Changing my model July 11, 2012 at 7:01 am
A recent Bloomberg story about VAR model changes points out
Wall Street firms routinely give only broad outlines of how their mathematicians calculate VaR, according to data compiled by Bloomberg, and almost nothing about changes in statistical assumptions or the prices they choose to feed into their models… The skewed comparisons can leave investors guessing about whether the potential for loss is rising or falling, according to risk analysts
Um, yeah, and?
OK, let’s be a little more precise. There are three types of VAR model changes:
- Data series changes. Here the model is simply updated with new market data. Some firms used to do this infrequently (e.g. quarterly), but the new Basel standards require this to be done at least monthly. Depending on how different the market is in the new period, and how long the data series used to calculate VAR, this change can be material; usually, though, it isn’t.
- Risk factor changes. Here a new risk factor is added (or occasionally subtracted). The mapping of products onto risk factors is changed at the same time to accommodate the new risk dimension. This change might be immaterial but often isn’t as typically the change is made to better measure risk.
- True model changes, such as going from a variance/covariance VAR to a historical simulation one. These are almost always material.
None of these changes have to be disclosed, and they never have been since the dawn of VAR in the 1990s. Moreover, even if they were disclosed, it would be very very difficult for an outside analyst to understand exactly what was going on. Modern VAR models are so complex, and the portfolios they are applied to have so many risk factors, that getting your head around a single bank’s VAR requires working with it for months if not years.
So what? We, for me, two lessons.
First, VAR tells you little about relative risk. If firm A’s 99% 10 day VAR is $50M and firm B’s is $80M, you can conclude nothing about the relative riskyness of A and B. Indeed it is perfectly possible to find two different VAR models, both approved by regulators, one of which says A is riskier than B and the other says B is riskier than A.
Second, it is astonishing that investors are only just working this out, if indeed this really is the case. Or perhaps Bloomberg is trying to make a story out of something that has been well known in the risk community for twenty years…
So, with apologies to David Bowie
(Turn and face the SEC)
Don’t want to be a richer man
(Turn and face the SEC)
Just gonna have to have a different model
Time may change my P/L
But I can’t trace time