As the investor appetite for non-qualified mortgage (non-QM) U.S. residential mortgage-backed securities (RMBS) grows, competition among originators and sellers appears to be driving down the effective interest rate charged to borrowers, as measured by the spread to a relevant benchmark. How are various loan-level attributes contributing to this spread? We performed an analysis of the underlying loans of non-QM transactions rated by S&P Global Ratings to determine this. The analysis also allowed us to understand how the spreads are changing over time. We discuss our methods and findings in this publication.

Non-QM's Meteoric Rise

Non-QM loans collateralize the fastest-growing segment of the non-agency RMBS market. The surge in the number of non-QM issuers and loan sellers/originators (which we discussed on in our Sept. 20, 2019, commentary, "Non-QM's Meteoric Rise Is Leading The Private-Label RMBS Comeback") has propelled the exponential growth in non-QM issuance over the past four years. Although annual doubling of volume is unsustainable long term, we are forecasting continued non-QM issuance growth this year, with an expected $35 billion in 2020, up from $25 billion last year.

The broader non-agency RMBS market has shown stable growth over the past several years, with issuance increasing from $70 billion in 2017, to $95 billion in 2018, and then to $124 billion in 2019. Initially, credit risk transfer (CRT) and seasoned re-performing loan (RPL) transactions drove this growth. If the non-QM market continues to expand, however, it will surpass other segments of non-agency RMBS in terms of annual issuance volume this year. The combination of increased non-QM issuance, potential government-sponsored entity (GSE) reform, and the impending expiration of the QM patch (which may limit the types of loans the GSEs can buy depending on what changes may prevail), could further accelerate the growth of the non-agency RMBS market.

The Data In Our Study

In order to understand the drivers and trends in non-QM interest rate spreads, we analyzed the underlying loans of 74 non-QM securitizations rated by S&P Global Ratings with issuance dates ranging from February 2017 through November 2019 (note that some of the loans themselves were originated before 2017). We excluded interest-only (IO) loans and adjustable-rate mortgages (ARMs) with fixed periods other than five years in order to be consistent with available benchmark data. Our final population had over 40,000 loans, all of which were either five-year ARMs or 30-year fixed-rate mortgages (FRMs). Although investor property loans typically fall outside the scope of the QM rules, we included those in our data set because they are considered part of the broader non-QM market. As a benchmark to gauge the market perception of marginal risk embedded in non-QM loans, we used the Freddie Mac survey rate (30-year FRM or five-year hybrid ARM as applicable). Subtracting the benchmark from the rate gives a "spread" that incorporates the additional interest rate premium associated primarily with credit risk and relatively limited liquidity associated with loans designated as non-QM. In our analysis, we assumed it takes about two months to originate the loan and therefore lagged the Freddie Mac rate accordingly, assuming that the non-QM interest rate was set close to the time of application.

ARMs dominate non-QM space

Relative to the conforming market, the non-QM space has a greater proportion of ARM loans. ARMs make up about 10%–15% of the total mortgages in the agency space. In the case of non-QM, however, the figure is roughly 65%. The dominance of ARMs in the non-QM space may be partly driven by lenders. In recent years, the GSEs have reliably securitized roughly $1.5 trillion in conforming mortgages, which allows them to reduce exposure to interest rate risk associated with FRMs. The much smaller non-QM market, with multiple small-scale lenders, has less certain deal flow and is thus exposed to interest rate risk if FRMs cannot be securitized in a timely fashion. ARM originations enable non-QM lenders to mitigate this interest rate risk.

Another reason for the prevalence of ARMs in the non-QM space lies with the borrower. It is possible that certain borrowers turn to the the non-QM market to acquire a short-term bridge loan, which will be prepaid once the borrower "cures" a credit attribute that had prevented them from acquiring a conforming mortgage in the first place. In this case, there is no point in paying more to lock in a long-term FRM. It is possible that these borrowers expect to obtain agency financing in the near future. Alternatively, they might intend to acquire another non-QM loan with better terms. Affordability, as well as active cash management by seasoned investors, could also be driving the prevalence of ARM loans in this space.

