Seven months into the pandemic, the forbearance and delinquency rates of the broader mortgage market appear to have crested and are now slowly working their way back down. Within the agency credit risk transfer (CRT) sector, the forbearance plan periods range from six to 12 months. (CRT is considered non-agency securitization and comprises loans that fell under the Coronavirus Aid, Relief, and Economic Security [CARES] Act.) These relatively protracted plans imply that loans in these CRT transactions will exhibit longer times to exit forbearance relative to other non-agency transactions.

In this commentary, we examine forbearance populations and provide various scenarios to determine the stress levels that would impair a generic B-1 tranche (oftentimes the bottom-most rated class) in representative hypothetical transactions. CRT transactions possess structural features that insulate them from the likelihood of interest shortfalls to bondholders. Furthermore, credit score is one of the main predictors of a borrower's need for assistance. Sensitivities to loan-to-value ratio (LTV) appear to be less influential, which is likely due to the soundness of fundamental characteristics currently underpinning U.S. residential real estate.

CRT Delinquencies Versus Other Non-Agency Subsectors

Chart 1 shows a monthly total delinquency rate for CRT relative to prime jumbo 2.0 (issued after 2008) and nonqualified mortgage (non-QM) residential mortgage-backed securities (RMBS). As with other non-agency mortgages, the majority of delinquent CRT loans are also in a forbearance plan; therefore, the delinquency information in the chart is a good proxy for the overall forbearance population.

Chart 1

While the delinquencies in CRT pools have been lower than what we have observed in non-QM pools, they are higher than those of jumbo 2.0 pools. This ranking is due to the relative credit strengths: average FICO scores for prime 2.0, CRT, and non-QM pools are 760, 740, and 700, respectively. Non-QM is also a sector with large exposures to self-employed borrowers who use alternative income documentation, which is for the most part weaker than that of CRT collateral.

The relatively long forbearance time frames applicable and offered under the CARES Act to the government-sponsored enterprise (GSE) loans (which the CRT transactions reference) compared to non-agency pools is another factor that could be affecting the CRT delinquency numbers. This would cause a slower decline in delinquencies in CRT transactions relative to the other sectors if borrowers do not voluntarily exit the forbearance plan early. That said, the degree of delinquency decline for CRT transactions may be biased in chart 1 (above) because Fannie Mae uses a two-month reporting lag (i.e., the dates in the chart pertain to the securitization reporting month, whereas the borrower due date is typically the month prior).

When observing delinquency levels for some low-LTV and high-LTV CRT transactions as of the September reporting period, we found that the share of 30+ day delinquencies for low-LTV CRT transactions was roughly 5.5%, while the share was closer to 9.0% for high-LTV ones.

The geographic presence of forbearance

Generally, CRT pools are geographically diversified. This feature has been beneficial during the pandemic because COVID-19 has had disproportionate impacts in certain regions. For example, the pandemic has hit the states of California, Florida, and New York particularly hard (in terms of number of cases and deaths, as well as unemployment). This is relevant for jumbo 2.0 RMBS because we observe high California loan concentrations in these pools. Moreover, roughly 70% of non-QM collateral in transactions we have evaluated is concentrated in these three states (see "Non-QM RMBS And COVID-19: Locking Down States' Exposure," June 1, 2020).

Charts 2a and 2b display the ranking of core-based statistical areas (CBSAs) based on percentage of loans with borrower assistance (forbearance, deferral, modification, etc.) for the top-50 CBSAs (by population) represented in sample low-LTV and high-LTV CRT collateral pools. Regions more reliant on the hospitality industry have increased exposure to the effects of the pandemic. Although certain CBSAs have higher proportions of borrower assistance, we reiterate that CRT pools benefit from stronger geographic diversification than most other non-agency pools.

Chart 2a

Chart 2b

The Near-Term Stakes

A structural snapshot

The jumbo 2.0 and non-QM sectors have seen losses trickle in for some securitizations as forbearance plans are resolved. However, the degree of losses to date has been relatively minor and predominantly covered by excess interest for certain structures. In some non-agency securitizations, the use of forbearance has resulted in reduced interest from loans in forbearance not advanced on, or servicer advance reimbursements for those that are advanced on (see "Can COVID-19 Cause A Cash Crunch For Certain U.S. RMBS?," Aug. 21, 2020). The good news for CRT is that its synthetic structure mitigates this interest shortfall risk because interest payments do not depend on interest collections from the reference obligations, and are ultimately the responsibility of the respective GSE.

