Equity UCITS in Morocco: Conceptual Foundations, Financial Risk Considerations, and the Contribution of Artificial Intelligence

Abstract

We adopt a hybrid approach that integrates traditional risk assessment methods with cutting-edge artificial intelligence techniques.The motivation for comparing three distinct models—XGBoost (gradient boosting), LSTM (recurrent neural networks), and Random Forest (ensemble learning)—stems from the need to evaluate their respective abilities to capture non-linear dependencies and long-term temporal patterns, which traditional GARCH models often fail to reflect in emerging markets. While conventional models are often inadequate for capturing the unique characteristics of emerging markets—where non-Gaussian distributions and asymmetric returns prevail—our study seeks to address these limitations. Standard methodologies, including likelihood function-based GARCH models for volatility clustering and Value at Risk (VaR) measures, frequently fall short in accurately reflecting market behavior during crisis periods. Our research delineates three distinct phases in market evolution, which illustrate an increasing maturity in financial markets and fund management practices. Our findings reveal that machine learning models, particularly XGBoost, substantially outperform traditional econometric techniques in volatility forecasting, although the performance of LSTM and Random Forest models varies across different risk applications. SHAP analysis highlights lagged volatility and market index returns as primary drivers of risk predictions. Ultimately, our findings demonstrate that XGBoost provides the most robust volatility forecasts, offering significant improvements for risk management frameworks and providing a resilient decision-making tool for regulators in the Moroccan context.