The financial crisis has provoked a lot of naval gazing and blame in different proportions. One of the factors that got a lot of blame was "Value at Risk" or "VaR" for short. This statistical measure was presented by some as a cure-all for identifying areas of risk and so when things went badly wrong they pointed at the model they had used and said "It wasn't my fault; it was the model and VaR that was wrong." Some added that according to their model the sort of losses that were recorded would only happen once in ten thousand years.
The most common use of VaR is to gauge what is the worst loss 99% of the time. Note the 99% of the time. That means that 1 day in 100 you would expect to lose more than VaR. At StatPro we use 520 days of history to help determine VaR. We don't try to extrapolate the worst loss 99.9% of the time (1 day in 1,000) as this falls into the realm of pure guess work and as such is unhelpful.
If you have a portfolio with 20 assets, knowing the individual VaR for each asset is very useful, knowing the VaR of the combined portfolio is even more useful as this can reveal how risky assets put together can actually reduce your overall risk. Given that generally if you have less risk you will probably have less potential performance and vice versa, understanding the level of risk in a portfolio enables the fund manager to increase or reduce his overall VaR by carrying out "what if?" scenarios. There are so many numbers for a fund manager to check, it is easy to make a mistake and not see an underlying risk. VaR helps keep things in order.
In many ways the absolute level of VaR calculated is not as important as the difference in VaR between the assets in your portfolio or the relative change in VaR of your assets or portfolio over time.
It is also important that your model for producing VaR works properly for the type of assets you are investing in and it is important that all the assets in your portfolio are covered by the same model. There were many people who calculated VaR on only 95% of the assets in their portfolio as their model did not cover more complex assets.
Finally, to have an idea about what the potential risk is from extreme events (i.e. the 1% of times when the result is worse than VaR) it is important to run scenario tests. At StatPro we run approximately 1,000 scenario tests for all the 200,000 assets covered every day for our clients. These scenarios cover events like the 2001 crash, the 2008 post Lehman events and Q1 2009 and many others. (Read more in our whitepaper section) The prudent fund manager will then see a realistic potential outcome for his portfolio in the event of a sudden crisis. By comparing the VaR and scenario tests, the fund manager can make an informed judgement about his portfolio's risk. He should also remember that it is just a statistic to help quantify risk and not a prediction of risk.