With COVID-19 vaccines now available, attention will soon revert to some of the other existential risks facing humankind. Top of the 2021 list for bank risk modelers, in the absence of new black swans, will undoubtedly be climate change. A number of key regulators have scheduled pilot stress testing projects in the coming year and the Bank of England has taken the bolder step of initiating a fully fledged regulatory stress test, known as the Climate Biennial Exploratory Scenario (CBES).
For large British banks, the 2021 CBES will focus on assessing the effect of global warming on the credit risk of major non-financial counterparties. Banks will be required to model the impact of three distinct scenarios on relevant credit risk metrics for major corporate exposures. The scenarios, due to be published in June, will cover forward looking assumptions about the path of global temperatures and the nature of future temperature-lowering policy initiatives. Because of the long-run nature of climate trends, the exercise will consider a 30-year prediction horizon.
The key question we address here is whether existing empirical strategies for assessing climate risk are mature enough to yield a meaningful stress test. In my view, while pilot projects may help us to formulate the right approach, a lot more research is needed before a truly compelling climate stress test will actually be possible.
When the first standard stress tests were proposed and implemented, the industry was acutely aware of the dangers of cyclical recessions for banks. Industry professionals and academics had amassed decades of experience modelling the effects of business cycles on credit performance. It was understood that recessions caused by an erosion of credit quality – like the 2008/09 event – were far more dangerous than those with causes external to financial markets. Regulator-led stress tests based on cyclical scenarios were a natural extension of these accumulated research efforts.
The Globe is Warming, What About Credit Risk?
Global warming, meanwhile, has been happening for at least the past half century. The structure of the CBES implicitly assumes that climate-related credit risks have been building over this period, and that banks have gradually adjusted their business models to cope with an increasingly climate affected economy. Note that while global temperatures exhibit clear long-term trends, aggregate credit default rates do not, implying that the opposing forces here are very closely balanced
A valid empirical strategy would be to isolate and quantify this historical buildup of climate-related credit risk and then use the resultant models to extrapolate over forward looking scenarios. These models will be exceptionally difficult – but not impossible – to statistically identify.
Just disentangling the effects of climate on a legacy book from the dynamic effects of a portfolio in transition will require impressive data analysis skills. As climate makes some exposures riskier, funding costs will rise for these loans, making the underlying investments less attractive for banking clients. Climate sensitive loans will then, over time, become less prominent on bank balance sheets. This endogenous response suggests that structural modeling techniques will be needed to effectively isolate legacy climate factors.
Assuming that this research goes well and a trending factor can be reliably estimated, considerable doubt will remain about whether global temperature increases are actually driving the observed result. Over the past 30 years, for most developed countries like the UK, the population has aged, interest rates have declined and labor productivity has stagnated. Any or all of these variables could be the real drivers of the identified credit risk trend, fatally undermining the veracity of analysis based on global temperature projections.
Identifying the Relationship
How can we build confidence that we are actually measuring the impact of climate and not one or more of these other trending factors?
If credit risk was an experimental science – and if we had plenty of time – we would simply define control and treatment groups, introduce appropriate climate shocks and then test their impact on default likelihood. In the real world, we need to find historical examples that mimic this experimental structure – so-called ‘natural’ experiments. At least in terms of physical climate risk, the most obvious natural experiments occur when unusual climate-related disasters strike particular regions, leaving adjacent areas unharmed.
For the U.S., major hurricanes provide the cleanest examples of this – the UK, by way of contrast, has a famously moderate and uniform climate. There has been a fair amount of academic and industry research looking at the effects of Hurricanes Katrina and Sandy on the performance of various financial instruments, including credit risk metrics. The results, though, have often been counterintuitive. Sometimes climate related shocks force or encourage lifestyle changes that actually boost the long term solvency of households and businesses.
Because the data science behind climate credit risk is difficult, and because a core research base has not been fleshed out, I fear that the CBES will yield overly simplistic models that lack empirical support.
We can all postulate theories about how climate may impact the performance of particular entities. For example, it seems obvious on face value that oil exploration companies will exhibit increased credit risk over the next decade as society increasingly moves toward renewable energy sources. Companies operating in declining industries can, however, remain solvent and profitable until their last day of existence, as long as expenses are forced to decline at least as rapidly as revenues and provided debt loads are carefully controlled as markets shrink.
The moral of the story is that face value speculation is never a substitute for rigorous empirical analysis.
Finding compelling empirical support for a link between credit quality and climate is a problem for the ages. A number of regulators have taken the plunge into this new field and many industry analysts are confused about how to best model the impact of climate on a range of bank exposures. Bank modellers are well used to dealing with cycles, but climate is a long run trend that is much harder to model.
This year’s work should advance our understanding of the interplay between credit and climate. It will, though, take many more years of research before we truly know how dangerous climate risks actually are for banks.