Abstract
In this study, we propose a two-step methodology for the creation and optimization of investment portfolios utilizing machine learning techniques. Our analysis focuses on S&P 500 stocks from 2008 to 2022, incorporating fundamental data, stock prices, and macroeconomic variables as features in our models. The initial phase involves feature selection through Principal Feature Analysis (PFA), a dimensionality reduction algorithm. These selected features are subsequently used to predict stock price movements (upward or downward) via the XGBoost classification algorithm. The subsequent phase entails selecting stocks based on their forecasted return, and constructing optimal portfolios through both mean-variance and mean-entropy optimization approaches.
Download the PDF to read the full paper.
0 Comments