Stock Selection Strategy Based on Support Vector Machine

Abstract

Stock traders nowadays attach increasing importance to artificial intelligence and machine learning techniques to construct better-performing stock portfolios. In this paper, a stock-selection model based on support vector machine (SVM) is applied to the data of selected technical indicators. Also, principal component analysis (PCA) is brought into the SVM model in order to cancel out the correlation and reduce the complexity of technical indicators. The model is carried out weekly on 12 years of historical data from 2008 to 2020, based on the component stocks of the Shanghai and Shenzhen 300 Index (CSI 300). Experimental results show that the annualized return yielded by our model reaches 14.5%, which significantly outperforms the return of the CSI 300. By comparing the results before and after employing PCA, the study suggests that PCA performs well when dealing with complex and non-linear data regarding investment securities, and PCA is especially beneficial for investments with relatively higher risk tolerance. It can be concluded that the proposed stock-selection model, which combines SVM with PCA, is of practical value for investors.