Risk Transfers in the Real World

Earlier this month Jim Parrott of the Urban Institute published a piece titled “Clarifying the Choices in Housing Finance Reform,” in which he compares three proposals for possible replacement of the current system built around Fannie Mae and Freddie Mac: the Urban Institute’s “Promising Road,” the Milken Institute’s “Toward a New Secondary Market,” and a work in progress from the Mortgage Bankers Association (MBA) called “GSE Reform Principles and Guardrails.”

In his piece, Parrott seeks to dispel the notion that the three proposals are “entirely different versions of reform,” saying instead that “they actually share a great deal and…deciding among them is not prohibitively complex, but a matter of assessing two or three key differences.” He identifies those differences as “what to do with Fannie and Freddie,” and whether the Common Securitization Platform or Ginnie Mae should be used to issue mortgage-backed securities. He also states that among the features common to each proposal is that they “distribute all [Urban Institute and Milken Institute] or most [MBA] non-catastrophic credit risk” to capital market investors “at time of origination or after pooling” [bold type in original]. That is, all three proposals—which Parrott claims represent “dominant versions” of the “strong majority view” on mortgage reform—rely exclusively or extensively on risk transfer mechanisms to absorb credit risk.

Conspicuously absent from Parrott’s review, however, is any objective assessment of the relative merits of the risk-transfer model compared with Fannie and Freddie’s equity-based model and, equally if not more importantly, any description of how the risk-transfer model would work in practice (we know how the Fannie and Freddie model works). Instead, Parrott dismisses the Fannie and Freddie model out of hand, saying, “The problem with this system was its dependence on a privately owned duopoly. Fannie and Freddie handled almost all of the securitization in this segment of the market and took almost all of the credit risk. Given our reliance on them, everyone in the market knew that we would bail them out if they ever stumbled. Their shareholders were thus incented to take excessive risk to chase greater profits, knowing that if their bets did not pay off, the taxpayer would step in to cover them. And that, of course, is precisely what happened.”

Except that’s not what happened. What happened was that our last experiment with “transferring non-catastrophic credit risk to capital markets investors”—private-label securitization—resulted in Fannie and Freddie losing their ability to set national underwriting standards. This led to the quick collapse of those standards (while regulators stood by passively), and a subsequent flood of easy credit that pushed housing activity and home prices to unsustainable levels, from which they inevitably collapsed. Fannie and Freddie were pulled into this vortex as were commercial banks, but unlike the banks—whose mortgage credit loss rates were triple those of the companies—Fannie and Freddie were not thrown lifelines; Treasury effectively nationalized them, without statutory authority and against their will.

A fictionalized version of the financial crisis is a useful device to avoid having to compare the Fannie and Freddie model with the risk-transfer model, and a written advocacy piece also affords the luxury of being able to discuss the risk-transfer model at the 36,000-foot level, where there is no visibility of what would have to happen for it to operate successfully on the ground. But when mortgage reform efforts in Washington get serious and focused, and Secretary Mnuchin becomes fully engaged with them, neither fiction nor vagueness will suffice. With the fate of a $5 trillion market at stake, the Fannie and Freddie model, the risk-transfer model and other models will be reviewed and analyzed in meticulous detail by people with real world mortgage experience. Under that scrutiny, the flaws and unworkability of the Urban Institute, Milken Institute and MBA risk-transfer models will be apparent.

I’ve written extensively about securitized credit risk transfers (CRTs), focusing primarily on Fannie’s Connecticut Avenue Securities, or CAS, but I haven’t brought that work down to the ground level either. In retrospect, I should have. My three primary criticisms of CAS (and Freddie’s similar STACRs) have been that they are too expensive, Fannie has too much first-loss exposure before they take effect, and because the risk-transfer tranches can prepay and amortize over time, they are unlikely to be outstanding long enough to absorb many of the losses that exceed the first-loss limit. My source for this last claim was the risk sensitivity tables in Fannie’s CAS prospectuses, which show virtually no losses transferred to investors in 64 different scenarios of combined credit loss and prepayment rates. But as a commenter on this site recently pointed out (and I thank him for that), Fannie’s loss sensitivity tables “assume that the Delinquency Test is satisfied” in each of the scenarios. In a severe stress scenario it would not be, which means Fannie’s sensitivity tables understate loss transfers in those scenarios, and thus are not a reliable measure of the overall effectiveness of CAS in absorbing credit risk.

