Perspective on Risk - Jan. 27, 2025
Systemic Risk Exemption; AI & Bank Supervision; Thinking About Risk; Stress Tests;
Systemic Risk Exemption
Readers will know that I have long been interested in the evocation of the systemic risk exception in the resolution of SVB, Signature and the other regional banks. This exemption allowed the regulators to avoid using the statutorily-mandated resolution process, and in this way extend protection to uninsured depositors. I’ve gone so far as to FOIA the Fed for information on how they reached the determination that the crisis was systemic1. Maybe now, after 18 months, they’ll respond.
The GAO is out with a review of the use of the exception: Federal Deposit Insurance Act: Federal Agency Efforts to Identify and Mitigate Systemic Risk from the March 2023 Bank Failures (GAO)
Background
The GAO has a nice description of the exemption:
FDICIA created an exception to the least-cost rule, known as the systemic risk exception. Under this exception, FDIC may resolve a troubled depository institution without complying with the least-cost rule, but only if the Secretary of the Treasury determines that (1) FDIC’s compliance with the least-cost rule would have serious adverse effects on economic conditions or financial stability, and (2) other authorized action or assistance would avoid or mitigate such effects. The Secretary of the Treasury must make the determination on the written recommendation of the FDIC’s Board of Directors and the Federal Reserve Board, in each case, on a vote of not less than two-thirds of their respective board members. The Secretary of the Treasury’s determination must also be made in consultation with the President of the United States. In 2010, the Dodd-Frank Wall Street Reform and Consumer Protection Act (DoddFrank Act) narrowed the systemic risk exception to be used only to wind up the affairs of an insured depository institution for which FDIC has been appointed receiver.
Agencies Complied With Legal Requirements
The document describes in broad strokes compliance with the requirements:
FDIC and Federal Reserve staff established the bases for recommending the systemic risk exception for SVB and Signature Bank. This involved conducting analyses on financial and economic conditions, including deposit outflow and funding analyses.
Collectively, these actions helped FDIC and Federal Reserve staff evaluate whether complying with the FDI Act’s least-cost requirements would have serious adverse effects on economic conditions or financial stability
FDIC and Federal Reserve staff shared their analysis on the potential effect of SVB and Signature Bank failures on financial markets and broader economy with their management. FDIC and Federal Reserve management then shared staff analysis and updates of market conditions with their respective Board members and legal division staff. This information helped Board members decide whether to recommend the systemic risk exception to the Secretary of the Treasury.
Treasury staff evaluated the recommendations from FDIC and the Federal Reserve and consulted with these agencies and OCC to assess the systemic risk stemming from the deposit runs … [Treasury] Staff recommended that the Secretary of the Treasury, after consulting with the President, invoke the systemic risk exception for both SVB and Signature Bank.
Substantive Regulator Fears
The report nicely summarizes the fears;
deposit outflows → liquidity pressure → bank runs & failures
deposit outflows → liquidity pressure → reduced credit availability
uninsured losses → corporate disruptions
reduced market confidence → broader negative economic effects
FDIC and Federal Reserve staff were concerned that not guaranteeing uninsured deposits at SVB and Signature Bank could trigger runs on other banks, leading to further bank failures, according to agency documentation. …
Federal Reserve documentation indicated that many banks funded largely by uninsured deposits were under considerable pressure and that the disorderly failure of these banks could lead to greater losses in deposit markets.
FDIC and Federal Reserve staff determined that a least-cost resolution of SVB and Signature Bank could result in higher lending costs. Federal Reserve staff told us that they anticipated that widespread deposit outflows and subsequent bank failures would reduce the number of banks willing or able to lend to U.S. households and businesses. This would raise lending costs for borrowers.
FDIC and the Federal Reserve determined that imposing losses on uninsured depositors at the two failing banks could cause widespread disruption across the U.S. economy and further destabilize U.S. banks, according to agency documentation. Many of these uninsured depositors were businesses, and regulators anticipated that their inability to access funds, even for a short time, would lead to payroll delays and other disruptions.
Did Events Pose ‘Serious’ Adverse Effects
Look back at the language of the exemption. It requires a determination of “serious adverse effects on economic conditions or financial stability.” It is unclear whether the methodology devised defined what would constitute serious adverse effects, as compared with just regular old adverse effects.
Do liquidity pressures at other banks make the event serious? Aren’t there other mechanisms for handling this?
