Risk Weighted

Would DFAST Have Saved SVB?

There is a school of thought in the banking industry that Silicon Valley Bank would not have failed had it been subject to comprehensive stress tests. Is this true or just a fallacy?

Friday, March 24, 2023

By Tony Hughes


What lessons can we learn about the true nature of stress tests from the failures of Silicon Valley Bank (SVB) and Signature Bank? There is at least some belief that these banks would have avoided collapse if they had been required by regulators to more rigorously examine their “stresses.”

Is this accurate, or have these incidents taught us that the U.S. stress-testing system needs to be revamped, with a greater emphasis on concentration risk?

Before much of the Dodd-Frank Act was rolled back in 2018, banks like SVB and Signature Bank were subject to what was tautologically known as the DFAST stress test. This exercise was a watered-down version of CCAR that applied to all banks with more than $30 billion in assets under management.

tony-hughes-Nov-04-2021-01-01-50-00-PMTony Hughes

Now that the aforementioned banks have failed, a reasonable question is whether a continuation of the DFAST program would have made any difference.

At the time the rollback was proposed, I argued that the exercise was not holding back smaller banks, which, relative to larger CCAR firms, were enjoying increased market share, taking on a riskier profile and achieving improved profitability – despite the obligation to test. DFAST itself was similar in spirit to CCAR in that banks had to report scenario-based stress projections, but the exercise was largely toothless in terms of capital enforcement.

The most positive aspect of the DFAST test was that it extended the benefits of stress testing across a broader swath of the U.S. banking industry. Banks that engage in such exercises are forced to build additional models of their portfolios and to improve their data collection and retention abilities. Moreover, their risk managers are required, at minimum, to annually consider existential threats.

Given its toothless nature and low cost, I thought the exercise was well worth retaining.

However, I still don’t think the DFAST would have saved SVB. The core premise of the test, shared by most other stress testing regimes around the world, was that deep recessions cause banks to fail. Neither the current episode nor our experience during the GFC are consistent with this core premise.

SVB vs. the GFC: Stress-Testing Flaws

If we start with the GFC, I would argue – with 20-20 hindsight – that recession became inevitable when the cohort of poor-quality subprime mortgages was originated. The subsequent default of these loans was then a link in a chain of events that led to the deepest recession in more than half a century. In other words, the recession did not cause the subprime crisis – both the bank failures and the recession were instead caused by the concentration of poor-quality mortgages that were in place on the eve of recession.

Despite this, regulators opted for stress tests where banks are assumed to be passive victims of events that are outside their control.

Turning to the current situation, with U.S. growth positive on a year-over-year basis and with a 3.4% unemployment rate in play, the standard stress test would suggest that banks like SVB and Signature should be doing just fine. It's true that house prices, the third macro focus of stress testing, are rather shaky in many jurisdictions, but the banks that failed were not particularly active in mortgages.

The "stress" that upended SVB was that successive Fed rate hikes eroded the value of Treasury bonds held by the bank. When jittery uninsured depositors started withdrawing their funds, SVB was forced to realize large losses on its Treasury holdings, further spooking depositors.

The interesting thing – pardon the pun – is that rising interest rates are usually positive for bank margins and profitability. It was the unusual concentration of SVB's assets that led to its downfall. Had the bank made a different choice when deploying its capital, or spread its bets more widely, it would still be alive today.

The Concentration Risk Factor

The word that is common to these disaster narratives is "concentration." In my view, stress testing should focus more on identifying various forms of concentration and then assessing whether such imbalances require correction by regulators, with a view to avoiding systemic crises. 

Where the current situation differs from the GFC is in the nature of the concentration. During the subprime crisis, the size and prominence of mortgage portfolios were growing rapidly for many banks, including several global systemically important banks (GSIBs). The associated growth in securities markets added leverage and complexity, increasing the risk of a systemic crisis and making a recession inevitable.

These are the most dangerous concentrations, those that extend to many or most banks – including those that are too big to fail. If it's growing like a weed, it's probably a weed.

The weed at SVB was rather more idiosyncratic. The question going forward is whether we can identify unusual clusters of banks that may be jointly susceptible to unexpected confluences of economic forces.

If we could stomach the situation where all banks in a given cluster fail simultaneously, the threat would not be systemically important, and further action would be unnecessary. If such an outcome would be intolerable, the imperative would be to understand the vulnerability of the cluster and the likelihood that the vulnerability will be exposed at some point in the future.

The technology and data required to identify clusters of banks is available right now. The techniques used are sometimes difficult to interpret, but are often described as unsupervised machine learning.

Parting Thoughts

I tend to dwell on regulators' overuse of scenario analysis for stress testing purposes – but it’s not without good reason.

There are a million ways bank behavior can be rigorously analyzed to improve our understanding of systemic risk. As an industry, we should be exploring a much broader range of analytical techniques, including methods to classify banks in terms of their business practices and structure. 

If a few firms adopt an unusual business model that is later found to be brittle or flawed, this is fine – as long as the banks involved are systemically irrelevant. We frankly need such banks to fail. If, on the other hand, radical practices are found to be sound and profitable, society will benefit from the boldness demonstrated by the pioneers.

One final point: regional banks are not normally systemically relevant, but they can sometimes be harbingers of broader issues in the industry. 

I continue to view the DFAST test as worthwhile, but I'd like to see it fortified with a much broader set of analytical techniques.


Tony Hughes is an expert risk modeler. He has more than 20 years of experience as a senior risk professional in North America, Europe and Australia, specializing in model risk management, model build/validation and quantitative climate risk solutions.


BylawsCode of ConductPrivacy NoticeTerms of Use © 2024 Global Association of Risk Professionals