A Deeper Dive into a Quarter Century of Mortgage Risk | American Enterprise Institute


In this email I will dig deeper into the key findings of a. immerse Research paper 2021, Findings that shed a new light in the run-up to the 2008 financial crisis. The paper is based on the first comprehensive data set of its kind for the period 1990-2019 and was published by the Federal Housing Finance Agency (FHFA) in collaboration with the Senior Adviser of the American Enterprise Institute (AEI) Housing Center, Steve Oliner (and AEI Associate Scholar Morris Davis).

Background and key findings:

  • The paper provides a summary measure of mortgage risk by estimating a Stressed Mortgage Default Rate, or Stressed MDR. The Stressed MDR takes out a loan at any point in time 1990-2019 and measures the risk of that loan as if it were incurred in 2006-2007, just before the 2008 financial crisis.
  • In the mortgage market, the Stressed MDR is the analogue of crash tests for motor vehicles or wind assessments for doors and windows in hurricane areas. In all of these cases, the goal is to assess performance under heavy stress. This measure of mortgage risk should be of primary interest to investors, homeowners, and policy makers as the stability of the mortgage market depends on its ability to withstand extreme events.
  • The Stressed MDR is derived from the Stressed Mortgage Risk Index, which was developed by the AEI Housing Center in 2013.
  • New research from the AEI Housing Center confirms that the Stressed MDR is highly correlated with post-financial default rates, including loans experiencing the COVID-19 stress event.
  • The paper begins by depicting historical market share trends from each of the major market participants / executions.
    • For home loans granted between 1990 and 2007, the companies (Fannie Mae and Freddie Mac), the FHA and the VA had a combined market share of 50% or more in all years except 2005 and 2006, over which time the same agencies had an average joint purchasing market share of around 60%
    • In relation to the refinancing loans granted between 1990 and 2007, the study documents the enormous volatility from year to year. During this period, the companies (Fannie Mae and Freddie Mac), the FHA and the VA had an average combined market share of about 55%.
  • The next step to cover is the build-up of mortgage risk as a result of credit easing, and it turned out to be starting to build up in the mid-1990s.
    • In the case of the housing loans granted in the period 1994 to 2007, the stress default rates would rise steadily and would more than double over the period.

  • For the refinancing loans granted between 1994 and 2007, the adjusted stress default rates would steadily increase and triple over the period.
    • Previous research has failed to determine that a refinancing boom from 2000-2003 masked the accumulation of mortgage risk. The authors explain: ‚ÄúDuring the refi boom, higher quality borrowers with no immediate need for cash step into the market to lower their mortgage rates. Their occurrence reduces the average risk of the borrower pool, which pushes the stressed default rate down. In contrast, when the refinance option is generally out of the money, these borrowers are marginalized, thereby targeting the pool towards higher risk, constrained borrowers who are withdrawing some of their accumulated equity or their monthly payment by extending their mortgage term even if the new interest rate is not lower than the old one. This disadvantageous shift in the borrower pool increases the stress-default rate. “

  • In relation to all loans granted in the period 1994 to 2007, the combined and adjusted stress failure rates would rise steadily and would more than double over the period.

  • Next to address is the accumulation of mortgage risk resulting from credit relief provided by the companies (Fannie Mae and Freddie Mac) and the FHA and VA (left panel), and portfolio lenders and private label securities (right panel).
    • An important finding was that lending easing began for all groups except FHA / VA, which remained roughly constant in the mid-1990s.
    • Both the company and the PLS adjusted stress default rates (all loans roughly doubled over the period 1994-2007, while portfolio loans increased 170%.
      • It’s also worth noting that the buy and refinance proportions of the portfolio fell from 40% and 45% to 25% and 30%, respectively, over the period.
    • Over the same period, the company’s stressed default rate rose from about 40% for the FHA / VA to 66% (all loans for both).
    • In the same period, the purchase credit volume of companies rose from around 1.7 times to 8 times the FHA / VA volume.
      • The Enterprises were by far the most leveraged of the persecuted groups.

Stress Default Rates for All Loans, by Market Segment
(Dashed lines adapt to changes in refi volume)

  • Next, the researchers documented the accumulation of mortgage risk resulting from credit easing by product characteristics and found that easing kicked in for all major credit mitigation metrics except creditworthiness in the early to mid-1990s (subject to data availability).

Average DTI by credit type
(Dashed lines for refis and all credits adjust to changes in refi volume)
Average CLTV by credit type
(Dashed lines for refis and all credits adjust to changes in refi volume)
Proportion of loans with little or no documentation, by loan type
(Dashed lines for refis and all credits adjust to changes in refi volume)
Share of investor loans
(Dashed line adapts to changes in refi volume)

  • As mentioned earlier, the researchers found that credit loosening, as evidenced by average creditworthiness, did not occur.

