Reassessing the Current State of Bank Safety

The recent and rapid confluence of tail-risk events – from the deadly global pandemic to Russia’s invasion of Ukraine to the subsequent supply-chain disruptions and inflation spike – has triggered many in our industry to muse about the risk of what I'll call "compound" threats. Instead of concentrating our attention on the possible impact of individual problems, the notion is that we must consider combinations of potentially unrelated issues as a part of routine risk management.

Tony Hughes

A decade ago, this may not have been possible, but the rise of machine learning (ML) and an explosion in available data has changed the equation. Through ML methods, for example, we can now mine data for subtle associations between seemingly oblique threats – allowing us to produce realistic simulations of tail events, as well as more accurate baseline projections.

Compound risks could very well be embedded in these simulations, albeit with very low probabilities attached. Artificial intelligence (AI) gives us answers to complex questions, but we have no assurance that the output will be valuable or accurate or even interpretable.

Let’s now unpack these ideas to determine whether a risk analysis approach that considers multiple (possibly unconnected) threats simultaneously is likely to be fruitful.

Is Banking Increasingly Dangerous?

After the events of the past four years, the natural tendency is to imagine that banks face an increasingly volatile and hostile environment.

But over the past few years we haven't had a full-blown banking crisis – and most major economies have avoided a post-pandemic recession. The handful of bank failures that have occurred seem to have been caused by self-inflicted wounds.

So, despite the epochal events of recent years, the vast majority of banks have muddled through them pretty effectively. Perhaps the world is not as dangerous for banks as it appears, at face value, to be?

The Arc of History and Systemic Banking Risk

It's interesting to view these ideas alongside a theory posited by Harvard psychology professor Steven Pinker's acclaimed 2011 book, The Better Angels of Our Nature: Why Violence Has Declined. He observed that the physical risk of violence we face as human animals has gradually declined over time. Pinker wrote that, at the time, we were living in the “safest period, in terms of our physical wellbeing and the threat of a violent death, in all of human history.”

Is it possible, across the broad arc, that financial and banking risk are now also in decline? 

The explosion in available data certainly means that our awareness of various perils is rising. (Ignorance is no longer bliss.) But whether this implies that the actual threat level is increasing is highly questionable.

Indeed, the recent robustness of the economy and of the banking system in the face of remarkable external upheaval may be a sign that Pinker's theory also applies in the world of banking.

A look at supply chains, disrupted by COVID and the troubles in Ukraine, gives some insight into why overall risk may now be lower.

Today, globalized supply chains are certainly more complex than local ones, but they are also better diversified. With suppliers in various countries, companies are now exposed to a different, more complex set of risks – like the threat of foreign war or a blockage in the Suez Canal. However, there are also likely to be workarounds that raise costs without threatening company extinction. From bankers’ perspective, they see a borrower with a temporarily increased cost base and a business that remains broadly viable.

This is a stark contrast from supply-chain risk we saw in the 20^th century, when a company manager might have personally known the person supplying the widgets. A simple blackout or train strike would have halted production for both companies. Moreover, the failure of a supplier would have, in many cases, been existential for a local manufacturer, with few outside options.

Walk Before Run

In statistics, there are several principles that should make us wary of trying to analyze compound risks. Most critically, Occam's razor tells us that when choosing between simple and complex explanations, the simple one should generally be preferred.

We need to ensure that we can walk before we try to run. The analysis of combinations of threats implies that complex, nonlinear models will be central to the analysis. If the occurrence of event A changes the potential impact of event B, this implies that a non-linear method is required to capture the behavior of the system. 

We don't yet have a proper handle on the impact of many individual threats. It's therefore unrealistic and premature to imagine that we are ready to tackle the effect of combinations of perils on our various portfolios.

The reason for my pessimism is that we get so few opportunities to test theories of what constitutes a genuine risk to bank survival. Recessions are rare and bank crises are even rarer. The closest we can get to a validated theory is that insufficient capital combined with avarice in banks (either collectively or individually) is a very dangerous mix.

Most external threats, meanwhile, seem to be benign – but, again, I can’t say this with sufficient statistical justification. It's possible that two or more apparently benign shocks, occurring simultaneously or in sequence, could cause a malignancy, but it will never be possible to support such a hypothesis with actual data.

Parting Thoughts

Machine learning and prodigious amounts of data allow us to point our sails in the direction of the most fickle of zephyrs. It’s now tempting to wonder what might happen if the zephyr ever becomes a gale – but, from the perspective of effective risk management, it makes more sense to study prevailing trade winds.

The events of the past few years have been tumultuous, tragic and fast moving. Notably, however, they have not caused an elevated risk to banks. Amounting as it does to a single data point, the evidence on compound risk therefore suggests that banks are safe enough.

The only thing we know for certain is the next time we have a global pandemic, followed by a European war, our data stocks will double.

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.