There's a very interesting new study on mortgage loan modifications out from the Boston Federal Reserve staff. This sort of study is long-overdue and from an academic standpoint, there's a lot I really like about this study. But the study is going to get a lot of policy attention, and I think it's important to point out some of the problems with the study that limit its ability to serve as a policy guide.
The study has two big claims. First is that the reason we aren't seeing many mods is because of high redefault and self-cure rates for borrowers in general, basically type I and type II errors respectively. If a mortgagee modifies a loan and it redefaults in a declining real estate market, the mortgagee's recovery from the foreclosure or REO sale will be diminished. Thus, there is a danger of modification actually reducing value for mortgagees. Conversely, some the defaults on some loans that are modified would have been cured without modification. The modification is thus a give-away from the mortgagee's perspective. Mortgagees are scared of both of these possibilities (or maybe rationally recognize that they are quite likely) and therefore aren't doing mods.
The other claim, based on an empirical analysis of a sample from the LoanPerformance database, is that there is no statistically significant difference in the percentage of portfolio and securitized loans being modified. From this the study concludes that securitization is not an important factor in the paucity of loan modifications. Instead, the authors' identify the common factors of redefault and self-cure as limiting mods.
From this, the study reaches two conclusions. First, that there are far fewer preventable foreclosures than assumed. Second, that servicer safe-harbor provisions to allow servicers to modify loans without fear of litigation are unimportant.
So what's wrong with this picture? Regarding the claim that redefault risk and self-cure risk are limiting loan mods, I think that as a pure matter of theory, it makes a lot of sense. But when the claim is tested against the actual numbers produced in the study, it doesn't hold up--there's still plenty of room to do value-maximizing modifications.
One of the very strange things about this study is that it has some empirical data and a model, but it never puts the two together. Instead, the study assumes that the empirical data and the model support the interpretation advanced simply because the model indicates that high levels of redefaults and self-cures would make modifications no longer worthwhile, and the numbers of redefaults and self-cures look really large. But just because a number looks large doesn't mean that it necessarily shifts the modification calculus. I've tried putting the numbers together with a model. Bear with me on the math--it's entirely possible that I've overlooked something in my calculations, and if I have please comment to let me know--but if my math is right, then redefault/self-cure risk just isn't what's limiting mods.
The question a rational mortgagee with no outside interests should ask when faced witha defaulted loan is whether the net present value (NPV) of a modified loan is greater than the NPV of a unmodified loan. If so, a modification would be value maximizing for the mortgagee. My NPV modeling is somewhat crude, not least because it tries to avoid discount rate and refinancing horizon issues by treating UPB and NPV as equivalent, which they are not, but I think it captures the essential point.
Let's use some very conservative assumptions--that there will be high redefault and self-cure rates, and that foreclosure losses will be high, and much higher for redefaults. For a modified loan, lets assume a 30% chance of self-cure (from the study), a 40% chance of redefault (conservative from the study), resulting in a 75% loss severity (quite conservative), and a 35% chance the modified loan will perform as modified. For an unmodified loan, there is a 30% chance of self-cure and a 70% chance of foreclosure, with a 55% loss severity (conservative--it's more like 60% now). Let's assume a mortgage with a $200,000 NPV if it performs unmodified, and that M is the maximum NPV of a modified loan such that it will be greater than the NPV of the loan unmodified. Thus:
[Value if modified, but would have self-cured] + [value if performs as modified] + [foreclosure value if redefault]≥ [value if self-cured without mod] + [value if foreclosed without mod]
To put the numbers on it:
.3M + .3M + .4*.25*$200,000 ≥ .3*$200,000 + .7 * .45 * $200,000
.6M + $20000 ≥ $123,000
.6M ≥ $103,000
This means that even using the Boston Fed's most conservative assumptions, the principal and/or interest could be written down such that the NPV of the loan would go to $171,666.67 and the modification would still maximize net present value for the mortgagee. To put this in slightly different terms, a modification would still be value maximizing, even with a 15% write-down in NPV. And that's with some very conservative assumptions. Loosen these assumptions (e.g., FDIC's 15% mod-in-the-box self-cure rate or a 30% redefault rate), and there are even more generous modifications possible.
