This paper evaluates the use of parametric generalized autoregressive conditional heteroskedasticity (GARCH) models for risk-management purposes on the daily spot prices of natural gas traded on the NewYork Mercantile Exchange. The model set includes nine linearGARCHmodels with alternative distributions and five nonlinear GARCH specifications based on the normal distribution. The out-of-sample volatility predictive performance is studied according to value-at-risk loss functions that mimic the preferences of portfolio managers for penalizing large forecast failures and opportunity costs from overinvestment. The results suggest that the GARCH model with generalized exponential distribution appears to outperform the competing models both in terms of failure criteria and risk-management loss functions. The model evaluation also indicates that it is important to account for the opportunity cost of capital.