We received the first loan-level performance data for the GSE’s a few months ago, so it’s about time to see what tentative observations can be drawn from this new data set. As a popular theme for this blog is the bank/nonbank share this seems a good place to start. In general, we have noticed that nonbank DQ’s tend to be higher than those for banks, and that this distinction is correlated with the relatively more generous credit terms available in the nonbank sector. Below find a table that demonstrates this for 2018 and 2019 vintage mortgages:

While market commentary is focused on developments such as inflation and house price increases, the key housing policy issues in the post-Covid world are financial inclusion and climate change. Our agency loan-level data provide us with many insights into market trends, but these do not contain demographic or geological details that are necessary to perform in-depth analysis in these areas.
​
On the topic of financial inclusion, the key supplemental data set is the Home Mortgage Disclosure Act (HMDA) dataset, an annual disclosure made by lenders in support of fair lending. HMDA data contains relevant data points such as income, gender, and race. Any assessment of fair lending practices requires an analysis of how these factors influence the availability of credit. To accomplish this, Recursion has applied a proprietary matching algorithm to create a robust dataset consisting of loans with both underwriting and demographic characteristics. Over the period 2008 – 2020 the data set consists of about 20 million loans. Below finds a chart of average credit scores by race (as measured by race of the first borrower) over the 2008 – 2020 period from this matched data set:

The increase in the number of cashout refis has led to some concerns about the implications for the quality of household balance sheets, similar to what occurred in the run-up to the Global Financial Crisis[1]. At that time, many homeowners were tempted to use their houses as a “piggy bank” as the national savings rate hit all-time lows. In the runup to the Global Financial Crisis, the national savings rate hit a sixty-year low of 2.2% while over the 2016-2019 period the rate averaged a much healthier 7.6%[2].

It is useful to scale the magnitude of assets being cashed out and Freddie Mac releases a very useful file documenting this quarterly for their book of business back to 1994[3]. In the fourth quarter of 2020, the amount cashed out reached $48.4 billion, about 58% of the $84.0 billion peak attained in the second quarter of 2006. As a share of household net worth, the most recent data point is 3.7%, well below the peak of 12.7% reached in Q2 2006.

​It’s useful along these lines to ask about the credit profile of cashouts compared to other refinancings. Freddie Mac didn’t report cashout refi separately until 2008 Q3, but the following useful picture can be obtained.

In general, lenders tend to “lean against the wind” by loosening credit conditions when demand for credit declines, and vice versa. Interestingly, it appears that the share of noncashout loans follows a pattern in which the share rises when credit in general is tightened. At the present time, the average FICO score in March for noncashouts was a tight 764, vs 753 for cashouts. It is difficult to pin the rise in equity cashouts in the current cycle to loosening credit conditions.
​
Of course, the proof is in performance, and now that we have loan-level DQs for the GSE’s beginning last month, we can look at this broken down by loan purpose for the Freddie Mac book:

​The performance of cashouts is mildly worse than that for noncashouts, but more in line with purchase mortgages. These statistics will bear watching in future months, particularly as forbearance programs begin to expire towards the end of the year.

In a recent post, we discussed the relative performance of loans with property inspection waivers vs those with traditional appraisals that qualified for a waiver[1]. We commented that the observed out-performance of loans with waivers as measured by lower total delinquency rates (DQs) was likely influenced by relatively tighter lending standards (eg higher credit scores, lower DTI) for these loans compared to eligible loans that received a traditional waiver.
​
A fully rigorous examination of this issue would be an extensive undertaking outside the scope of these brief posts. But let’s do a quick example as a demonstration of what our tools can produce along these lines. To make for an apple-to-apple comparison, below find two grids containing the difference between the total delinquency rates for purchase loans with PIWs compared to those that are eligible but obtain a traditional appraisal. The first is for loans originated in 2019 and the second is for those originated in 2020:

​We find that PIW’s are more extensively used in 2020 than in 2019. In addition, in 2019 the range of PIW takeup across cells was 8%-13%, while for 2020 it was 14%-28%. In both cases, takeup tends to rise with credit score. Lenders appear to be more willing to allow a waiver for borrowers with better credit.

For 2019, there are a number of outliers, but there is no clear pattern across the grid. Many lenders were just beginning to implement their waiver programs that year. By 2020, PIWs became a standard part of the toolkit. For most of the center of the grid, loans with waivers very slightly outperform those eligible loans using appraisals. Bigger outperformance can be seen, however, along the edges, i.e. loans with credit scores less than 720, and DTIs greater than 47. It appears it is not the waiver itself that leads to outperformance, but likely that underwriters are more careful and pay more attention in general to these riskier classes of loans.
​
Further work would look at performance across the largest servicers, and by state.

In a recent post, we discussed our comment letter to FHFA regarding policies and procedures related to property inspection wavers (PIWs)[1]. In that note we commented that one of the best ways to assess the impact of the program is to look at the performance of loans with appraisal waivers vs those eligible to obtain waivers but did not. At the time the note was posted (late February 2021) the loan-level data needed to perform such a calculation was not available, so we used a sample obtained from the reference loans in the pools used by the Fannie Mae Connecticut Avenue Security (CAS) Credit Risk Transfer (CRT) program.

Earlier this month we obtained the loan level DQ data for the books of the GSEs as of the end of February 2021 so a more comprehensive analysis is now possible. As stated in the comment letter, the eligibility rules to obtain a PIW vary by product type and agency, so to obtain an apples-to-apples comparison we need to look at the performance of loans with waivers against those that are eligible to use them but did not, as opposed to all loans. Since waivers are generally a recent development, we look at performance for loans originated in 2019 and 2020.

​On Tuesday February 23, FHFA released its monthly purchase-only HPI for December, showing a 1.1% rise from the prior month, and a striking 11.1% increase from December 2019, the record-high annual growth rate reported since this data was first released in the early 1990s.

Our last post demonstrated that Fannie Mae performance at the pool level has been lagging that of Freddie Mac since the start of the pandemic[1]. The question remains as to why. The challenge in answering this question is that unlike the case for Ginnie Mae programs, Fannie Mae and Freddie Mac have not been releasing performance data on the loan level[2]. Those who subscribe to our monthly risk reports know that we have been tracking relative underwriting standards between the two mortgage giants for some time. We do this not by looking at the average levels of underwriting characteristics, but rather at looking at the tails of these characteristics. Our experience is that this is a far superior method for this as distinct policy about risk come in much clearer this way. We focus on the share of GSE deliveries ​with LTV>95, DTI>45, and credit score<680.

Our regular readers will be aware that an ongoing theme is the collapsing bank share of mortgage deliveries to the Agencies[1]. Our recent monthly download shows that the bank share of deliveries to the GSEs fell sharply again in December, collapsing by over 7.0%(!) from November to a record low 22.3%. A year earlier this figure stood at 41.7%. The plunge witnessed over the past year marked an acceleration in a long-term trend, as banks face a heavier regulatory burden relative to nonbanks, and as nonbanks have made inroads into the market through their development of superior technology interfaces with their clients. Covid-19 has served to accelerate this trend by pulling customers out of bank branches and putting them in front of their laptops and smart phones.

The latest drop incentivized us to dig a little deeper; we didn’t have to peer too deep to find an interesting result. Below finds a bank of the bank share of GSE deliveries, and the same chart excluding Wells Fargo and JP Morgan Chase.