by Tanta on 10/12/2007 11:14:00 AM

Courtesy of Thomas Zimmerman of UBS, whose PowerPoint presentation is available here. There's quite a bit of interesting data for the nerds.

These charts are mini-vintages (quarterly rather than annual) of 2/28 subprime ARMs.

The first shows serious delinquency (60 or more days delinquent, FC, or REO) for first lien purchase money loans using 100% financing (CLTV greater than or equal to 100%) with less than full documentation in states with "stable" HPA. (I don't know exactly what universe of states that is.)

The second chart shows the same loan type for California properties only:

To put this into some context, the third chart shows what we might call the more "traditional" subprime loan: a 2/28 ARM cash out, with full doc and CLTV less than or equal to 80%. This third chart is California properties only.

I think I've said this before, but it bears repeating: I have never, in my hundreds* of years in this business, worked with any mortgage model--pricing, credit analysis, due diligence sampling--that did not consider cash-out an additional risk factor. That is, historically speaking, cash-out refinances always performed worse than purchase money or rate/term refinances, and the models therefore would give a worse risk-weighting to a pool with a majority of cash-outs than a purchase-heavy pool. There were two main reasons for this: cash-out does correlate with heavy debt use (obviously), and also, historically speaking, cash-out refi appraisals were the least reliable, most subject to "hit the number" pressures. This was true even when lenders allowed substantially lower LTVs on cash-outs than recently has been the case.

In my view, a whole lot of the failure of the rating models to adequately account for the risk of these recent pools is that they used "historical" assumptions about the risk of purchase transactions.

*Mortgage years are like dog years, only worse.

by Tanta on 10/10/2007 09:40:00 AM

More unattractive little snips from my unattractive spreadsheet collection (earlier posts here and here). What can I say? UberNerds don't need no steenkin' fancy formatting.

Item one gives you some sense of the size of the residential first lien securitization market since 1988.

I have been avoiding the terms "agency" and "nonagency" on this blog, but I'm breaking down and using them here. These are an established and pretty old-fashioned way of describing things inside the biz, but they are traps for the unwary. In this particular context, "agency" means Ginnie Mae (which securitizes FHA, VA, and a few other government-insured loans), Fannie Mae, and Freddie Mac, even though only Ginnie Mae is actually an agency of the government (Fannie and Freddie are GSEs, Government-Sponsored Enterprises, not actual agencies). But we used to call them all agencies, and the term survived reality by about a generation and a half, so there. "Nonagency" just means any private issuer.

The column "Issues / Originations" is simply that: one annual number divided by another annual number. That is a very, very approximate way to describe the rate of securitization of originated loans. You would get a number much closer to reality if you used quarterly numbers with a one quarter lag, but I don't have quarterly origination numbers handy. So do throw this number around with a high degree of caution.

What we learn from this spreadsheet is something like the approximate size of the segment of mortgage outstandings that have been in the news lately. The nonagency category (in these charts) includes Jumbo A, Alt-A, and Subprime, primarily first liens. (It includes some MBS that have a small percentage of second liens in them, but excludes MBS that are exclusively second liens. I complain regularly about the "lumpy" or Bridge Mix nature of recent nonagency MBS issues, and this is one reason why.) Basically, all the reporting you are seeing that is based on securitized nonagency loans is discussing around a third of securitized loans outstanding, or 19% of all loans outstanding (as of Q4 2006). Because there is so little data available on unsecuritized loans, it is extremely difficult to answer the question of the extent to which "nonagency" unsecurtized (these are mostly but not exclusively bank and thrift portfolio loans) will perform like their securitized brethren. Most of us believe that the securitized loans were written to much riskier standards than the unsecuritized loans, although as I noted yesterday in reference to the C of the C's last exasperated speech, I do believe that portfolio lending standards have loosened significantly in the last several years. You may in any case draw your own conclusions.

Item two is all the information I have on the break-out of the nonagency category. I got nuthin' on outstandings prior to 2000, but you can guess from what's here that they were rather modest in relation to total mortgage outstandings in those years.

I do not have a refi mix breakout by product for Jumbo A and Alt-A, so I didn't include it. But you can get a sense for how much of new origination is refinance (turnover in the outstandings rather than net additions to it) by comparing issues to the change in outstandings in a given year.

You can also get an idea for why people like me have been snorting derisively for years over this claim that "Alt-A" has a stellar performance history. It barely has a "history" at all. Furthermore, the definition of "Alt-A" in 1995 bears little resemblance to the definition of "Alt-A" in 2006. Remember that "Alt-A" means "alternative" to "A," and so whatever it is, its composition will change as the definition of "A" or "prime," to which it is an "alternative," changes. Back in the mid-90s, SIVA (stated income/verified asset) or--gasp!--CLTVs of 95% were the big "alternatives" and "interest only" was the sort of thing you ran into in commercial lending. Not only do you have, nowadays, IO SIVA with 100% CLTV in "A" (conforming or Jumbo), you have stuff in Alt-A that was simply unimaginable in 1995. So as "A" gets more "alty" over time, "alt" gets waaay more "alty" over time. What people are trying to get at by asking how "Alt-A" can "revert to normal" is, as far as I'm concerned, not very clear. I have no idea what other people think "normal" Alt-A is.

by Tanta on 10/09/2007 10:04:00 AM

More stuff from the spreadsheet collection. This one looks at characteristics and some performance measures of securitized subprime loans from 2000-2006. Unfortunately, there is very little publically available data on unsecuritized subprime.

Comments:

1. Total MBS issued on this chart is mostly, but not exclusively, first liens. (It includes securities that have some second liens, but excludes securities that are exclusively second liens.)

