Fat-tailed distributions (leptokurtic distributions) have more extreme values in the tails than the normal distribution. When using VaR models that assume normal distribution, the actual risk is underestimated because extreme losses occur more frequently than predicted by the normal distribution.