Beyond ARIMA: A GARCH Framework for Forecasting Kijang Emas Gold Price Volatility in Pre- and Post-COVID Analysis

Abstract

This study applies Generalised Autoregressive Conditional Heteroscedasticity (GARCH) models to examine volatility in Kijang Emas 1oz gold prices in Malaysia from 2012 to 2022 across three analytically distinct phases: pre-COVID-19, post-COVID-19 and the full observation period. Stationarity is confirmed via the Kwiatkowski-Phillips-Schmidt-Shin (KPSS) test following first-order differencing, while the Jarque-Bera test and excess kurtosis are identified across the three phases, formally reject normality and confirm leptokurtic, heavy-tailed return distributions, providing the statistical justification for GARCH over constant-variance alternatives. Autoregressive Conditional Heteroscedasticity (ARCH) effects are confirmed in all phases via the Ljung-Box and ARCH-LM tests on squared first differences. Following systematic model selection, the optimal models are identified as GARCH(1,6) for the pre-COVID-19 phase, GARCH(2,1) for the post-COVID-19 phase, and GARCH(1,6) for the full period. These models achieve persistence parameters Σ(α+β) of 0.998999, 0.993806, and 0.998146 respectively, indicative of near-integrated variance processes. Critically, ARCH-LM tests on GARCH standardised residuals confirm complete elimination of conditional heteroscedasticity for all phases and all lags up to 10, whereas Autoregressive Integrated Moving Average (ARIMA) residuals retain highly significant ARCH effects, formally demonstrating GARCH's superiority in capturing volatility dynamics. The GARCH models achieve post-COVID mean absolute percentage error (MAPE) of 0.42308%, representing a 5.69% improvement over the ARIMA benchmark. Student’s t-distributional GARCH further reduces Akaike information criterion (AIC) by approximately 0.4311 to 0.4545 units across all phases with shape parameter, ν ≈ 2.12 to 2.38, confirming the presence of extreme fat tails consistent with the high excess kurtosis observed. These findings have direct implications for gold risk management in emerging markets under crisis-driven regimes.