Very good and refreshingly open presentation given by Alan Smillie of Citi Group on whether Value at Risk (VaR) has been discredited as a risk management tool (see earlier post for Taleb on VaR).
He started by generically describing VaR as being composed of both:
- A statistical model of the market
- Pricing models/sensitivities to translate the market movements to P&L
He said that much of the recent losses/write downs at Citi are in ABS CDOs at levels of 10x/100x larger than those predicted by the VaR models that they use. He summarised by saying that whilst VaR has not performed well, it should not be dismissed since their experience is that their losses (and inaccuracies in VaR calculation) were not (mainly) due to failings in the statistical model of the market, but rather in major failings in the pricing methodology for pricing ABS CDOs.
It seems that the traders and risk managers at Citi regarded ABS (Asset Backed Securities, mainly mortgages) as equivalent to a bond, and so they regarded an ABS CDO as equivalent to a CDO of bonds. In fact the ABS was itself a securitised product, so effectively they were dealing with a CDO of CDOs (CDO^2). This did not offer the diversification of risk to justify the AAA rating given to the ABS CDOs – apparently the ABS prices in each CDO were 90% correlated. Alan clarified that given the problem was in the pricing model, not in the model of the market, better scenario analysis would not have helped Citi to avoid these losses either.
Alan refered back to an article in Risk magazine in 2006, saying that banks were not experiencing any “exceptions” (breaches of the VaR loss level) at the level that should be expected (2 to 3 per year for 99% VaR). The article suggested that banks were therefore being charged too much regulatory capital and this should be reduced. He said that Citi experienced 20 VaR exceptions in 2007, and expected much more in 2008 and as such VaR is not working well given the current market volatility.
He expects that future risk capital calculations required by the regulators will be based upon VaR combined with subjective, non-probabilistic stress testing (apparently something that Deutsche Bank have been doing internally for years according to a later speaker). He didn’t seem to address the issue of how to avoid pricing model risk, but it was a good talk with a lot more openness over Citi’s losses and problems than I expected.