Is VAR broken? Frequently debated, it is the equivalent of trying to determine whether the entire past 10 years of financial innovation has been a mere column of smoke. The answer is Yes and No. It really depends on the application and the user.
Let's take an example of a firm that used a 95% VAR so that 1 day in 20 the P&L should exceed the VAR number in terms of gains and losses. This firm saw over the end of 2007 20 days in 60 where P&L exceeded VAR and 18 of the 20 times it exceeded VAR in the plus column. Management chose to ignore this fact because gains are far more appreciated than losses and why stop a trader on a roll. Management consciously avoided tough risk management decisions - can we really blame VAR for this failure?
VAR is not a holy grail. There are many shortcomings due to the many assumptions in the number. But much of the blame for VAR should fall on the shoulders of management who did not understand or chose to ignore the many signals that VAR was giving.
skip to main |
skip to sidebar
Posts
Blog Archive
My website
About Me
- JJ
- Actuary, trader, portfolio manager, I know both sides of the balance sheet better than most. And having worked at a pension plan, insurance company and bank I have seen them all.
VaR is not broken. (apart from many simplifying approximations on the way to the bank).
ReplyDeleteIts value has been overstated. The TREND in VaR and the systems required to support the calculation of VaR are far more important.
Also, bank wide VaR is much more approximate than can be calculated for subsets of its exposures. When the crisis hit, (and previous ones) they typically were initiated in one area (for example securitized asset portfolios) Greater supervision and more accurate BI and simulations relating to key areas would have done more to protect the system.
.. on another topic, it is important to consider the "parent population" from which 2008 is regarded as a sample. Statistically it is possible to have increased no of VaR excesses. However if the parent population is all possibilities in 2007/8 of which a limited number actually occurred this argument can be made.
ReplyDeleteThe truth is the underlying population changes several times a year, volatility clustering is just a symptom. The use of the Gaussian distribution (as imperfect as it is) then becomes downright wrong! Forecasting is generally inaccurate how much more so can we build models using mean reversion and the like.
There is a difference between management playing by the rules of "best practices" and performing risk management. Reporting VaR is an act of compliance, not real risk management. Real risk management requires a much higher standard than the regulators. Everyone seems to forget that the regulators are only setting the minimum level of the bar.
It's analogous to learning to build a house by reading all of the municipal and architectural regulations and by-laws. Doesn't ANYBODY get it?