[The] FDIC found that certain banks had turned to wholesale funding sources to offset deposit outflows. Because wholesale funding sources, such as brokered deposits, are generally more expensive than retail deposits, FDIC staff told us that they expected this would increase banks’ funding cost and, in turn, reduce their lending activities.
Note sure this justifies the exemption.
Would the resulting increase in bank failures make this serious? The optimal number of failures is not zero and perhaps the failure of the banks living on the edge is a good thing.
FDIC staff told us that banks with high concentrations of uninsured depositors and unrealized losses in securities were particularly susceptible to losing investor confidence.
Aren’t these exactly the kind of risky banks that should fail?
Do higher credit spreads and perhaps a degree of reduced credit availability reach the serious threshold? How much would spreads have to spike or lending drop?
Federal Reserve staff told us that they also considered economic theory and prior experience on how bank strains can have spillover effects on the broader economy.
I guess we have to trust them as there is no evidence provided for the possible broader spillover effects.
There is only one sentence I found that seems to pose broader systemic risk concerns, and it is in the section on approving the Fed’s Bank Term Funding Program:
Treasury staff specified that the potential run risk on uninsured deposits posed a broader financial stability concern, rather than a localized issue limited to a small number of regional banks.
Fear of Regional Bank Failure
While, it does not appear that the systemically-significant banks were at risk2, there is a lot of discussion, but little direct evidence, that the failure of regional banks was the motivating force.
FDIC and Federal Reserve staff obtained and analyzed real-time information on depositor composition of other regional banks. … In the event that other large regional banks failed, they concluded that the inability of businesses to access funds would likely lead to similar payroll and payment delays
FDIC staff observed that following SVB’s failure, the S&P regional banks index had its worst week since 2009 …
[The FDIC and Federal Reserve] obtained information on the credit spread movement of regional banks, observing a widening of credit spreads for these banks. This indicated that investors perceived large regional banks as riskier.
Treasury staff also used public regulatory filings and other public reporting to assess the condition of … large regional banks
Market Discipline & Moral Hazard
The ultimate question is whether the risks here warranted the response, and the reduced market discipline and increased moral hazard that results. Kudos for the GAO for even raising these concerns, albeit obliquely starting on page 29 with its “Proposed Changes Might Address Risks Posed by Use of Systemic Risk Exception.”
[The] systemic risk exception can increase moral hazard and the risk of bank failures. By providing explicit deposit guarantees and implicit guarantees, such as backing uninsured deposits, FDIC may create incentives for banks to engage in riskier behavior, such as investing in riskier assets or increasing risky lending, according to literature and academics.
Furthermore, the systemic risk exception may reduce depositor monitoring of their bank’s activities, encouraging banks to take on excessive risk, according to literature. This is because deposit insurance expansion may cause depositors to be less careful in selecting and monitoring their banks. If depositors feel less afraid of losing their deposits, they may also be less likely to withdraw their funds from risk-taking banks weakening the relationship between risk and bank funding.
My Take
I’m still not sure the juice was worth the squeeze. We are now in a world where large uninsured corporate deposits are no longer at risk. I have not seen any credible proposals to reintroduce market discipline. Moral hazard rises unabated.
AI & Bank Supervision
Two of my favorite things, together.
Dan Davies of the FT writes Do androids dream of financial crises? (FT) which discusses a paper by some ECB economists Predicting Bank Distress in Europe - Using Machine Learning and a Novel Definition of Distress (ECB).
As I have stated before, it is reasonably easy to develop a model of (non-SIFI) bank distress based on published financial and call report data. I’ve gone so far as to suggest that the supervisors should rely on these models rather than the subjective judgment of examiners. Dan Davies seems to agree:
[The] team at the European Banking Authority is indeed getting serious — it’s just published a distinctly better than workmanlike piece on the possibility of using random forests, neural networks and similar techniques to potentially make a bit of progress towards one of the great dreams of central bank economists. That is to say, the possibility to automate the dreary and unprestigious job of bank supervision, using machines to monitor the data rather than going through supervisory returns yourself.
The EBA researchers actually do quite a bit better than previous efforts. “Breaches of supervisory concern levels on a few key ratios” are their better source of training data points than “actual failures”, meaning they can train a variety of models and use an ensemble approach.
They use private regulatory data along with country-specific economic data to predict breaches of regulatory metrics (Tier 1 common equity risk-based capital, leverage ration, liquidity coverage ratio). They find a Random Forest model, a fairly basic AI approach, outperforms the other models. They under-sample non-distress banks, and over-sample distress banks; they also synthetically create some distress banks.
