About one in seven first mortgages. Roughly 7.5 million first-lien subprime loans were outstanding in the spring of 2007, about 14 percent of all first liens, with near-prime lending adding another 8 to 10 percent. The harder question at the time was not the size of the market but what sat inside any given loan.
In the spring of 2007, roughly 7.5 million first-lien subprime mortgages were outstanding, about 14 percent of all first liens. Near-prime lending, sometimes described as alt-A, accounted for another 8 to 10 percent.
Subprime described the borrower rather than the loan. It referred to people seen as carrying higher credit risk, often because they had a thin credit history or other characteristics associated with default. The category had emerged two decades earlier and expanded in earnest from the mid-1990s, helped along by credit scoring, which lowered the cost of assessing a borrower, and by a secondary market that meant a lender no longer had to hold a loan in order to make one.
The size of the market was straightforward to state. What a given loan actually contained was harder to see.
Risks could stack. A thin credit history, limited documentation of income, and a high combined loan-to-value ratio could all appear in the same file. The industry called this risk-layering, and it became more common as origination volume slowed while investor appetite for yield held up.
Each element was recorded somewhere. Whether they landed together in the same loan was a different question, and one the public record of the day could not answer directly. HMDA had no field for any of it. Beginning with the 2004 data year it collected one measure of price, the gap between a loan’s annual percentage rate and the yield on a comparable-maturity Treasury, and only when that gap was wide enough to trigger reporting. Loans above the line were called higher-priced, and that flag became the closest available proxy for nonprime lending.
The economists who published those figures were candid about the limits. Some prime loans were higher-priced, and some subprime loans were not.
About two-thirds of subprime first liens carried adjustable rates, roughly 9 percent of all first mortgages, and that is where delinquencies concentrated.
The reason has less to do with the rate than with the shape of the loan. A payment could be affordable at the starting rate and scheduled to reset later, with the plan being to refinance before it did. That plan rested on conditions outside the loan and outside the borrower: house prices that kept rising and rates that stayed low. When both turned in 2006, borrowers who had counted on refinancing found they lacked the equity to qualify for a new loan.
Other features worked the same way. Interest-only periods, negative amortization, and balloon payments all moved the cost of a loan into the future rather than removing it.
Most of those structures have since been ruled out. The ability-to-repay standard, which took effect in January 2014, requires a creditor to make a reasonable and good faith determination that a borrower can repay, based on verified income and assets and on the fully indexed rate rather than an introductory one. The qualified mortgage definition built on top of it excludes negative amortization, interest-only payments, balloon payments, and terms beyond 30 years.
The reporting changed alongside the rules. Since the 2018 data year, the public HMDA record carries as separate fields what once had to be inferred: balloon payment, negative amortization, interest-only payment, other non-amortizing features, loan term, introductory rate period, combined loan-to-value, debt-to-income, and rate spread measured against the average prime offer rate rather than a Treasury yield.
The useful question was never how large subprime lending was. It was what sat inside a given loan, and whether several things sat there at once.
That question can now be asked directly. The features are individually reportable, so whether they appear together in the same loan is something to count rather than something to estimate.
That is the layer we work in. HMDAVision structures the full HMDA record so those features can be examined directly, by lender, market, borrower, product, and year.
A note on the estimate. The 2007 figures were estimates rather than counts. Loan totals and foreclosure starts came from an industry survey, adjusted for incomplete coverage of the market, and delinquency rates came from a private loan-performance database. No public source held the number directly.
A note on what is still not reported. HMDA carries debt-to-income but does not record whether income was verified. The documentation dimension of risk-layering remains outside the public record.
