The time is ripe to reflect on whether we’ve extracted all we can from traditional stress tests, and to then consider some alternative strategies for future installments.
On the heels of the recent release of the 2022 CCAR scenarios, the U.S. stress testing season is now in full swing. It’s fair to say that banks are not on tenterhooks – no one has “failed” the test since 2015, and no one is expected to this time around.
This mirrors the sentiment in the industry across the Atlantic: no UK bank has failed since 2016. Indeed, the most recent British stress-testing exercise, the results of which were released in December, was met with a collective yawn.
Banks have mastered the art of running portfolio stress tests involving an external, severely adverse, macro scenario. The test has consequently become a straightforward compliance exercise; bankers and regulators no longer learn anything new.
Indeed, everyone does roughly the same thing, every year, and everyone passes. Moreover, no one expects anything to change in 2022 and beyond.
Before exploring a potential solution to the existing malaise, it’s helpful to consider how the pandemic has changed the stress-testing game.
The COVID-19 Effect
Through the pandemic, stress tests were criticized because they had not previously considered a global shutdown-style event. Since SARS and MERS provided ample warning of the pandemic threat, these critics had a point.
But it’s unclear that a prior stress test that considered the potential impact of a pandemic would have helped bankers or regulators to prepare for COVID-19. Though there were many doom-and-gloom prognostications about a complete economic shutdown, the reality turned out to be far milder for the financial system.
If this pandemic-driven criticism of conventional stress tests was met head-on, the solution would involve simply running more scenarios. However, I don’t think more of the same cuts the mustard when the system is showing such clear signs of atrophy.
We should instead be trying to re-imagine the whole exercise, asking ourselves: What should be the design of Stress Testing 2.0?
There are several elements of traditional stress tests that have always bugged me.
The premise of the existing framework is that banks are passive victims of external shocks. I’m sure there are historical examples where well-managed, conservative, law-abiding banks failed because external conditions rapidly deteriorated. Much more common are situations where poor individual or collective decision-making by the banks precipitated severe financial crises.
For example, banks were not passive victims during the GFC: they underwrote and then actively traded the subprime mortgages that ultimately skewered the economy. They were also not passive victims during the Asian financial crisis.
On the other hand, banks were passive victims during the pandemic. I haven’t heard any conspiracy theories involving evil banks this time around – but associated credit losses were low and bank failures were basically non-existent.
The Next Generation of Stress Testing
Stress Testing 2.0 should therefore make each bank the architect of its own potential demise. This style of analysis involves a much deeper consideration of the behavior of banks during the preceding boom, as opposed to the recession that inevitably follows.
The focus of my proposed next-generation stress test would therefore fall on originations as opposed to defaults.
Data would be collected on sectoral lending volumes at a granular level, and the parts of the business growing most quickly would then be studied very closely. The stress test would ask questions about whether industry growth rates were similarly vigorous, whether the specific bank’s market share was expanding or declining and whether such growth was justified by broader economic conditions.
Put simply, Stress Testing 2.0 would try to identify lending products and segments experiencing bubble-like behavior during expansions. If something is growing like a weed, it’s probably a weed.
If a particular bank was found to be expanding its market share in a certain category, this may be considered perfectly healthy and desirable; there is no systemic risk if the odd bank makes a risky foray into a niche market. If, on the contrary, the same bank with the same growth rate was experiencing a declining market share in that niche market, this may be a sign of irrationally exuberant lending across the industry.
Regulators could make this determination by lining up results from a number of banks to see whether there is commonality of purpose in current lending practices.
The regulator’s subsequent response would not be straightforward. They would need to decide whether the weed-like sector was systemically important and whether higher-than-normal sectoral lending growth constituted a positive economic development.
This alternative stress test has two major benefits relative to the existing approach. First, it recognizes that the best time for banks to control future credit risk is at the application table. (You can always avoid a default by refraining from making the loan in the first place.) Second, it augments the existing methodology by looking at an entirely different stage of the lending process.
It is based on actual decisions taken by banks, as opposed to hypothetical future scenarios that may never happen.
Existing stress-testing methodologies (like CCAR and DFAST) have brought many benefits. In the early days, for example, the stress tests provided a modicum of assurance that counterparties could be relied upon in the midst of a raging financial crisis.
Subsequently, stress tests provided the justification for a significant increase in regulatory capital for big banks – though, admittedly, this could have been achieved without the theater of the annual test. Throughout the period, stress testing forced banks to modernize data and modeling resources to meet the exacting standards of supervisors.
These benefits, however, have now been fully realized, and it therefore makes sense to consider alternatives – whether it’s the credit-originations-driven approach discussed here or something else. The critical takeaway is that there are many potential ways to skin the stress-testing cat.
There is no doubt the current approach is stale: everyone is too comfortable with it and the results are terminally boring. If alternative stress-testing approaches are not tried, the system will soon lose all relevance.
Tony Hughes is an expert risk modeler for Grant Thornton in London, UK. His team specializes in model risk management, model build/validation and quantitative climate risk solutions. He has extensive experience as a senior risk professional in North America, Europe and Australia.