Consensus Subprime Mortgage Loss Estimates: Mathematically Impossible?

by: Tom Brown

We’ve been saying for awhile now that the cumulative credit losses from the subprime mortgage market won’t be as high as the consensus seems to think. Judging by how the financial stocks have been acting lately, not a single person on the planet believes us.

Oh well. These things take time--so let me take another stab at this. In particular, allow me to walk you through some numbers that I believe show, compellingly, why it is that the consensus subprime loss numbers being thrown around are nearly mathematically impossible to achieve.

Ready? For the purposes of this discussion, let’s use as “consensus” the loss estimates lately being published by the analysts at UBS. UBS has been publishing numbers for as long as anyone on the Street, and the analysts’ work there is especially thorough. (If anything, in fact, the “real” consensus loss number might even be higher than UBS’s estimates.)

At a conference call earlier this week, UBS said it believes the cumulative loss on the ABX 06-1 subprime mortgage index will come to 19.5% when all is said and done, and will be 29.6% on the ABX 06-2. As I say, that’s too pessimistic.

I’ll explain why in a minute. First, a quick review of how we come up with our estimates. To get to expected cumulative losses, we look at the loans that comprise the ABX indices and add up a) realized losses to date, b) estimated losses from loans that are seriously (like, more than 60 days’) delinquent and real estate owned, and c) estimated losses from loans that are still current. As it happens, estimating a and b above isn’t all that hard. Essentially all of those loans will go bad, or have already. It’s just what will happen to c, the loans that are still current, that that’s the area of conjecture.

Servicer reports filed monthly

Anyway, to how we come up with our numbers. Recall that each ABX index consists of 20 securitized mortgage trusts.  The servicers of those trusts file reports on the 25th of each month that update the performance of the loans, through the last day of the month. The servicer reports filed June 25th capture loan performance through the end of May.

We model each trust individually, then roll up the totals to arrive at a loss estimate for each ABX index. 

Now to the numbers, using the a-b-c method of analysis described above.

First, the sum of the realized losses to date incurred by the 20 trusts that make up ABX 06-1 represents 2.8% of the sum of the trusts’ beginning balances.

Next, we estimate losses that will come from seriously delinquent loans.  We assume that 75% of loan dollars 61 to 90 days past due become real estate owned [REO], that 90% of loans 90 days past due go to REO, and 95% of loans in foreclosure go to REO.  We then add these numbers to the REO total and assume 55% severity to arrive at our estimate of losses for past-due loans.

 OK so far? The roll rates we assume are well above historic averages and even a little higher than what has occurred in recent months, so I feel comfortable that they’re conservative. Using these assumptions, we get to a loss rate on delinquent loans of 7.5%.

Our story thus far: realized losses come to 2.8%, while “pipeline” losses on delinquent loans are another 7.5%, for a total of 10.3% in cumulative losses.

Getting to 19.5%

But UBS’s loss estimate for 06-1, remember, is 19.5%. Where will those losses come from? Well, one place they won’t come from is the loans in the trust that have already been repaid--which account for fully 58% of the index’s original balance. Rather, the 9.2% incremental losses UBS expects have to come from the loans in the trusts that are still current.

We’ve studied those loans closely. We’ve looked at their underwriters, the locations of the properties, loan-to-value ratios, levels of documentation, and borrowers’ FICOs, and have come up with an estimate that still-performing loans in the trusts will generate a cumulative loss of . . . 2.5%.

That brings our estimate of total cumulative losses for 06-1 to 12.7%, rather than the 19.5% UBS expects.

Wait a minute!, I hear you saying. Losses of just 2.5% from the performing loans? That seems way, way too low.

No, it’s not. If anything, it’s likely too high. Here’s why. Remember, 61% of the beginning balance of the ABX 06-1 has either paid off or charged off, while another 14% of the original balance is 60 or more days delinquent or in REO. That leaves just 25% of the original balance as performing. 

Higher than the loans already gone bad

Again, if you assume 80% of loans 60 days past due roll all the way though to REO, and then 55% loan severity, that 2.5% loss estimate means that 22% of loans still performing will eventually go delinquent. That is a very conservative number. Why? It’s higher than the cumulative delinquency rate that has occurred already.  And those loans, recall, include the weakest credits in the trust, including the legions of speculators and con artists who walked away as soon as their properties were underwater.

So we’re assuming the performance of the still-current loans will turn out to be even worse than what’s occurred with the loans that are in serious trouble already and have been charged off!

So, then, what would have to happen to get to UBS’s 19.5% cumulative default rate? The bank doesn’t share the details about how it gets to its number. But we can back into it using our model, to see what their estimate implies. I do know UBS assumes severity of 60%.  That would raise the cumulative losses from the past-due loans to 8.1%.

That means that the loans still performing have to create an incremental 8.6% cumulative losses.


Which gets us to the incredible-number portion of the discussion. If you assume an 80% roll rate and 60% severity, to get to the loss estimate UBS has in mind, 72% of the currently performing loans would have to default. That is not a typo:72%.

I somehow don’t think that’s going to happen. As you see, the vast majority of the difference between our loss estimate for 06-1 and UBS’s boils down to how many of the loans still performing (for 2½ years!) will default. Given that the cumulative delinquency rate to date has been just 20%, and includes the frauds, speculators, and weakest credits, I have a high degree of confidence that our number, not UBS’s, will turn out to be closer to the mark.

Even so, Wall Street seems to be laboring under the impression that losses will zoom to stratospheric levels. Oh, they’ll be high, there’s no doubt about that. But even the numbers put out by relatively sober-minded analysts have essentially no chance of happening. Eventually investors will sooner or later figure that out.

Tom Brown is head of