Research On Forecasting Stock Index With Hybrid Functional Link Artificial Neural Network And Wavelet Mutation Based Particle Swarm Optimization

As the stock m,arket has become more complete and mature,the stock index reflecting the general trend of stock market are regarded as the parameter for economy movement.Therefore,stock-index-forecasting has long been a researeh focus in financial fields as it of great importance both to the government in market regulation and individual investors in investing decisions.However,as the movements of stock index are quite complex,it is very difficult to have ralatively accurate prediction.In this paper,by absobing results from previous literature,I carried out research on theory an method for foreforecasting stock index in the following aspecets:Firstly,it summarizes the domestic and foreign research about stock index forecasting,and then briefly introduces two types of neural network,the Multilayer Perceptron(MLP)and the Functional Link Artificial Neural Network(FLANN),two parameter optimizing algorithms,the Back Propagation algorithm(BP)and the Particle Swarm Optimization algorithm(PSO),and their comparisons.Secondly,it introduces the Particle Swarm Optimization algorithm with Wavelet Mutation(WM-PSO),and a enhancement of its existing shortcoming as Improved Particle Swarm Optimization algorithm with Wavelet Mutation(IWM-PSO).The result of the data stimulation of five benchmark functions shows that the proposed methods have better performance in both accuracy and steady compared to other methods.Finally,it introduces the empirical study on Shanghai-Shenzhen 300 Index(CSI 300).We use data from 2016/1/21 to 2017/3/15 up to 278 daily transaction data to construct the traing dataset and the testing dateset.The IWM-PSO-FLANN gets the best performance among all three standards when forecasting the stock index,in the mean time,it shows that it is better to use trigonometric function as the functional expansion function than the chebyshev functions.When it comes to forecast earning rate,the IWM-PSO-FLANN model has the same performance as forecasting the stock index.Moreover it also has more favorable characteristics like signs prediction and earning performance.