My previous post ended with a call to the other 18 essayists in the Urban Institute’s “Housing Finance Reform Incubator” series to “come back with their ideas on capital and pricing” for their proposed system of secondary mortgage market credit guarantees, with a focus not just on the amount of capital they would require but also on how that capital amount is determined, why they believe it is optimum, and how it would affect the availability and affordability of mortgages to a range of different borrower types. In this post I take my own advice and elaborate on the approach to capital I propose and my rationale for it.

As I said in my essay “Fixing What Works,” I believe credit guaranty companies should be required to hold sufficient capital to withstand a defined, worst-case stress scenario. My proposal was to have administration policymakers pick that scenario, but I recommended that they require guarantors to hold sufficient capital to survive a 25 percent nationwide decline in home prices over five years. I noted that while such a price decline did happen between 2006 and 2011, the two factors that fueled the bubble that preceded it—very risky mortgage types and lending practices now prohibited by the Consumer Financial Protection Bureau (CFPB), and an unregulated private-label financing mechanism that placed few limits on the risks of the mortgages it accepted—will not be present going forward, making the chance of a future 25 percent decline in home prices exceedingly small.

We can get an approximation of the amount of capital a credit guarantor would need to hold to survive such a scenario by drawing on what happened with Fannie Mae’s single-family credit guaranty business during the previous crisis. In the analysis below, I look at Fannie’s single-family business over its worst five consecutive years of financial results, using its actual credit losses, administrative expenses and guaranty fee income from 2008 through 2012 (after which the business returned comfortably to profitability). I ignore results from Fannie’s multifamily credit guaranty business and its business of managing interest rate risk on its mortgage portfolio, both of which were positive throughout the period we’re examining.

We’ll start with the raw numbers. At the end of 2007, Fannie held the credit risk on $2.55 trillion of single-family mortgages (about a quarter of which were in the company’s own portfolio, with the balance held by other investors), at an average guaranty fee rate of 26.5 basis points. On an economic basis Fannie’s single-family credit guaranty business barely broke even in 2008, and it lost significant amounts of money over the next four years. For 2008-2012, single-family credit losses totaled $75.7 billion, administrative expenses were $7.8 billion, and guaranty fees were $39.3 billion, resulting in a cumulative loss of $44.2 billion. To survive that, Fannie would have needed about 1.75 percent capital against its single-family book going into 2008 ($44.2 billion in losses divided by $2.55 trillion in outstanding credit guarantees).

Using the raw numbers from 2008-2012, however, does not give us the best approximation of the amount of capital a guarantor ought to be required to hold to survive a similar crisis in the future. For one thing, the 2008-2012 results include guaranty fee income (and credit losses) from business put on after 2007. A traditional “stress test” is done on a liquidating book, and does not assume any replacement or new business. The single-family credit guaranty book Fannie had at year-end 2007 paid off at 20.1 percent annual rate though December 2012, ending at $833 billion. The guaranty fee income from these loans during 2008-2012 was only $21.4 billion. Making this adjustment, along with a related reduction of $5.6 billion for credit losses on loans acquired after 2007 (Fannie publishes those data), the cumulative loss on the $2.55 trillion in single-family guarantees Fannie had going into the crisis rises to $56.5 billion, or about 2.25 percent of that book.

But there is one more important adjustment that needs to be made: at least one third and as much as three fifths of the single-family credit losses experienced by Fannie during 2008-2012 came from loan types and underwriting practices that no longer are permitted by the CFPB, and thus will not be present (and causing losses) in a future stress environment. Fannie publishes data on this, too. They show that 32% of its 2008-2012 credit losses came from Alt A (no- or reduced-documentation) loans and 28% came from interest-only adjustable-rate mortgages. These categories are not mutually exclusive, but they allow us to put a range on the amount of losses involved. If there were no overlap at all in the categories—that is, if no interest-only ARMs also had reduced or no documentation—then 60 percent of the $70.1 billion in losses from Fannie’s December 2007 book, or $42.1 billion, would have come from loans the company no longer accepts. At the other extreme, even if there were a total overlap between the categories, which we know there is not—not all interest-only ARMs also would have had no or reduced documentation—it still would be true that a minimum of 32 percent of the credit losses from Fannie’s December 2007 book, or $22.4 billion, would not have happened had those risk categories been prohibited back then.

So, where does that leave us on an experience-based estimate of the amount of capital Fannie (or another credit guarantor) would have to hold against its single-family credit guaranty business to survive a repeat of the 25 percent home price decline of the past crisis? Using actual 2008-2012 administrative expenses ($7.8 billion), guaranty fees only from loans on the books at December 31, 2007 and outstanding through the end of 2012 ($21.4 billion), and reducing that book’s $70.1 billion in credit losses by the smallest possible estimate—32 percent, or $22.4 billion—of high-risk loans the company no longer is permitted to acquire (resulting in adjusted credit losses of $47.7 billion), Fannie’s required stress capital on December 31, 2007 would have been $34.1 billion, or 1.34 percent of its $2.55 trillion book of business. Adjusting Fannie’s 2008-2012 credit losses to remove a more realistic 50 percent of losses on high-risk loan types and combinations would lower the stress capital requirement to 84 basis points.

