|
|
|
|
|
|
Name: Richard Hall
Affiliation: Algorithmics (UK) Limited
E-mail: richardh@uk.algorithmics.com
Date: 24 Jan 1997
Time: 06:37:46
Given the regulatory demands and the preference of many major institutions to use historic VaR, what degree of error should we be comfortable with? Using a three/five-year window for approximately one-hundred IR/FX risk factors seems to give very accurate results when comparing historic P&L against the usual VaR confidence intervals - but clearly this has no statistical validity.
A secondary phenomena where care must be taken is to recognise extreme cases, and anticipate the impact on limits when these scenarios mature out of the rolling time-window. Dramatic drops in VaR levels accompany such events.
|
|
