DFAST monoculture is its own test

Drop in frequency and scope of stress test disclosures makes it hard to monitor bank mimicry of Fed models

Key points Summary Q&A

• In the wake of the 2008 financial crisis, the Dodd-Frank Act produced stress tests that have informed equity prices and credit spreads, inter alia.
• But 2019 reforms led to fewer data points being collected less frequently and by a reduced number of participating banks.
• Critics say the changes compromise the transparency and effectiveness of the stress tests in assessing banks’ risk profiles
• There are concerns this might result in what former Fed chairman Ben Bernanke called a ‘model monoculture’ – where banks rely on mimicking Fed methods instead of developing their own.
• It could mean banks’ calculations – and their risk deficiencies – won’t be as easy to discern.

These summaries are generated by an LLM, but edited by humans.

From what constitutes adequate capitalisation to how more opaque trading activities should be risk-weighted, the regulatory response to the 2008 financial crisis marked a paradigm shift on many fronts.

When it comes to public interest, however, the most consequential rethink arguably addressed the question of who stress tests serve – and what information they provide.

When Washington officials announced, in early 2009, that they would subject banks to what they initially termed a supervisory capital assessment program, the objective was in part to size up each lender’s specific toxic-asset exposure by vetting their balance sheets in a standardised, microgranular fashion.

A cave-in to banks’ demand for transparency may result in public disclosures that yield fewer genuine insights into banks’ risk profiles

“We’re going to do a tough scrub of the balance sheets,” said Ben Bernanke – then US Federal Reserve chairman – in a February 7 meeting. “We’re going to require enhanced disclosure so that there will be as much information as possible, with as much consistency as possible, about the balance sheets of the banks.”

Industry insiders and commentators, however, feared the Fed was about to compound the very problem it was trying to solve. Banks outed as capital-deficient – the argument went – would be required to seek extra equity from private sources. But with market spigots still tightly shut, Wall Street executives would have no choice but to turn to Washington, accept an equity injection by the Treasury and throw whatever shareholder value was left onto the pyre.

The following months – and years – proved the supervisors’ camp right. Not only did Scap manage to stabilise markets, but annual results from its successor – the Dodd-Frank Act stress test – began informing equity prices and credit spreads, inter alia. The Fed’s new way of stress-testing opened a window onto a trove of newly minted information for shareholders, investors and analysts.

Related

DFAST fashion: emerging trends from 12 years of US stress tests 

Fed agrees to unveil stress-test models in transparency U-turn 

But that potential to generate information is slowly getting eroded. Reforms enacted in 2019, during the first Trump administration, have resulted in DFAST publications that cover fewer data points for a smaller crop of banks. And a cave-in to banks’ demand for transparency may result in public disclosures that yield fewer genuine insights into banks’ risk profiles.

From DFAST’s early days, the degree to which the test’s performances are predictable – thus potentially pre-empting risk deficiencies from being spotted – has nagged academics and regulators.

A 2017 working paper from the Office of Financial Research found a “nearly perfect linear relationship” between loan-loss rates in DFAST’s ‘adverse’ scenario and those under the ‘severely adverse’ simulation across all participants. Highly predictable, routine DFAST outcomes, the authors noted, could incentivise banks to optimise balance sheets and processes to output the best possible performance in stress tests, while neglecting to incorporate more idiosyncratic or harder-to-detect risks.

The analysis, based on 2015–2017 DFAST data, is impossible to replicate today. Since the first reformed DFAST in 2020, banks have only been stressed under the ‘severely adverse’ assumptions, albeit federal regulations ostensibly allow for additional scenarios. 

The loss of information was compounded by the move to biennial participation for most banks, which made the results’ data series highly discontinuous and therefore tougher to monitor across years.

Model monoculture

The reforms also decreased the onus on most banks to generate their own DFAST figures, using in-house assumptions and models. Under the original framework, all banks above $10 billion in assets were required to run internal DFASTs, even if they weren’t part of the Fed-run exercise. Post-reform, only the eight global systemically important banks must do so every year, while banks in categories II and III of the Fed’s tailoring framework are only subject to biennial requirements. The majority of banks – meaning those with less than $250 billion in assets – were relieved of the requirement altogether.

 

Lurking in the background is the risk of a drift into what Bernanke, in 2013, called a “model monoculture”, where banks are incentivised to mimic the Fed’s stress-testing methodology, rather than invest in internal frameworks tailored to their own risks.

The Fed’s decision – until this year – to not fully lift the hood on its stress-test engine, and the requirement for lenders to conduct their parallel simulation, were intended to stave off this risk. In the words of then-New York Fed president Bill Dudley in 2011: “If all we ever did was run supervisory stress tests in which we instruct banks in detail how to perform the test, we would be in the position of a parent who shows his child how to solve each problem in her homework – and never discovers whether the child can do the work on her own or not.”

Even back then, when the mood was still very much in favour of tighter regulation, bank executives would have chafed at that characterisation. 

For sure, it may be a fallacy to assume higher transparency on the Fed’s part will automatically encourage complacency. The problem is, anyone outside of the Fed and the banks’ top echelons may not have a way of telling.

To stretch Dudley’s metaphor, the students have been progressively given ever finer inputs to feed into an AI assistant – and the end-result may soon be indistinguishable from organic work.