On the Lack of Reliability of Subprime (and Just About Any) Mortgage Delinquency Stats

A very informative (albeit also a bit rambling at points) post from Tanta at Calculated Risk on how subprime mortgage delinquency figures are derived. Answer: by small scale sampling of institutions that have very different standards for reporting.

I anticipate some howls of outrage about now. You mean to tell me that these delinquency numbers are just ballpark estimates and are not historical? These servicers are allowed to lie to us? Scratch a free-marketer, and you’ll find someone who secretly believes that a vast, heterogenous, discontinuous, non-centrally-planned and inconsistently regulated industry with participants who enter and exit over time not only can be but has been reporting uniform historical data to some central database which is freely and unproblematically available to the public, and therefore any inconsistency or incommensurate data must be another Enron.

The implication is that mortgage delinquency statistics aren’t as meaningful from either a practical or a policy standpoint as many, particularly journalists, would like to believe.

Tanta highlights another fallacy of the debate over what to do about subprimes. A popular position is, “Gee, on balance they are still good because a lot of people got housing through them than would have otherwise.” She (I presume it is a she) says you can’t know that for certain, because many of these may have been prime mortgages refinanced into subprime as the borrower’s condition deteriorated (i.e, the subprime did not enable them to buy housing, but allowed the bank, perhaps not successfully in the end, to, depending on your view of bank motives, either hold off foreclosing or exploit the opportunity to collect more income).

From Tanta:

I’d like to take the opportunity to remind everyone what mortgage delinquency numbers actually tell you—and what they don’t.

In general, a delinquency or default rate on a given book of mortgages is calculated as the balance of delinquent or defaulted mortgages at the end of a reporting period divided by the total balance of the book as of that reporting period. For anything other than a mortgage security with a static pool, any given reporting period will involve new loans added, old loans paid off, existing loans paid on time and catching up, and existing loans becoming or continuing as delinquent.

In a static pool, such as a REMIC, no new loans are added, and so a calculation on current balances might be useful to an investor wanting to buy a seasoned security, but is less useful to anyone who wants to use delinquency rates to answer some general question about credit quality. Therefore, with static pools, one generally needs to calculate the delinquency rate in reference to the original balance, or to apply the “pool factor” (the percent of original balance remaining) to the delinquent balance, in order to account for the effect of prepayments of performing loans, which will, over time, push up the ratio of delinquent balances to total current balances even if no more loans become delinquent.

Prepayments, of course, create their difficulties even in calculating a delinquency rate on a non-static universe of loans like a servicer’s portfolio. In a static pool, a prepayment is a net decrease in the total pool balance; in a non-static book, a prepayment can simply be a rollover within a portfolio, or a movement of a loan from X’s book to Y’s book.

It does appear, sadly, that some people think that “national delinquency rates” are based on some total database of all outstanding loans, rather than on a sample of various reporters, some of whom report on static pools and some on active pools or books. If you’ve been thinking that, please adjust: these reported rates, such as the MBA’s, are based on a sampling of servicers, and they are adjusted statistically, in some manner, for the effect of prepayments.

The point here is that all you are getting is a percent of delinquent balances to total balances as of a date. First of all, “balance” is not “units.” Without further data-crunching, you cannot leap from a balance of delinquent loans to a number of delinquent borrowers; even if you have an average balance per loan to work with, you need an average loan per borrower to work with as well. That’s not a trivial issue: not only do individuals quite frequently carry two or even three mortgages per property, individuals can own more than one mortgaged property, and what we do know so far suggests that speculators have been getting caught in a delinquency problem in droves lately. Casey Serin may be—we hope—an extreme case, but he is a case of one borrower leaving a long trail of many defaulted loans in his wake.

Second, these are not “historical” numbers. Once a loan is removed from a book, through refinance, sale, final amortization or REO liquidation, it is no longer a balance on the original lender’s book. If a servicer liquidates REO in June, there is no longer a delinquent balance to report in July. Therefore, if the delinquent balance in July is identical to June’s, you have at least one new delinquent loan.

Some servicers may not even report loans in foreclosure—a state a loan may be in for many months before it becomes REO and the REO is liquidated—in the “delinquent” category, which is why you find nerds like me getting occasionally rather anal about the terms “delinquent” and “defaulted,” or “seriously delinquent,” or “non-accrual,” or “non-performing,” or “collateral-dependent,” or any of the other categories problem loans may be found in, depending on what one is up to and whose book is involved (servicing a loan, accounting for assets or loan income, taking an impairment, etc.). Even with a “vintage anaysis” that separates loans into year of origination, you have the potential problem of a purchase loan that closed in January being refinanced in September and becoming delinquent in December. That’s two loans, one early payoff, one EPD, one year, and one borrower.

Third, loans go in and out of delinquency. You need the rate of “conversion” of mildly delinquent to seriously delinquent loans, just as you need the rate of conversion of NOD to FC, to use these numbers to predict eventual loss of a home. In markets where unemployment rates are low, for instance, loans that are delinquent due to sudden job loss can catch up in a month or two when the borrower finds new work. Over time, once-delinquent loans tend to converge on foreclosure; the recovery is often only temporary. But it does mean that any given month’s delinquency rate can be the same as the prior month’s, but a new set of loans has rotated into the delinquent pool.