Climate Change Scenarios and Risk Management

SMITH BRAIN TRUST – They say that all models are wrong, but some are useful. And that’s true in the finance sector’s climate change scenario analysis, according to Maryland Smith’s Clifford Rossi.

In a recent interview, Rossi discussed how institutions can prepare meaningful climate change scenarios for decision making and risk management. The interview was conducted ahead of the latest webinar in the Center for Financial Policy’s Risk Leadership series, an event co-sponsored by the Center for Global Sustainability and SEBA International. In its ongoing series, CFP has been tackling some of the financial sector’s biggest environmental, social and governance (ESG) challenges.

Rossi is an Executive-in-Residence and Professor of the Practice at the University of Maryland’s Robert H. Smith School of Business. Before joining academia, he spent 25-plus years in the financial sector, as both a senior executive at top financial institutions and a federal-banking regulator. Just prior to his academic work, he was Chief Risk Officer for Citigroup’s Consumer Lending Group.

Here is an excerpt of that interview.

Question: Would you tell us a little about how scenario analysis and stress testing is applied today in the banking and insurance industry?

Answer: I like to think of stress testing as putting a bank or bank holding company (BHC) on a treadmill and monitoring its heart rate on an EKG. The treadmill starts out at a moderate pace (call that the baseline scenario if we are talking about the Fed’s annual stress test), then has the bank speeding up the pace under an adverse scenario, to finally a flat-out sprint for a short time (nine quarters), which is called the severely adverse scenario.

Banks have been conducting scenario and stress testing analyses for years and come in two basic forms:

1. Institution driven scenarios selected by the firms themselves – e.g., liquidity contingency planning scenarios:

2. And those that are regulatory-required (e.g., the Fed’s CCAR for large BHC’s or the OCC’s DFAST for large national banks). Back during the crisis, for example I led Citi’s SCAP stress test (precursor to what eventually became CCAR) efforts for their consumer portfolio.

We use scenario analysis in a wide range of risk applications from credit, operational (Quantum Dawn for cyber), liquidity and market risks.

In addition to these types of what-if analyses companies conduct category risk assessments such as the ones I did nearly 20 years ago with companies to assess the loss exposure of various types of collateral such as residential or commercial real estate.

Question: At a high level, can you expand on the potential pitfalls of scenario analysis and stress testing. For example, where can risk models go wrong and lead to suboptimal decisions?

Answer: First, the old adage “all models are wrong, but some are useful” really rings true especially for our climate change scenario analysis. Scenario and stress testing models, particularly for large institutions are complex systems of statistically based equations and financial relationships that are highly dependent on the data, assumptions, and model specification applied. This is why the largest banks have large model risk management organizations to validate these results. Furthermore, the accuracy of the scenarios tends to decline the further out in time you project. Today, for example, CCAR analyses run out nine quarters. Climate scenarios (SSPs) can go out years if not decades depending on what test you need to run. Imagine the model risk that exists with climate models.

As to underlying data requirements; there are two basic kinds:

1. Physical climate data such as for flood/sea level rise; hurricanes, wildfires.

2. Financial and operational data on facilities and properties underlying loan portfolios for example.

The data produced by the climate models generates a number of physical outputs such as CO2 emissions and temperature change, among others. This is perhaps one of the most challenging parts of integrating climate scenarios in bank stress tests today.

The trick is integrating the former with the latter data and trying to make sense of it.

Data and tools exist for the physical climate data though companies need to take care and ask questions of their vendors regarding the resolution of their underlying models which for the most part are simply taking existing publicly available models and downscaling them.

Question: As someone who led teams implementing large bank stress tests could you comment on what exactly these tests entail, that is some of the mechanics, data and models and what’s involved to perform them?

Answer: A scenario or stress test analysis. Let’s say an effort by a large systemically important financial institution (SIFI) such as for DFAST is an extraordinarily labor-intensive effort that spans across the organization, with the Finance and Risk teams along with Business groups engaged in pulling together the data and models to be used.

The CCAR stress tests – i.e., the quantitative portion – are designed to test the bank’s ability to withstand various severe economic events such as a Great Financial Crisis-like scenario in the form of calculating changes in capital.

A regulator provides the firm with a set of updated macroeconomic assumptions over the stress period that the bank integrates into its financial and risk models – e.g., what is the statistical effect of a 30% peak to trough decline in home prices on mortgage default and prepayment. Then those relationships flow through the financials such as pre-provision net revenue and eventually to equity capital.

