We get a lot of requests at Recursion, the bulk of which never make it to these posts, but one that struck home recently was from a regulator who asked what we know about Lahaina. Given the scope of this tragedy, we thought it worth the effort to talk about what we do and don’t have. Specific to that location, the answer is relatively limited. The Agency disclosure data is provided at the state level. The population of Lahaina was 12,700 as of the 2020 Census, out of a population of the Island (County) of Maui of 164,000 and 1.4 million for the state of Hawaii overall. So, this level of detail seems unlikely to be sufficiently granular to provide a basis for analysis. However, it seems we can take a bottom-up approach that may yield something of value. This would be based on the HMDA data.
HMDA data has the advantage of granularity down to the census tract level. Out of over 84,000 Census Tracts, we can identify 6 for Lahaina. We can then pull-down originations from HMDA:
The release of the Agency performance data in early May provided confirmation that the dip in Early Payment Defaults we have witnessed over the last three months ended a 16-month long uptrend in this statistic for FHA loans. A similar but far more muted pattern can be seen for VA and conventional mortgages. In a previous post, we speculated that the uptrend was correlated with the higher inflationary trend observed since early 2021. Below please find an update of the chart:
In a previous note, we looked at mortgage trends derived from the recent release of 2022 HMDA data. Of course, HMDA is a prime data source for analysts and policymakers who seek to understand how social and economic trends interact. The most discussed issue is the distribution of originations by race. Below find a bar chart for the share of originations by race annually from 2004-2022 by loan count:
The cherry blossoms are blooming, which means it’s time for the HMDA preliminary data set to be released. The dataset provides a social underpinning to the nation’s mortgage market and enhances our understanding of the behavior of borrowers and lenders. The 2022 dataset has been particularly eagerly awaited, as we get our view on the new world of high inflation and mortgage rates for the first time in decades. We start with origination volumes and get not just confirmation of the onset of mortgage winter, but some breakdown of its characteristics.
In a recent post, we discussed the attributes of manufactured housing data that came with the final 2021 HMDA release in July, which were not available in the preliminary release. Another important data point from this final release is the conforming flag, identifying which loans satisfy the requirements for delivery to Fannie Mae and Freddie Mac. Obviously, any loan sold to a GSE is conforming, so its main use is to enable analysts to examine these loans which are held on bank balance sheets.
The lag between the preliminary and final releases of HMDA data can be four months or longer, so it would be useful to be able to identify these loans right after the preliminary release. One way to approach this is to flag non-government loans with balances above the conforming limits as “jumbo”. How does Recursion’s “Jumbo” flag compared with HMDA’s Conforming flag? If the information is perfect, “Jumbo” loans should be all the loans that are not “Conforming”. However, the exact original balance of a mortgage is not provided by HMDA to protect privacy. For those loans close to the conforming boundary, our program can misjudge which category to assign. Given all that, going back to 2018, there is still a very strong negative connection between the two measures:
Usually, when we talk about financial institutions in our posts, we focus on sellers and/or servicers as we have a clear view from the Agency disclosures. An interesting distinction in this regard is to break down originations between those sourced through a retail channel within the lending institutions and those purchased from other lenders, known as third-party originations (TPOs). We are often asked the question in the case of TPO lending, where only sponsors of the mortgages are reported, who are the originators? This information is not reported in the agency loan-level disclosure. We can supplement this information by examining originators in the HMDA data by observing the fact a TPO (correspondent or broker) loan is often reported twice, one record reported by the originator and another reported by the sponsor. At Recursion, we conducted an exercise by matching the pairs together, and we were able to identify the counterparty pairs for about 50% of the mortgages marked as “purchased”, and also made this revealing data point to our HMDA Analyzer users.
According to the 2021 HMDA preliminary release, about 2.65 million loans were purchased from other lenders that year, about 18% of all originations. Roughly half of these purchases were made by 10 institutions:
In a recent post, we discussed findings obtained with the recent release of 2021 HMDA data. Among other things, we looked at the share of mortgage originations by income group and product type. In this note, we look at the difference in lending patterns between the banks and nonbanks.
The incentive behind this approach is policy driven. There is a long history of measures taken to encourage lenders and builders to foster economic development in low-income areas via the housing market. For example, the Community Reinvestment Act (CRA) stipulates that a bank’s performance with regards to compliance of their regulatory requirements depends in part on:
“the geographic distribution of loans—that is, the proportion of the bank's total loans made within its assessment area; how these loans are distributed among low-, moderate-, middle-, and upper income locations”
To assess this issue, we assign a flag to each of the census tracts designated by HUD as having a greater than 51% share of households with incomes in the Low-to-Moderate (LMI) range in the larger MSA the tract is part of, which are called LMI area by HUD, or “low income” tracts by FHFA. Below find a chart of the 10-year trend in the share of loans originated in this category by institution type for conventional and FHA loans: