Research On Exchange Rate Forecasting Methods Based On Wavelet Analysis And PSO-WLSSVM Model

With the rapid development of financial globalization, the risk of financial and foreignexchange is complex and changeable, the subprime crisis and the euro zone’s sovereign debtcrisis have triggered the global financial crisis and economic downturn, which aggravate thedifficulty of exchange rate forecasting. The exchange rate has evolved into a complexnonlinear dynamic system, so it is difficult to use conventional time series prediction methodto analyze the change of exchange rate effectively. In order to analyze the inherent law ofexchange rate and reduce the risk of currency fluctuation, this paper tries to study andconstruct a forecast model for exchange rate based on wavelet analysis and PSO-WLSSVM.Firstly, The paper reviews the research and development of exchange rate forecasting,also reviews some time series forecasting methods comprehensively; And then expounds thetheoretical basis and application of wavelet analysis, least squares support vector machine(LSSVM) and WLSSVM; In order to solve the parameter selection problem in the applicationof WLSSVM model about the forecasting and regression, this paper discusses the selection ofkernel functions and regularization parameters, including the parameter sensitivity analysis,the improvement of particle swarm optimization (PSO); On this basis, we construct theexchange rate forecast model, including the selection of wavelet function and the lag of timesequence and designs a parameters optimization method based on CV-PSO-WLSSVM;Finally through the applied research to verify the validity of the model.This paper focuses on two study aspects: one is the denoising and reconstruction ofexchange rate data under the wavelet analysis; the other is to construct the PSO-WLSSVMmodel to regress and forecast the exchange rate. This paper adopts the exchange rate of RMBagainst the pound data for empirical analysis to verify the validity of the constructed model.The empirical results show that using wavelet analysis for the data preliminary treatment cananalyze the exchange rate volatility and change rule efficiency, so it can improve theforecasting accuracy of the exchange rate; Then the forecasting accuracy based on improvedPSO-WLSSVM model is better than the average of support vector machine (SVM) model,and it also has better fitting and forecasting effect. Further shows that the constructed Wavelet–PSO-WLSSVM model is feasible and effective in terms of exchange rate forecasting, so asto provide scientific and reasonable tools for exchange rate forecasting, thus it has importanttheory value and practical significance.