COMPARATIVE ANALYSIS AND RANKING OF HYBRID MACHINE LEARNING AND GARCH MODELS FOR STOCK MARKET VOLATILITY IN SADC FINANCIAL MARKETS

Abstract

This study investigates the forecasting performance of machine learning models and traditional econometric volatility models in predicting daily stock price volatility across selected Southern African Development Community (SADC) markets from 02 January 2015 to 08 May 2026. Using data sourced from Yahoo Finance, the study evaluates Random Forest, LSTM, GRU, Ensemble, and XGBoost against EGARCH and GJR-GARCH models. The results reveal that machine learning models consistently outperform traditional volatility models across multiple accuracy metrics. Random Forest achieved the best overall performance with the lowest MAE (0.0121), lowest RMSE (0.0166), and highest ranking score, while LSTM and GRU followed closely with strong predictive stability. In contrast, EGARCH and GJR-GARCH exhibited higher forecasting errors and weaker explanatory power. These findings confirm that nonlinear machine learning models better capture the complex dynamics of SADC financial markets. The study provides important implications for risk management, portfolio optimization, and financial forecasting in emerging markets.