Does a low savings rate predict defaults?

The savings rate did not establish a reliable warning

No reliable warning showed up in our test. We tested the national savings rate against five measures of loans going unpaid. The gap that fit best ran from seven to 16 quarters, depending on the loan, and none got as far as our prediction test. In the financial crisis, the rate rose while card payments went unpaid.

The savings rate is the share of Americans’ after-tax income that isn’t spent. In August 2026 it was 4.1%. People read a low rate as a sign that households are running on empty, and that missed payments will follow. If that were true at the national level, the rate would turn down a set time before missed payments turn up, and you could use it as a clock. That clock is what we went looking for.

The test measures delinquency: the share of loan balances behind on payments, including loans on which banks have stopped accruing interest. It doesn’t measure defaults, charge-offs or the share of people who missed a payment. The U.S. savings-rate statistics page carries the rate’s full history and monthly figures.

What our savings-rate scanner found

We run a program, our leading-indicator scanner, that lines up every pair of measures we track at every gap from zero to 16 quarters and asks which one moves first. A real lead has a clear best gap that shows up for related measures, holds in past recessions and helps predict data the test never saw.

The savings rate didn’t establish that. Four of the five best matches point in the direction the theory expects: less saving, more missed payments later. Here’s what the scanner recorded against five ways of measuring loans that have fallen behind:

• Credit card balances: best fit 7 quarters out, with a correlation of -0.29.
• Other consumer loans: best fit 8 quarters out, with a correlation of -0.19.
• Card balances, smaller banks: best fit 16 quarters out, with a correlation of 0.17, at the very end of the search.
• All loans: best fit 15 quarters out, with a correlation of -0.54.
• Home mortgages: best fit 16 quarters out, with a correlation of -0.50, at the very end of the search.

The same kind of trouble, people falling behind on what they owe, and the best gaps disagree. When the best fit lands at the end of the search, the scanner hasn’t pinned down a lead. It ran out of room. Two rates that both drifted for decades can match better the further apart you line them up.

Then the checks. Every pair passed the financial-crisis check, and none passed the pandemic check. None got far enough to be tested on whether it helps predict anything. An untested pair hasn’t failed a prediction test; it hasn’t established a prediction either. The full results appear below.

One measure isn’t in the frozen table: FHA mortgages behind. The scanner skipped it in August for lack of history. The later run, dated 2026-10-08, does test it over 50 quarters: the closest match is in the same quarter, with a correlation of 0.53. Saving and FHA late payments rose together, the opposite of a warning from falling saving. The shorter history still limits the test.

The financial-crisis test: saving rose as card payments went unpaid

The financial-crisis check is the one the savings rate passed, so it’s worth looking at what passing meant.

The check asks whether the savings rate fell into the bottom quarter of its own history before card delinquency rose into the top quarter of its. It did. But the revised savings history first fell below that cutoff in the second quarter of 1999 and stayed below it for 33 of the 36 quarters from the start of 1999 through the end of 2007. A signal that’s switched on for years doesn’t tell you when anything will happen.

Here’s what the two did once the trouble came. At the end of 2007, Americans were saving about 2.2% of their income, and 4.6% of card balances was behind. By the second quarter of 2009, card delinquency had climbed to 6.77%, its highest point in the chart’s history. The saving rate climbed too, to 6.8%. The rate rose with the damage. That doesn’t rule out a lead from an earlier low, but the rise itself wasn’t a warning.

The pandemic ran the other way. BEA’s account of April 2020 shows relief payments lifting income while spending stalled. In our stored history the monthly saving rate reached 31.8% in April 2020, its highest recorded level. Card delinquency later fell to 1.53% in the third quarter of 2021, its lowest recorded level. The changes overlapped the same period of aid; neither established a warning of the other.

Who the savings rate misses

The savings rate isn’t anyone’s bank balance. The Bureau of Economic Analysis defines personal saving as a share of disposable personal income. It takes everyone’s after-tax income added together, subtracts everyone’s spending added together, and divides. What’s left over is mostly what higher-income households didn’t spend.

That isn’t a guess. In a BLS working paper published in 2024, Marina Gindelsky and Robert Martin found saving was “negative for the bottom half of the distribution.” The bottom half was spending more than it took in. The national rate stayed positive because of the top.

That helps explain why a positive national rate can coexist with people running short. One likely reason the rate rose during the financial crisis is that higher earners who kept their jobs cut spending and saved more while people with less fell behind. It fits what we see, but we haven’t tested that explanation. Pandemic relief payments offer another reason the rate can rise.

The cushion that matters. The Federal Reserve’s yearly survey of household finances, called SHED, asks how people would pay a surprise $400 bill. In 2025, 37% said they wouldn’t pay it entirely with cash or its equivalent. The Fed’s report on saving and investments counts a credit card paid off at the next statement as cash-equivalent payment. It also warns that choosing another way to pay doesn’t necessarily mean someone lacks the cash.

That survey and the savings rate don’t move together. In 2020 the savings rate averaged 15.1%, against 7.3% the year before. The share who wouldn’t cover the surprise bill with cash barely moved, from 37% to 36%. Then the saving rate fell to 3.5% in 2022 and bounced back to 6.1% in 2023. The survey share stayed at 37% in each of the four years through 2025. The national rate swung; the survey answer didn’t.

The gap by income is wide. In the same Fed survey, 21% of adults in families earning under $25,000 had three months of expenses set aside, against 75% in families earning $100,000 or more. That question measures an emergency cushion, not the national saving rate.

