A Machine Learning Integrated Portfolio Rebalance Framework with Risk Aversion Adjustment

This paper studies the evaluation of S&P 500 trend movements and their influence on portfolio optimization strategies, emphasizing the goal of achieving diversification across 24 stocks from eight distinct industry sectors, along with the inclusion of a risk-free asset. Through the meticulous construction of risk aversion-adjusted portfolios applicable to multi-period analyses, the paper employs variance and Mean Absolute Deviation (MAD) as the risk metric. The portfolio optimization process is enhanced by the combined use of Python for computational modeling and AMPL (A Mathematical Programming Language) for implementing complex mathematical formulations. To effectively predict risk aversion behaviors, we employ five classification models—Logistic Regression (LR), Support Vector Machines (SVM), Decision Trees (DT), Random Forest (RF), and eXtreme Gradient Boosting (XGBoost)— integrated with 29 indicators derived from technical analysis. The findings of this research show that the methods used perform differently in various optimization problems. Among the mean-variance and mean-MAD portfolios, the equal-weight strategy delivered the highest Sharpe ratio (26.75%), significantly outperforming all other models. Conversely, the Decision Tree (DT) model exhibited the weakest performance within the mean-variance portfolios (8.34%), while the Logistic Regression (LR) model yielded the lowest Sharpe ratio in the mean-MAD portfolios (5.40%).