Volatility Forecasting and Value-at-Risk: An Application to Cattle Feeding

Abstract

Based on mean-squared error criteria, the overall conclusion of the volatility forecasting exercise mirrors that found in the literature: performance of any volatility forecast is both data and horizon specific. However, composite techniques, especially simple composites that combine both conditional time series and implied volatility forecasts, perform well here. Interestingly, correlations, not volatility forecasts, were found to be the dominant factor influencing various Value-at-Risk measures ability to forecast large losses in the cattle feeding margin. Variances and correlations developed using the JP Morgan's Risk Metrics method with a decay factor of 0.97 provided superior Value-at-Risk estimates among other well calibrated specifications. This research is one of the few known empirical applications of composite volatility forecasting and the first known application of Value-at-Risk in the context of agriculture.