Date of Award
Doctor of Business
In this paper, I conduct a comprehensive study of using machine learning tools to forecast the U.S. stock returns. I use three sets of predictors: the past history summarized by 120 lagged returns, the technical indicators measured by 120 moving average trading signals, and the 79 firm fundamentals, which helps to understand the weak-form market efficiency, algorithm trading and fundamental analysis. I find each set independently has strong predictive power, and buying the top 20% stocks with the greatest predicted returns and shorting bottom 20% with the lowest earns economically significant profits, and the profitability is robust to a number of controls. Econometrically, neural network generally improves forecasting over linear models, but makes little difference with firm fundamental predictors. Ensemble method tends to perform the best. However, when combining information from all the predictors, traditional machine learning improves little the performance due to perhaps not enough time series for too large dimensionality. In contrast, simple forecasting combination and portfolio diversification approach provide large gains.
Chair and Committee
Guofu Zhou (Chair), Thomas Maurer, Ngoc-Khanh Tran
Yi, Yingnan, "Machine Learning and Empirical Asset Pricing" (2019). Doctor of Business Administration Dissertations. 7.
Business Administration, Management, and Operations Commons, Finance and Financial Management Commons