Modeling Coupling Relationships In Financial Markets On Deep Learning

The academic community began to study methods for predicting financial markets very early.Scholars proposed many different time series models and various statistical models,but the prediction effects of these models are often not satisfactory.The occurrence of the financial crisis and the interaction between different financial markets have made the academic community realize that the prediction of financial markets is very difficult and the importance of understanding the coupling relationship between financial markets.These coupling relationships are not only directly observable from financial market data,but also very difficult to reflect through the model.The deep learning model can fit a variety of complex nonlinear functions and can obtain high-dimensional complex feature vectors by learning and extracting simple features.This paper describes three different types of financial market coupling relationships:homogeneous relationship,heterogeneous relationship,autoregressive relations,and then uses the deep neural network composed of CRBM and GCRBM to model the coupling relationship.We predict different financial market price indices in eight countries,and our models achieve more accurate results than traditional time series models and shallow or deep neural network models that directly predict.