Empirical Study On Stock Price Forecasting Based On Hidden Markov Model And Support Vector Machine

Stock prices seem to be uncertain,but policies,industries,financial indicators and other conditions will affect the stock market.According to this characteristic,if investors can find out how these conditions affect the law of stock prices and predict the price trend beforehand,It can reduce the loss of investment caused by large fluctuations in stock price uncertainty,and even achieve unexpected returns.For the government,stock price forecasting is also one of the aspects that the government regulates the economy and prevents unexpected situations.If the stock market can be analyzed and predicted in advance,it will have great theoretical and practical significance for retail investors,enterprises,stock investment institutions and the formulation of relevant government policies.With the arrival of the era of financial big data,financial quantitative analysis will play an increasingly important role in future financial research and application in China.Based on CSI300 index,this paper makes empirical research on hidden Markov model and support vector machine model,and uses empirical method to test the model.As an attempt,this paper tries to provide some new methods and ideas for the study of stock price index.The empirical process of this paper mainly includes data selection,determination of optimal prediction factors,parameter estimation and prediction.At the beginning of this paper,the multi-day weighted hidden Markov model is used to forecast the price of CSI300 index.Through the comparative analysis of the experimental results,the 10-day weighted hidden Markov model is determined to be used for the next study.Then,the 10-day weighted hidden Markov prediction model is used to predict the combination of 120 prediction factors,and different gradient prediction factors are selected according to the prediction effect.Then,under different gradient prediction factors,support vector machine regression is used to predict stock prices.We choose the Gaussian kernel as the kernel function of support vector machine,so we need to determine the parameter g of the Gaussian kernel function and the penalty parameter C of the support vector machine before making prediction.This paper chooses the method of cross-validation to let C and g run in a certain range(e.g.C= 2-5,2-4,...,25,g= 2-5,2-4-...,25)and then choose the most accurate one.After determining the optimal parameters,this paper uses the rolling prediction method to predict.Then we use mean square error(MSE)and square correlation coefficient(DS)to evaluate the prediction results.From the results,we can find that 40 days are often the optimal forecast days when predicting different days,and the combination of the optimal prediction factors(turnover,MA10,MA60)obtained by HMM model also performs very well in the support vector machine prediction model,and their performance even exceeds the prediction results with the opening price,the highest price and the lowest price as the feature vectors.The empirical analysis in this paper is aimed at specific object and time interval.Generally speaking,in the actual investment,we need to think about the factors that may affect the returns in all aspects,and then adjust the model to adapt to the changes of the market.