Forecasting commodity market volatility in the presence of extreme observations

 

Extreme observations in commodity returns time series data occur as the result of shocks to a market through macroeconomic news, or market-specific events including fundamental and speculative pressures. However, outliers can have a dominating and deleterious effect in empirical models. This paper examines the forecasting of returns volatility in the presence of extreme observations using an AR(1)-GARCH(1,1) model for a non-ferrous metal futures contract. A simple method of accommodating extreme observations is applied that involves squeezing outliers to various thresholds. The forecasts obtained using this method are compared with a simple model in which all observations from the sample are used, and no adjustment for atypical observations is made. Estimates from the rolling one-step ahead models are presented graphically, and a number of forecast evaluation criteria are used to compare the forecasts generated under different outlier regimes.

Keywords: Extreme observations, Outliers, GARCH, Futures, Volatility, Metals