Securitization

I'm reading Too Big to Fail, Andrew Ross Sorkin's account of the financial crisis. It's good so far, but it stuns me that Sorkin of all people doesn't understand how securitization works. I want my readers to understand, so I will teach you.

It's actually not very complicated in broad strokes. I'll describe how a securitization might work and then I'll explain why it can be a useful thing to do.

I start by buying a bunch of mortgage loans. (This is a very pre-crisis example. Today there is essentially no market for private-label (that is, not Fannie or Freddie) residential mortgage-backed securities. But since residential mortgage-backed securities, RMBS, basically destroyed the world, I might as well use them for my example.)

So anyway I've got these mortgage loans, which I've purchased using financing from a "warehouse lender" (don't worry about it). I put the loans in a special-purpose vehicle (a legal entity that has no other purpose beyond its use in the securitization), which I own. The SPV issues notes (bonds) to investors, and I use the proceeds to pay off my warehouse lender.

Now we've got a bunch of notes owned by investors. The principal and interest on those notes will be paid with money coming from the mortgages. Money in, money out. Already the investors have gained an advantage from the securitization: the pool of mortgages is diversified, so a single default isn't going to be ruinous. By contrast, if an investor buys a whole mortgage loan, he loses a lot of money if that particular homeowner defaults. Much better to own 1/30th of 30 mortgage loans.

But that's not the real magic. The real magic is that the notes are "tranched." What this means is that they are divided into different classes ("tranches"), and each class has a different level of priority. In other words, if there isn't enough income from the mortgages to pay the principal and interest on all the notes, they don't suffer equally. The lowest tranche gets hit first (it is in the "first loss position"), and then the next tranche, and so forth. (The analogy people use is a waterfall—as water (cash) flows in, it fills one tank (the highest tranche) completely before it spills down to the next tank, then the next, then the next. In practice this is not exactly how it works, since even the lowest tranches need to get paid principal and interest until the losses wipe them out, but it's a pretty good metaphor.) There is a fair amount of loss-absorbing capacity before you get to the highest tranche. The highest tranche, then, can be much safer than the underlying assets as a whole—it can even be safer than the safest of the underlying mortgages. In fact these notes were designed so that the top tranche would get a very high rating from the rating agencies.

An important point here is that while everyone likes to say that mortgages were "sliced and diced" in the process of securitization, this is a highly misleading metaphor. In fact the mortgages were pooled together and the resulting flow of cash was divided without regard to which mortgages generated it. It's not as though the top tranches were backed by the best mortgages in the pool, while the shitty tranches were backed by subprime garbage. They were all backed by all the mortgages, it's just that the high tranches got paid first. If by some weird circumstance the subprime mortgages happened to be the ones that didn't go into default, the top tranche would still have gotten paid first. The mortgages weren't divided (as "sliced and diced" implies), they were pooled.

Of course there's no free lunch—the safety of the highest tranche comes at the expense of increased risk in the lower tranches. But the point is that you can engineer the risks to meet investor demand. If investors really want AAA-rated bonds, you can generate them, whereas no mortgage would ever be AAA-rated. And that was basically the game. You could generate AAA bonds from mortgages of almost any quality. So for instance, infamously, subprime mortgages were put into securitizations that issued AAA securities.

I want to emphasize that this could have worked. With enough loss absorption built into the structure, you could absolutely support AAA bonds with subprime mortgages. (In the extreme case, imagine a senior tranche so small that it will pay out so long as there is a single performing mortgage in a pool of hundreds.) But of course the loss absorption generally proved to be insufficient, the bonds had to be downgraded, and so here we are.

The problem was twofold: mortgages were shittier than most people anticipated, and their shittiness was more highly correlated than most people anticipated. Think about it this way—if mortgage performance is perfectly correlated, then the securitization has accomplished nothing, because all of the mortgage loans will fail simultaneously and all of the tranches will be wiped out. You might as well not go to the trouble and expense of tranching. Of course in reality mortgage loans weren't that correlated, but their performance was much more correlated than you might have expected based on historical data. (So for instance, historically mortgage performance in Houston (a function of oil prices) would have been mostly uncorrelated with mortgage performance in Los Angeles (a function of... whatever). But this time the housing bubble was national and so was the crash.)

Anyway, now you know more about securitization than Andrew Ross Sorkin does.