The model is balanced between false positives and negatives; I might try and skew the analysis to minimize calling a bank safe when it should be in the distress category.
Developing a sophisticated methodology to “predict” bank balance sheet and income statements could be a game-changer for supervision.
For example, with this approach, the regulators might be able to eliminate the DFAST stress-testing requirement by conducting their own “what-if” analysis. While the regulators should have (and did) completed analysis of the risks to regional banks from rising rates, this would speed their ability to test a variety of different economic outcomes.
Thinking About Risk
Minimum Levels of Stress
As the world improves, our threshold for complaining drops.
In the absence of big problems, people shift their worries to smaller ones. In the absence of small problems, they focus on petty or even imaginary ones.
A nice little discourse from Morgan Housel on the Collabfund site: Minimum Levels of Stress.
Psychologist Nick Haslam once described what he called Concept Creep. It’s when the definition of a problem expands beyond its original boundaries. It often gives the impression that the world is getting worse when what’s changed is our definition of what counts as a problem. It happens two ways:
Things previously considered normal are redefined as risks. Like a child being bullied at school, or mild anxiety being diagnosed as mental illness.
Less severe instances of a risk are recast as major risks. Like having to delay retirement from age 65 to age 67.
In each case, the world can get better but people don’t feel it – they can even feel like they’re going backwards – because once a problem is solved it’s replaced by a new one, often with the same level of anxiety, fear, and anger.
Two Papers on Decision-making Under Uncertainty
I’m going to discuss two papers that I’ve read recently. Both papers deal with decision-making under uncertainty, but they come at it from different angles. Both papers suggest that standard economic models of risk are flawed.
The first paper argues that ambiguity aversion is a rational response to deep uncertainty. The second paper argues that what looks like risk aversion might just be people struggling with complexity—not a true preference for avoiding risk.
Are Unknown Risk & Long Term Risk The Same Thing
Here is an sophisticated and interesting paper with a simple title, Ambiguity and the Language of Long Run Risk, and a pretentious abstract. But despite this, it is worth reading and considering.
This paper explores a surprising connection between two different ways of thinking about uncertainty: how people deal with ambiguous situations (where they don’t know the exact probabilities) and how we evaluate long-term risky investments (like financial assets or public projects with uncertain future returns).
This paper formalizes Knightian uncertainty in a long-run setting using tools from ambiguity aversion and stochastic processes. It argues that as time goes to infinity, all risk becomes Knightian uncertainty, so we need a different approach than standard probability-based models.
Ambiguity aversion and discounting expected long-term future cashflows are two sides of the same coin. Specifically:
Any model of ambiguity aversion can be seen as a model of long-term risk.
Under reasonable conditions, any model of long-term risk can be rewritten as an ambiguity aversion model.
This has a number of implications for financial risk managers, some of which we intuitively already do. We need to:
Complement VaR with worst-case scenario modeling (e.g., min-max expected utility).
Use stress tests that assume different possible probability distributions, rather than relying on one estimated distribution.
Account for stochastic discount factors that reflect evolving ambiguity rather than assuming a fixed risk premium.
Include ambiguity aversion in portfolio allocation models and shift from static diversification to robust portfolio optimization, which considers the worst-case impact of model uncertainty.
Identify assets with exposure to Knightian uncertainty and apply higher risk buffers to them.
Consider long-term risk correlations, not just short-term historical correlations.
On long-term positions, shift from maximizing expected returns to minimizing regret under extreme uncertainty.
If you want to simplify this list:
Stop assuming you can estimate probabilities for long-term risks—instead, plan for ambiguity and unknown unknowns.
Behavioral Anomalies May Just Be Complexity-Induced Mistakes
Here is another interesting paper, which eventually I will relate to the last paper discussed.
The author of Decisions under Risk Are Decisions under Complexity (American Economic Review) posits that many well-known behavioral anomalies in decision-making under risk are not actually about risk at all. Instead, these behaviors appear just as strongly in situations where there is no uncertainty, only complexity.
This paper potentially makes a significant contribution by challenging long-held assumptions in behavioral economics. The paper presents strong evidence of a correlation between complexity and decision anomalies (supported by experiments).
