The Big Picture View on Mortgage Delinquencies: Update

In a recent post, we discussed macro factors driving delinquency rates across the mortgage landscape.[1] In this brief update, we make two comments, one fundamental and the second technical: First, we look at the possibility that the relatively greater financial distress observed among FHA borrowers compared to conforming borrowers at present is due, at least in part, to the relatively higher debt service levels held in the former cohort.  Second, we perform an exercise matching FHA loan in HMDA to those in the Ginnie Mae disclosures to measure natural disasters’ impact on loan performance.

Debt and Delinquency [DQ]
​
In our prior post, we noted the large differential in DQs across consumer sectors: autos, credit cards, and mortgages since interest rates began to rise in early 2022. We now have an additional quarter of data from the Federal Reserve Bank of NY which is summarized here:

There is a clear distinction between the soaring credit card DQs and those in the other two sectors. A key factor behind this discrepancy is that credit card debt is floating while autos and mortgages are mostly fixed-rate obligations. Unanticipated higher interest payments on their cards naturally lead to distress for many borrowers.
​
The main point we want to make here is that this distress has the potential, in some circumstances, to spill over into mortgages. As we commented previously, most mortgage borrowers are incentivized to remain current on these payments as they have considerable equity in their homes. However, households with rising costs of debt service may struggle.

We start by looking at weighted average DTIs for FHA and conforming loan issuance broken down between banks and nonbanks:

RUN UNDERLYING QUERY

RUN UNDERLYING QUERY

​In each case, nonbank DTIs are greater than those of banks, although the gap is bigger for FHA than conforming loans. Our supposition that loan performance depends importantly on debt service and that credit card debt might be particularly impactful can be tested by looking at the gap in  DQ rates between banks and nonbanks for the two mortgage types and relating this to the path of credit card delinquency rates:

RUN UNDERLYING QUERY

RUN UNDERLYING QUERY

We can see that the gap rises with CC delinquency rates for both banks and nonbanks. Due to higher DTIs, the nonbank gap grows faster than banks. These relationships are suggestive, if not conclusive regarding the effectiveness of the theory we proposed. We leave it up to our modeling and trading clients to dig deeper into these relationships.
 
Climate risk assessment based on matching technology
​
Often, there are multiple sources available for a particular data set containing different characteristics and we wish to combine these to generate a more robust view of the available information. Within the mortgage universe, the Home Mortgage Disclosure Act (HMDA)[2] and the agency loan-level disclosures[3] are the two major sources. These are massive data sets with very different structures. The HMDA data are released annually, providing a comprehensive picture of mortgage issuance in that year, including lender, loan size, note rate, income, demographic characteristics, such as race and age, and importantly, for our purpose, geography down to the county level. The Ginnie Mae loan tape is released monthly and contains all the loans contained in active Ginnie Mae pools. The characteristics here also include issuer (both retail and third party), loan size, note rate, borrower characteristics such as credit score, geography down to the state level, and of importance here, loan performance.
​
Using Recursion’s proprietary algorithm, we create a combined data set containing characteristics from both data sets for over 80% of the outstanding FHA loans. Our aim is to see if we can track the impact of a natural disaster on the mortgage market in an impacted geography. Recall that back in July, Hurricane Beryl hit the Gulf Coast of Texas. We want to see if we can see an impact by looking at the DQ rate for FHA loans in the five main impacted counties compared to the state overall[4]:

RUN UNDERLYING QUERY

Indeed we can.
In both of these examples, we are just scratching the surface of what can be done with the tools at hand, and we are building additional ones and incorporating new data into the analysis. Stay tuned.

---

[1] https://www.recursionco.com/blog/the-big-picture-view-on-delinquencies
[2] https://ffiec.cfpb.gov/
[3] In this exercise, we look at FHA loans from the Ginnie Mae loan tape (https://www.ginniemae.gov/investors/disclosures_and_reports/Documents/mbs_loanlevel_dictionary.pdf)
[4] Since the latest HMDA release is 2023, we exclude mortgages originated in 2024 in this analysis.