It is both surprising and disappointing that Fannie would publish tables in its CAS prospectuses that omit a key variable and thus understate potential loss transfers to investors, but since it does, the only way to realistically assess the effectiveness of CAS is to simulate their performance in an actual stress environment. I now have done that. I’ve taken Fannie’s book of business at December 31, 2007, by origination year, assumed all of its $2.53 trillion in single-family mortgages were covered by CAS risk-sharing tranches upon acquisition, then analyzed that book’s credit performance through the end of 2016—specifically, which origination years took what losses, and whether CAS investors or Fannie would have picked them up.

Before I get to the results of this exercise, I should summarize how the current CAS structures work. Fannie takes the first 100 basis points of credit losses on each insured pool. (Recently it has begun sharing a portion of those first losses on a pari passu basis with investors, but the price it pays for this loss sharing—more than 12 percent per year—makes the economic effect little different from keeping all of the credit risk itself.) If cumulative credit losses on any insured pool exceed 100 basis points of its initial unpaid principal balance (UPB), those losses are transferred to the holders of the CAS M-2 tranche until either cumulative losses exceed 275 basis points of the initial UPB or the M-2 tranche is paid off. If losses exceed 275 basis points of the pool’s initial UPB, they are transferred to holders of the CAS M-1 tranche (assuming it’s still outstanding) until the loss rate reaches 400 basis points. Above 400 basis points of the initial pool UPB, Fannie takes all further losses.

Both the M-1 and the M-2 tranches are reduced by prepayments and amortization in the insured pool, unless the Delinquency Test is not met. In that case, the tranches no longer pay down with prepayments, although they do continue to be reduced by amortization. There is a complicated formula for determining when the Delinquency Test is not met, but it can be approximated by when the six-month average of the 90-day delinquency rate for any pool exceeds 1.5 percent.

With that as background, here are the key findings from the CAS stress exercise, using actual Fannie data.

Through 2016, the $2.53 trillion in single-family loans Fannie had on its books at the end of 2007 suffered a total of $85.4 billion in credit losses. Of these, $27.4 billion would have fallen below the CAS first-loss threshold of 100 basis points (and been absorbed by the company). Of the $58 billion in losses above that threshold, fewer than half, or $28 billion, would have been absorbed by holders of either the CAS M-2 or M-1 tranches; $15.9 billion in losses would have fallen within the coverage range of the CAS tranches but not been transferred to them because they would have paid off, and a further $14.1 billion would have been above the CAS coverage maximum of 400 basis points of initial UPB. Thus, even were Fannie to have insured all of its 2007 single-family loans with CAS risk-transfer securities—i.e., were it to have fully adopted the “risk-transfer model”—it still would have had to cover more than two-thirds of its $84.5 billion in losses with a combination of revenues and equity capital.

It’s also instructive to examine these credit losses, and their coverage, in groups: loans from the 2000-2004 origination years taken together, and loans from 2005, 2006 and 2007 individually. (We can safely ignore the small remaining number of pre-2000 loans).

The 2000-2004 books had an aggregate initial UPB of over $3.7 trillion (the 2003 book alone was more than $1.3 trillion); its $11.1 billion in 2008-2016 losses represented a credit loss rate (credit losses as a percent of initial UPB) of about 30 basis points. Had Fannie been required to cover all of its 2000-2004 loans with CAS, it would have issued over $110 billion in face value of M-2 and M-1 securities, with interest payments in excess of ten billion dollars, but received no benefit from them. The CAS benefit for the 2005 book, with an initial UPB of $602 billion and 2008-2016 losses of $15.7 billion (a 260 basis point loss rate), would have been only somewhat better. Fannie’s 2005 CAS M-2 tranches would have picked up $4.3 billion in losses, but another $5.4 billion that fell within the M-2 coverage range would have kicked back to the company because those tranches would have continued to amortize to zero even after prepayments to them were suspended in 2008.

