A Quarter Century of Mortgage Risk
Morris A Davis, William D Larson, Stephen D Oliner, Benjamin R Smith
Review of Finance, Volume 27, Issue 2, March 2023, Pages 581–618, https://doi.org/10.1093/rof/rfac034
This paper provides a comprehensive account of the evolution of default risk for newly originated home mortgages over the past quarter century. We bring together several data sources to produce this history, including loan-level data for the entire Enterprise (Fannie Mae and Freddie Mac) book. Our dataset includes more than 200 million home mortgage loans originated from 1990 through 2019. All the results presented in the paper and many other series are available for download at https:/www.fhfa.gov/papers/wp1902.aspx.
We calculate a stressed default rate for every loan in the dataset based on the observed default experience of similar loans originated nationwide in 2006 and 2007. The stressed default rate for a given loan represents its expected performance had it been hit shortly after origination with a replay of the financial crisis and experienced the national average decline in house prices.
Figure 1 shows the stressed default rate for all loans in our sample. The figure clearly shows that the riskiness of mortgage originations in the aggregate rose over the entire period from the mid-1990s to 2006, indicating that seeds of the financial crisis were planted far in advance of the event itself. This finding cuts against the common view that mortgage lending conditions were normal in the early 2000s. After the crisis, mortgage risk fell precipitously and has remained low since.
We analyze the characteristics of mortgage loans that account for the increase in risk over 1994-2006 and find more than half of the rise was due to “plain-vanilla” risk factors like rising loan-to-value ratios and increasing debt-payment to income ratios. Thus, a narrative that focuses primarily on risky product features, such as loans with low or no documentation of income, overstates their role during the boom and underplays the risk-increasing effect of more prosaic forms of leverage.
We also use the data to evaluate the role of “subprime” borrowers (borrowers with a credit score below 660) in the housing boom. We find that the share of mortgage loans made to subprime borrowers was flat on net from 2000 through 2006, the primary years of the housing boom, though this share rose somewhat in the 1990s. We also show that over the entire period from 1994 to 2006, the stressed default rate for subprime borrowers moved in close alignment with the stressed default rate of borrowers with higher scores. These findings are not favorable to a “subprime-centric” view of the financial crisis.
Although our full dataset ends in 2019, we use a subset of the data to provide an update extending into the COVID-19 episode. This update includes Enterprise, FHA, and VA loans and shows that the riskiness of mortgage originations fell as the mix of both home purchase and refinance loans shifted toward lower-risk borrowers.
Figure 1: Stressed Default Rate, All Loans, 1994-2019
Note: The series shown covers first-lien home purchase and refinance mortgage loans secured by 1-4 unit properties. It includes a regression-based adjustment to control for changes in refinance volume. Shading is for 2000-2003.