The Savings Rate Does Not Lead Delinquency

The story is easy to tell, which is the first thing that should have made us suspicious of it.

Savings go first. A household stops adding to the account, then draws it down, then runs out. Only after that does a payment get missed. Only after that does the missed payment reach a bank’s quarterly filing. Three stages, three delays, and at the end of them a number that arrives late enough to be useless as a warning.

If that is how it works at the national level, the personal saving rate should turn before delinquency does, by an interval you could measure once and then reuse. That interval is the whole claim. Without it, “buffers lead defaults” is a sentence about one household wearing the costume of a national indicator.

We went looking for the interval. It is not there.

What a lead has to look like

A real lead has a shape. Line the two series up at every plausible offset, correlate them at each one, and the correlation should climb to a peak at some lag and fall away on either side. The peak is the interval. Its steepness tells you how sharply the relationship is identified.

A shared trend has a different shape. Two series that both drift over thirty years will correlate at every offset you test, and the correlation will keep improving the further you push the lag, because you are increasingly comparing the start of one long slope to the end of another. Push the search far enough and the “best” lag lands wherever you stopped looking.

That distinction is the whole test, and it is easy to run.

The level column carries the sign the thesis wants: more saving now, less mortgage delinquency later. It also gets stronger at every single lag out to the end of the search, and it is still strengthening when the table runs out of room. Nothing in it says four years. It says two series drifted across their whole shared history, and a drift is not an interval.

The year-over-year column does not even manage the sign. Its strongest alignment is positive and close in, which reads as saving and delinquency rising in the same short window rather than one preceding the other.

Same leader, different clocks

That is one pair. The leading-indicator scanner that produces our validated research runs the same test across every indicator pair in the catalog, and it keeps the failures alongside the passes.

Followers of the same kind — households failing to pay lenders — and the recorded lags do not agree with each other. The ones that land on the ceiling of the search are the tell described above. The scanner did not find a lag there. It ran out of lags.

One follower is missing from that table, and its absence is worth naming. The FHA Signal is the delinquency series most people reach for when they want to watch thin-margin borrowers, which makes it the natural place to look for a buffer effect. The scanner never tests it. It requires a minimum run of quarterly observations before it will evaluate a pair, and the FHA series does not have enough of them. That is a coverage limit, not a result — the buffer thesis has not been tested against FHA delinquency here, and nothing in this piece should be read as ruling it out.

Then the gates. A pair that survives the correlation filter still has to hold during a crisis window, then pass a Granger test showing the leader’s lagged values add information beyond the follower’s own history, then replicate on data the model never calibrated on. Nothing starting from the savings rate reaches the end of that funnel. The relationships that do — initial claims into unemployment, credit-card delinquency into all-loan charge-offs — clear every stage. The buffer story clears the first one and stops.

The 2008 case, in the direction nobody expects

The averages could still be hiding a real relationship that only shows up under stress. So look at the stress directly.

Going into the last credit crisis the savings rate was near the floor of its recorded history: 1.9% in November 2007, against a trough for the whole series of 1.4% in July 2005. Buffers were thin. The thesis says missed payments follow.

They did. The Late Fee, the delinquency rate on credit card loans at commercial banks, went from 4.6% in the fourth quarter of 2007 to 6.77% in the second quarter of 2009.

And the savings rate over the same stretch? It went up. 7.8% by May 2009, a climb of 3 percentage points across the two and a half years from July 2007 to January 2010.

The buffer measure and the missed payments rose together, through the worst household credit event on the modern record. Whatever the savings rate was doing in 2008, it was not draining ahead of the damage.

What the savings rate is actually measuring

Here is the part I keep coming back to, because it reframes the failure as something other than noise.

The personal saving rate is not a reading of anyone’s account balance. It is a residual in the national income accounts: what is left of aggregate disposable personal income after aggregate outlays, published by the U.S. Bureau of Economic Analysis and defined as personal saving as a percentage of disposable personal income. One national numerator over one national denominator.

That construction has two consequences, and both of them break the buffer story.

It is dominated by the households with the money. Saving is concentrated at the top of the income distribution, so the aggregate rate mostly reports what high-income households did with their income this month. The households running out of cash contribute almost nothing to the numerator, which is precisely the population whose delinquency you are trying to anticipate.

And it rises in a downturn. When the labor market turns, higher-income households cut discretionary spending while their income holds, and transfer payments raise disposable income for everyone else. Both effects push the rate up in the same quarters that missed payments are climbing. That is the 2009 pattern above, and it is a mechanism hypothesis consistent with the evidence rather than a tested claim — but it explains why the sign of the relationship is not stable, and an unstable sign is fatal to a lead.

So the savings rate is a real measure of a real thing. It is a poor proxy for the cash cushion of the household that is about to miss a payment, because those households barely appear in it.

The measure that does show an ordering

If a buffer is the wrong place to look, the obligation is the obvious next place.

That is what the other shape looks like. The year-over-year column rises to a peak roughly a year out and falls away after it, which is at least the profile of an interval rather than a trend.

Read it carefully, though. This is an association at a repeated offset, not a validated lead. In the scanner’s record the debt-service pair held during the 2008 window and failed during the pandemic window, and it never reached the out-of-sample stage. Debt Service stood at 11.16% of disposable income in Q1 2026. What the profile supports is that the burden measure and the delinquency measure move in a repeated order. It does not support putting a date on the next turn.

Where the buffer still belongs

None of this retires the savings rate. It relocates the claim.

The Buffer read 2.7% in June 2026 against a pre-pandemic average of 6.1% over January 2015 to December 2019. That is a level statement, and levels are what this series is good for. It carries the Safety Net & Buffer domain in the American Distress Index at a 20.0% share, one of five equal domains, and it belongs there for the same reason it fails as a clock: it describes the condition of the national household balance sheet, not the sequence in which one breaks.

The lag we wanted would have been genuinely useful. A measurable interval between buffers and defaults would let anyone reading a savings print know something about a delinquency report two or three quarters out. Instead the series that looks most like a cause turns out to move with the damage rather than ahead of it, and the honest version of this piece is shorter than the one we expected to write.

The buffer drains. It just does not announce anything on its way down.