The PCA-SVM Quantitative Timing Strategy Based On Emotional Factors

The development of behavioral finance provides new research ideas for us to analyze phenomena that cannot be reasonably explained by traditional financial theories in the market.According to behavioral finance,in the process of forming investment decision,the investor's emotion will make the investor have cognitive deviation to the stock price,and make the stock price deviate from its intrinsic value for a long time,that is,the investor's emotion has a significant impact on the fluctuation of the stock price.At the same time,combined with the "weak effectiveness" of China's stock market and the high proportion of individual investors,it is doomed that the research on investor sentiment cannot be ignored in the process of the research on the returns of China's stock market.Therefore,this paper takes investor sentiment factor as the analysis object and HS300 index as the investment target,aiming to build a stable and effective quantitative investment strategy model.In the construction of quantification strategy,data mining algorithm is not used in emotion quantification timing,and compared with other algorithm models,SVM shows better adaptability and accuracy in the prediction and modeling of financial time series,so this paper uses SVM model to construct emotion timing strategy.First of all,by reading the literature and comparing the advantages and disadvantages of the three methods of investor sentiment measurement,this paper selects 13 indirect indicators,such as turnover rate,stock market futures premium,financing balance growth rate,to form investor sentiment factor system,and selects agency indicators with stronger impact on the index through single factor back testing;then,use principal component analysis to reduce the dimension of the indicators,remove the multicollinearity between indicators and improve the efficiency of model learning.Single factor backtesting is used to verify the synthesis effect of principal component factors.Finally,the support vector machine model is introduced,the radial basis function is selected,and the ergodic search method is used to determine the optimal model parameters by estimating the prediction accuracy of the period data model,so as to complete the prediction of the rise and fall signals of the HS300 index.In order to conform to the logic of real deal,the rolling training method is used to optimize the model.In this paper,the original model can achieve an annualized yield of 11.63% in the back testing period,with a winning rate of 52.7%.After optimization,the annualized yield of the model is 17.23%,with a winning rate of 57.13%.Compared with the return of the CSI 300 index in the same period,the model shows a clear winning trend before and after optimization.The results show that it is effective to apply investor sentiment factor to financial market and predict the rise and fall of index price by data mining.The model of this paper provides some reference for quantitative investors in making investor decisions,and promotes the combination of data mining and emotion timing to build quantitative strategies.