Modeling Value at Risk of Agricultural Crops Using Extreme Value Theory

Modeling extreme risk in returns accurately due to volatility in agricultural prices is of utmost importance for both the farmers and policy makers. In this study, we compare and contrast performances of four EVT based methods in modeling extreme risk and VaR of three crops: US corn, soybean and wheat using daily frequency data covering the period 1986 to 2010 (i.e., using a total number of 7796 observations). Based on a rigorous process of backtesting, we conclude that the conditional GPD-normal model performs better than DPOT, conditional GPD-sst, and unconditional GPD. This is because the agricultural commodities have their own unique properties, such as, they are less risky, have seasonality effect, and move in response to both supply and demand information, which makes it quite different from other financial series. Therefore, relevant stakeholders should take into account these properties in order to improve the accuracy of forecasts.