Research On Exchange Rate Forecasting Based On A Class Of Comprehensive Neural Network Model

With the rapid development of global integration,business trade between countries around the world continues to expand.As a link to global finance,the exchange rate is increasingly affecting government,business and individual investors.For government decision-making,a better understanding of exchange rate changes,timely extraction of relevant economic and financial information,and the development of forward-looking financial and monetary policies will help promote the healthy and sustainable development of the domestic economy.For large multinational companies and individual investors,mastering the fluctuations of exchange rates will help them avoid all kinds of investment risks in a timely and effective manner and reduce losses caused by changes in foreign exchange rates.Therefore,accurate understanding of exchange rate changes is particularly important for government,business and individual investors.The complexity and volatility of the exchange rate determine that effective prediction research should start from a non-linear perspective.In recent years,with the introduction of deep learning algorithms,artificial neural networks have been further developed because of their strong nonlinear processing power,flexibility and robustness,which are obtained in various industries including financial forecasting.Therefore,this paper intends to use artificial neural network as the main technology to establish a composite artificial neural network prediction model.Considering the volatility and non-stationary nature of exchange rate data,firstly,the original foreign exchange rate data is decomposed into several orthogonal,sinple homogeneous data branches by EEMD,namely several intrinsic mode functions(IMF)and one residual.Secondly,an artificial neural network-based intelligent prediction model is established.Each IMF and residual is predicted by a model such as GA-SVM(GA-BP),and then all IMFs and residuals are used by another GA-SVM(GA-BP).The prediction results are integrated to get the output.Finally,the model's predictive ability and stability are evaluated and tested.The established models and other comparative models are compared and analyzed using MSE,MAE,MAPE and Dstat four error evaluation indicators.At the same time,the EEMD-GASVM-GASVM model is used as the standard,and DM test is performed with other models.The empirical results show that the proposed decomposition integrated composite model is an effective technical method,which can significantly improve the accuracy of exchange rate prediction,and its prediction error is much lower than other comparison models.The prediction results of the four exchange rates indicate the stability and effectiveness of this kind of decomposition integrated composite model,and can also be used for the prediction research of other financial data.