The FHA Gap Measures Concentration, Not Timing

The question underneath the FHA split is a good one, and it is worth stating precisely before looking at whether the data answers it.

FHA-insured mortgages are the entry-level channel. A borrower can buy with 3.5% down. Thin equity, thin reserves, a payment that works while everything else does. If household distress shows up first where the cash cushion is smallest, it should show up here first.

Two claims are hiding in that sentence. One is about level. The other is about order. They need separate evidence, and only one of them has any.

The level claim holds

The FHA Signal reads 11.79% in Q2 2026, against 1.86% for single-family mortgages booked on commercial-bank balance sheets, a Board of Governors of the Federal Reserve System series retrieved via FRED as DRSFRMACBS.

A gap of 9.93 points. A ratio of 6.3×.

That gap is not new and it is not an artifact of one bad quarter. It is present in every quarter the survey has recorded since Q1 2007. Whatever else is true, the entry-level channel carries a different rate of missed payments than the bank-held book, and it always has.

The order claim has no quarters to stand on

The Mortgage Bankers Association’s National Delinquency Survey is the only public source for the FHA rate, and the record American Default holds from it is thin.

The record holds 20 quarterly observations, spread across the span from Q1 2007 to Q2 2026. The longest break runs 33 quarters, between Q4 2011 and Q2 2020.

The break that decides this article is smaller and much worse placed. There is no observation for the 10 quarters between Q2 2007 and Q1 2010.

That is the entire onset of the last mortgage crisis.

We can see both endpoints. At Q2 2007, FHA was 12.58% and the Board of Governors series was 2.29%. At Q1 2010, FHA was 13.15% and the bank book had reached 11.48%.

We know where both series ended up. We have no observation of the order they got there.

This is also why The FHA Signal does not appear in American Default’s leading-indicator research at all. That scan covers 73 series and drops any series without enough quarterly observations to correlate before the first test runs. FHA is dropped at that step. It has never been tested for a lead, because it cannot be.

The continuous window cannot settle it either

The survey has been unbroken recently: 11 consecutive quarters from Q4 2023, every one of them matched by a published reading on the bank-booked side.

Across that window FHA rose 0.98 points and the bank-booked series rose 0.16 points. Roughly 6.1 times as far, which is the borrower-split thesis behaving exactly as advertised on magnitude.

On order it gives nothing. The bank-booked series reaches its lowest reading of the window in Q4 2023, which is the first quarter of the window. A turning point sitting on the boundary of the observation window is not a turning point you have observed. FHA’s own low arrives in Q1 2024, one quarter later, and a single ambiguous crossing is not a lead in either direction.

What the gap does measure

Take the widest and narrowest readings of the gap. The interpretation flips.

The widest it has ever been recorded is 13.11 points, in Q2 2020. The narrowest is 1.66 points, in Q1 2011.

Now put those in order of how bad things were. The narrow reading belongs to the worse quarter. The bank book had climbed to 10.37% by then, and the gap closed because everyone else caught up, not because FHA borrowers recovered.

So the gap is wide when distress is concentrated and narrow when distress is general. It is a concentration measure. A reader who treats a widening gap as a countdown has it backwards: the widest gap on record arrived in a quarter when the rest of the mortgage book was quiet.

There is a plausible mechanism behind that, offered as a hypothesis rather than a finding. The FHA population is held roughly constant by program design — low down payment, thin reserves, a credit profile the conventional channel declined — so its delinquency rate is high in good conditions and bad. Most of the movement in the gap then comes from the other side of the subtraction. The prediction that follows is testable: in any episode where conventional mortgage distress rises materially, the gap should compress. It did in Q1 2011. That is one episode, which is not a validation.

The comparison is not like for like

One caveat has to travel with every number above, because it limits what the level of the gap can mean.

The two series are collected by different institutions from different populations. The FHA rate comes from the Mortgage Bankers Association’s National Delinquency Survey and counts FHA-insured loans at least one payment past due, excluding loans already in foreclosure. The comparison series is a Board of Governors bank-condition measure covering single-family mortgages held on the balance sheets of commercial banks. FHA-insured lending is not confined to that universe.

So part of the level difference is definitional, and no amount of care in the arithmetic removes it. What survives the caveat is the movement: both series are internally consistent over time, so the direction and size of changes in the gap remain informative even where its absolute height is not a clean apples-to-apples ratio.

Where this leaves the thesis

The borrower split is real and it is measurable. Households with the least cash cushion miss payments at a rate the bank-held mortgage book never approaches, in every quarter we can see, including quarters when the aggregate looked calm. That is worth publishing on its own.

The early-warning version is not supported. It requires the survey to have recorded the quarters where the turn happened, and it did not. The strongest honest statement is that the gap tells you how concentrated the distress in a quarter is, which is a real thing to know and a different thing from knowing when.

The thesis said the thin-buffer channel would show it first. The record says it shows it worse. Only one of those is in the file.