Forecasting Exchange Rate With Deep Belief Networks

Forecasting exchange rates is an important financial problem which has received much attention. However, the foreign exchange market is a multi-variable nonlinear systems, and the factors’s mutuality is quite complex in exchange rate market. There-fore, forecasting exchange rates is an important and challenging task. Neural network is a nonlinear dynamical system, and has extensive adaptability and learning ability. So it has been used successfully for modeling and control of multi-variable nonlinear systems.Since the1990s, neural network has been widely used in the field of economy, finance, and has become one of the effective tools in this research field. Feed forward neural network (FFNN), which is the state-of-the-art method for forecasting exchange rate with neural networks, is easy to fall into local optimum. Deep belief network (DBN) is a new neural network model and is introduced in2006. Besides, DBN is able to converge to the global optimum. Consequently, we can obtain more accurate predictions.This paper summarizes the theoretical framework of exchange rate forecasting and DBN. We discuss the content of DBN algorithm and design the optimal architec-ture of a DBN through experiments to the forecasting exchange rate task. Then, we firstly propose DBN for forecasting exchange rate and we introduce the experimental process in detail and carry on the analysis to the experimental result. First, we take pretreatment for the three exchange rate series. During the training, a deep belief net-work (DBN) is combined with conjugate gradient method to accelerate the learning. In testing phase, four evaluation criteria is used to evaluate the performance. Finally, we compare the results with FFNN and some traditional methods. It has been proved by our experiments that DBN combined with conjugate gradient method has the best result, and it has a good prospect of development.