Average creditworthiness by credit type
(Dashed lines for refis and all credits adjust to changes in refi volume)

  • The researchers found that between 1994 and 2007, the stressed default rate for both loans to low credit borrowers (<660) and for higher-score borrowers (>= 660).
    • This was due to the significant credit relaxation (risk stratification) that had occurred in relation to both groups during the reporting period.
      • In 2007, loans with credit scores> = 660 had a stress failure rate of 28%, not much below the 32% rate for the group <660 in 1994.
      • In 2007, loans with credit scores <660 had a stress failure rate of 52%, compared to the group's much lower stress failure rate of 32% in 1994.
      • The credit easing, reflected in the previously noted increases in CLTV, DTI, low-doc / no-doc and investor credit, resulted in higher stress default rates as they were stratified year after year beyond 1994 risk levels.
      • It is also clear that the stress default risk level for both groups had returned to the 1994 level by 2011.

In more recent studies by the AEI Housing Center, stressed MDRs, which are calculated exclusively on the basis of the credit characteristics available when the loan was granted (in this case, conventional fixed-rate loan in 2018), strongly correlated with the default rates after the financial crisis:[1]

  • Credit performance data through November 2019 were used to test the performance of the Stressed MDR in a non-stress default environment. The MDR had an R2 of 93.7%.
  • To test the performance of the Stressed MDR in a stress delinquency environment, credit performance data up to April 2021 was used. The MDR had an R2 of 99.1%. This finding is of additional importance as it reflects an event considered an unprecedented and complex stressful event, namely the COVID-19 pandemic. Still, the Stressed MDR correlated almost perfectly with the actual pandemic failure results.

These results are not surprising, as decades of mortgage research confirm that the probability of a mortgage defaulting is strongly correlated to five main risk factors, all of which are measured at the time of issuance, and the degree to which those factors are risk-layered. The factors are: (i) combined loan-to-value ratio, (ii) creditworthiness, (iii) total debt-to-income ratio, (iv) repayment period and (v) loan purpose (purchase, no refinancing, and refinancing payout).

Current FHFA press release identified

key lessons that shed new light on the financial crisis of 2008 and beyond, and the failures of Fannie Mae and Freddie Mac

  • Past accumulation of mortgage risk: Mortgage risk began to accumulate in the mid-1990s. This was much earlier than some researchers previously thought. The new data shows that the build up of mortgage risk in the 1990s was a precursor to the market failure in 2008.
  • This previous investigation failed to recognize that the refinancing boom of 2000-2003 masked the accumulation of mortgage risk.
  • Loan relaxation risk increased for borrowers of all credit ratings: Before the 2008 financial crisis, mortgage risk increased for all borrowers, not just those with low credit ratings, as some have previously suggested.
  • Loose lending standards: In the mid-2000s, mortgage spreads between non-risky loans and very risky loans narrowed for portfolio and private label securities mortgages, suggesting an expansion in credit supply just before the 2008 financial crisis.
  • The mortgage risk in America is increasing again today: the sustained rise in house prices is increasing the mortgage risk.


The research convincingly demonstrates the erroneous nature of the widely held view that the early 2000s represented a normal period in the mortgage market and could therefore be used as a benchmark to assess whether lending standards are loose or strict in other periods. The study shows that mortgage risk began to pile up in the mid-1990s and continued until it peaked in 2006-2007.

The study reflects the conclusion of the study I wrote in 2010:[2]

The main cause of the financial crisis was the accumulation of an unprecedented number of weak or nontraditional mortgages (NTM) in the US financial system. NTMs were characterized by low or no down payments, increased debt ratios, compromised credit, reduced loan repayments, and other changes in underwriting standards. These NTMs were no accident. In the 15 years after the 1992 GSE Act was passed, trillions of dollars in under-subscribed loans would spike and then topple the housing market.

Finally, the stressful event related to the COVID-19 pandemic demonstrated the usefulness of the MDR as a tool that can be used to assess loans at risk at or after being granted, based only on the standard data known at the time of issuance .

[1] When calculating the default rates, forbearance loans were taken into account based on their actual default status, regardless of whether they were in forbearance.

[2] Pinto, 2010, Government Housing Policy Ahead of the Financial Crisis: A Forensic Study. Note: Because the data was not available to perform credit level analysis or to examine risk stratification in detail, I used the concept of NTM, credits with features that were traditionally absent in the early 1990s. I then tracked the expanded use of these NTM features over time.

Please click here to view them PDF