Foreign exchange prediction and trading using restricted boltzmann machines

Abstract

Forecasting the exchange rate is a challenging task since the exchange market is a very complicated system with many latent variables and great randomness. In this report, by stacking up Continuous Restricted Boltzmann machines (CRBMs) which is a modified version of Restricted Boltzmann Machines (RBM) designed to model continuous data, a Deep Belief Network (DBN) is built. Trained on the historical data, this Deep Belief Network is able to learn high dimensional abstract features and patterns of the exchange market, and then a Long Short-term memory (LSTM) network is trained based on the features extracted in order to perform one-step-ahead prediction. The architecture and hyperparameters of our model are determined by experiments. To evaluate the performance of our model as a forecasting tool, three exchange rate series are tested with the comparison of other widely-used forecasting techniques in terms of there evaluation criteria. The results show that our model is applicable to forecast foreign exchange rate and outperforms traditional method in some aspects.