Importance of better risk models

I have been writing for some time now about better risk models (my SSRN paper is here and my blog post on that paper is here). Phorgy Phynance has a fascinating graph about the difference between the normal distribution and a fat tailed (stable) distribution in computing the 1% daily VaR for the S & P 500 going back 80 years (hat tip Felix Salmon).

His second graph shows that using stable distributions would not by itself have provided any better warning during the boom years of 2005-2007. But the early warning that it provides from early 2007 onward is truly impressive. During 2007-2009, the stable distribution VaR gives about 6-9 months advance warning about where the normal distribution VaR will be. In the world of financial markets, that is more than enough warning even if you were holding a bunch of illiquid stocks.

This also means (via the Merton model) that one would have had several months advance warning of corporate credit market stresses. That good models are better than bad models might look like an obvious statement, but too many people that I talk to seem to have convinced themselves (or let Taleb convince them) that all models are useless in times of crises.

But the performance of even the stable distribution during 2005-2007 highlights the need for using data going back several business cycles. This is also a point that I emphasize in my paper.

Posted at 12:13 pm IST on Fri, 7 Aug 2009         permanent link

Comments