%0 Objeto de conferencia (Conference Object) %A Figueroa-García J.C. %I Springer Verlag %D 2017 %G Inglés (English) %T A three-step deep neural network methodology for exchange rate forecasting %U https://repository.urosario.edu.co/handle/10336/22520 %U https://doi.org/10.1007/978-3-319-63309-1_70 %X We present a methodology for volatile time series forecasting using deep learning. We use a three-step methodology in order to remove trend and nonlinearities from data before applying two parallel deep neural networks to forecast two main features from processed data: absolute value and sign. The proposal is successfully applied to a volatile exchange rate time series problem. © Springer International Publishing AG 2017.