It is now clear that a deep COVID-19 recession will hit the global economy this year. This event, completely distinct from anything that has happened in living memory, will bring a new set of challenges, and a host of opportunities, for risk modelers around the world.
Last year, back in the pre-COVID-19 era, the threat of a global pandemic was (along with cyberattacks) the most existential operational risk considered by banks. Following the 2002/03 SARS outbreak, regulators in Asia were particularly cognizant of the threat – Hong Kong, for example, ran a detailed pandemic stress test, focused on payment systems and banking operations, only a few weeks before COVID-19 started to unfold.
For those engaged in the more prosaic world of credit modeling, however, a consideration of pandemic risk was not really possible.
If a time traveler had offered a preview of actual 2020 events and data, they would have been met with befuddlement from 2019-era stress testers. A 20+% U.S. unemployment rate was so far outside the domain of historical banking data that any analysis of such an event could have only been described as speculative.
Those conducting scenario analysis last year, perhaps for stress testing, CECL or IFRS 9, were instead focused on the trade war and the apparent likelihood of a mild 2020-21 recession. Severely adverse scenarios still mimicked a rehash of the 2008/09 downturn.
There’s little jeopardy for credit modelers in such an environment. Specifying models to capture the contours of the 2008/09 recession, as well as the subsequent expansion, was a well-established practice by about 2012. Models developed during the subsequent period have never been tested by an actual recession.
Pandemic and Post-Pandemic Modeling Challenges
In the next few months, however, the domain of the data will expand considerably. While most 2019 risk modelers had information covering just one recession, the 2021 cohort will have data covering at least two. In both theory and practice, access to more information is always beneficial – but it can also cause headaches, especially if the new data exhibit unexpected or unusual behavior.
Alas, 2020 is a unique beast in almost all respects. It will require a truly exceptional credit risk model to be able to capture observed losses in both recessions, as well as during the intervening expansion.
Most discussion in the industry at the moment is that, despite the severity of the economic data, big bank failures are unlikely to occur in the COVID-19 recession. Given various forbearance programs, few think that the coronavirus will cause credit losses much in excess of those seen during the global financial crisis. If these projections hold – a big ‘if’ no doubt – we will have doubled, at least, the severity of the 2008-09 economic downturn, without appreciably increasing observed industry credit losses.
This will take some deft modeling.
In addition to accounting for the impact of forbearance and for major differences in credit quality at the outset of recession, the distinct dynamics of the two downturns will also need to be reconciled.
By comparison with the current recession, the 2008/09 event happened in slow motion. The time between the first subprime-related headline (December 2006, by my reckoning) and U.S. initial unemployment claims rising above 500,000 (November 2008) was 23 months. Immediately prior to the Lehman collapse, in September 2008, you could still cling to hopes of a relatively mild recession. In contrast, it took around three months from the first reported COVID-19 case in China to initial U.S. unemployment claims in excess of 3,000,000!
If credit losses are disproportionately low this time around, it may, in part, be because the start of the current recession was less drawn out. A 2019-era risk modeler had no way of capturing this effect, since they only had access to slow-onset recessions. A 2021 modeler, expanded domain in hand, will at least be able to investigate this phenomenon.
The shape of recovery will also be critical.
The last two recessions (2001 and 2008/09) have been marked by U-shaped recoveries in most countries; the slow exit mirroring the slow onset of recession. Some have suggested a “swoosh” shape this time – a rapid onset followed by a painful recovery. The optimists among us will hope for the sharpest possible V-shaped recession – the exit mirroring the precipitous downfall – with subsequent credit losses only slightly elevated relative to normal.
Parting Thoughts: Peering Into the Future
In 2021, or when models covering both recessions have fully bedded down, analysts will be able to run an alphabet-soup of different alternative scenarios – and have at least some relevant credit data to back up their analysis. This should make 2021-era credit modeling both extremely challenging and highly rewarding.
If we send our time traveler forward to, say, 2023, I suspect that the 2020 recession will again be viewed with a certain kind of befuddlement. We can all hope that the response to the next pandemic – and the remainder of this one – will be informed by 2020 policy successes and failures. If the lessons are learned, by 2023, the notion of 20% (or 40%) unemployment will again be considered ludicrous.
There will still be interest in running the 2020 scenario in years to come, but I suspect the humble 2019-era severely adverse scenario will remain the gold standard for most applications.
Risk Weighted 