A Portfolio Strategy Based on XGBoost Regression and Monte Carlo Method

Abstract

In this research, XGBoost algorithm was used to choose stocks. The stock data was downloaded from Yahoo Finance. The volumes, the differences of open price and close price, the differences of high price and low price, the adjusted close prices of the previous three days were considered as factors. Based on XGBoost, the data were segmented and trained to obtain the importance of each factor for each stock. The price of the previous three days is the most important factor for most stocks. In addition, RMSE and MAPE were calculated. After selecting the stocks with the minimum MAPE, the mean variance portfolio optimization model and the Monte Carlo method were used to find a range of portfolio weights of each stock in the stock pool. The return was calculated under the condition of reducing the risk. When the weights of the stock portfolio with the maximum Sharpe ratio are applied to the next year, the portfolio will achieve higher returns. Therefore, the model can be considered as a suitable tool to help investors implement better portfolio strategies.