Mean-variance Portfolio Optimization with Stock Return Prediction Using XGBoost

Abstract

Portfolio optimization is one of the most concerning issues in finance and its success relies on accurate prediction of future stock market, which is challenging due to its dynamic, non-stationary, chaotic and noisy nature. This paper studies the performance of a portfolio optimization model when combined with stock return prediction using a machine learning model. In this study, two portfolio optimization algorithms are proposed. The first algorithm performs the eXtreme Gradient Boosting (XGBoost) for stock return forecasting and the meanvariance (MV) model for portfolio selection. The second algorithm modifies the MV model by introducing an additional penalty term based on the prediction error of XGBoost. The empirical tests using the historical data from 2010 to 2016 of the component stocks of the Korea Composite Stock Price Index show that the proposed algorithms are superior to traditional methods.