Emerging markets, such as Morocco’s stock exchange, face challenges including low liquidity, sectoral concentration, and economic sensitivity, which render traditional portfolio optimization methods inadequate. This study introduces a hybrid framework that integrates machine learning (ML) for stock selection with a novel Mean Variance Complex-Based (MVCB) optimization to enhance performance in the Moroccan All Shares Index (MASI). Four ML models named Stepwise Regression, Random Forest, Generalized Boosted Regression, and XGBoost are used to predict returns based on fundamental and technical indicators, with XGBoost achieving superior accuracy. The MVCB method leverages complex returns derived from the Hilbert Transform, capturing dynamic market correlations and phase-amplitude relationships to optimize weights under volatility. Backtesting reveals that the MVCB portfolio outperforms traditional mean-variance (MV) and market benchmarks, yielding a 10.48\% annual return with 3.52\% volatility and a Sharpe ratio of 2.48 (compared to 1.12 for MASI). Sector diversification and reduced left-tail risk (19.3\%) mitigate crisis-driven correlation breakdowns. By synergizing predictive ML with adaptive optimization, this framework addresses instability in emerging markets, offering a robust, scalable solution for risk-adjusted returns. The results highlight the viability of data-driven strategies in volatile, resource-constrained environments.
BOUHMADY et al. (Wed,) studied this question.