Board of Governors of the Federal Reserve System

I cannot support the package of changes to the supervisory stress test presented here today. Disclosure of the models and scenarios will make the stress test weaker and less credible. The proposed changes to the framework and models risk turning the stress test into an ossified exercise that will provide illusory comfort in the resilience of the system. They will lead to overly optimistic projections in the stress test, both because of less conservative modeling choices and because of the potential for gaming by banks. This will lead to weaker bank capital requirements, a result that is inconsistent with the Board's previous statement that the proposed changes would not be designed to materially alter capital requirements. In addition, the more routinized nature of the stress test will lead banks to reduce the capital management buffers that they hold above required capital. It will also disincentivize them from paying adequate attention to independent risk management. Altogether, these changes will leave banks in a worse position to weather stress. What's more, the steps taken today will be difficult to reverse. For example, if public comments demonstrate that the costs of model disclosure outweigh the benefits, there is no way, unless and until the models change materially, to un-disclose them.

I've spoken elsewhere about the history of the stress test, its importance, and what I would do as an alternative to today's proposal.¹ I stand by those comments and continue to believe that a better path is available and beneficial for all. Today, I will focus my remarks more narrowly on the issues with the proposal before us.

I have ten points to make.

First, subjecting the models to the proposed disclosure and comment process could lead them to ossify, causing the dynamism and rigor of the stress test program to fade. Stress test models are inherently complex. They require nimble adjustments to maintain accuracy and relevance as banks innovate and create new products and risks. Instead, they will be forced to undergo a lengthy and complicated process of public review each year. The dynamism previously found in the models may disappear and they may fall behind.

Second, this process will likely lead to lower bank capital. The model and scenario review process could create a one-way ratchet on bank capital requirements, in which bank commenters object to the aspects of the models or scenarios that result in higher capital requirements, while ignoring areas in which the models underestimate or scenarios downplay downside risk. As time goes on, responding to such comments could lead to weaker capital requirements. Indeed, the process has already started before the comments have come in, as I will describe.

Third, banks are likely to game the capital requirements using the newly revealed details of the models. Specifically, banks are likely to shift to riskier assets where the disclosed models underestimate the risks compared to the banks' assessments. They are more likely to use window dressing—balance sheet adjustments to make the banks appear safer—and to strategically and temporarily deploy hedging strategies that reduce losses under the test but are unrealistic to have in place and maintain under stressed conditions.

Fourth, with more routinized stress testing, banks will likely reduce their management buffers, the capital cushion that they hold over required minimum capital and regulatory buffers. Such a reduction will weaken banks' resilience when a significant event actually happens.

Fifth, the new restrictions in how the yearly hypothetical scenarios can be structured will also make it easier for banks to game the tests and reduce the ability of the tests to probe salient or unusual risks. That means stress testing will be less useful in identifying and managing emerging risks. And I also note that the scenario being proposed for the 2026 cycle is at the low end of stressfulness compared to prior tests.

Sixth, because the proposal routinizes the stress test, banks are likely to invest less in their own risk management processes. Instead, they are likely to focus on managing to the stress test using the Fed's models and the limited universe of conditions that may appear in the scenarios. Consequently, banks' ability to independently identify and manage risks may erode. A "model monoculture" may develop that undermines our ability to understand growing risks.

My seventh point is related to the model monoculture problem. By incentivizing banks to focus excessively on the Board's models and position themselves to game the stress test, the proposal may encourage concentration across the system in certain assets for which the models give relatively favorable treatment. This could create new risks to financial stability.

Eighth, the proposed model changes will make the stress test less risk sensitive and less conservative, likely resulting in lower capital requirements than apply under the current stress test models.

One of the hallmark strengths of the stress testing program is its risk sensitivity. Today's proposal would reduce risk sensitivity by eliminating some of the granularity that was previously considered in certain models.

In addition, one of the key stress testing principles is conservatism.² Conservatism is important because it helps compensate for gaps in what the stress test captures. For example, the test does not consider the contagion effects that would occur if multiple large banks came under stress at the same time. Because we know that contagion effects would likely increase the severity of the stress for each bank and the system, it is appropriate to make conservative assumptions in our modeling choices to account for this gap.

