Screening for credit stress across a bank cohort — and the merger that almost fooled us

Ahead of Q2 2026 bank earnings, we wanted to answer a specific question: is there evidence that bank customers — consumer and commercial borrowers alike — are under rising credit stress? Not for one bank, read off a single 10-Q, but systematically, across the sector, using Calcbench's standardized data.

This post walks through the method, what it found, and a wrinkle along the way that's arguably the more important lesson: a systematic screen is only as good as your willingness to double-check what it flags.

The method

Provision for loan loss (PLL) is the natural starting point for a credit-stress question — it's the expense banks book each quarter in anticipation of loans going bad. But raw PLL dollars are a noisy signal on their own. A bank's provision grows simply because its loan book is growing, independent of whether borrower quality is deteriorating. To separate “more loans” from “worse loans,” we normalized provision by the loan book itself:

PLL ratio = Provision for Loan Loss ÷ Loans Receivable

Using Calcbench's standardized metrics ( ProvisionForLoanLoss and LoansReceivable , both available as clean, comparable fields across filers), we built:

• A universe of national commercial banks, SIC code 6021, with total assets of $20 billion or more — pulled live via the Calcbench API, so it stays current as banks cross the threshold or merge away. You could also use the Calcbench Excel Addin or the Calcbench Multi Company page.
• The trailing 16 quarters of data (Q2 2022 through Q1 2026, the most recent quarter filed as of this writing) plus full-year annual totals for additional context.
• A minimum-coverage filter, requiring at least 12 of 16 quarters of valid data before a bank is included in any chart, so recent IPOs or fiscal-year mismatches don't clutter the picture with broken, partial lines.

First pass: the mega banks diverge

Looking first at the largest, most closely watched banks — JPMorgan, Bank of America, Citigroup, and Wells Fargo — raw provision dollars already tell a real, if modest, story of divergence rather than a uniform trend. (These four have loan books of a broadly similar scale, so the dollar figures are directly comparable here; we apply the normalized ratio once we broaden to the full bank universe below, where asset sizes vary far more.)

Citi and BAC show a sustained multi-year build in provisioning — consistent with, though not proof of, rising credit stress in their books. JPMorgan is essentially flat. Wells Fargo, notably, is heading the other direction entirely, with provisions declining each year. That's not what a uniform “the consumer is under stress” narrative would predict — it's a genuinely mixed picture, and worth watching whether the Citi/BAC trend continues into Q2 2026.

Broadening the screen — and a red flag

Expanding the universe to the full set of $20B+ SIC 6021 banks, one name jumped out immediately: Capital One , whose PLL ratio looked like the most stressed loan book in the entire cohort.

On the surface, that's plausible — Capital One's business is concentrated in credit cards, a structurally higher-loss, higher-yield lending category than traditional commercial banking. But the size and shape of the signal didn't look like ordinary credit-card seasoning. It looked like a single, enormous spike.

The Discover effect

Pulling Capital One's PLL ratio quarter by quarter tells the real story:

In every quarter shown outside Q2 2025, Capital One's ratio sits in a steady 0.65%–1.30% band — comparable in kind to the other mega banks. Then, in a single quarter, loans receivable jumps by $107.7 billion and provision expense jumps to $11.43 billion, more than four times the typical run rate. The very next quarter, both numbers snap back to normal.

That $107.7 billion jump in loans receivable is not organic loan growth — it's Discover Financial's loan portfolio landing on Capital One's balance sheet at the close of the Capital One–Discover merger in May 2025. Under CECL accounting, acquiring a loan portfolio typically requires booking a large “day one” provision against the acquired book, even though those loans aren't newly risky — the reserve reflects an accounting requirement at the moment of acquisition, not a change in the borrowers' behavior.

We also checked whether this distortion showed up elsewhere. It does: Capital One's net income briefly went negative in the same quarter (a loss of roughly $4.3 billion, against revenue that fell to just over $1 billion, driven by the same provision mechanic flowing through Calcbench's standardized bank revenue definition). Every downstream metric that touches provision or net income in that quarter tells the same distorted story for the same underlying reason.

Why this matters more than the finding itself

Excluding the merger quarter, Capital One's credit trend actually looks unremarkable — stable, in a similar range to the other mega banks, with no clear deterioration. The “riskiest loan book in the cohort” read was, in the end, a single quarter of merger accounting, not a multi-year trend.

That's arguably the more useful lesson here. Standardized data makes it possible to run a systematic credit-stress screen across dozens of banks in minutes rather than reading 10-Qs one at a time — but a screen that ranks a ratio and stops there will happily flag M&A activity as “stress” right alongside genuine deterioration. The value of the data is in the breadth it unlocks; the value of the analysis is in knowing which spikes deserve a second look before they go in a headline.

What we're watching into Q2 2026

• Does the Citi / Bank of America upward provisioning trend continue, or has it plateaued?
• With the Discover integration quarter now a year behind it, does Capital One's PLL ratio look clean and back in its normal band, and is that number, going forward, a meaningful signal on card-lending stress?
• Do any other banks in the broader $20B+ universe show a Citi/BAC-style multi-year build worth flagging before earnings land?

We'll revisit this screen once Q2 2026 filings are in.