Salmon is getting close to the dirty secret: no one has an answer. The most they have are some interesting datapoints, factoids, and analyses.
At the risk of having someone prove me wrong, let me offer a high-concept explanation. And I invite amplification and corrections.
The short answer is that no one has the right data to perform any kind of reasonable statistical analysis because the historical data is not comparable. So what you have is tweaking, guesswork, and fingers in the air. And that applies to subprimes that wound up in fairly conventional structures. Imagine how the problem is multiplied when you get to CDOs, CDO squared and cubed, and synthetic CDOs that have a fair bit of underlying subprime assets.
The subprime market is a relatively new phenomenon. As recently as 1996, most mortgages to lower-income families came through the FHA, which offered fixed rate loans to qualified buyers. According to a GAO report, the FHA’s share of the home purchase market was 31.6% in 1996, and fell to 6.9% in 2005, while commercial subprime loans had only a 2% share in 1996 versus 26.0% in 2005. And that 26% figure understates the level of overall subprime activity, since 50% of the subprimes were “refis”, meaning they not for home acquisition, but to repay higher-cost consumer debt.
Now we have two problems with subprime data. The minor one is that it doesn’t go back very far in history, only to a small level of activity in the mid-1990s. And that paper hasn’t been tested in a serious credit downturn. Yes, the post-9/11 period was soft, but it wasn’t a real housing recession. The last one we had was 1990-1991, when you saw distress and banks chary with credit.
But the second, and vastly more serious problem, is that that the subprime loans made recently are simply not comparable to anything that preceded them. The superheated period was the latter part of 2005 and all of 2006. As Lew Ranieri said:
The subprime mess is simply – and first, I think the important thing to understand, is this a creature of a very narrow window. It starts at the end of the third quarter of ’05 and carries through, principally, the fourth quarter of ’05 and ’06. And what it is, is that in those five or six quarters, a series of attributes which were largely in existence already, took on a life . . . of their own and in combination, created risk layering, which on one hand, enabled many, many people to get into housing who might not otherwise have.
Now it may also be that loans made earlier, 2003, 2004, the earlier part of 2005, are also not comparable to earlier subprimes They may not have been as God-awful as the later stuff but still worse than historical patterns. And ironically, even if they were as awful, they still can’t be used as a basis for forecasting what the losses from the latest batch might look like. Why not? The housing markets were still rising, so the worst of them were refinanced, generating more fees and keeping zombie borrowers alive.
The reason it’s quite possible that even earlier subprimes don’t correspond to earlier paper is that, contrary to the image this mess has created, prudent lending to lower-income and dented credit borrowers doesn’t have to produce big default rates. In fact, as my colleague Doug Smith (no relationship) pointed out in a recent article in Slate:
Here’s a snapshot of the quality of loans made by two kinds of lenders to aspiring homebuyers who are financially strapped: The record for the first kind of lender is that one out of every five or six borrowers are late on payments, and foreclosure rates are rising. For the second kind of lender, at the most one in 20 borrowers pay late, and foreclosure rates are holding steady.
Which group do you want lending your money or financing your home purchase? Obviously, the second kind. And yet if you picked them, you’d be running against the tide of American capitalism. The lenders with the strong record are America’s nonprofit housing enterprises.
The nonprofits do old-fashioned things like verify income and help borrowers prepare household budgets to make sure they can afford the mortgage. Many prospective customers decide not to go ahead because they can see the purchase is too much of a stretch.
So, depending on one’s point of view, the rating agencies were either snookered, or more likely, complicit in a “don’t ask, don’t tell” game. They were running all sorts completely irrelevant analyses. They had absolutely no basis for forecasting the performance of the later 2005-2006 paper, and that may have been true of earlier vintages too. And to Salmon’s issue, they still have no basis. They are reacting and updating, but this is driving a car by looking out the rear view mirror.
Now the second complicating factor is that this is on the way to being the worst housing recession since the Great Depression. The US housing market in aggregate has not seen a nominal price decline since then. In its press release Tuesday discussing its new ratings methodology, S&P predicted a price decline of 8% from 2006 levels by 2008.