Beyond disclosing the models, the Board today proposes several changes to the credit, market, and operational risk models that cut against that principle of being conservative. For example, mortgage loss projections incorporate fewer non-linear effects because of the lack of regional variation in losses; the methodology for estimating probability of default for the corporate loan model is less conservative; loss given default and line of credit draw rate assumptions are weaker for market risk; liquidity horizons for adjusting positions under stress are shortened in ways that weaken the effect of the global market shock; and operational risk is no longer macro-sensitive, meaning that it does not take into account macroeconomic changes in the scenario. These are but a handful of the proposed changes to the models.

If these new credit, market, and operational risk models were used for the 2024 or 2025 stress test, they would have led to lower projected losses, leading to projected capital ratios that were roughly 100 basis points higher than using the existing models. That's approximately $115 billion of additional capital more optimistically projected by these three models. These higher projected capital levels would translate to lower stress capital buffer requirements.³

The proposal frames the changes to the pre-provision net revenue (PPNR) model as largely offsetting this reduced conservatism. However, I have concerns that the projected impact of the PPNR model changes may not be durable. For example, the PPNR model will rely on banking industry estimates of noninterest income and expense for its projections. This reliance will incentivize banks to make rosier estimates than they otherwise would, biasing the models over time. This aspect of the proposal is inconsistent with the Board's principle of independence in modeling.⁴ I am skeptical that the PPNR model will provide a meaningful long-term offset to the reductions in conservatism in the other models.

In sum, the model changes proposed today make the stress test less conservative, which will lead to lower bank capital. The Board previously said it would not use this rulemaking to materially lower capital requirements, but making the models less conservative has that effect.

Ninth, the proposal is premature in revealing all of the details of the models up front. Once published, the details of the models cannot be taken back. The Board should have taken comment on whether model and scenario disclosure was a good idea compared to other alternatives before jumping ahead to the actual disclosures themselves.

I am also concerned that the Board has not had time to conduct a comprehensive economic analysis on this proposal. For instance, the proposal considers the effect of the model changes on capital in comparison to only the 2024 and 2025 stress tests, rather than over a longer timeframe. The proposal also makes no attempt to quantify the anticipated effects of gaming on lowering capital, though it acknowledges that gaming will increase. These gaps leave us with an incomplete picture of the effects of the proposal.

Moreover, it would have been appropriate to consider these changes in light of other contemplated changes to the capital system, including the enhanced supplementary leverage ratio, the global systemically important bank surcharge, and the Basel III capital reforms. The Board also has a pending proposal on averaging stress tests results, which curiously has not been brought back to the Board for finalization. A comprehensive review would permit the Board to understand the interactions among all of these capital changes.

Tenth, the operational risks of this proposal are substantial and put the staff in a difficult position. The proposal lays out an ambitious annual process for publishing the scenarios and models for comment and responding to such comments before conducting the stress test. I am concerned that this process is unrealistic from an operational perspective, especially at existing staffing and resource levels. This situation will create risks to the Board's ability to administer a key aspect of risk-based prudential requirements for the largest firms.

In conclusion, it is critical that the stress test remains a dynamic, rigorous process that allows supervisors, banks, and the public to understand vulnerabilities in the banking system and at individual banks. I believe today's proposal will significantly weaken the stress test and, consequently, bank resilience. I have deep concerns that the set of changes the Board is now considering risks reliance on a stress test that can no longer effectively assess the resilience of the largest banks and lacks credibility, thus putting our financial system and economy at risk.

I find the Board's choice today regrettable and I cannot join it.

1. See Board of Governors of the Federal Reserve System, "Statement on Stress Test Proposal by Governor Michael S. Barr," press release, April 17, 2025; Michael S. Barr, "Preserving the Dynamism and Credibility of Stress Testing," speech at the Peterson Institute for International Economics, Washington, D.C., September 25, 2025. Return to text

2. 12 CFR 252, Appendix B, § 1.6. Return to text

3. The model changes also do not include the Current Expected Credit Losses (CECL) standard for calculating allowances, even though banks are required to use the standard. The failure to incorporate CECL into the stress test is inconsistent with both the Board's approach to adhering to U.S. generally accepted accounting principles and the principle of conservatism, and results in much lower capital. Return to text

4. See 12 CFR 252, Appendix B, § 1.1. Return to text