Machine Learning–Enhanced Risk-Adjusted Portfolio Optimization under Volatility Clustering

Abstract

Portfolio optimization is one of the main tasks of financial engineering and becomes more complex in scenarios of high volatility. This research contributes a hybrid framework based on both mean–variance optimization and volatility forecasts from GARCH and LSTM models using daily price data from global equity and bond indices (2018–2024). The model leverages the capabilities of econometric and deep learning models to account for short-term volatility clustering and non-linear complexity. Following the application of the hybrid model to daily price data (2018–2024) from global equity and bond indices, the model consistently outperforms static Markowitz portfolios in the Sharpe ratio, Sortino ratio, and maximum drawdown. The findings have realworld applications in risk management and adaptive asset allocation.