Quantitative Stock Selection Strategies Based on Kernel Principal Component Analysis

Abstract

Any stock is exposed to many different risk factors simultaneously. Collinearity among risk factors often makes it difficult to identify effective factors. Based on multi-factor quantitative stocks selection, 60 factors are selected including the fundamental, technical, macroeconomic factors and so forth. Then we map, process, and identify the data of Shanghai and Shenzhen 300 constituent stocks in high-dimensional space and extract characteristics of factors by kernel regression. The required model factors are determined by the variance contribution rate and the sample data are transformed by the feature vectors and kernel function, which are regressed with the stock returns to construct a multi-factor stocks selection model. The back test of historical data in 2017 indicates that the model has a lower back test and a higher return than Shanghai and Shenzhen 300 Index. Based on the bootstrap method, the robustness of the model is verified in the robustness test finally. The result shows that the combination of kernel principal component analysis and multi-factor stocks selection model is well verified, which can beat the market with high probability and effectively overcome the problem of random stocks selection.