The paper argues that behavioral anomalies traditionally attributed to risk preferences (like probability weighting and loss aversion) are instead largely driven by complexity-induced mistakes. This shifts the focus from risk preferences to cognitive limitations as a primary driver of these behaviors. By demonstrating that these anomalies occur in deterministic settings without risk, the findings question the normative relevance of behavioral theories like prospect theory. These theories often assume that observed behaviors under risk directly reflect preferences, which the paper disputes.
Let’s Think About The Two Papers Together
Considering the two papers together, a behavioral economist might suggest that if people struggle to evaluate uncertain situations, they might default to avoiding them not because of deep uncertainty, but because their brains take shortcuts. System 1 thinking in Kahneman theory.
Kahneman laid out System 1 and System 2 thinking, whereby System 1 is fast, intuitive and automatic - your first reaction to a stimulus, and System 2 is slow, analytical, deliberate, and effortful—used for careful reasoning.
The second paper could challenge part of the first paper's argument by suggesting that ambiguity aversion might not always be a true preference—sometimes, it’s just cognitive overload. Put more simply, when faced with a complex problem some people turn off System 2 and rely only on their System 1 approach.
Ambiguity aversion and risk misjudgment are both driven by System 1 heuristics, and the real question is “Are people truly ambiguity-averse, or are they just avoiding the effort of System 2 thinking?”
If that’s the case, then long-term risk management shouldn’t just focus on ambiguity aversion but also on helping decision-makers simplify complex risks to avoid mistakes. Ambiguity aversion might be a fixable cognitive bias.
Stress Tests
Fed
So the Fed was sued (on behalf of the big banks) by American Bankers Association and the U.S. Chamber of Commerce. The suit alleges that the current stress testing framework lacks transparency and violates federal statutes by not allowing public input, leading to erratic results and regulatory uncertainty. Welcome to the new post-Chevron world.
Unfortunately, dropping the stress test requirements (as I have advocated in the past - the requirements no longer serve the original purpose - and just advocated above) will look like capitulation to the industry, unfortunately possibly prompting Fed legal to react like a cornered animal. Maybe Bowman can change things.
Bank of England (BoE):
The BoE conducts two types of stress tests - Bank capital stress tests and System-Wide Exploratory Scenario (SWES) exercise aimed at understanding the behaviors of both banks and NBFIs under stressed financial market conditions.
The BoE is scaling back the frequency of its Bank capital stress test, and has not yet announced the next SWES, however it is clear that this will be an integral part of their approach going forward.
Starting in 2025, the BoE will transition to a biennial schedule for its main bank capital stress tests, moving away from the previous annual assessments. These tests will focus on risks associated with the financial cycle and will inform the setting of capital buffers for both the overall banking system and individual institutions. In the intervening years, the BoE plans to conduct targeted stress tests to explore specific vulnerabilities or emerging risks as needed.
In January, Nathanaël Benjamin, the BoE’s Executive hDirector for Financial Stability Strategy and Risk gave a speech titled Joining the Dots.
A central theme of the speech was the need to integrate findings from different stress tests across sectors—such as banks, insurers, and central counterparties—to gain a holistic view of system-wide risks. Benjamin emphasized that as financial activities increasingly shift towards non-bank institutions, understanding the interconnectedness and potential systemic implications of these entities becomes crucial.
In it, he articulated a three-part approach to stress testing:
[We] will continue to run a financial resilience test of risks … involving submissions from participating banks, but it will be every other year. It will otherwise be akin to the previous annual cyclical scenario test and, will inform capital buffer setting.
[In] the intervening years we will supplement our assessment of the resilience of the banking system using stress testing when appropriate, but in a way that is much less burdensome for banks than our full Bank Capital Stress Tests.
[The] Bank will continue to use exploratory exercises involving participation from banks, akin to the BES of the previous framework, but not strictly biennially. These tests will be a means of assessing other risks, including structural and emerging risks that are not closely linked to the financial cycle.
The BoE has adapted its approach to the condition of the industry, in ways the Fed has not. As I stated above, conditions are different from when DFAST requirements were imposed.
A key reason why we are comfortable reducing the frequency of the full cyclical tests involving bank submissions is because banks are now much better capitalised.
Risk-taking has shifted across the financial sector since the early days of stress tests, and we must use our toolkit to widen our understanding of the financial system and gain an increasingly comprehensive view of the risks affecting it.
Much of this is defined by the mandate. The Fed, as a regulator, still is blinkered by its focus on the banks; the BoE is not.
The only reference to the SIFI banks is that the Fed monitored their liquidity.