Was JP Morgan’s Loss Caused By “VaR” & Fancy Risk Management Orthodoxy?
May 14, 2012 2 Comments
Idle, probably ignorant and ill-advised question I’m mentally toying with as the whale-a-lanche dies down: Was JP Morgan’s loss caused by the “VaR” approach to risk itself?
I keep coming back in my mind to the fact that JP Morgan’s announcement of the $2 billion loss was paired with an admission of an ‘error’ in their VaR (and/or some ‘new’ risk measurement model that impacted their VaR calc). This implies they were related. But, to ask a perhaps naive question, why would they be related? Why would CIO’s trades have anything to do with VaR or ‘models’?
A lot of electrons have been harmed trying to tie The Financial Crisis™ to models. You’ve read articles and books trying to blame it on Black-Scholes or the Gaussian copula or string theory or some such. Well, although it is something of a poor hick cousin, VaR is a model number too. Sometimes in pop finance journalism, ‘VaR’ gets translated as ‘risk’ for the general public. But that is misguided; VaR is not ‘the risk’ of your portfolio. It is some opaque number that’s, like, um…about your portfolio, which you (or rather some faceless team upstairs) have to calculate, black-box style. Oh, and it is always wrong, every day. (Remember the first rule of RWCG: All Large Calculations Are Wrong.)
More to the point here, even if calculated ‘correctly’, VaR is certainly not ‘the risk’ of a portfolio as a trader would see it – it is not a measure of how much the portfolio will move if prices do such-and-such. You can’t meaningfully ‘hedge’ your portfolio with any accuracy ‘based on VaR’; it doesn’t give you any hedge ratios per se (and that’s, like, the first thing you’d need). What VaR is, is a (simulated) measure of its almost-worst-case-scenario losses: a ‘tail’ measure. This is true of VaR and a host of similar other VaR-like ‘tail’ metrics that banks calculate – and in some cases are forced to calculate and pay attention to by regulators, the government, and Basel – because those ‘tail’ based metrics, which seem to be a staple of modern risk management orthodoxy, are so ubiquitous now that they have been written into the rules governing a bank’s capital requirements.
Now. Let’s go along with the claims elsewhere that the core of JP’s trade was short HY/long IG. Ignore tranches, and ignore whether the IG was meant as a DVA hedge or simply was used to pay for the short. Question: Why was the short HY needed in the first place?
Well duh. As adult-nonBritish-male-named-Jamie Dimon says, it was put on to hedge some unspecified ‘overall credit exposure’. Ok fine. So you have credit exposure (long) and you put on a HY short against it. Question 2: How exactly do you screw up the ‘risk’ if all you’re doing is shorting against your longs? What does some ‘risk model’ have to do with anything? There’s nothing overly complex about calculating ‘how long are we all this stuff?’ and thus ‘how much short HY might we need to offset it?’. Craig Pirrong alludes to them maybe getting simulated inter-security correlations wrong but I don’t get it. What do those correlations** have to do with sizing a simple macro delta hedge of a net long position?
Unless what they were ‘hedging’ was not their (actual, market) risk, but a black-box measure like their VaR. Or (more precisely), they had come up with a trade hedge package they kinda liked, and fed it into VaR to get an answer as to whether it ‘reduced risk’, and when it did, used that as a reason to justify the trade. Either hypothesis leads to a story of them feeding some trades into a black box, and results coming out saying ‘Yes These Trades Reduce Risk. Do Them!’ And so they did them. If this is the explanation, it’s hard to for me to believe they could be so…dumb as to have drunk their own model kool-aid (I just read somewhere that JP invented VaR); perhaps, at best, they were trying to do some fancy optimization/overtuning of hedges using VaR.
Either story though would point to the real culprit behind this loss being the methodology and mentality that treats VaR – and things like it – as a highly useful risk-management measure that senior management should focus a lot on and actively use in day to day management and trade selection. I’ll give you one guess what I think of that mentality. In short, there’s a chance it would have been easier for JP Morgan to see how dumb their trade was if they didn’t have “VaR” or similar black-box model-risk-metrics to distract them in the first place.
Of course, I fully expect one outcome of this incident will be an increased regulatory push for calculating and backtesting all sorts of better, stronger, more perfect tail-type metrics. This is another rule of RWCG, policy-induced failures inevitably lead to calls for more hair of the dog.
(Charles Rowley has related thoughts on VaR.)
**There is a second possible meaning of ‘correlations’ in this context, which I’m ignoring here, since if that ‘correlation’ is the culprit, everyone has to rewrite all their commentary anyway. (UPDATE Speaking of, someone pointed me to this Zero Hedge post which appears to tell such a correlation-based story. But, I read it and cannot understand their steps 10-11, postulating as an explanation of their IG long that JP ‘had to’ go long(er) to rebalance a certain tail hedge. If anything I would expect such rebalancing of their hypothesized tranche to mean going shorter IG not longer, but maybe I’m tired, if you know better feel free to correct me…)