Preparing for a stress test requires the dedication of significant resources by the bank – the infrastructure required to run these tests is daunting and time consuming. Imagine for a $2 trillion bank like Citigroup, running these scenarios through the various asset portfolios and you get an idea of the complexity and breadth of scope of these tests.

Question: As a follow up, Cliff, what issues do you see in the process for banks to implement these climate change stress tests?

Answer: Banks are facing a number of challenges in standing up their climate change programs. U.S. firms seem to be a little behind their European and Asian counterparts at this point but they will need to redouble their efforts. One of the biggest challenges is solving the square-peg-round-hole dilemma. That is how they can build the systems and models and databases needed to run these climate scenarios.

Where do you even start? The Task Force on Climate-related Financial Disclosures (TCFD) is already starting to push banks in this direction. My former employer Citigroup I think has a great game plan for developing its climate change risk management and scenario analysis framework and would be well worth looking at as a model. I would focus exclusively on building the data and models to conduct physical risk assessments perhaps with emphasis on my credit/counterparty and operational risks.

With that in mind I would look to prioritize my exposures and work on analyzing the biggest ones first. I would want to set up a governance structure that aligns with a company’s existing ERM framework, build in climate change and ESG as part of the risk taxonomy or catalog; establish policies and procedures around these processes. What banks will be going through isn’t all that different from when they had to build cyber risk programs.

I would then turn to cataloguing the intersections of climate change on other risk types; credit, market, liquidity, operational.

Once that’s in place, I would identify partners that could help size up my potential exposures; (e.g., consumer and commercial loans), but also investment portfolios, facilities, etc., and importantly provide climate data and tools that could be ingested by the risk and financial models.

Then start to work on developing models that build various climate change effects into them – e.g., for mortgages estimated the effects of hurricane intensity and frequency on default – similar approaches could be used for other asset types and risks for example. I’ve demonstrated the feasibility of such an approach to that in a research paper published by GARP last summer.

Question: Cliff, can you comment on what’s needed to improve our understanding of the linkages between climate change and economic and financial performance?

Answer: If you look at the three Networking for Greening Financial Systems (NGFS) scenarios – Orderly, Disorderly and Hothouse – we are provided various assumptions on emissions, temperature, energy mix and economic damages.

When we build our risk and financial models, we do so by establishing statistically reliable relationships between a risk outcome and a risk factor.

I would ask the audience, how exactly would I come up with solid data to inform my models of how default would be affected by a change in carbon price when we have virtually no data to derive that relationship?

Moreover, how do I assess risk out 10, 20 or 30 years in some cases on a portfolio when the scenarios imply massive structural changes in society and economy that will undoubtedly affect the composition of balance sheets from what they look like today?

The linkages between climate and socioeconomic outcomes have been established by way of the Integrated Assessment Models (IAMs) that are in use today. But understanding the way those outcomes link back to risk and financial performance outcomes is an area ripe for analysis. Sure, we have established linkages between various macro effects and loss for example, but I would argue that there’s a structural change assumption underlying those relationships that would require not a simple re-estimation of a loss model using past historical data that do not reflect the kind of climate scenarios we are testing. In other words the loss distributions for our standard credit, market and operational risks are likely to have a different shape than the ones used today by banks due to the effects of climate change affecting the underlying structure of macroeconomic to loss outcomes.

Question: Before we close out our Q&A session and take questions from the audience, what are your closing thoughts about the precision and challenges of climate change stress tests, and their current relevance to the financial services sector?

Answer: In summary I think the challenges for climate change scenario analysis come down to the following issues:

1. Climate and IAM models that have a good deal of uncertainty associated with them. This in my mind implies that we should be looking for more tractable solutions to climate scenario and stress testing than to start trying to boil the ocean by developing very intricate models that produce outputs with significant uncertainty around them. The analogy is we would build a Porsche Carrera on top of a Chevy Spark – looks great, feels great but performs terribly.

2. Concern that in our quest to do something about climate change that we simply rush in and require the investment of huge amounts of time and resources to conduct climate scenarios that simply aren’t ready for prime time.

3. We need to prioritize our risk exposures and start with some top-down modeling adaptations to inform our climate scenario analysis focusing on physical climate change effects first – in other words learn how to walk before we can run.

In the end I fear that applying these scenarios as given may in the end amplify financial risks in the future by decisions made today that are informed by these weak models. I guess what concerns me the most is that we are moving from models that were designed and built in a pure research and academic environment to implementation in business applications that have hard dollar consequences.