Investment Portfolio Optimization Using Technical Indicators and White-Box Models

Quantitative trading has revolutionized in recent years with the integration of machine learning. However, most proposals are complex models that often need help with model understanding and feature importance identification. This study presents a methodology for optimizing investment portfolios using the XGBoost algorithm and a comprehensive set of technical indicators. The primary objective is to maximize returns by accurately predicting stock prices and selecting the most profitable stocks. Our proposal is based on decision trees, eliminating the need for recurrent neural networks or time series representations of data and enabling white-box machine learning models that are easier to interpret. We tried our proposal with real data corresponding to a collection of stocks of the 500 most influential companies in the United States of America, utilizing historical data such as open prices, highest and lowest prices, and trading volume. Experimental results demonstrated that our approach successfully identified the most profitable stocks, outperforming random portfolios and showing significant profit accumulation over time. This approach recognizes the most feasible indicators and facilitates the automatic design of investment portfolios and the analysis of the importance of technical indicators in complex data.