In doing this analysis I do not mean to imply that one percent capital is the right amount for a reformed single-family mortgage credit guarantor (although it may be). FHFA would need to redo the exercise using detailed Fannie and Freddie data, and produce a capital requirement not just for the books of business the companies had at the end of 2007 but also for categories of risk by loan type. And it’s possible that policymakers may choose a stress scenario other a 25 percent drop in home prices as the basis for the credit guarantors’ capital requirement. But I do want to emphasize that credit guaranty capital requirements have to be based on something that can be objectively evaluated; they can’t just be made up. Capital requirements have too important an impact on which types of borrower can get a fixed-rate mortgage financed in the secondary market, and at what price.

One of the aspects of the essays submitted for the “Housing Finance Reform Incubator” series that struck me was how little attention was paid to the way the recommended or assumed capital requirements and structure of the proposed credit guaranty systems would affect the mortgage rates quoted to different classes of borrower. In most essays, the authors simply state that their proposed systems would be good for affordable housing because the guarantors will be given housing goals. Yet housing goals will have little effect if the guarantor has no practical way to turn them into business that can be financed on the ground.

In evaluating how efficiently a proposed credit guaranty system might work, Fannie’s pre-crisis configuration is a useful reference point. Fannie had a risk-based capital requirement for its single-family business, subject to a 45 basis point minimum. Everyone focused on the (very low) minimum figure, but in fact the binding constraint was the risk-based standard. That standard was based on work done in conjunction with former Federal Reserve Board Chairman Paul Volcker in 1989 and 1990; it required Fannie and Freddie to hold enough capital against their guaranteed single-family loans, by product and risk category, to survive a scenario in which the worst credit defaults and loss severities experienced by the companies over any two-year period in any region of the country with at least 5 percent of the U.S. population were repeated nationwide, for all books of business. (This standard proved adequate until private-label securitization became the dominant form of residential mortgage financing in 2004, and underwriting disciplines collapsed.)

By managing its product menu and underwriting so that its risk-based requirement remained at or below the minimum, Fannie was able to keep its average charged guaranty fee at around 20 basis points through 2004. At this level, Fannie could do what it called “profile pricing” for its larger lenders: a lender would tell Fannie what its expected mix of business would be in the coming year, and based on that mix Fannie would quote a fixed guaranty fee (typically with a provision for adjustment should the profile delivered differ materially from what was envisioned). With a known average guaranty fee, lenders were free to set mortgage rates for individual borrowers as they pleased. Generally, they would heavily or totally subsidize higher-risk borrowers by quoting them rates close to or the same as the rates for lower-risk borrowers. Lender cross-subsidization was critical in enabling Fannie to meet its affordable housing goals during this time.

What made profile pricing possible, and cross-subsidization effective, prior to the crisis were that Fannie’s (and Freddie’s) average guaranty fees were relatively low, and the differences between capital and fees on the companies’ lower- and higher-risk loans were not that large. That no longer is true. Fannie and Freddie now price to about a 3.5 percent average capital requirement, and data from FHFA show that in the first quarter of 2014 their “fully priced” guaranty fee for the riskiest third of their business was 112 basis points, while the fully priced fee for the least risky third was 43 basis points. The 69 basis point difference between the riskiest and least risky one-third of Fannie and Freddie’s business overwhelmed their ability to use cross-subsidization effectively. The fee they actually charged for the riskiest one-third of their business was only 70 basis points, while their charged fee on the least risky one-third was 53 basis points. Because they had to cut fees by 42 basis points to attract higher-risk business, but only were able to raise fees on lower-risk business by 10 basis points to avoid losing it, Fannie and Freddie together undershot their “fully priced” average fee by 12 basis points in the first quarter of 2014. Even at that, just 16 percent of their business had credit scores of less than 700, and only 23 percent had down payments of less than 20 percent (with those borrowers having to pay an additional cost for private mortgage insurance).

This is a crippled system, and it shows how difficult a challenge a “reformed” credit guarantor will face if it has an average capital requirement even close to 3.5 percent. If it underprices its credit guarantees (as Fannie and Freddie now are doing), it will not hit its return target. If it prices all its business at the average fee (which for Fannie and Freddie is 72 basis points), bank originators will keep their lowest-risk mortgages in portfolio and sell their highest-risk loans to the guarantor, sending its risk profile soaring. The guarantor will have no real alternative to pricing all of its business close to the fees dictated by its risk-based capital requirements, because the extreme spread between the “fully priced” fees on its higher- and lower-risk loans will limit its scope for lowering fees on higher-risk loans before the offsetting fee add-ons to lower-risk loans cause that business to drop off significantly. Inevitably, the guarantor will end up with very small amounts of affordable housing business, at very high mortgage rates to the affected borrowers.

It doesn’t have to be this way. Capital is the key. In theory, adding “cushions” of extra capital—or being unrealistically conservative in designing a stress test (by not counting any guaranty fee income as an offset to credit losses, for example)—may seem to be a good thing. In practice, however, these impose very real costs on the low- moderate- and middle-income borrowers the system is intended to serve. The unavoidable reality is that to give maximum flexibility to guarantors and lenders to provide the greatest amount of affordable financing to the widest possible range of borrowers, capital requirements must be set at realistic levels—high enough to meet a rigorous, defined, stress standard, but no higher. That should be the goal of all reformers.