Skip to main content
6 events
when toggle format what by license comment
Aug 4 at 8:31 comment added Yves The ML estimates of the GP parameters $\xi$ and $\beta$ are known to be biased for small samples. Yet since the two estimates are correlated it may be more relevant to study the bias on a high quantile rather than on one parameter. My findings (using R code) are that the GP approximation of the tail of the Student distribution with $4$ dfs works quite well with a threshold chosen as the quantile with probability $0.97$.
Aug 2 at 16:26 comment added Alex Craft @JimB left and right tails should be estimated independently. Ideally, left estimator should use only <5% and right >95% of the data.
Aug 1 at 5:17 comment added JimB What do you mean by "without using the full distribution"?
Jul 31 at 10:04 history edited Alex Craft CC BY-SA 4.0
added 19 characters in body
Jul 31 at 9:20 history edited Alex Craft CC BY-SA 4.0
deleted 21 characters in body
Jul 31 at 8:34 history asked Alex Craft CC BY-SA 4.0