2. The average loan amount is based on first liens.

3. WAC is weighted average coupon or "interest rate" in English.

4. "Reported" DTI simply means that's what was reported. While I have some doubts about the accuracy of that number when the full doc percentage is dropping, do notice that it is climbing even so. The historical maximum acceptable DTI for conforming agency-quality loans was 36%.

5. Historically, subprime was a refinance business, not a purchase money business. This chart shows that very clearly.

6. "Serious Delinquency" means 60 or more days delinquent, FC, REO, or BK. Because this is calculated on the current balance of these securities, this number will be much higher than what you see reported based on original balance. You should be aware that the remaining current balance of these older vintages is very low; the average "pool factor" or balance remaining for 2000, for instance, is around 5%, as opposed to 83% for 2006.

7. "Default" is a very specific technical measure here. A loan is reported as a default in a month when its balance is reported as zero and its last reported status was in foreclosure, REO, delinquent more than 150 days, or any other status and a loss of more than $1000 was recorded at payoff. In other words, "default" is the final disposition of a loan, and it includes things like short sales and short refis as well as foreclosures. It does not include active modifications or forbearances, since these loans still have a reported balance. It is a loss measure, and because it involves the final disposition of a loan, it is always much lower for new issues than for older issues, even if they are performing equally.

8. Cumulative loss is based on the original security balance, and is equal to default times severity.

Now, about that FICO average. On the one hand, the fact that the average FICO is rising can be filed under "I sure as hell hope so." When you look at the steadily rising risk factors of CLTV, documentation level, DTI, and so on, you would certainly expect that higher FICOs were being required as some kind of risk offset.

On the other hand, those average FICOs are getting awfully close to near-prime or even prime territory, depending on your definition (620-660 being the usual floor for prime). That means that a lot of these loans have FICOs clearly in prime range. In order to rule out the possibility of predatory steering, you have to trust that the subprime industry has been scrupulous about giving subprime loans to higher-FICO borrowers only when the other loan characteristics are clearly non-prime. This question cannot be solved by looking at averages or even really good stratifications; it takes loan-file-level reviews to really understand what's going on. As those loan-file-level reviews were, apparently, not done by aggregators and raters and investors, they are now being done by servicers and courts.

by Tanta on 10/08/2007 11:30:00 AM

I've been updating some old spreadsheets of mine, and I thought some of you might be interested in having some of the numbers. Data like the following, which involve national averages over entire years, are awfully blunt instruments for a lot of analytical purposes; I'm not offering these as "proof" of any particularly detailed claim about the world, nor am I suggesting that any particular set of numbers in this table can "explain" any other set in completely reliable ways.

Nonetheless, broad-brush numbers like these do provide a kind of context for certain discussions of the mortgage market. I tend, personally, to cringe a lot when certain numbers are reported in the press, because I possess a sense of context that, frankly, non-insiders don't have. It is second nature to me, for instance, to distinguish between origination volume and level of loans outstanding at the end of a period, but you will find press reports moving back and forth from originations and outstandings in blithe disregard of the issues.

So make of this what you will. A few nerdly observations about the data:

1. Total originations are hard to pin down; there's often a lot of vapor as well as volatility in those numbers. I pick what I think is the most reliable, but you should know that data collection and reporting practices change over time, and so a lot of the older numbers are pretty approximate. That's one reason why I don't care to go back before 1988.

2. The ownership rate statistical calculation changed significantly in the early 90s. The pre-1993 numbers should be thrown around with even more caution than the 1993-2006 numbers.

3. There are a lot of different rates you can use to establish an average mortgage interest rate. I chose the FHFB conforming fixed contract rate because it can be considered an index of "refinance incentive."

4. Mortgage FOR is a statistical measure of mortgage debt, property taxes, and insurance divided by disposable personal income for all homeowners with a mortgage. As a level, it's not particularly helpful, but it does help establish trends, and it is certainly more consistently calculated than DTI.

5. Refinance percent is all refinances. I am not yet ready to try to sort out the cash-out issue over this time period, and I may never be. But general refi share is a useful bit of context for those changes that you see in the other numbers.

A couple of general observations I would make about this data:

1. Notice the volatility of origination volume compared to mortgage outstandings. A large part of some of the knotty issues we've talked about on this blog, like use of brokers, barriers to entry (or the lack thereof) for mortgage originators, and historical changes in quality of mortgage origination personnel (including loan officers and appraisers), has to do with that volatility. The short version is that originators staff up and staff down in the cycle, and that loan quality (not just borrower credit quality, but accuracy of paperwork, depth of documentation, clarity of disclosures) zigs and zags along that cycle. In the early part of those big booms, for instance, you can see a lot of novice work. In the troughs, you can see a lot of desperate commission-paid people doing desperate things. That's something to take into account when you look at, say, vintage charts of loan performance. I believe, for instance, that a lot of the problems with the (in)famous 2001 vintage had to do with a huge crop of brand-new brokers and loan officers and appraisers getting into the business to cope with the volume. A lot of the problems with the 2000 vintage is that a bunch of originators who had been, in the past couple of years, making plenty of money off the refi boom started scraping the bottom of the barrel when volume dropped off. The 1993-1994 period had a similar problem.

2. There has always been much more stability on the servicing side; the problems there in terms of expertise are more a function of technological "productivity" changes and outsourcing reducing the cadre of gray-haired veterans of past crises. The thing to notice here is the level of turnover in the outstandings. In 2003, for instance, around a third of the outstanding mortgage book turned over in refinances. Mortgage sevicers can do a lot of work to stay in the same place, let alone to grow a servicing portfolio.

3. Whatever generalizations you want to make about earlier periods, the 2004-2006 period confounds them.