Distribution of non-QM interest rates is wide

There is a reasonably wide distribution of interest rates observed in the non-QM market. For FRM and ARM loans, the range is mainly between 5% and 9%, with the arithmetic mean for both around 6.8%. Charts 1A and 1B show the distributions of interest rates for FRM and ARM non-QM loans in our dataset. This wide dispersion in non-QM rates is attributable to the broad array of credit attributes, such as FICO and CLTV characteristics, income documentation, loan purpose, and occupancy type that we see in this space.

Chart 1A

Chart 1B

Spread by non-QM subsectors

In our Sept. 20, 2019, commentary, "Non-QM's Meteoric Rise Is Leading The Private-Label RMBS Comeback," we defined several categories for non-QM loans (which are neither mutually exclusive nor exhaustive). Given the array of categories in the non-QM space, we provide the following breakdown:

• Foreign national (FN): Loans made to foreign nationals or non-permanent residents.
• Alternative income documentation (Alt Doc): Loans primarily underwritten using Bank/P&L statements.
• Prior credit event (PCE): Loans to borrowers with a history of housing related credit event.
• Other documentation: Typically includes debt service coverage ratio (DSCR) loans, which are investor properties for which the loan is underwritten to rental cash flows from the property instead of to the borrower's income. Also includes asset depletion loans.
• Debt to income (DTI) over 43%: Loans to borrowers with DTI ratio over 43% and that are ineligible for GSE purchases.

Charts 2A and 2B show the average mortgage interest rate over time for each of these categories for both FRM and ARM loans. For reference, we have also plotted the relevant (fixed- or floating-rate) Freddie Mac survey rate, lagged by two months, which lies well below the other curves.

Chart 2A

Chart 2B

In both charts, the spread-to-benchmark of FN loans stands out as being typically higher than the other categories during 2017. However, it is difficult to conclude anything from this pattern because there were relatively few 2017 vintage FN loans in our dataset, meaning this could be a statistical anomaly. The PCE category, on the other hand, generally lies above the other categories. This suggests that lenders are particularly sensitive to weaker credit as indicated by the borrower's past behavior (note that the borrower's FICO score should reflect much of this marginal risk). In our categorization, PCE includes loans for which a housing event (e.g., foreclosure, short sale, deed-in-lieu) transpired less than three years (two years in the case of or bankruptcy) from the securitization cut-off date.

The Results Of Our Study: The Spread Trend

The average spread to the Freddie Mac rate for our analyzed loan population is roughly 250 basis points (bps) in the case of FRMs and 315 bps in the case of ARMs. Most of the spread in the non-QM note rate derives from perceived credit risk, which depends on non-QM loan credit characteristics. Limited liquidity for non-QM loans relative to that of conventional/agency loans, as well as the current regulatory regime for non-QM, may also influence the note rates. As described in our earlier publications on the topic, non-QM loans can range from prime to non-prime depending on why the loans are designated non-QM in the first place. We therefore performed an analysis (including a regression) of how different non-QM credit characteristics affect spread over time.

Chart 3A

Chart 3B

Charts 3A and 3B show the spread-to-benchmark of the average non-QM rates for FRMs and ARMs over time. The pattern suggests that for both non-QM FRMs and ARMs, spreads have tightened since early 2017. By fourth-quarter 2018, however, the trend appears to have reversed, especially in the case of FRMs, and spreads have widened back to the levels seen between fourth-quarter 2017 and first-quarter 2018.

One can interpret the spread as a measure of marginal risk: as it decreases, the market perceives borrower repayment risk as lessening, and vice versa. As part of our ratings process, we determine measures of risk based on collateral and borrower characteristics (such as the FICO score and the LTV ratio) for each loan and determine expected losses at different rating levels (which correspond to economic environments). We have plotted our 'AAA' loss coverage (LC) at the time of issuance, alongside the spread in charts 3A and 3B. It is interesting to note that the LC compresses to some extent in 2017 for both FRMs and ARMs. More striking, however, is that in 2018 and 2019, when spreads increase, the LC also increases. This suggests—but does not prove—that our models are attributing increased credit risk to these loans for reasons similar to those of the non-QM lenders.

