[R-SIG-Finance] Framework for VAR allocation among traders

[Brian G. Peterson]

On Saturday 15 March 2008 08:37:46 kriskumar at earthlink.net wrote:
> My two cents on this... to paraphrase taleb using var to manage risk or
> allocate capital is like driving an airplane with a faulty altimeter.
> Probably ok in quieter markets but quite risky in these turbulent markets.

I recommend against using a normal distribution assumption at any time.  One 
of the reasons I like the Cornish Fisher expansion for incorporating 
non-normality is that it directly ustilizes skewness and kiurtosis, which 
have meaning independent of the VaR calculation.  Extending this to Component 
VaR (or ES) incorporates the higher co-moments as well.

I also think that measuring VaR (preferably without assuming the normal) and 
stress or scenario testing the portfolio are two separate and complimentary 
activities.

Kris has mentioned Taleb ("Black Swans"), and I can also recommend 
Rebonato's "Plight of the Fortune Tellers" as an excellent deconstruction of 
many current risk practices (without resorting to the math).

> I might be mistaken but I think  comp VaR is not additive?

Component VaR and Component ES are coherent risk measures (per Artzner), and 
subadditive.

What most authors call Marginal VaR is *not* subadditive.

> Finally if you are going to use this because of Basel-2 etc I think it is
> key to make sure that it is not gamed.

Wholeheartedly agreed.  Note that the regulators require both VaR calculation 
and stress tests, as described above.

Regards,

   - Brian