Credit Card Delinquency and Charge-Offs

Two Board of Governors series describe different aspects of commercial-bank credit-card portfolios. They are aggregate loan-dollar measures. The aggregate releases do not identify or follow individual accounts from delinquency into charge-off.

The Late Fee counts balances that are behind: card loans past due and still sitting on the bank’s books, as a share of card balances outstanding. It stood at 2.92% in Q1 2026.

The charge-off rate on credit card loans reports gross charge-offs less recoveries, annualized, as a share of average credit-card loans. It stood at 3.84% in Q1 2026.

A delinquent balance may later be charged off, cure, or leave the reported stock for another reason. The aggregate releases do not show which outcome occurred for any account. That is why historical timing between the two series must be read as a statistical description, not as a traced pipeline.

Definitions do not create a fixed aggregate lag

The delinquency rate is the value of credit-card loans that are past due or in nonaccrual status divided by end-of-period credit-card loans. The net charge-off rate is an annualized flow—charge-offs net of recoveries—divided by average credit-card loans.

Those different numerators, denominators, and time bases prevent a one-for-one conversion. The definitions do not establish that a movement in one quarterly aggregate must appear in the other after a fixed number of quarters.

The measures can still be compared historically. The careful question is whether their recorded movements aligned at a particular offset—not whether the data prove why they aligned.

What the series actually show

The lag profile below tests the historical correlation at several quarterly offsets. It is descriptive: a peak at one offset does not identify a transition rule, a borrower pathway, or a forecast.

Lag profile

Credit-card delinquency against credit-card charge-offs

Delinquency rate on credit card loans correlated against the charge-off rate on credit card loans at each quarterly offset, in levels and in year-over-year change. Both series are quarterly and come from the same Board of Governors collection.

Pearson correlation between The Late Fee at quarter t minus the lag and the card charge-off rate at quarter t, computed at build time from the committed indicator bundles over 141 shared quarters. Unit: correlation coefficient, −1 to 1.

Lag	Levels	Year-over-year change
same quarter	0.752	0.723
1 quarter	0.797	0.851
2 quarters	0.805	0.855
3 quarters	0.784	0.768
4 quarters	0.747	0.62
5 quarters	0.683	0.394
6 quarters	0.612	0.183
7 quarters	0.54	0.01
8 quarters	0.469	-0.13

Strongest level alignment: 2 quarters at r = 0.805. Strongest year-over-year alignment: 2 quarters at r = 0.855. The level column turns over before the last lag tested, which is what an interval looks like as opposed to a trend.

Series: The Late Fee and the card charge-off rate, quarter-aligned, latest observations Q1 2026 and Q1 2026. Correlation is an association at the stated alignment. It does not establish that the same households, accounts, or balances moved from one series into the other.

The profile shows how the aggregate series aligned in the observed history. It does not make the timing mechanical, and it does not show that the same balances account for both readings. Two additional limitations matter.

Where it stops holding

The first is the size of the move. Over the last credit crisis, from July 2007 to January 2010, card delinquency rose 1.37 percentage points. The charge-off rate over exactly the same window rose 6.61 percentage points.

The calendar window is the same, but the measures are not on the same scale: delinquency is a stock measured against balances outstanding at a point in time, while the charge-off rate is an annualized flow measured against average balances over the period. A percentage-point move in one therefore cannot be converted into the size or timing of a move in the other.

The second is whether the timing survives a crisis. The leading-indicator scanner tests every pair it finds against the 2001 recession, the 2008 crisis, and the pandemic, then against a Granger test and an out-of-sample split.

Scanner record

Credit-card delinquency as a leader

What the scanner recorded when it tested the credit-card delinquency rate against the card charge-off rate, the all-loan charge-off rate, consumer-loan delinquency, and card delinquency at commercial banks outside the 100 largest by consolidated foreign and domestic assets.

Delinquency Rate on Credit Card Loans as the leader, against the followers named here. Best lag is the alignment with the strongest correlation in levels, searched to a ceiling of 16 quarters. Source: American Default Research leading-indicator scanner leading_indicator_scanner v3.0, run August 21, 2026.

Follower	Best lag	r at that lag	Quarters compared	Crisis windows held	Validation
Charge-Off Rate on Credit Card Loans	2 quarters	0.781	139	2001	not carried forward
Charge-Off Rate on All Loans	3 quarters	0.657	138	2001, GFC	passed out of sample
Delinquency Rate on Consumer Loans (ex credit card)	1 quarter	0.902	140	GFC	not carried forward
Credit Card Delinquency Rate — Banks Outside Top 100	16 quarters	-0.55	125	2001, GFC	failed out of sample

Recorded best lags run from 1 quarter to 16 quarters across 4 followers, taking 4 distinct values, with 1 sitting at the 16 quarters search ceiling. Of the 12 relationships that cleared every gate in this run, 1 involve Delinquency Rate on Credit Card Loans.

The pair that clears every gate in the research artifact is card delinquency against the all-loan charge-off rate, recorded at 3 quarters with r = 0.66. Its crisis-window result is 2 (GFC, 2001), and its out-of-sample correlation is r = 0.83. That is a historical temporal association. It is not evidence that the same balances moved from one series into the other, and it is not a forecast.

The other tested pairings do not clear the same gates. Differences across crisis windows could reflect definitions, policy changes, sample variation, or other factors this aggregate test cannot distinguish. Separating those explanations would require evidence at a finer level than these series provide.

What a falling delinquency rate cannot tell you

Here is the part that matters most for reading a headline, and it follows from the measures’ construction rather than from any statistical test.

The delinquency rate can fall when balances leave its numerator, when fewer balances enter it, or when its denominator changes. The aggregate does not report how much each channel contributed.

So a quarter where The Late Fee declines is genuinely ambiguous on its own. It does not identify how many accounts cured, were charged off, entered delinquency, or changed the balance denominator.

The charge-off rate adds separate context; it does not resolve those account-level outcomes. Both series turned down after their 2024 highs, delinquency from 3.22% in the second quarter of 2024 and charge-offs from 4.64% in the third quarter of 2024, one quarter apart at the top. That sequence is observed timing, not evidence of why either rate changed.

That is the defensible use of the pair: two separate aggregate views, read together without treating one as a deterministic predictor or explanation of the other.

Where each one sits in the index

The American Distress Index keeps the two stages apart deliberately. Card delinquency is one of the members of the Delinquency domain; the card charge-off rate is a member of Default & Legal. Each domain contributes 20.0% of the composite, one of five equal shares.

That separation is not bookkeeping tidiness. The two measures have different constructions and belong to different analytical domains. Combining them as though they were repeated observations of a known account-level sequence would claim linkage the aggregate data do not provide.

They are distinct aggregate signals. Their historical co-movement can be measured; the underlying account paths and causes cannot be read from these series.