Bankrate’s poll asks a different question, about a $1,000 bill. In its January 2026 emergency-savings report, 30% said they’d pay from savings. Bankrate changed the answer choices and the survey panel that year, so that result can’t be compared directly with its earlier years or with SHED.

Debt payments and missed payments

The Fed’s debt service ratio tracks the share of after-tax income needed for debt payments. It tracks the bill, where the saving rate tracks what’s left after everything. The stored source is the Fed’s Financial Accounts debt-service measure.

It does show an order, though a loose one. In the history since the method changed, debt payments hit their highest share of income, 15.8%, in the fourth quarter of 2007. Card delinquency peaked later, in the second quarter of 2009. In the supplemental scanner run, changes in debt payments matched changes in card delinquency best four quarters later.

That’s an association, not a proven lead. The pair held in the financial-crisis check but not in the other recession checks, and it hasn’t been through the prediction test. Debt payments fell to 9.05%, the low in the comparable history, in the first quarter of 2021, while card delinquency also fell. The national payment burden still doesn’t put a date on the next turn.

The Fed changed its calculation method, so the old and new debt-service histories aren’t directly comparable. The scanner’s full-history result spans that break; it needs a separate test of the comparable years before it can support a dependable warning. In the second quarter of 2026, debt payments took 11.1% of after-tax income, and 2.85% of card balances was behind.

The strongest case against our reading of saving

The card result isn’t nothing. The saving rate’s best match with card delinquency is seven quarters out, and it points the way the theory says: less saving, more missed payments later. The pair passed both the financial-crisis and earlier recession checks. The pandemic was a one-time shock, and it could be hiding a real relationship.

The years since fit the theory too. The savings rate fell to 2.4% in June 2022. Card delinquency then rose from 1.69% in the first quarter of 2022 to 3.22% in the second quarter of 2024. Years of thin savings before the financial crisis may also have left households with little to fall back on when the trouble hit.

The strength depends on the window. The frozen table uses the August scanner’s full available history: card observations from Q1 1991 through Q1 2026, matched to saving seven quarters earlier. Its correlation is -0.29. We keep that run because it is the dated comparison across all five loan measures, not a new test chosen to fit the chart. The scanner averages monthly observations into quarters and clips each series’ extreme tails before testing.

Using the October chart data from Q1 2005 through Q2 2026, without that clipping, the correlation is -0.52 at eight quarters. That is a stronger association. It changes both the period and the source vintage, so it doesn’t reproduce the frozen run or establish prediction. The total-debt and mortgage matches in the frozen table are stronger too, but sit at or near the search limit.

Here’s why we still don’t treat it as a warning. The card match in the full-history run is weak, the best gaps disagree across loan types, and none of the savings pairs reached the prediction test. Card delinquency had already turned up before the saving rate reached its post-pandemic low. Both were coming off the same pandemic aid. A long stretch of low saving may describe vulnerability without telling us when payments will be missed.

What the national savings rate cannot tell us about a household

A monthly savings-rate headline isn’t a dependable forecast of missed payments. A drop doesn’t establish that late payments are coming, and a rise doesn’t establish that they aren’t. The national payment burden and the share of card balances already behind describe different parts of the strain.

The national rate doesn’t show an individual household’s income, expenses or available cash. Households with the same income can have different debts and cash reserves. The aggregate rate cannot identify which household will fall behind.

See also: Being behind on mortgage payments.

Every loan measure we tested against the savings rate

“Held in” means the savings rate fell into the bottom quarter of its own history before the loan measure rose into the top quarter of its, inside that recession’s window. Only the scanner’s strongest eligible pairs go on to the prediction test. “Not reached” means the test wasn’t run, not that the pair failed it.

How the savings-rate test works

• The saving rate is BEA’s monthly personal saving rate via FRED. The chart averages the months in each complete quarter. The October source vintage is stored separately from the frozen scanner run; BEA’s revisions mean the two use different versions of history.
• The card and mortgage measures are Federal Reserve Board delinquency rates at U.S. commercial banks. The dollar amount behind is divided by total dollars outstanding. The Fed includes loans past due and still accruing interest, as well as loans in nonaccrual status, in its charge-off and delinquency release.
• The total-debt measure comes from the New York Fed’s Household Debt and Credit Report, a credit-report dataset rather than banks’ loan filings. It includes more kinds of household debt.
• The scanner tests gaps from zero to 16 quarters, both in levels and in quarter-to-quarter changes. It checks whether warning thresholds were crossed in order inside recession windows. Selected pairs reach a prediction test and a test on unseen data. The main table preserves the full-history run of August 18, 2026, including its clipping of extreme values; the card window and the separate, unclipped chart-window comparison are disclosed above. Other pairs use their own available histories. The main table is not restricted to the chart period; the FHA and debt-service discussion uses the supplemental run of 2026-10-08. Those results are kept separate.
• The survey comparison uses SHED’s yearly share who would not cover a surprise bill with cash or its equivalent and the saving rate’s calendar-year average. Willingness to use cash differs from ability to use it.
• The chart’s recession bands use NBER’s U.S. business-cycle dates, marking the financial crisis and the pandemic. The peaks are calculated from the stored observations.
• The limit is a national test. It doesn’t measure whether an individual household’s savings protect that household from a missed payment.

The cash emergency-expense survey carries the household-survey history. The other source histories appear in the related indicators below.