Extend and Pretend? — Q3 2026

Modifications Outran Loans

Modified loans were 60‱ of the panel's loans at September 30, 2026, against 52‱ a year earlier and 41‱ at September 30, 2024. They reached $10.6B, up 23.40% over the year while the same credit unions' loans grew 7.61%. The panel is the 4,143 credit unions that filed all nine quarters from Q3 2024 to Q3 2026, so each figure compares the same credit unions at every date.

A modification here is the Call Report's: principal forgiveness, a lower rate, a significant payment delay or a longer term, granted to a borrower in financial difficulty. A credit union that has adopted CECL (the current expected credit loss standard) reports each one for 12 months from the modification date, and a later modification restarts the count, so the line holds about a year of modifications, not every modified loan on the books. NCUA applied the standard to credit unions' reporting years beginning after December 15, 2022, so the line has held a full year of modifications since the end of 2023. A modified loan's delinquency is measured against its new terms, so a modification can return a late loan to current.

More of the trouble on the panel's books is modified rather than late. Modified loans were 37.1% of modified and 60+ day delinquent loans together at September 30, 2026, against 35.3% a year earlier and 31.1% two years earlier, while 60+ day delinquency rose from 95‱ to 101‱ of loans over the year. A smaller share of modified loans was failing on its new terms: 21.9% were 30 days or more past due or in nonaccrual, against 26.8% a year earlier. A pool that holds each modification for a year is young, and its newest loans have had little time to fail.

Modified, Delinquent and Reserved: the table gives the panel's modified loans, their share of loans, their share of modified and delinquent loans together, the allowance against those loans, and the share of modified loans failing on their new terms, at each quarter-end from Q3 2024. The modified share of the trouble and the allowance's cover moved in opposite directions over the two years.

The Largest Holder

One credit union holds much of the panel's modified loans. The credit union with the largest modified-loan balance at September 30, 2026 held 43.2% of the panel's, and its rise over the year was 62.6% of the panel's rise. Its modified loans were 3.0% of its loans, against 2.4% a year earlier. Its allowance was 345‱ of its loans at September 30, 2026, against 349‱ a year earlier; it fell to 310‱ by June 30, 2026 and was rebuilt in the third quarter, and at September 30, 2026 it covered 73.1% of the credit union's modified and delinquent loans, against 87.6%. The page sets that credit union aside by this test, not by name.

Without it, modified loans grew 14.26% over the year against loan growth of 7.70%, to 37‱ of loans from 35‱. The rest of the panel modifies at a fraction of the largest holder's rate, and its modified loans still grew faster than its loans. The allowance's cover of modified and delinquent loans fell without the largest holder too, to 87.8% from 91.8%, while the allowance rose as a share of loans, from 114‱ to 116‱. Reserves grew, and the modified and delinquent loans they stand against grew faster.

Without the Largest Holder: the table compares the three third quarters for the whole panel and for the panel without the largest holder. Without the largest holder the modified share of the trouble rose over the two years but held over the last one, and the cover fell in both years.

The Account Tested

The account is extend-and-pretend: a lender modifies a loan its borrower cannot pay, by extending the term or cutting the payment, sometimes more than once, instead of charging it off or reporting it as nonperforming, so the loss stays off its books. It was raised about banks' commercial real estate loans after the pandemic and again after the bank stress of spring 2023, and studies of large banks' loan-level supervisory data disagree: Crosignani and Prazad (2024), as Glancy (2026) describes them, found worse-capitalized banks more likely to extend loans whose income had fallen, while Glancy, on the banks as a whole after 2023, found that extensions mostly bridged temporary payment problems and that extended loans performed well afterward. In a credit union's Call Report the account predicts a sign the panel can be checked for: where modifications rise, delinquency and charge-offs fall or hold while the allowance falls behind.

