Monday, May 19, 2008

FHA Rolls Out Risk-Based Premiums

by Tanta on 5/19/2008 10:28:00 AM

Ken Harney reports in the LAT:

The FHA, which for decades has used a one-size-fits-all approach to pricing its insurance on home loans, plans to shift to a "risk-based" system keyed to FICO scores and down payments, beginning as early as mid-July. Private-sector lenders and insurers have priced interest rates and premiums using sliding scales of FICO scores and down-payment amounts since the mid-1990s.

The agency's move, which will cover new applications including "jumbo" loans up to $729,750 in high-cost markets through December, will bring the FHA in line with the private sector's main approach. . . .

Under the old approach, [Montgomery] noted, buyers with stellar FICO scores paid the same premiums as borrowers with poor scores. That amounted to a pricing inequity for applicants who presented a low risk of default on loans and an inappropriate subsidy of applicants who were likely to default.

A study of an entire year's applications turned up the additional fact that the FHA's lower-income borrowers typically had higher FICO scores than those with larger incomes.

"Is it counterintuitive? Yes," Montgomery said.

According to the study, applicants with FICO scores of 680 to 850 had a median income of $48,756 last year, while those with low scores of 500 to 559 had a median income of $53,388. Fair Isaac Corp.'s FICO scores range from about 300 to 850 -- the higher, the better -- and are predictive of future defaults and foreclosures. Even at rock-bottom down payments of 3%, applicants with lower incomes had higher credit scores than applicants with bigger incomes making similar-size down payments.
I would like to get my hands on that study, which I haven't yet found online. (If any of you have a link, please drop it in the comments.) I did locate this HUD document that outlines the actual premium schedule currently proposed; it provides only one chart with aggregated information on income/FICO breakouts for 2007 FHA applicants in the Appendix. You might be interested to know that the original proposal for risk-based premium pricing distinguished between source of downpayment funds, with borrowers using such things as the notorious DAP (seller-funded "assistance") paying higher premiums than borrowers making downpayments from their own funds or a gift from relatives. That provision has been eliminated.

As far as I know, there is better data on relative performance of FHA loans with or without DAP than there is on relative performance of FHA loans with FICOs just over or just under 600. But HUD is forging ahead with FICO-based pricing while pulling back on downpayment-source-based pricing. I'm having some problems with that.

I think it's important to concede that FICOs do indeed have an established track record of establishing relative default probabilities. The trouble we have had recently with FICOs is mostly, in my view, that they were relied on to offset extremely high risk characteristics in loans with a lot of "risk layering." The problem is not that a 90% LTV loan with a 720 FICO won't outperform, statistically, a 90% loan with a 620 FICO. It will, although it isn't always clear that the difference in performance is all that substantial. The problem is that a 100% loan with a 720 FICO will not necessarily outperform a 90% loan with a 620 FICO. The idea that a high(er) FICO offsets lack of downpayment or high DTI is currently dying a painful death.

The difficulty, however, with deciding that a (theoretical) 90% loan with a 720 should pay a lower risk premium than a 90% loan with a 620 (all other things being equal) is the problem of calibration. How significant is the difference in default probability? Of course, with the FHA premiums the question is how significant the difference in default probability is in some pretty small FICO buckets that are already well within the "subprime" or high-default-probability range. For instance, in the highest-LTV group of loans, the new upfront premium is 200 bps for FICOs from 560 to 599 and 225 bps for FICOs from 500 to 559. But is there really a significant difference in default probabilities in these two FICO buckets? Enough to warrant 25 bps in premium? I would really like to see more information on how HUD calculated all this.

Part of what is bothering me is this set of charts provided in a recent Moody's investor presentation. These numbers are based on a representative sampling of loans closed between the first quarter of 2006 and the second quarter of 2007.

This is of course only one data point, but it certainly raises a question in my mind about making FICO-based premium distinctions within the general category of subprime FICOs, particularly since in the new premium scheme a loan with a downpayment made by the builder gets treated the same as a loan with a downpayment coming from the borrower's own funds. I would really like to see the work on this one.

It gets Nerdier from here . . .