A two-stage multi-view prediction method for investment strategy

Scholars and industrial professionals are committed to integrating traditional financial economics models and machine learning models to improve the prediction model for stock prices, which is still a challenging topic. However, there is few acceptable results reported. This study proposes a two-stage multi-view prediction method that provides a new integration perspective for the integration of finance theory and machine learning technique. The first stage provides stock price prediction from one kind of model or a hybrid forecasting model, and the second stage adopts machine learning technique to improve the prediction accuracy. This study makes empirical analysis in Chinese A-share stock market. We adopt a statistical arbitrage that is designed according to the detection of the financial misevaluation opportunities in the first stage, which is a common investment strategy. And we build a gradient boosting decision tree model with the use of multiple views of features in the second stage to improve the performance of investment strategy. Our results show that the two-stage multi-view prediction method can optimize the prediction accuracy and enhance the outcome and profit of original trading strategy.