Only for the very poor-performing 2006 and 2007 origination years would the CAS M-2 tranches have provided their full amounts of loss absorption ($10.3 billion and $9.3 billion, respectively). Yet because of rapid payoffs, the M-1 tranches would have provided no protection in 2006 (leaving Fannie to pick up all $6.8 billion in losses that fell within their range), and for the same reason the M-1 tranches would have absorbed just $4.1 billion of the $6.7 billion they should have taken in the terrible 2007 year. And that’s not the whole story. Credit losses in both years exceeded the 400 basis point CAS coverage cap, meaning Fannie would have picked up the $4.4 billion in 2006 losses and $9.7 billion in 2007 losses above that cap. So even in the two years when CAS were most effective, Fannie still would have had to cover far more credit losses ($34.9 billion) with its own revenues and equity than would have been transferred to CAS holders ($23.7 billion).

One obvious conclusion from this exercise is that there is no such thing as a “risk-transfer model” in the real world. Even if CRT securities are issued against every guaranteed loan, a credit guarantor still has to cover all the losses on its good books of business, and a sizable majority of the losses on its bad and very bad books of business, with earnings or equity capital. CRTs can at best be a supplemental source of credit protection.

In addition, compared with the Fannie and Freddie model of backing guaranteed loans with upfront equity, issuing CRT securities against all guaranteed loans is extremely inefficient. Interest payments on CRTs issued against high-quality loan pools—or pools that pay off rapidly—are simply wasted, whereas equity put up to back these same groups of loans can be used to cover losses elsewhere. Moreover, CRTs do not offer predictable coverage even for bad or very bad loan pools, because of their caps. The loss caps on CRTs (in the case of Fannie’s CAS, 400 basis points of initial UPB) apply to all loan pools. Just as these caps are too high, and wasteful, for most pools, they are too low, and insufficient, for the worst ones. As we saw with Fannie’s 2006 and 2007 books, once the CRT cap is breached on a bad or a very bad pool, a credit guarantor using CRTs has open-ended exposure to those pools’ credit losses, and can’t reach back to revenues from other pools to cover them.

Yet the most damning argument against the CRT-based model is the extreme and unnecessary risk it would pose to the financial system. With this model, by the time a credit guarantor realizes how much equity it really will need to cover the losses on its worst performing pools in a stress scenario, raising that equity won’t be possible.

When a CRT security is issued “at time of origination or after pooling,” the ultimate credit performance of the insured pool is both unknown and unknowable. In most years a credit guarantor will be able to cover its losses on good pools with retained earnings (guaranty fees, less administrative expenses and CRT interest payments), but in a stress scenario it will need large amounts of equity to absorb the losses on its bad or very bad pools. The insoluble real-life problem for a CRT-based guarantor is: as a stress scenario unfolds, how does it determine the amount of capital needed to survive it, and how would the guarantor then obtain that capital?

Here again, our exercise with Fannie’s December 2007 book is instructive. With one brief exception, Fannie’s 90-day delinquency rate stayed within a range of 55 to 71 basis points from January 2003 through August 2007, while over that same period the company’s single-family credit losses averaged $0.5 billion per year. Then, in just 16 months, 90-day delinquencies shot up to 242 basis points in December 2008, on the way to a peak of 559 basis points in February 2010. At $14.0 billion per year, the average single-family credit loss for Fannie’s December 2007 book during 2008-2012 was nearly 30 times the average for 2003-2007. Had Fannie been relying solely on CRTs to protect that book, it would have had very little time to guess the magnitude of those 2008-2012 losses, and in a collapsing housing market no chance at all of raising the capital required to cover them. At the same time, the market for CRTs for its acquisition of new loans would have dried up as well.

In “Clarifying the Choices in Housing Finance Reform,” Parrot leaves out the most important commonality of the CRT-based proposals from the Urban Institute, the Milken Institute and the MBA: none of them would actually work. The Fannie and Freddie equity-based model does. It relies on equity capital put in up front, not uncertain credit loss transfers in the future, and in the equity model, capital from the best books doesn’t disappear; it’s available to cover losses from the worst books.

The Urban Institute, the Milken Institute and the MBA risk-transfer proposals may have superficial appeal to members of Congress and the media, but to mortgage market professionals they are desperate, and dangerous, attempts to argue against the only sure path to mortgage reform. Secretary Mnuchin is experienced with and understands the mortgage market. He can differentiate between a secondary market model that works and one that does not. When it comes time to make a call on how to structure the U.S. secondary market of the future—with his name attached to the outcome—there can be little doubt that he will make the right choice.