The credit profile of borrower assistance loans

Federal government consumer stimulus plans offered additional fixed amounts of unemployment compensation, often to those borrowers who were entering into forbearance plans. Stratifying by whether or not loans received borrower assistance within both low-LTV and high-LTV sample transactions, we see marked differences in FICO score and debt-to-income (DTI) ratio (see table 1). It is therefore no surprise that such factors correlate with forbearance rates. In order to differentiate and identify borrower/loan characteristics that were more influential in forbearance rates, we analyzed samples of low- and high-LTV CRT transactions in our loan-level credit model to compare the populations with borrower assistance against populations without (delinquencies that began in March 2020 and after were not considered as factors when projecting the defaults). Table 1 shows some of the differentiating loan characteristics and the resulting default, loss severity, and loss projections under a 'B' rating scenario in our credit model for the two sample pools.

Table 1

Our loan-level credit analysis considers the somewhat weaker characteristics of loans needing borrower assistance, which is reflected in the 'B' loss coverages. Loans needing borrower assistance are in general around 1.7x-2.0x more likely to default compared to loans not needing borrower assistance.

These results indicate that loss severity is largely unchanged under the comparison. This suggests that LTV is of secondary importance as a predictor of forbearance-related defaults, possibly taking a backseat to FICO score as a predictor in the current environment. This could be due to the relatively healthy U.S. residential real estate market and the fact that home values are generally supported by housing market fundamentals.

It is also possible that FICO score is a more direct measure of consumer behavior (e.g., it may indicate whether the borrower exercises the option of forbearance when immediately made available) compared to LTV. The LTV, however, may influence an ultimate default scenario as time passes should residential real estate prices experience downward pressure.

Borrowers with assistance had balances roughly 15% higher than those without, suggesting that borrowers with higher monthly debt burdens are more inclined to seek assistance. For those with assistance, there was a significantly larger percentage of loans with DTI ratios greater than or equal to 43%. Borrowers with assistance also had slightly higher weighted average coupons (WACs), suggesting that credit risk factors were already somewhat priced into the mortgage rate.

The Credit Impact

We recently resolved CreditWatch negative placements on classes from five CRT transactions (see "Various Rating Actions Taken On 139 Ratings From Five U.S. RMBS Credit Risk Transfer Transactions," Aug. 20, 2020), resulting in one-notch downgrades for 72 classes, most of which were modifiable and combinable real estate mortgage investment conduit (MACR) classes, with four root classes also being downgraded (ratings were in the 'B' to 'BB' range). For our analysis, we surmised that our revised archetypal foreclosure frequency assumption (see "Guidance: Methodology And Assumptions For Rating U.S. RMBS Issued 2009 And Later," April 17, 2020) adequately accounts for forbearance levels and that no additional adjustments were necessary. As such, loans that we deemed to be delinquent solely due to being on a temporary forbearance plan did not get a delinquency adjustment in our credit analysis. The tranches with the most downgrade exposure were the B-1 and M-2 classes and their associated MACR classes, which affected more recently rated transactions that had less time to de-lever and less time to benefit from home price appreciation since closing.

On average, the population of CRT in forbearance is roughly 5.0%-7.5%, depending on the remittance month. Given the general credit enhancement for M-2 tranches in conjunction with our base-case loss expectations, we focused on the sensitivity of more junior B-1 bonds to get a sense of what percentage of outstanding forbearance would be needed to default at a given loss severity in order to incur a principal write-down. As an illustration, we consider hypothetical tranches with average B-1 thickness (bond size as a percentage of collateral balance) and average B-1 credit support using the representative low-LTV and high-LTV transactions.

