America's Banks Passed a Stress Test for the Wrong Stress

The 10-year Treasury closed Friday at 5.18%. That is up 22 basis points in a week, up more than 60 since mid-July, and — in Yahoo Finance's daily series — the highest close in at least two years, clearing even the October 2023 spike near 5% that marked the top of the last tightening cycle. The 30-year finished at 5.50%. The Federal Reserve raised rates on September 16 for the first time since 2023, to a 3.75%–4.00% target range, and by midweek CaixaBank Research counted markets pricing almost four more Fed hikes over the next 12 months.

Two days after that hike, on September 18, the Fed's Vice Chair for Supervision, Michelle Bowman, stood in London's Mansion House and announced that the Board will vote "in the coming weeks" to finalize the most consequential rewrite of the bank stress test since it was created after the 2008 crisis. In her words, the reforms will "finally close the book on an opaque and unnecessarily unpredictable framework."

Hold those two facts next to each other. The bond market is delivering the sharpest interest-rate stress the banking system has faced since 2022–23. And Washington is about to finalize rules that make the test of that system more transparent, more predictable, and — by explicit design — half as volatile in the capital requirements it produces.

Whether those two things are compatible is now a live question for anyone who owns bank stocks or bank debt.

What the Fed is about to change

The overhaul has been coming since December 2024, when — amid a legal challenge from bank trade groups led by the Bank Policy Institute — the Board committed to the basic contours of reform. Two proposals followed, in April and October 2025. Bowman's speech previewed how they land:

• Model disclosure. The Fed will publish the equations, variables, coefficients, assumptions, and limitations of the models that generate hypothetical losses — plus the rationale for design decisions, alternatives it considered, and planned model changes for 2027.
• Scenario transparency. The scenario design framework goes out for public comment, with expanded guides beyond the current two variables (unemployment and home prices).
• Two-year averaging. A bank's stress capital buffer will be set on the average of its two most recent annual results, and the effective date shifts from October 1 to January 1. Bowman says the package cuts buffer volatility "by half without materially changing aggregate levels of required capital."
• Two refinements from the comment file: a balance-sheet freeze date before scenarios are released, and two global market shocks run on the same as-of date, with the larger loss binding.

None of this is scandalous. Much of it is genuinely good administrative practice, and the two-shock change is a real improvement for trading books. The problem is not what the new framework adds. It is what the test has never measured — and the timing.

Briefings like this land in members' inboxes before the market prices them in. Join free →

The test everyone passed

On June 24, the Fed released the 2026 stress test results. All 32 large banks passed. The severely adverse scenario put unemployment at 10%, cut home prices 30% and commercial real estate 39%, and generated more than $708 billion in hypothetical losses. Aggregate common equity Tier 1 capital fell from 12.8% to a minimum of 11.2% — a 1.6-percentage-point dent, comfortably above minimums. Headlines called the system resilient.

Buried in the scenario was the assumption doing quiet work: the 10-year Treasury yield was floored at 2.3%. In the Fed's model recession, long rates fall — from roughly where they were toward 2.3%, against floors of 1.0% in the 2025 test and about 0.8% in 2024. As The Asian Banker noted in its analysis of the results, that higher rate path meant the Fed's models projected substantially more net interest income across the nine-quarter horizon — enough to offset growing modeled credit losses. Falling rates also mean bond portfolios rally in the scenario, cushioning capital.

In other words: the assumed direction of interest rates didn't just fail to stress the banks. It helped them pass.

That is how every Dodd-Frank stress test has worked since inception. Recessions in the model look like 2008: demand collapses, the Fed cuts, yields fall, and a bank's bond book becomes a shock absorber. But the defining banking stress of this decade ran the other way. Silicon Valley Bank died in March 2023 of rising rates — unrealized losses on a long-duration securities book meeting a deposit run. It wasn't in that year's test cohort, and the scenario that year, like every year, modeled rates falling.

The numbers say the same configuration is rebuilding. FDIC data show $326.7 billion in unrealized losses on bank securities portfolios as of June 30 — up for a second consecutive quarter, with $109.8 billion sitting in available-for-sale portfolios and roughly $217 billion parked in held-to-maturity books, where the losses are real but invisible to regulatory capital. And those marks were struck at June 30 yields. The 10-year has risen more than 60 basis points since mid-July. Unless the long end rallies hard before September 30, the third-quarter number — due from the FDIC in late November — is going one direction.

So the open question: the one stress actually accumulating on bank balance sheets is the one the capital test is structurally built not to see — and the Fed's proposed answer to that gap isn't in the capital test at all.

Get the next briefing free in your inbox: Join AlphaBriefing

The rest of this briefing is free — it just requires a free AlphaBriefing account: the arithmetic linking the summer's yield move to the next unrealized-loss print, what two-year averaging does to capital requirements if 2027 turns ugly, where the duration exposure is concentrated (and the deposit math that decides which banks wear it), and the catalyst calendar through November.

Create your free account → — 30 seconds, no card.