This turned out to be a much more lengthy post than I intended. I’ve condensed it and left notes. You can just read the first few paragraphs for the general idea, but I suggest you read the notes as well because they will give you the full story.
I listened to This American Life while running two days ago on Magnetar and their CDO bet. And their story is very one-sided on how Magnetar played a hand in the housing crash, ignoring the actual mechanics of CDOs and Structured Credit. So I thought I’d share my view.
I also got into a bit of a discussion at work yesterday on how this obviously (to me) was a bet on correlation increasing (Note 1). The point that was made to me was that, “any time someone risks 10MM to earn 1B there are going to be questions asked, and when 99.99% of the world can’t understand, let alone explain Structured Credit, that leaves people standing outside the castle with pitchforks and guns.” The SEC moved on Goldman yesterday, and the drama is just starting to unfold.
And so is the misguided view on CDOs.
So what did Magnetar, Paulson, et. al. do? One of two things (or both, perhaps):
TRADE 1:
They were long the equity tranche (buyer of equity risk, or seller of protection), short the mezzanine tranche (seller of mezz risk, or buyer of protection).
Think of it in bond terms for simplicity. At t=0 you (the hedge fund) bought a really risky bond for 10 cents (the equity tranche) and sold a much less risky bond (the mezz) for 90 cents. At t=0, this trade was the stupidest trade in the world to the real money insurance companies, banks, and pensions, because you just bought a risky, unrated bond that could definitely default and sold a risky bond that was unlikely to default (house prices always increase), yet you were paying too much for it.
From Magnetar’s perspective, you paid very little for the equity tranche and received a chunky yield that could fund your short. Magnetar was paying out a steady, pricey stream on the mezz tranche (albeit in basis points), and waiting for the housing bubble to burst. (Note 3)
What Magnetar did was (1) throw out the Guassian Copula Structured Credit model and figure out what correlation really meant and (2) realized that banks were in a bind b/c they needed equity investors in these deals because no real money account wanted to take equity CDO risk.
If Magnetar could get involved in a deal in which they knew the whole thing was likely to go bust (in Structured Credit terms, correlation to 100%) they could profit immensely, and get this, since they were the deal sponsors (they bought the equity tranche) they had a heavy say in what bonds went in (Note 4)! They were net flat, because the cash in and out was the same on both tranches.
This was the smartest trade of the 2000’s.
TRADE 2:
They were long the equity tranche (buyers of equity risk) and delta hedged (Note 2). They were strictly long the CDO and hedged. They made more on their hedge specifically because at the time of pricing, delta was very high and the offsetting index hedge amount was very large. If a bank or a pension was long the same equity tranche, they would have been hedged the same exact way.
In this case, Magnetar was betting that correlation would increase. They hedged the trade at appropriate delta levels, which meant if they were long the equity tranche w/ 10x leverage, at t=0 you would buy 10x of protection against that tranche.
This was another smart trade, but not as clean as long equity short mezz.
Notes:
(1) Think of correlation like a minefield in a bay (credit goes to Merrill Lynch for the idea a while back). This means that w/ low correlation, the mines are disperesed evenly all over the bay. However, when correlation goes up, the mines are all clustered together. So what does that mean for an equity tranche? Well, you the investor are 1 of 3 yachts, sailing through the bay. You’re the first yacht puffing through, and if you hit a mine, you blow up. And you’ve presumably taken care of the mine problem so yachts 2 and 3 can sail through w/ east.
In a low-correlation environment, the mines are scattered everywhere, and there’s a good chance you hit one. However, when correlation goes up, all those mines are clustered somewhere — but you can probably random walk (sail) through the minefield-bay w/out hitting a mine. This means an equity tranche has more value (relative) b/c it means there is a chance that you don’t hit any of the mines.
(2) Now, how do you get to that delta? That’s where the model comes in. Tranches are priced off correlation — a funny concept that doesn’t tie back to what correlation means in statistics, but it basically comes down to, an equity tranche is long correlation (Note 1). One of the first things you learn about the equity tranches in any CDO, synthetic or cash, is that it has a huge delta — i.e. you buy the equity tranche w/ a delta of say, 10x. That means if you are long the equity tranche, the model implied hedge ratio is 10x that amount. You buy the equity tranche (long risk) then must buy TEN TIMES that amount in CDS as protection (short risk). Risk and volatility are huge in the equity tranche, and that means you have to buy a lot of CDS to hedge your long equity tranche exposure.
(3) Correlation in this case went from 30% to 100% very very quickly, and there was no chance for the CDO waterfall to trickle along slowly. The deal went from solvent to bust. From the real money perspective, it was reasonable that in a risky deal, some defaults occur, and some of the lower tranches get hit, but it was outside the mindview at the time that housing could decline so rapidly and fiercely. Magnetar saw that people weren’t thinking about this and capitalized on this.
(4) Now, about the selection of bonds for a CDO. At t=0, the equity investors (hedge funds) and senior investors (real money insurance and pensions) GOALS WERE ALIGNED. They both wanted the most yield possible in a CDO. There was no fighting by the senior investors over the bonds that were included. They wanted the most yield on their “safe” senior tranches. They thought their tranches were safe because of the Guassian Copula model and the fact that housing never goes down with correlation of 100%.