BLOG VIEW: In three months investor concern for mortgage losses went from distant thought to major valuation worry.
An unprecedented virus spread has driven U.S. unemployment to depression levels. Mortgage security and whole loan sales have exploded, and the credit markets are now supported by Federal Reserve purchases.
The best benchmark for mortgage loss prediction today is the 2004-2013 period, during which, albeit at a much slower pace, residential foreclosures went from a trickle to the highest level on record. The challenge today is to adjust history in order to evaluate mortgage assets in an environment where the pace of economic change is rapid.
The Great Recession Mortgage Loss Story
Residential mortgage losses were very low to nonexistent in the early years of the 21st Century, as depicted by the 2004 non-agency prime 30-year fixed-rate graph below. Only about one in 100 loans defaulted in this “prime” 2004 book year. The economy was strong and underwriting standards required larger down payments and higher credit scores. The 2004 book FICO scores averaged 725 and loan-to-value ratios averaged 72%.
The vast majority of loans were fully documented and full appraisals required. Prime securities required very little credit support to achieve high AAA bond concentrations of 95-97% of the total securities in a transaction. Even mezzanine bonds were sold at low yields as mortgage loss expectation was extremely low. However, mortgage originators and security dealers were expanding their offerings and transactions, known as “Alternate A” and “subprime,” due in part to the lower origination volumes prior to the refinance wave of 2004.
Appendix B contains a number of historical data sets including 30-year mortgage interest rates since 2000. Early originations of Alt A and subprime loans had low LTVs as higher down payments were required to offset lower FICOs and easier borrower documentation.
The 2004-2005 Alt A book had average LTVs of 75%, FICO scores were 723, and most loans were owner occupied. Loss levels remained lower as cumulative losses for the 2004 Alt A production year were only 3-5% of total originations. The primary and secondary markets for mortgage securities exploded. As did derivatives of those securitizations such as collateralized mortgage and debt obligations, the ABX subprime index and credit default swaps.
Underwriting standards remained conservative pre-2005, as mortgage originators were refinancing large volumes of loans and booking significant gains on sale. Higher margins in non-prime mortgage production profitability were at all-time highs. Home price appreciation was nearly 10% annually for a number of consecutive years.
When the refinance boom ended, the industry went after mortgage purchase business with a vengeance. Loan standards were relaxed across all products as low down payments, lower FICO scores, streamlined documentation and appraisal were “layered” on the same loan.
Appendix A contains 2006 subprime book year statistics. The theory being that borrowers, no matter the number of derogatory loan attributes, would pay to stay in their primary residences. More investor loans were made to support the “flipping” market; however, it is unclear what the justification was for that underwriting expansion. Cumulative losses in the 2006 subprime 30-year fixed-rate origination year reached 10-15%; nearly one-in-three loans defaulted. The adjustable rate subprime books approached 50% cumulative losses with some pools having default percentages and severities of 70%. Conventional agency loans reached annual default rates of 10-14%; levels no one expected. The “stacked” underwriting experiment ended very badly as housing prices declined rapidly; borrowers lost their homes and security investors, mortgage insurers, and federal mortgage agencies went bankrupt. The non-agency market is still not what it was and agency loans are purchased by entities still in receivership.
A Virus Triggers Economic Chaos and Mortgage Market Concerns
In March 2020 the U.S. economy essentially shut down, as most of the world’s businesses closed and citizens “locked down” to reduce the spread of the coronavirus. A terrible event that in two short months resulted in the following:
- More than 350,000 worldwide deaths; more than 100,000 in the U.S.;
- Asset liquidation and bankruptcies from highly leveraged and low margin businesses;
- U.S. unemployment rates approaching 20%;
- Unprecedented worldwide monetary and fiscal support to combat frozen credit markets, job losses and a shutdown in consumer purchases; and
- Mortgage deferment programs for borrowers unable to make payments.
A recent snapshot of economic results shows the quick deterioration. Of note is the significant decline in retail sales (in this case represented by an annual change in a retail index) and large increase in unemployment.
Mortgage investors are faced with determining the potential for loan loss in the fastest economic decline in modern history. The “science” is knowledge taken from the Great Recession of 10 years ago, while the “art” is how to adjust that history based on recent economic turbulence.
The number one question is: What does 20% unemployment adjusted for some recovery mean for mortgage defaults?
To start, let’s select a data series such as the 6-2006 to 2-2009 period, where we had both housing appreciation and depreciation. That dataset shows some remarkable correlation between the annual change in retail sales, unemployment and housing values. When housing was going strong, retail sales were growing and unemployment (already pretty low) was the same or declining; when housing took a downturn, the change in retail sales was negative (on a normally ever increasing index) and unemployment was increasing.
Nationally, home prices went down approximately 35%. Doing a simple fit of changes in retail sales (CHRS) and the change in unemployment (CHUN) we get a projection of housing price appreciation (HPA). The 30-year mortgage rate, believe it or not, isn’t predictive of home purchase activity (but rather refinance activity) as interest rates are slower to react to a struggling economy. The 30-year mortgage interest rate data supports this notion.
How do we use this very simplistic housing market projector when retail sales and unemployment are already well outside the bounds of our model’s worst cases?