Tables 2a and 2b show the degree to which a hypothetical B-1 bond (assuming representative credit support and class thickness from 2019 vintage transactions) would be written down at different terminal default rates across different loss severity levels. For example, if 6% of loans are in forbearance for this representative low-LTV transaction and 33% of them end up defaulting, resulting in a 2% terminal default rate (33% * 6%), the hypothetical B-1 bond would not incur a loss from them, given a 20% loss severity (roughly our base-case severity assumption for low-LTV pools).

Table 2a

Table 2b

If one were to assume a 3% default rate for a low-LTV pool from the nonforbearance portion of the pool, adding to it the aforementioned 2% default rate for loans in forbearance would give a total terminal default rate of approximately 5%. Given a 20% loss severity, this would cause an approximate 11% write-down of the hypothetical B-1 bond. Similarly, for the hypothetical B-1 bond in the high-LTV transaction, a 5% default rate from the nonforbearance portion plus another 2% default rate for loans in forbearance would give a total terminal default of approximately 7%. In this case, a 25% loss severity would cause a 46% write-down for the B-1 tranche.

In both instances, assuming loss severities of 20% and 25% for low-LTV and high-LTV, respectively, roughly two-thirds of the loans in forbearance (assuming 7.5% of loans in CRT pools are in forbearance) would need to default to impair the B-1 bond, absent additional defaults from nonforbearance loans.

We have observed that the 2019 vintage transactions have recent conditional prepayment rates (CPRs) over 50%. All other vintages have remained below 45%. This could be due to the temporary increase in mortgage rates that occurred at the end of 2018 to the beginning of 2019. Unlike borrowers who took out mortgages in the lower rate environment, these borrowers are now more "in the money" and incentivized to prepay in the currently record-low rate environment. This is noteworthy because prepayments can offset some potential lifetime defaults as they exit the pool prematurely and reduce the population available to default.

The Path To Forbearance Resolution

As the first six months of forbearance winds down (or perhaps is extended), the prevalence of the various possible forbearance outcomes will unfold. The ultimate credit impact will depend on how many of these borrowers in forbearance can be reinstated to their regular payment, or even some modified payment plan. Therefore, the forbearance resolution will be a key factor in future credit performance. Information from Fannie Mae suggests that the most common resolutions to date have been reinstatements, followed by deferrals, repayment plans, and modifications. Reinstatement implies that borrowers either exit forbearance after having not missed any payments or (in some cases) make lump-sum repayments.

Several months ago, the proportion of forbearance loans that were producing cash flows (forbearance notwithstanding) was about 40%. More recently, that number fell to roughly 25%. Data from several months back indicate the percentage of borrowers reinstating was higher, which suggests that the remaining forbearance borrowers may not necessarily possess the same credit/financial status needed to be reinstated. Furthermore, certain borrowers may have requested a forbearance plan early in the pandemic as a safety net. The approval requirements for COVID-19 related forbearance for GSE borrowers are typically more lenient than those for non-agency borrowers. Non-agency borrowers may need to prove a greater threshold of income loss or curtailment due to COVID-19. Agency borrowers, on the other hand, may only need to provide an attestation of hardship.

Absent new lockdowns across the country, we expect that a portion of borrowers will continue to be reinstated and taken out of payment forbearance. There may be an initial roll-off bias, however, meaning that the first wave of borrowers to exit forbearance will likely be those who needed forbearance/assistance the least. Such adverse selection suggests that the profile of resolution will begin to differ from what we have observed thus far. Forbearance extensions and subsequent payment deferrals may be more prevalent in the medium and long term. The good news is that, given low interest rates and strong residential housing fundamentals, a portion of the loans have exited forbearance as a consequence of prepayments. As long as these fundamental underpinnings of U.S. real estate continue, prepayment activity could bode well for credit as it relates to forbearance. In any case, the end state for CRT forbearance (and other non-agency transactions) will largely depend on economic conditions, which, in turn, will depend on the course of the pandemic.

S&P Global Ratings acknowledges a high degree of uncertainty about the evolution of the coronavirus pandemic. The current consensus among health experts is that COVID-19 will remain a threat until a vaccine or effective treatment becomes widely available, which could be around mid-2021. We are using this assumption in assessing the economic and credit implications associated with the pandemic (see our research here: www.spglobal.com/ratings). As the situation evolves, we will update our assumptions and estimates accordingly.