The Application of Ridge Regression, Random Forest and Mean-Variance Model in Portfolio Optimization

In recent years, with the increasing volatility in global financial markets, which is driven by factors including geopolitical events and policy uncertainty, the need for more effective portfolio optimization techniques has highly intensified. This study explores the use of machine learning, specifically Ridge regression and Random Forest model, under the framework of the Mean-Variance Model to optimize portfolio returns. Historical stock between July 2014 and July 2023 was used. Key technical indicators such as moving averages (MA50, MA200), volatility, and volume-based metrics were utilized as input features. The specific models used in this study were Ridge Regression and Random Forest for comparing the performance of the linear and non-linear models. The predicted returns from both models’ test sets were incorporated into the Mean-Variance portfolio optimization, aiming to maximize the Sharpe ratio. The back-testing results between July 2023 and July 2024 showed that both machine learning-enhanced portfolios outperformed the benchmark portfolio based on actual market returns, with the Random Forest portfolio achieving superior risk-adjusted returns and lower volatility. These findings represent the potential of utilizing machine learning models to enhance traditional portfolio optimization strategies.