After JPM announced a $2B mark-to-market loss in a credit portfolio, U.S. financials suffered across the board. With such company-specific news, it's difficult to understand the correction in the sector as a whole - until you consider the post-2008 regulatory questions still looming in the background.
Here's the obligatory quote du jour, coming from Sen. Levin, one of the original authors of the Volcker Rule:
The enormous loss JP Morgan announced today is just the latest evidence that what banks call "hedges" are often risky bets that so-called "too big to fail" banks have no business making.
This statement does not reassure me that regulators have a clear grasp of the relevant questions here - what is a hedge, what is a bet, and how to tell the difference. Perhaps we can chalk this up to the whimsy of politicians and their followers, for whom dedication to an ideal trumps actual practical relevance any day of the week.
To me, the difference is simple. A hedge reduces the magnitude of some net risk exposure. It could be a credit spread DV01 - as intended in this case - or a delta, or a vega, or a beta, or whatever.
In this case, they're dealing with credit spread duration, or "CS DV01." This number represents how a 1bp change in the credit spread of an issuer -- and keep in mind, credit spreads are literally a rewriting of default probabilities over time -- would impact a particular credit derivative or bond. This allows you to aggregate the risks across different bonds and CDS of varying maturities into a single sensitivity, which then, like a bond's DV01, can be hedged. And, like a bond's DV01, a hedge is any situation in which you can diminish the DV01 associated with a particular spread move.
Sen. Levin's remark makes the distinction sound more complicated, but I don't think it needs to be. In the course of normal business, an investment bank will end up with exposures. Naturally, their decisions of what to hedge will be predicated on their market views. Proprietary trading is not synonymous with taking views!
That said, I think I have an idea what may have gone on here. From the sounds of it, they were hedging their CS DV01 on a bunch of CDS they sold to clients. They did so by way of CDX.
Now, I don't know whether they were using tranches or the whole structure. The primary difference is, with tranches, you are also making (extremely significant) bets on correlation of defaults. But either way, one thing is excruciatingly clear: the hedge was not a perfect match. Otherwise, these collateral questions wouldn't be relevant: for every dollar they lost on the hedge, they would make at least a dollar on the original position. Thus, the question of collateralizing the hedge would boil down to transferring capital from one account to another.
The reality is, whatever positions they had due to client flow, there's no way they just happened to exactly match the weightings of a CDX basket. It seems what they did is approximate their total CS DV01 from the flow desks, and duration-match an equivalent amount of CDX. Thus, seemingly, they reduce their CS DV01 exposure.
But as they learned, this isn't really the case. At first blush the problem seems to be that, although they decreased a net exposure, they increased a gross exposure. But that's not exactly the case. The problem is, you can only safely add up numbers that are similar in some manner - to be a true hedge, ideally you would only sum CS DV01 by issuer.
The mistake can be more easily understood by converting it into equity terms. This move is the equivalent of going long a bunch of names and putting on a beta hedge (short SPX). So really, it's an outperformance bet - the hedge doesn't actually neutralize the risk in the portfolio. It simply shifts it to a slightly different location.
If JPM's intention was to decrease their credit market exposure, they would need to understand where the big exposures lie, and in real time or near-real time. But at a firm the size of JPM, this is highly nontrivial. Their trading desks cover the globe and trade a huge variety of instruments.
Naturally, I can understand how they had trouble doing this. The $24,000 question in the reader's mind, I imagine, is: what would I do if JPM allowed me to try to fix the problem with my software? I'd suggest running CS DV01 analysis across JPM in real time and allow the risk managers to build their hedges with issuer-by-issuer, real-time precision. But, I would imagine, like many investment banks, JPM hasn't yet sought out a unified, real-time technological platform -- and until they do, I would imagine "mistakes" like this will continue to happen.