« Defending an argument against VaR |
Main
| Well Intentioned is often not so in practice »
May 25, 2007
A reason to defend VaR
In continuing on with the debate on VaR, I agree with some of the comments that were made to the previous blog article and I accept I was hard with my critique. Sure VaR may not be the panacea of risk measurement that it is proclaimed to be but it has some good attributes, starting with the way in which it translates relatively well from market to other disciplines of risk such as credit and this is kind of neat. What is the saving grace is that it is parametrically based and focuses on losses. Up until VaR became a relative standard some banks used what I would classify as exciting alchemy to define the chemistry of causality for their potential loss with the occasional Fibonacci related approaches really taking us to the exotic edge of measurement. Many organisations didn't adopt anything cognisant at all and some still operate this way.
What is important to the industry as a whole is that it finds a reasonable measure of exposure that can be commonly accepted, stochastically sound (whatever that is) and is not refined for one investment strategy over another otherwise transparency and exposure rating is lost.
VaR meets the majority of the criteria to take the world of capital measurement global however many analysts claim expected short fall is a better approach for tail work. Expected Short Fall is the conditional expectation of loss given that the loss is beyond the VaR level and it again is still working in a coherent domain.
The real catch that so many oversee is not that VaR underestimates or overestimates exposure, it is the requirement of including correlation and aggregation that is complex and without these two the VaR calculation is likely to be lower in weight. When a risk analyst drops aggregation from the calculation only to do it post simulation the VaR calculation could lead an organisation to position where it might measure exposure in a manner that lacks subadditivity. Subadditivity implies that aggregating individual risks does not increase the overall risk and that is mathematically expressed in the following way. For all X and all Y, p(X+Y) <= p(X) + p(Y)
What we have here is a deficiency where a decentralised risk framework may be flawed because VaR is calculated on individual portfolios which cannot be aggregated to produce an upper bound of loss and Rudiger Frey and Alexander McNeil published a paper on this quite a few years back when VaR was all the rage in Europe. Another issue with VaR is that investment portfolios have become exponentially complex and contain exposure of many natures including a blend of credit, market and operational risk. This can be seen in the new hybrid derivatives or mitigation alpha investment strategies that are now popular and which for what its worth are fantastic investment approaches with a real purpose but from a risk perspective; is the overall exposure being measured correctly? Is the inference between the arbitrage space of the pure risk disciplines being reported? Perhaps the real concern is that advanced investment strategies are being used by an ever growing investment community and I fair I am not the only one with this fear as many regulators are showing apprehension over the increase in popularity of hedge funds globally.
To close off on this, I see VaR as a stepping stone to a more sophisticated technique of exposure measurement and I agree with you entirely that there may be a better way but we are on a path and in reality the industry as whole has a long way to walk yet. Some banks are stuck on Basel I, some are adopting the basic approaches to capital estimation which is a tragic representation of failure, barely risk based at all and puts expression of loss and capital reserves as nothing more than a proxy of revenue. So now the industry has accepted VaR, we might all just be in a risk renaissance here where VaR is the Euclidean equivalent of risk reasoning just as Greek mathematics opened the space for geometry 320BC and while banks are moving onto risk based capital regimes, VaR might be a likely vehicle to embed a risk based approach in the financial sector and while its doing that, I will defend it.
Posted by CausalEvents at May 25, 2007 04:14 PM
Dear Mr. Davies,
thank you for your follow-up on your last article. I don't really know how to contribute further to the discussion. Maybe it's not 'to VaR or not to VaR'.
I think with you that VaR makes sense, and I think with you too that the value of VaR depends highly on the implementation.
We will have to look at what we use VaR for. VaR was invented to have a day-to-day overview of the actual risk position under normal market situations. This generally works fine, I think, in the investment banking environment where it was invented at the time it was invented.
The Basel committee's intention is a broader one though. I don't know whether VaR should be the only measure, when it comes to stabilizing the international banking system in non-normal situations. Referring to your article here: under normal circumstances your correlation matrix is important. But correlations are not always causal relationships. So they might vanish when the markets get wild.
And I think that this is what Chris Whalen wass referring to. (Quoting Nassim Nicholas Taleb means automatically that you are in the realm of the Black Swans, the not-normal, the sudden change, that generally is not covered by VaR.) VaR is not for non-normal situations. Maybe VaR is not good for non-normal markets neither. Enron's Jef Skilling used the word VaR quite often.
What is VaR in relation to hedge funds, with their sometimes wild portfolio structures?
So maybe the discussion has to fork. One discussion might be: what is the use of VaR? And how do we implement it properly? (And this might be extremely difficult.) The other discussion is: what kind of risk measures do we need to stabilize the international banking system. Maybe VaR should be part of it, I don't know. At least in my view it is not enough.
Posted by: Johan Steunenberg at June 4, 2007 06:18 AM
I agree with Mr. Steunenberg that VaR is not enough. It are extreme events that bankrupt a bank or institution not normal events. To calculate the Economic capital stess testing is better. There is also another problem with VaR and that is related to the models that we use, we make so many approximations that the results are not accurate (backtesting will help in finding the performance of internal models). Still models break down and in the end it is human judgement that is most important. I think that the organizational culture that rewards excessive risk taking needs to be curtailed.
Posted by: Deepankar at June 29, 2007 05:12 AM
Post a comment
