EVT incorporates the notion that extreme losses are more likely than standard distribution would suggest. In other words, empirical distributions often have "fatter" tails than assumed distributions. EVT therefore treats the tails of assumed distributions differently than the rest of the distribution by assuming this region follows a different distribution, such as the generalized Pareto distribution (most common). Rather than estimating losses up to, say, the 99th percentile as in value-at-risk (VAR) analysis, EVT estimates the expected value of the losses beyond the 99th percentile in the "fat" tail of the distribution. This approach yields estimates of operational risk that are much larger than standard VAR approaches. |
