Enhancing Time Series Forecasting: Applying Cross-Validation Methods to Hybrid Models for Predicting GCC Stock Market Indices

Abstract

Forecasting Gulf Cooperation Council (GCC) country stock market returns is challenging because these markets exhibit volatility clustering, heavy tails, oil-price sensitivity, and asymmetric responses to shocks. This study evaluates whether ARIMA–GARCH-family hybrid models improve one-step-ahead forecasting of daily GCC stock index returns relative to single econometric specifications and whether time-series cross-validation provides a more robust model-selection framework than relying solely on information criteria such as AIC. Daily index data for Saudi Arabia, Kuwait, Bahrain, Qatar, Oman, Abu Dhabi, and Dubai from 3 October 2012 to 3 November 2022 are modelled using ARIMA, GARCH, TGARCH, APARCH, and hybrid ARIMA–GARCH-family models with Student-t innovations. Forecasting performance is assessed using RMSE, MAE, and MAPE measures under holdout testing, block cross-validation, walk-forward cross-validation, and rolling-window cross-validation. Model differences are evaluated using Friedman and Diebold–Mariano tests. The results show that single ARIMA specifications do not adequately capture the conditional heteroscedasticity and heavy-tailed behaviour of GCC stock index returns. Hybrid ARIMA–GARCH-family models provide more stable forecasting performance, with ARIMA–TGARCH(1,1)-t specifications selected for most GCC markets. ARIMA(5,0,3)–APARCH(1,1)-t provides the strongest performance for the Saudi index. The findings show that time-series cross-validation can reduce the selection of potentially over-complex models indicated by the AIC criterion and provides a more reliable basis for selecting forecasting models in emerging, oil-dependent financial markets.