The objective of this paper is to investigate the performance of different VaR models in the context of risk assessment in hog production. Potential pitfalls of traditional VaR models are pinpointed and proposals to solve them are analyzed. After a brief description these methods are used to calculate the VaR of the hog finishing margin under German market conditions. In particular we apply Extreme Value Theory (EVT) to our data and compare the results with historical simulation (HS) and the variance-covariance method (VCM). Hill's estimator is used to determine the tail index of the extreme distribution of the gross margin in hog finishing and farrow production. A bootstrap method proposed by Danielsson et al. (1999) is adopted to choose the optimal sample fraction for the tail estimator. It turns out that EVT, VCM, and HS lead to different VaR forecasts if the return distributions are fat tailed and the forecast horizon is long.