Forecasting Of Stock Price Index Based On Wavelet Analysis And Support Vector Machine Algorithm

Since the establishment of the stock market,the ability to predict stock movements and achieve excess return through historical data has always been a core concern for investors.After Eugene Fama advances the effective market theory,thousands and thousands of scholars begin to study the effectiveness of the market.If the stock market is weak-form efficient,then it is futile to predict the stock price to obtain excess returns through historical data.Some researches indicate that the stock market is a nonlinear chaotic system.It's impossible to predict in the long run,but feasible to conduct short-term prediction.However,the traditional financial time series models,such as ARIMA,GARCH,which are based on the assumptions of stationarity and normal distribution,work poorly due to the complexity and variability of the stock market.Assumptions are often inconsistent with the actual.With the development of mathematics and computers,researchers introduce artificial intelligence algorithms into stock market forecasting,among which support vector machine is a very typical algorithm.The support vector machine algorithm has great advantages in dealing with small samples and nonlinear problems,and it works well in stock price forecasting.In reality,the stock price is often interfered by various “noise”.The support vector machine algorithm cannot eliminate the interference of these noises.Wavelet analysis can solve this problem.Wavelet analysis is also a common algorithm in the field of artificial intelligence.It can decompose the stock price into several details and approximation.By setting appropriate thresholds,it can remove most of noises,the stock price curve becomes smoother and can extract useful information without noises.In this paper,wavelet analysis is added into the support vector machine algorithm to improve the accuracy of prediction.Firstly,this paper selects the appropriate wavelet family and filter length to de-noise.The result shows that the CoifN is better than the DbN?SymN,and the optimal length of CoifN filter is 4.After that,this paper uses the Coif4 to decompose the stock price into 5 levels through a'trous algorithm,and forcibly de-noise the 5th and 4th detail level,keeping others unchanged.Finally,this paper studys the effects on how wavelet analysis preprocessed before support vector machine affects the prediction model.It shows that the application of wavelet de-noising can improve the accuracy.Through simple quantitative strategy analysis,it's found that wavelet de-noising can significantly improve the quantitative indicators,such as sharpe ratio and returns.