Double-Selection based High-Dimensional Factor Model with Application in Asset Pricing

This paper proposes a principal component analysis (PCA) approach after a double-selection Lasso and applies it to both Chinese and US stock market data. Similar to the idea of Post-Lasso, we perform least squares regression on the principal component factors. To accommodate the nonlinear nature of the data, this paper compares the support vector regression (SVR) model with least squares regression model. Empirical results show that the SVR method can improve the prediction ability, as evidenced by the superior accumulated rate of return using the test set sample of both markets.