The Comparsion of Stock Return Prediction for Random Forest, Ordinary Least Square, and XGBoost

Abstract

With the stock market growing larger and the violent fluctuation becoming more frequent after the COVID-19 pandemic broke out, investors and researchers urgently need a method to predict the behavior of the stock market accurately. This research is determined to find out the performance of random forest (RF), XGBoost and ordinary least square (OLS) models in terms of predicting the return of given subjects. This research uses tushare to collect data and Jupyter Notebook to run the models. Libraries such as numpy, pandas, scikit-learn, and stockstats are also used in this paper. According to the analysis, XGBoost and RF model outperformed OLS model in all three subjects and the difference between RF and XGBoost model is subtle. Meanwhile, the results also revealed that the choice of subjects may affect the performance of model. Finally, only technical indicators were included in the process of model setup and this may negatively impact the results. These results shed light on the performance difference of the three models and lay a foundation for future high-efficiency hybrid models.