We project large recoveries. Getting half lost jobs back gets us to Great Recession levels of unemployment of 10-12% – a historical period associated with housing price declines of 35%. Yes, there are other considerations, such as, residential building supply was higher in the Great Recession, and investor speculation today is not what it was then. The important point is we need a significant recovery to get back to troubling times.
So we should, at a minimum, make the Great Recession our worst case scenario (as a case can be made for higher housing depreciation).
Another scenario may be a rebound to high single-digit unemployment and a period of flat retail sales.
Last, we get back to 3-4% unemployment and retail sales are up 8-10% annually.
Approximating, that makes three cases for housing down 35%, down 17.5%, and no change. Our “baselines” determined from historical default rates and loss severities, now we can adjust those baselines for differences in loan characteristics as needed.
Different LTV ratios, FICO scores, documentation styles, loan purposes, and occupancies can significantly impact loss expectation. FICO and LTV ratios show the highest variance in mortgage losses. The 2006 30-year fixed rate subprime bar graphs reflect the poor performance of lower FICO scores and down payments.
Appendix A, model development, is a summary of the loan loss estimation techniques used.
A takeaway from the appendix is pool loss estimation is pretty accurate, however, individual loan loss prediction, better data science or not, is unreliable.
A regression fitted to the population of loss loans from the 2006 subprime book can be used to forecast pools (even if they are not subprime in current mortgage world parlance) altered by attributes that differ from history. For example, many LTV- FICO-balance-occupancy combinations may show similar 25% historical default expectation, but changing one or more of the variables can result in considerable estimate variance.
Loans backing securities originated in the past few years have underwriting standards superior to those of the Great Recession and, hence, better loan loss expectancy. An analysis of non-QM transaction DRMT 2017-2A adjusts loss projections by different underlying loan characteristics.
Example: Analysis of Non-QM DRMT 2017-2A
DRMT 2017-2A has not incurred any losses to date. In 2020, paying loans have quickly declined from 81% to 67% of the pool. The security has 224 loans, $78 million in current balances, 70% are adjustable rate loans and 30% are fixed rate. A high-level summary of other key DRMT 2017-2A statistics are as follows:
The calculation of loan losses for the worst case scenario utilized the following assumptions:
1) Baseline cumulative losses for the ARM portfolio and fixed rate portfolios of are 39% and 20%, respectively.
2) Long-term prepayment speeds of 20% for ARMS and 10% for fixed rate.
3) Adjustments to loss curves due to overall better portfolio characteristics for a decrease in cumulative losses to 36% for ARMS and 18% for fixed-rate loans.
4) Three servicer advance payment scenarios of 80%, 40%, and 10%. The mortgage market received a “gotcha” when regulators forgot to cover payment advances for mortgage servicers when mortgage deferment programs were launched. Questionable strategy as housing is a consumers’ number one expense. Regulators fixed the mistake on the agency side but non-agency buyers need to quantify liquidity pressures on loan servicers.
5) Two foreclosures to REO timelines of 24 months and 42 months are used. Foreclosures will take longer with judicial state processes, stressed servicer operations, and pro-borrower public sentiment extending timelines.
Five bonds in the transaction A1, M1, B1, B2, and B3 are evaluated. Just like in 2004-2009 investors will have to adjust their risk taking thinking as mortgage losses will change opinions on bonds considered “bullet proof” even in the case of higher credit enhanced transactions such as DRMT 2017-2A.
The collateral and bond evaluation of DRMT 2017-2A uses combinations of three loss scenarios, three advance scenarios, and two foreclosure timelines. The A1 has considerable credit support (but don’t be surprised when AAA spreads widen on higher loss expectancy) and the B2 and B3 bonds expire (from losses) in all scenarios. The M1 and B1 bonds are the focus. The M1 bond has losses only in our worst case scenario. The B1 bond is a complete loss in the worst case, has no losses if a pre virus loss expectation materializes, and varying degrees of loss in our “middle” case.
Losses are likely to be considerably higher in relation to current market expectation. For those who understand loss expectation versus credit enhancement, opportunity will surface when market sentiment “flips” and yields are attractive even in the highest realistic loss scenarios.
Will This Time Be Different for Mortgage Losses?
If I had to choose between an unprecedented economic decline and a large recovery or a slow economic deterioration that “peaks,” I would choose the latter. I believe we are, no matter the recovery scenario and rationalization of bad economic conditions, going to have lasting damage from one in five people being unemployed.
Will this time be different for mortgage losses? The strongest case suggests we will see economic deterioration consistent with the Great Recession or at best our bounce back will be insufficient to stop considerably higher residential mortgage loan losses.
Mortgage underwriting never stooped to the lows of the past, so that’s a positive. Generally, the “layering” of higher loss loan attributes (like high LTVs and low FICOs and reduced documentation) is not absent today, but better. Yield spreads on securitized mortgage product are tight, in my view, like other markets pricing in quick recovery from massive stimulus (even if there isn’t enough or quick enough money creation to cure all market and economic ills).
The opportunity to purchase bonds pricing-in harsher loss expectancy will happen but for now caution is advised.
Nick Krsnich is managing member of JMN Investment Management.