Research On Dimension Reduction Of Stock Market Return Prediction

Economic fundamentals and stock market performance have traditionally been considered closely linked.However,there is not enough strong evidence on the effectiveness of using macroeconomic variables to predict stock market returns.In this paper,time series frequency domain decomposition is innovatively introduced into dimension reduction prediction,and a low-frequency dependent variable PLS method is proposed.Based on the 14 classic macroeconomic variables,the macroeconomic index containing aligned information of stock market return prediction(AMEI)is extracted.Compared with the original PLS method,the low-frequency dependent variable PLS method uses the long-term component of the original dependent variable with low noise,so it can reduce the estimation error of the factor.The empirical results show that AMEI has advantages in in-sample and out-ofsample prediction accuracy compared with single original macroeconomic variables,equal-weighted average factor,principal component factor and PLS factor based on market returns and other frequency components on forecasting S&P 500 index return.These advantages are significant across the prediction horizon from one month to one year.Asset allocation application demonstrates that AMEI can bring substantial economic returns to mean-variance investors.Economic mechanism analysis shows that AMEI’s forecasting ability comes from both cash flow predictability and discount factor predictability channels.In addition,AMEI shows significant predictive power for future financial uncertainty,which represents long-term market risk.Robustness test reveal that using the same 14 macroeconomic variables,the low-frequency dependent variable PLS method is substantially more efficient than many existing multivariate prediction models in extracting forecast information.In addition,AMEI can effectively predict the monthly returns of most characteristic portfolios and is complementary to investor sentiment in predicting overall market returns.Using full sample data for PLS estimation can further improve the predictive power of AMEI.