Non-monotonous logic model of the CLP-USD exchange rate: An approach from Artificial Neural Networks (ANN)

In this work we propose a forecast model of the variation of the sign between exchange rate Chilean peso and American dollar (CLP-USD) at the close of two consecutive exchange days. This model is based on feedforward Artificial Neural Network development, whose structure considers a set of explanatory premises to describe the exchange behavior under study in a non-monotonous logical setting. The exchange rate CLP-USD at the beginning, at the closure, minimum and maximum values obtained during an exchange day from January 1, 2005 to July 9, 2018 are used. The obtained results show a good and robust percentage of sign prediction in different extra samples.