EVT

[higaurav]

Hi David,

Q36 in the practice exam 2008

Which of the following statements about Extreme Value Theory (EVT) and its application to value at risk
are true?

It mentions in answer for this that "For empirical stock market data, standard value at risk estimates at the 95 percent
confidence level tend to be fairly accurate, and generally only becomes inaccurate at the 99.5 percent
confidence level and beyond"... I am not clear how can estimate be inaccurate at the 99.5% level ?

Regrds,
OM

 

[David Harper CFA FRM]

Hi OM,

I checked the source (Lampros et al), they don't say that anywhere I can see. I have often seen, in practice, the EVT distributions tend to start at 99.5% and 99.9% (e.g., Carol Alexander has used 99.9% as an example of where EVT distributions start). But I do *not* know of empirical data to support this point.

The answer means to suggest that, for stock market data, the normal "body" is mostly accurate and then, the fatter tails start to become relevant at the extremes of the tails. So, I don't infer anything magic about 99.5% except some conventions.

But i would agree with the question that 95% doesn't qualify as extreme; my impression is that "extreme" (defined as: outgrows usefulness of normal) starts at 99% and beyond

David