That is a massive decline. And it is way way outside of any historical data set.
Now to digress to slip in an important concept: foreclosures are not simply a function of no or negative equity. A homeowner has to have both payment stress and an equity problem to be at risk of default and foreclosure. If he merely has negative equity but can make the payments, he will (no one wants a foreclosure on his credit record). And if he has a short-term payment problem, the bank will likely find some sort of loan modification. Conversely, if he can’t make the payments but has adequate equity in the house, he will sell it on his own. So only the most troubled situations wind up in foreclosure.
So we have a lack of reasonable data on two critical variables: what the defaults might look like, and what this very big price drop will mean for foreclosures. In certain communities where there have been particularly large price drops, sales may grind to a halt due to the huge inventory overhang. Vacation properties are generally harder hit, and I recall during the 1990-1991 recession that in certain very picturesque areas of Maine, virtually every other house was for sale. That means it is very hard to make informed estimates of what recoveries will be like. A very large-scale, detailed, and impressively thorough analysis of adjustable rate mortgages of all kinds by American CoreLogic in March came up with a slightly-under 40% loss on the face amount of mortgage foreclosures (I presume that also included lost interest payments through the foreclosure). At the time it was issued, I found some assumptions to be optimistic, and clearly, that assumption is now too low.
Now there are some things that appear to be known. One reader was so kind as to e-mail and tell me that UBS has used a program they call Intex to run assumptions on various securitizations to see what the impact on the various tranches will be. He says an 8% loss (cumulative loss) on 2006 vintage issues will break the BBB- tranche on many (most?) issues. I presume this is subprime ABS, since CDOs vary greatly as to how much subprime exposure they have (i.e., you couldn’t give a single point estimate). The number is higher for 2005 (9.5%-10%) due to greater seasoning and more equity due to appreciation in some markets.
Now what would it take to get to an 8% loss level? On an individual deal, merely bad luck, but let’s think on a more aggregate basis.
The problem again is that we in the lay public still don’t have the right data, although the pros might. We read in the papers about delinquency and foreclosure rates. Now I don’t know over what period of time conventionally is used to determine foreclosure rates (I assume it’s an annualized rate, but I don’t know if, for example, it’s a rolling 12 month figure, or annualized based on the latest month or quarter.) But the issue is that the relevant stat for assessing the state of subprime paper is aggregate losses. That means cumulative foreclosures, from which you could derive cumulative losses. For those of you who remember your calculus, it’s like having a couple of observations of the slope of the curve and trying to compute the integral. We don’t know enough to do that. If the foreclosure rates don;t change much, and we know the life of a typical deal, it might not be hard to do a back of an envelope calculation, but the more variability, the trickier it gets. The pros hopefully have enough data points to calculate the area under the curve.
With that admitted, drive-the-truck-through-it gap, I will still plow through what we have and see where it leads.
Hopefully our readers will help us fill in better data later (it’s late as usual and I don’t have the energy to go trolling around the Internet for a few hours to see if I might find what I need).
Foreclosures nationally are now running at under 1%. A New York Times story on how Atlanta is having what the powers that be regard as a troubling level of foreclosures reported their level at 2.7%.
Now remember, foreclosures don’t equal losses. If we assume that the American CoreLogic 40% loss figure is too low given rising inventory levels and the expectation of further price declines, let’s be pessimistic and assume 55% losses on the face value of mortgages. That means to get to 8% losses, you’d need 14.5% foreclosures. That looks like a massive number.
But that 2.7% Atlanta number was presumably across all housing, and not foreclosures on subprimes. We are seeing subprime delinquencies that are reported to be in the 13-14% range already, but in reality, the Mortgage Bankers’ Association uses a dubious classification scheme. Delinquencies as of early May for subprimes was in the 18% to 22% range, and of that, roughly 4.5% was foreclosures.
So the real issues are: how do these foreclosure rates relate to cumulative losses on specific subprime ABS? And how badly are things deteriorating? We don’t have the answers, but hopefully someone out there does.