals. And third, it is necessary consider as inputs not only the explanatory variables alone but also their interactions.