Dr Yue (Nina) Chen
Dr Yue (Nina) Chen was chief climate risk officer at US federal banking regulator the Office of the Comptroller of the Currency until March 2025 and is currently a senior visiting fellow at the Centre for Economic Transition Expertise at LSE
We all know the fable of the blind men and an elephant, and probably see it as a charming parable. Yet do bankers and investors truly grasp the limitations of a critical tool for assessing climate-related financial risks – climate scenario analysis (CSA) – as it is commonly practised?
Too often, CSA is treated as a compliance formality. I frequently hear from institutions that their CSA showed only mild losses or negligible impact on strategic asset allocation. The implicit – and sometimes explicit – conclusion? Their portfolios must face limited climate risk, so no further action is necessary.
That is a dangerous misread.
The problem with off-the-shelf models
For transition risk, most institutions use the scenarios produced by the Network for Greening the Financial System (NGFS) because they are “what regulators ask us to use” or “the norm”.
There is no question that NGFS has elevated the quality and visibility of climate risk assessments. Yet the integrated assessment models (IAM) underpinning its scenarios have known limitations, acknowledged in academic literature and NGFS documentation.
NGFS notes that its outputs may underestimate economic and financial system impacts (albeit in an easily overlooked disclaimer) and highlights two gaps: the exclusion of tail risks and tipping points. People tend to view “tail risks” and “tipping points” as highly unlikely and decades out (eg a climate tipping point like Amazon forest dieback), hence immaterial for risk assessments whose timeframe is often within several years.
However, even NGFS’s near-term outputs can underrepresent reality given structural model limitations. Process-based IAMs and other energy-environment-economy models currently used by NGFS use a general equilibrium framework and have difficulty in fully capturing nonlinear changes like rapid, S-curve technology adoption.
NGFS also uses a carbon price to represent all transition risk drivers, ignoring technological disruptions already enabled by past R&D and policy.
Take battery cost: NGFS’s short-term highly ambitious transition risk scenario – Highway to Paris – projects a 40 percent drop from 2024 to 2030. But reality is moving faster, with a 30 percent drop between 2022 and 2024, and another 15 percent drop by early 2025.
Cheaper batteries accelerate wider EV adoption, reshaping not only transportation but also the global value chain of oil with disparate geographical impacts. Without further policy support, battery costs (also solar and wind) are expected to drop further.
When NGFS’s most aggressive transition scenario underestimates historical trends, how can institutions rely on it to prepare for a more disruptive future?
Misjudging physical risks
For physical risk, institutions often model only the direct impact of one or several natural disasters on residential mortgages. This is a reasonable start, but far from sufficient.
If climate change only caused several more disasters, we wouldn’t need to worry.
Climate change is dangerous because it is systemic, global and persistent. Limiting analysis to property damages overlooks critical indirect impacts – housing price declines, employment shocks, weakened tax bases, and interactions between climate and non-climate risk such as an economic downturn (ie compound risk).
Past US experience – where rebuilding enabled by federal disaster aid cushioned financial impacts – no longer guarantees the same outcome in today’s political environment. Without rebuilding, property values may fall and local economies may falter, affecting credit cards, commercial real estate and small business lending.
Commercial mortgages and commercial and industrial (C&I) loans are similarly exposed. A property may remain intact, but if surrounding infrastructure fails, businesses without sufficient business interruption (BI) insurance may not operate. Only about 40 percent of small businesses carry BI insurance today.
Other often overlooked transmission channels with present-day effects include:
Supply chain disruptions from extreme weather;
Reduced labour productivity and power generation from heat waves;
Lower demand due to weaker growth and economic performance;
Deteriorating insurance availability and affordability, ultimately affecting banks’ mortgage underwriting in the absence of a robust public backstop.
Imperfect tools, false precision
Banks often use regulatory stress testing apparatus for CSA, which models short-term, one-time macrofinancial shocks. Climate change produces compounding, persistent impacts with strong regional and sectoral disparities. These impacts are missed when institutions implement CSA using the same tools and assumptions without careful adjustment.
Granted, some institutions recognise the limitations. But once a result is produced – a tidy number, a graph – our cognitive bias towards numerical certainty overrides our knowledge of model limitations, leading us to complacency.
We need a warning label on such CSA outputs: “Partial estimate. Likely underestimates. Use for education, not risk quantification.”
A shift in the wind
Two cross-Atlantic forces may change CSA practices. In the US, with federal regulators rolling back climate risk supervision, climate risk teams need to prove their value by showing CSA’s business relevance.
In Europe, consultations by the European Banking Authority and the Bank of England call on firms to justify scenario selection and acknowledge limitations and uncertainties.
These developments will rightly raise the bar.
Better practices are emerging
Some institutions are tailoring scenarios to portfolio composition and business models, analysing both direct and indirect impacts of physical risks, and adopting nonlinear models better suited to capture technology S-curves. Some differentiate scenarios by use cases – distinguishing those for risk management versus those for risk appetite setting.
Thoughtful qualitative approaches are also useful. Customised table top exercises with cross-functional stakeholder inputs – like those for operational risk – can identify vulnerability, interconnections and blind spots. For example, an investor with concentration in a certain region can think through how their municipal bonds and small-cap equity portfolio would fare amid a regional recession and severe natural disasters.
These improvements take resources. But one simple shift is free: change your interpretation of the outputs from a limited CSA. Treat it not as a full risk assessment, but as a learning tool. The outputs should be viewed as likely lower bounds with high uncertainty, not precise forecasts. The most valuable insight might not be the number itself, but the questions that follow.
The upside of seeing clearly
Done right, CSA is not just a risk management tool but a strategic advantage. If the market is full of blind men misreading the elephant, those with clearer sight can seize opportunities – selecting assets better, anticipating disruptions and mitigating risks.
Don’t touch the elephant’s leg and conclude that it is a tree. That leg can trample you. A smart general would tame the elephant and ride it to win wars.
The author would like to thank Paul Champey, Andrew Eli, Matt Goldklang, Elizabeth Jacobs, Jean-Francois Mercure, Willemijn Verdegaal and others for their inputs on the piece.