This paper assesses model effectiveness by considering three models namely RiskMetrics’s Exponentially Weighted Moving Average (EWMA), ARMA-GARCH with normal and Student’s t-distribution, and ARMA-APARCH with normal and Student’s t-distribution. These models have been applied to spot price returns of 7 commodities (3 nonagricultural commodities i.e aluminium, copper, gold; and 4 agricultural commodities i.e soyabean, guar seed, chana and cardamom). For these seven commodities, daily Value-at-Risk (VaR) has been computed for different time horizons and VaR exceptions at 99% confidence interval have been calculated. The forecasted values of VaR are compared with the actual returns data and the number of times returns data exceed the VaR is monitored. For traders who are long, an exception occurs when the return for a particular day is less than the model predicted VaR. Correspondingly, for short position holders, a VaR exception occurs when the return on a particular day exceeds the model predicted VaR on that day.
The paper finds that ARMA-GARCH model with Student’s t-distribution is the most preferred model to forecast 1-day VaR for five commodities except gold and guar seed. ARMA-APARCH model with t-distribution was able to model the VaR exceptions for gold while the same model failed for guar seed. In fact, it can be concluded that volatility of guar seed is too high to be captured by any of the models considered for the study. These models are then compared on the basis of two metrics i.e. number of VaR exceptions and loss function. Commodity prices tend to exhibit higher volatility during certain times of the year due to supply and demand mismatch as well as seasonality in production and consumption. Hence, this paper also tests whether VaR exceptions have any relationship with seasonality in spot prices.