Tables 1 and 2 provide more detail into the credit attributes, 'AAA' LC, and spread. Over the past few years, the 'AAA' LC has steadily risen. This is attributable, in part to the upward drift in LTV ratio. Documentation type, including underwriting to property rental income (i.e., DSCR loans), is another important factor that influences the LC. As the prevalence of alternative and other loan documentation types has increased in the non-QM space over time, our 'AAA' LC levels have risen even when FICO and LTV and have remained relatively stable. Meanwhile, the spreads decreased over time, perhaps due to lenders' increased comfort with the non-QM product, as well as to competition in the space (although they have ticked up in the past few quarters).

It is important to understand that both the 30-year FRM and five-year ARM conforming benchmarks respond relatively quickly to the broader economy, typically moving with the 10-year Treasury yield. The non-QM rates, on the other hand, are slower to respond to macroeconomic variables, and are therefore somewhat "stickier." This means that our definition of spread may reflect this inertia in addition to credit and liquidity risk. For example, the recently observed spread widening could be a result of the sudden decrease in the Freddie Mac conforming rate during 2019 (see charts 2A and 2B), as well as sources of market risk specific to the non-QM market.

Applying a statistical analysis

One obtains a more systematic understanding of the non-QM spread via a statistical analysis. We used a linear regression to relate the spread-to-benchmark to various loan-level collateral characteristics (which are a subset of characteristics we use in our credit modeling). As such, we determined the extent to which individual covariates influenced the spread.

Controlling for other covariates in the model, we drew the following conclusions:

• For every 100-point increase in FICO, the spread decreases by 90 bps.
• An FRM spread is on average 55 bps lower than an ARM spread, both spreads being measured relative to the appropriate benchmarks. This makes sense in that ARMs are generally thought to be risker.
• For every point increase in combined LTV (CLTV), the spread widens by 3 bps.
• Loans originated in California or New York have spreads that are 25 bps lower than those originated elsewhere in the U.S. This could likely be due to the higher concentration of non-QM loans in these states.
• Loans for a purchase are 20 bps lower than rate-term/cash-out refinance loans.
• Self-employed borrowers have spreads that are 3 bps higher than those who are not.
• Relative to fully documented loans, "other doc" loans, primarily DSCR loans, pay up to 50 bps. Loans with alternative documentation type (such as bank statements) pay up only 10 bps relative to fully documented loans.

Interpreting the evolution of non-QM spreads

While the regression accounts for the variation of spread over time, it does not provide insight into why--controlling for loan-level credit characteristics--the spread changes from one quarter to the next. It is likely that the compression derives from competition among lenders, as well as the increased degree to which lenders understand and are comfortable with the non-QM product. An increase in originators complements the borrower's ability to shop around for a more competitive loan interest rate. In addition, the liquidity provided by an active securitization market can have a knock-on effect, further reducing the asset spreads.

The increase in spread over the past six months could be due to a weakening in one or more of credit factors, as reflected by our 'AAA' LC (in addition to the inertia/stickiness described above). For example, table 1 shows that the average CLTV ratio has been drifting up over time, which has influenced the rising 'AAA' LC. Given that we controlled for the CLTV in the regression, the recent spread widening may also be due to other factors, such as the increasing presence of investor loans and the varying levels of DSCR associated with these loans.

The key mortgage credit variables: FICO and LTV

Our analysis showed that FICO and LTV ratio were the greatest drivers of interest rate spread to the benchmark, as would be expected from traditional mortgage credit modeling. Therefore, we illustrated the relationship between mortgage rate, FICO, and the original LTV ratio in charts 4A and 4B with heat maps indicating the gradation. The charts are consistent with our regression results.

Chart 4A

Chart 4B

The Borrower Is The Biggest "Winner"

Although the spreads for non-QM loans have been increasing in recent months, the general trend over the past few years has been spread compression. Perhaps the biggest winner of this trend is the borrower. As the market for non-QM grows, more would-be borrowers are accessing credit. Consequently, more lenders may start originating the product, especially in light of potential changes on the GSE front in connection to the QM patch expiration and housing finance reform in general (there has been some mention of using a spread to prime rate to determine the QM designation). As additional volume flows into the non-QM space, the competitive framework of mortgage origination could put downward pressure on spread. If this coincides with additional liquidity for the loan, the spread to the conforming rate may more closely resemble a pure credit spread. From a bond structure perspective, compression of asset spreads could reduce excess interest in bond structures and, all else being equal, result in greater subordination.