Hello, It seems like in some of the material, ES is calculated using ranked historical VaR and using n-1 (observations beyond the confidence interval) observations to calculate ES and in some other GARP questiosn they simply use n ranked observations (observations including confidence interval)? Thank you, Glen

Hi Glen, Shannon and I were recently discussing this here @ http://www.bionicturtle.com/forum/threads/p1-t4-30-expected-shortfall-es.5700/#post-16577 My summary view is: In regard to discrete HS VaR, you DO find different approaches. In particular, if n = 100, Jorion tends to use the 5th worst loss for 95% VaR while Dowd uses the 6th worst. I prefer Dowd because (1) it's more conservative [a lower VaR; ie, if VaR is on a + scale, i mean] and (2) it's more in the spirit of characterizing VaR as "significance% of the time we do expect the loss to exceed the VaR". However, more importantly, GARP is totally, keenly aware of the fact that there are several definitions of the quantile in a discrete HS, valid include the 5th, the 6th and interpolation between the two. But I do not perceive such an issue with expected shortfall (ES). I perceive neither theoretical ambiguity, nor am aware of any recent ES FRM questions that should cause confusion. If n = 100, the 95% ES is the average of the worst 5 as the worst 5 are unambiguously the 5% tail. Even in the XLS, when the quantile straddles discrete outcomes, it's easy to find the ES. I think the ES is unimpeachably clearly defined. Thanks,

Hi Glen, Thank you for your feedback. It's greatly appreciated. I'd like to recommend that you take a look at this thread and change your username in the forum: http://www.bionicturtle.com/news-an...mend-you-change-your-screen-name-in-the-forum Were recommending this for multiple reasons but one of them being security! Thanks, Suzanne Evans