Non parametric approach

[[email protected]]

Hi

In the non parametric approach, under "Historical simulation usng non parametric density estimation"
dowd says if we have 100 P/L observations, to obtain var at 95% confidence level,we need to take the 6 largest loss. Is that right?

Should't it be the fifth largest loss?

Regards
..

 

[David Harper CFA FRM]

Hi vinoth,

Technically either, and I've been quite back and forth with GARP on this (search this forum for many threads), so that from an exam-perspective please understand GARP is going to accept (in this case) either 5th or 6th. Jorion assumes the 5th largest, and would agree with you.

But i prefer Dowd's 6th largest because then: the other 5% (5/100) is cleanly in the tail. (it is also technically the case that interpolations between 5th and 6th are okay too).

So, neither is wrong (at least is the understanding we've reached with GARP after citing several sources). I hope that helps, David