Adjusting VaR by a factor, depending on backtest results

[yanko]

Hi,



[url=/Show%20Post.aspx?PostIDKey=151207]Aaaron wrote[/url] a couple of years ago, that VaR which doesn't go up with backtest breaches, since they cluster regardless of model quality, is useless. This is kind of obvious and in my experience it's a quick and dirty solution which performs really well.



I was wondering, if this kind of VaR-adjustmen is sort of an "industry standard"? I am looking for arguments to support this method, but can't really find anything. I suppose it's not scientific enough to write a paper about it.



Can anyone point me to a paper/book/article/interview, which suggests this approach is state of the art, i.e. brute force make-fit, risk modeling?



Thanks,



yanko