Mutual Fund Selection Strategies Based on Machine Learning

The present study trains and tests the support vector machine (SVR), eXtreme Gradient Boosting (XGBoost), and Light Gradient Boosting Machine (LightGBM) machine learning models by using the data of 684 mutual funds from January 2014 to December 2019 that have been established for more than 5 years. The mutual fund products for which the models have a higher performance prediction accuracy are selected. The ROI of the selected mutual funds are verified by using common investment strategies, such as the lump-sum investment (LSI), dollar-cost averaging (DCA), Bollinger Bands (BBands), and momentum (MOM) strategies. According to the empirical results, the SVR, XGBoost, and LightGBM machine learning models effectively select funds with growth potential and lead to ROI higher than those of the included 684 funds. Furthermore, the positive predicted growth rate of net asset value is identified as a favorable criterion for selecting funds. This study discovers that the SVR model paired with any of the four investment strategies leads to an average rate of return higher than that of the 684 funds.