Conditional Variance Dynamics of Gold and Other Precious Metals - Forecast Comparison with Intra-Day Data

Abstract

The conditional variance of returns of Gold, Silver, Palladium, and Platinum is modeled incorporating GARCH, APARCH, FIGARCH, and FIAPARCH for different underlying distributions. These returns are found to be non-normal. The results suggest that Gold and Silver feature different econometric properties in their dynamics than Palladium and Platinum. While Gold and Silver are dominated by asymmetric effects, Palladium and Platinum feature a preponderant long memory. This behavior is verified by taking into account recent shocks like the Volkswagen Diesel scandal or the 'Brexit' poll. The forecasting comparison of these models for one, five, and 20 day-ahead forecasts is compared with two different proxies for the daily realized variance -- squared daily returns and a proxy based on intra-day data. It is found that incorporating a more realistic measure for the realized volatility, the loss functions are significantly reduced and the decision on the best performing model changes. For Gold and Silver, APARCH and partly FIAPARCH are found to be the best performing.