Bowman details stress test overhaul to halve capital volatility
Federal Reserve Vice Chair for Supervision Michelle Bowman announced upcoming final revisions to the U.S. bank stress testing framework during a speech in London on September 18, 2026. The reforms will mandate full model disclosures and average test results over two consecutive years.
Two-year averaging and full model disclosure
The finalized framework establishes two binding rules alongside new model updates.
Under the Enhanced Transparency and Public Accountability rule, the Federal Reserve Board will publish model equations, variables, coefficients, underlying assumptions, and scenario design guides.
The second rule targets stress capital buffer (SCB) volatility by averaging results across two consecutive years and shifting the annual compliance date from October 1 to January 1. Together, these measures reduce SCB volatility by half without altering aggregate capital levels.
Bowman also confirmed plans to propose a revised noninterest income model for the 2027 test cycle and finalize G-SIB surcharge reforms before year-end.
Uncoupling risk discovery from capital buffers
Beyond capital calculations, Bowman outlined an expanded role for exploratory stress testing as a confidential supervisory tool.
These forward-looking exercises—including reverse stress tests and idiosyncratic scenario analyses—will not alter capital requirements but will help supervisors spot emerging vulnerabilities earlier.
Bowman cited the collapse of Silicon Valley Bank, noting that a hypothetical test incorporating rising interest rates and fair-value securities accounting would have revealed capital falling below required minimums as early as the fourth quarter of 2021.
Transparency achieved, responsiveness diluted
Opening Fed stress test models to the public finally tackles persistent legal and operational opacity.
Yet averaging buffers over two years risks muting rapid balance-sheet signals during sharp downturns.
The shift delivers regulatory predictability, but its safety value hinges entirely on supervisor follow-through.