IMPROVING THE FORECASTING ACCURACY OF DYNAMIC CONDITIONAL CORRELATION (DCC) GARCH MODELS FOR EXCHANGE RATES OF EMERGING MARKET CURRENCIES

Abstract

This study examines the performance of Dynamic Conditional Correlation (DCC)-GARCH and hybrid DCC-GJR-GARCH models in modelling and forecasting exchange rate volatility in emerging markets, with a focus on Nigeria and South Africa. Daily exchange rate data for USD, GBP, and EUR against the Nigerian Naira (NGN) and South African Rand (ZAR) from January 2020 to May 2025 were analysed. The models were estimated under both multivariate normal and multivariate Student-t distributions to account for the stylised facts of financial returns, including volatility clustering, heavy tails, and potential asymmetry. Empirical results reveal that exchange rate returns exhibit significant leptokurtosis and time-varying volatility. The Student-t distribution consistently outperforms the normal distribution across all models, as evidenced by higher log-likelihood values and lower AIC and BIC criteria, indicating the importance of capturing fat tails in exchange rate modelling. While the DCC-GJR-GARCH model incorporates asymmetric effects, the leverage parameter was largely insignificant across most currency pairs, suggesting weak asymmetry in exchange rate volatility. Forecast evaluation using RMSE, MAE, and the Diebold-Mariano (DM) test shows that the DCC-GJR-GARCH model provides superior predictive accuracy compared to the standard DCC-GARCH model for most exchange rate series. The findings highlight the relevance of combining dynamic correlations with heavy-tailed distributions in improving volatility forecasts, offering valuable insights for risk management and policy formulation in emerging financial markets.