A quick-and-dirty analysis suggests that one of the GSEs consistently outperforms in asset recovery rates over a sustained period, based on the historical performance data of 30-year fixed-rate Fannie and Freddie mortgage portfolios.
Top-down or bottom-up?
Going "bottom-up" for ~45 million loans is not practical, so we could dive straight into 10-100 pages of stratification tables. But that is still a lot to review manually, and more importantly, that approach only shows me the data slices that had to be determined before we actually sliced the data-- which may well turn out not to be what I will think is most useful once I start seeing some results.
A better way, I think, is a top-down approach. I can start by simply looking at losses by vintage (see endnote for details), broken out by GSE in a side-by-side manner, and then shooting from the hip. I don't know what analysis I'll want to see next, but I don't have to-- I will decide that at each step based on the information I just received.
So, here's what that first cut might look like (with the help of Liquidaty software, of course-- and I promise, that will be the only plug I put in here):
One pattern I notice right away is that for each year, FNMA's loss rates seem to be the same or lower than FHLMC's, which isn't something I would have necessarily expected.
Loss = Default x Severity
Since a loss rate can be broken into default and severity rates, the natural next question is how those sub-components compare. So I broke the losses rates out into defaults and severities, and below shows the differences (FNMA value less FHLMC value):
Generally, one would expect defaults and severities to move together, because borrowers are less likely to default when they are less underwater. So, in 2001-2004, the fact that one of the two GSEs is lower on both counts isn't surprising (had to be one of them right?).
However, starting in 2005, that changes—FNMA defaults begin to consistently, sometimes significantly, exceed FHLMC's, but severities continue to be lower (sometimes by a lot). Why would that be?
There are, of course, many possible reasons, and a couple that first come to my mind are geographical concentrations and loan sizes. Since loan size is an easy statistic to add to our prior graph, let's do that and see how it looks:
Bam!! Look at that big difference in average loan size (green bars), starting in 2005. Generally, people who think a lot more than me about these kinds of things will tell you that larger loan sizes, especially in this range, have lower average severities due to the fact that many costs associated with foreclosure are fixed (see endnote for more on this). And that seems to be exactly what is going on here.
Now we have a theory... or don't we?
So is that the end of the story? Maybe. If loan size really is driving the difference, we should find that, within like loan-size bands, severities are similar-- all else equal. So let's look at like loan sizes... but how do you make sure all else is equal?
The answer is, you do a lot more analysis than I will do here. But I can at least get started in muddying the picture by looking at four GSE-comparison stratifications (each strat contains multiple tables, each of which is a break-out by band or value of State, MSA, Occupancy, Maturity, Purpose, Rate, LTV, CLTV, FICO, DTI, or Property Type):
- 2005 vintage loans with original balance of 200-300k
- 2006 vintage loans with original balance of 200-300k
- 2005 vintage loans with original balance of 300-400k
- 2005 vintage loans with original balance of 300-400k only in the state of CA
Here's a summary of what that yields if you calculate the percentage of FNMA severity statistics that were lower than its FHLMC counterpart. The bottom line, at this point in the analysis, is that across the board (usually more than 90% of the time), even within the same-loan-size bands, FNMA severities are lower:
It's a terribly simplistic summary, but the result is so lopsided that perhaps it doesn't matter much. Here's another way to look at it (just as simplistic, but no need to read: just lean back in your chair and blur your eyes): the image to the left of this text is the stratification comparison, where every statistic where FNMA had a better recovery rate than FHLMC is colored green, and where the opposite was true, is colored red.
Look at how one-sided that story is! That is pretty amazing, I think. Maybe too amazing. Could it be true that FNMA is better in its underwriting, servicing or other secret sauce? Perhaps. Or perhaps the reported figures for one GSE includes certain items that are not included in the others, so the comparison is not apples to apples-- though, the results are identical if you compare reported "Net Sales Proceeds" divided by the original loan balance.
I don't know the answer-- this analysis is certainly not enough to draw causality conclusions from-- but for a couple of points across many billions dollars' worth of loan workouts, it might be worth finding out.
What do you think?
I am not a researcher and research is not my company's product (software is). The figures herein have not been independently verified or audited by anyone. I do not represent any of the information herein to be accurate or complete, and you should not rely its accuracy or completeness. Any information, opinions, estimates and forecasts contained herein is subject to change without prior notification. No part of this material may be (i) copied, photocopied or duplicated in any form by any means or (ii) redistributed without the prior written consent of Liquidaty.
1. If you're interested, some discussion of loan size impact on severities can be found here: http://www.urban.org/sites/default/files/alfresco/publication-pdfs/2000092-Loss-Severity-on-Residential-Mortgages.pdf
2. For the analyses here, "vintage" was based on the year of the origination date and "loss" was calculated from loan amount plus expenses, less net sale proceeds. Notably, the "loss" figures used herein do not including mortgage insurance proceeds and other non-MI recoveries, to make it easier to equate "loss" to the concept of the gap between the money spent on a property and the sale proceeds of that property. Weighted averages are weighted by original loan balance.