There's also another way to test the explanatory power of redefault and self-cure risk. Presumably redefault risk and cure rates also vary with other mortgage characteristics. For example, it stands to reason that an underwater investor property mortgage is less likely to be cured than an above-water owner-occupied one. The question, then is whether modification rates track the variations in redefault and cure rates by mortgage characteristic. If they do, then the study's conclusion would be much stronger; if they don't then either these factors don't matter or mortgagees only care about them in the very rough aggregate (which seems both unlikely and unfortunate). Hopefully this is something the authors will investigate in later versions of their study.
Regarding the second claim
, that securitization doesn't matter in terms of mods,
there's first a data question and then, regardless of how that is answered, a factual and a logical problem. I don't know the LoanPerformance data set in great detail. You can see some of its characteristics listed here
. I don't know if LP includes data on loans held in portfolio by credit unions and community banks. If it doesn't, that might be distorting the results as the portfolio loans for large banks might well be serviced by the same servicers as securitized loans. If so, the study wouldn't be comparing securitized vs. portfolio, as much as self-serviced vs. serviced-by-others.
Irrespective, there is a major factual issue overlooked by the study: there is a difference in how a securitization servicer and a portfolio lender view redefaults and self-cures. A portfolio lender is fully sensitive to both; a servicer, in contrast, does not care what the property brings in at a foreclosure or REO sale because the servicer is paid off the top. As long as there is just some land value left, the servicer will get paid. The servicer might have to make additional months of servicing advances on a redefault, but those are reimburseable too, off the top. The only cost to a servicer from redefault is some time value. That means servicers are less sensitive to redefault than portfolio lenders. Self-cure is also less of an issue for servicers because most of their mods involve interest rate reductions, and rate reductions have only a small affect on servicer compensation, unlike a principal reduction. In short, redefault and self-cure risk is not an equally applicable factor, so it cannot alone explain the similar rate of mods for securitized and portfolio loans.
So what is the explanation? There are two possibilities. First is that it is simply coincidence. The paper recognizes this as a possibility, although it quickly dismisses this. The second, possibility is that there is another common factor (or factors). I think servicer capacity is a major concern that applies across the board. To start with the bulk of servicer personnel at most companies aren't even in the US; they've been outsourced. Doing a mod is like underwriting a new loan in a distressed situation. That's a skill, and I don't think it's what servicers were looking for over the past decade when they moved operations to India. Instead, they were looking for low-cost labor for their routine ministerial tasks, and it will take a long time for the industry to acquire the workout talent it needs.
In any case, even if there is a common factor, that hardly means securitization doesn't create serious concerns. Even if issues like capacity can be addressed, there are layers of problems preventing modifications, all of which must be addressed, but the study dismisses this possibility a little too quickly, and based on a questionable analysis of the other literature on securitization. For example, the study claims based on a single empirical study of PSA provisions (which has its own limitations), that "suggests a small role for contract frictions in the context of renegotiation." This is a very strange statement, as it assumes that contract frictions are just a matter of formal contract provisions. My article with Anna Gelpern shows that in PSAs there are a variety of frictions to renegotiation, some formal, some functional, and some structural. Our article is cited elsewhere in the study, but doesn't seem to have informed the Boston Fed staff's study on this point. The study also claims that the Congressional Oversight Panel's foreclosure report
states that "none of the [contractual] restrictions [on loan modification in PSAs] were binding." The Oversight Panel said no such thing. The Panel merely observed that in some pools where there was a 5% cap on the number of loans that could be modified, that that cap was not yet limiting modifications.
In short, there is no reason to assume that contractual frictions don't matter. That said, I agree with the study's claim that servicer safeharbors are unlikely to do much good, but that is because there are contractual frictions that safeharbors don't address as well agency problems.
Explaining the failure of modification efforts will be an unresolved question for some time, but at this point I think it's really just academic to try and pinpoint why the mods aren't being done. Instead, we have to look for a method that we know will produce loan modifications. I hate to sound like a broken record, but there's a solution that cuts to the chase--bankruptcy modification cuts through all of the mess. And when even a conservative scholar like Stan Leibowitz can write a piece in the WSJ
arguing that negative equity is the driving problem, it's time to take another look at cramdown.
Finally, I wonder whether the goal of maximizing NPV for investors is the right metric. A foreclosure might maximize value for investors, but be socially detrimental. If the policy goal is improving social welfare, then we might want to discourage foreclosures that by themselves might be economical.