The quality of statistical risk models is much lower than often assumed. Such models are useful for measuring the risk of frequent small events, such as in internal risk management, but not for systematically important events. Unfortunately, it is common to see unrealistic demands placed on risk models. Having a number representing risk seems to be more important than having a number which is correct. Here, it is demonstrated that even in what may be the easiest and most reliable modeling exercise, Value-at-Risk forecasts from the most commonly used risk models provide very inconsistent results.