The Applied Research Of PCA And Random Forest In BARRA Quantitative Hedging Model

Recently,market participants begin to realize the limitation of traditional qualitative investment when there are more and more domestic listed companies showed up in this unpredictable market.That is why the qualitative investment strategy with good performance comes into the picture gradually.Based on the framework of BARRA theory,in this paper,we have made the following research:(1)First of all,we constructed pure factor test model and BARRA quantitative hedging model to support this article.These two models were all based on the BARRA multiple-factored structured risk model.(2)Secondly,to solve the problem that there is no highlight for important information in the BARRA quantification hedging model,which is caused by equal weighted multiple homogeneous factors synthesis,we need to use random forest to get the importance of factors,then regard them as the weighted weight of the corresponding factor.To achieve this purpose,firstly,we need to classify the given samples by the strategic goal of maximizing the moderate return.Then combined with the characteristics of the hedging model,which needs to select style exposure direction,the method of dividing the sample set to determine the weight of the factor is proposed.The empirical results showed that this method has improved experimental result markedly.(3)In order to solve the problem that factor yield is estim,ated inaccurately in the BARRA quantitative hedging model,which is caused by information overlapping among style factors got by weighting,we propose the method using PCA to extract the information in order to strengthen the accuracy of the estimation.The specific implementation process is as follows.Firstly,a number of factors in a kind of style were processed by PCA.Then we considered the linear relationship of components and the contribution rate of each component and then selected a principal component as the substitution of the style factor.The empirical results showed that although the approach made some styles lose a part of valuable information,the overall accuracy of the model is improved.Therefore,it derived a more reliable basis for hedging model in setting the style exposure.(4)We designed a corresponding system to avoid the faulty that might appear in multiple-factor model testing process.After the trial,the results show that the system does improve the accuracy and efficiency of the model in testing and application process.It proves to be a proficient stock picking tool for investors.