For that check the base is the 3,379 panel credit unions with loans and with $10M or more in assets at both September 30, 2025 and September 30, 2026. A credit union under that size need not adopt CECL, and one that has not reports its troubled debt restructurings on the line instead, a different quantity. In the base the sign runs the other way. Without the largest holder, the 472 whose modified loans rose by more than 10‱ of loans saw 60+ day delinquency rise from 100‱ to 111‱ of loans and annualized nine-month charge-offs from 71‱ to 83‱. At the 411 whose modified loans fell by more than 10‱, delinquency went from 131‱ to 129‱ and charge-offs went from 71‱ to 63‱. Modifications rose where credit worsened, and those credit unions reported the losses. Their allowance did not keep pace: it covered 64.5% of their modified and delinquent loans, against 82.7% a year earlier, the one part of the account's sign the group shows. We read modification there as a response to rising trouble that did not hide it from the delinquency and loss lines, with reserves lagging behind both. No comparison in the filings can show how high delinquency would have gone without the modifications.

Where Modifications Rose: the table groups the base, without the largest holder, by the change in their modified loans as a share of loans over the year, with each group's delinquency, charge-off rate and allowance cover at both dates. Delinquency and charge-offs rose and the cover fell in the group whose modifications rose; in the group whose modifications fell, charge-offs fell and the cover rose.

At a few credit unions the full pattern does appear. Of the 3,379 in the base, 276 raised modified loans by 25‱ of loans or more over the year, 135 of those also reported lower 60+ day delinquency, 62 also reported nine-month charge-offs no higher as a share of loans, and 41 also had an allowance that covered less of their modified and delinquent loans. Those 41 hold 3.7% of the base's modified loans and 1.3% of its loans. The filings cannot separate a modification that works from one that pretends until the loan is repaid or charged off, so the pattern is a reason to follow those credit unions' later losses, not evidence of what their lenders intended. The page names none of them.

Narrowing to the Pattern: the first table counts the credit unions in the base that meet each condition in turn; the second places the pattern's credit unions by size. The first two conditions after the rise each remove about half of those left, and the pattern reaches about one to two in a hundred credit unions in each band up to $10B in assets and none above it.

What Settles It

Test H-2026Q3-breakdowns-extend-01. Whether the four-quarter net charge-off rate of the 41 credit unions that met all four conditions above at September 30, 2026 rises by more than that of the rest of the base, from the four quarters to Q3 2026 to the four quarters to Q3 2027. The pattern's population is fixed at publication; the comparison population is the other 3,338 credit unions in the base, the panel credit unions with loans and with $10M or more in assets at both September 30, 2025 and September 30, 2026. The measure is net charge-offs over four quarters as a share of loans at the end of the period, the pattern's change less the comparison's, scored in whole ‱ against a threshold of zero. Settles with the Q3 2027 data; made after September 30, 2026, before the first of the four quarters ends. At publication the pattern's rate over the four quarters to Q3 2026 is 89‱ and the comparison's 76‱. A reading above zero would be consistent with losses arriving after modification at the credit unions that fit the pattern, as the account predicts; a reading at or below zero would be consistent with modifications that bridged a temporary problem, as Glancy (2026) found for banks. The evidence supports neither side, so the test is published alone.

Connecting the Lines

This installment opens the Extend and Pretend thread, which the Margin Note Extend and Pretend sets out and which Signals — Q3 2026 carries as a watch item. It extends that issue's second signal, on the allowance's cover of delinquent loans, to modified loans. Its companion for the quarter, The Ratio Fell, the Losses Did Not — Q3 2026, finds the auto delinquency ratio falling while late auto loans hold level in dollars and auto charge-offs do not fall; the Call Report does not split modifications by loan type, so neither page can say how much of that fall modifications account for. The thread returns with each quarter's filings, and Test H-2026Q3-breakdowns-extend-01 settles with the Q3 2027 data.

QuarterCall publishes analysis of publicly available NCUA Call Report data. Nothing here is investment, financial, or membership advice.