Introducing the Recursion Agency Multifamily Dataset Part 2: Outstanding Balance Benchmarking

In a recent post, we introduced the Recursion Agency Multifamily Dataset,[1] a complete accounting of multifamily loans securitized in Agency pools, going back to 2009. We provided a breakdown of issuance for the total market and by agency in that post. A natural question that arises is how can we benchmark our new dataset?
​
In this case, we look at outstanding balance rather than issuance volume. The Agencies provide quarterly data on total outstanding balances on their web sites[2]. When we compare this to our own data, we get this:

RUN UNDERLYING QUERY 1

RUN UNDERLYING QUERY 2

RUN UNDERLYING QUERY 3

These series are extremely close for the Fannie Mae and Ginnie Mae programs, and quite close for Freddie Mac.
​
A longer-term way to look at this issue is through comparison with the data available in the Federal Reserve’s “Financial Accounts of the United States” or Z.1 data[3]. Here we have:

RUN UNDERLYING QUERY 1

RUN UNDERLYING QUERY 2

RUN UNDERLYING QUERY 3

There are several interesting take-aways from this chart. For Ginnie pools, the data are spot-on between what we produce and the Fed’s numbers all the way back to 2009. For Fannie Mae, things are very close for the period provided in the first chart, 2016-present. Prior to that period, the Z1 data get progressively bigger than our data the further it goes back in time. The Freddie lines are particularly notable. First, for the period 2020-present, our data and the Z1 data match very closely, in contrast to the first chart where the Freddie data come in persistently, if modestly, higher than our data over that period. Prior to 2020, the pattern is similar to what we see in the Fannie comparison: our data increasingly falls below the Z1 data as it goes back in time.

Overall, we believe that this exercise validates our approach as most of the data post-2016 are very close across all three measures. As far as the pre-2016 period, we can only speculate. Looking at the pattern for Fannie Mae and Freddie Mac, what comes to mind between 2009-2016 is that what we might be seeing is shrinking holdings of multifamily loans on the part of the Enterprises themselves[4].

How to validate this speculation? The only source that comes to mind is the financial statements of Fannie and Freddie. From 2010-2016 we find the following table in each of the reports, this from the 2013 10-q:

We are many things at Recursion, but not accountants. But it appears that the column marked “Of Fannie Mae” consists of its own holdings. What if we scrape the 10-q’s and plot this figure against the gap between the Z.1 data and our Fannie data? This is what we get:

​Wow. Unfortunately, we could find no corresponding numbers in the Freddie Mac statements, but the above chart is suggestive that something similar is going on for them. So for now we declare victory and move on to more interesting analytics. Any comments are very much welcome.

---

[1]https://www.recursionco.com/blog/introducing-the-recursion-agency-multifamily-dataset-part-i-issuance
[2] Freddie Mac: https://mf.freddiemac.com/investors/data ;
Fannie Mae: https://capitalmarkets.fanniemae.com/mortgage-backed-securities/multifamily-mbs/multifamily-issuances-and-total-mbs-outstanding ;
Ginnie Mae: https://www.ginniemae.gov/data_and_reports/reporting/Pages/monthly_rpb_reports.aspx
[3] https://www.federalreserve.gov/releases/z1/
[4] See for example https://home.treasury.gov/news/press-releases/tg1684