Using machine learning tool in fund selection: review and empirical test

Abstract

Since the Finance Industry is, through the years, growing tremendously, the willingness to understand, predict and trade on financial markets has led to an urgent demand for developing innovative models. Hence, Machine Learning Methods have gained in popularity. Yet, considered underperforming compared to traditional statistical approaches, the real-life application of machine learning on financial decision-making is currently lagging behind its promise. Indeed, poor performance was demonstrated when dealing with a large amount of data with intrinsic complexity and high dimensionality. This thesis aims to scrutinise opinions on machine learning, especially on its potential of being a tool for selecting, predicting,and optimisingportfolios. The study will evaluate the machine learning method by selecting and forecasting United States open-ended mutual funds from June 1990 to January 2021. This evaluation will be made on the seven following Machine Learning Methods: OLS regression, Lasso, Ridge, Elastic Net, XG Boost, Deep Neural Network and Random Forest. To select the open-ended mutual funds, the characteristics used as predictors are the following variables: monthly realised alpha, alpha t-stat, total net asset, flows, value-added, R2, and the t-stats of the market, profitability, investment, size, value, and momentum betas. The results obtained were regarding cumulative returns that Random Forest gains the first position with 37.21%, followed by XG Boost with 30.76%, and closely by Deep Neural Network with 29.43%. Hence, the results demonstrated the efficiency and robustness of machine learning methods in the selection strategy of mutual funds. Thus, the findings support that those machine learning methods will find their way into finance due to their reliability and good performance. Yet, the results should also be nuanced as the high turnover and implying high transactions cost could adversely affect the investment strategy's performance.