Sandy Trust, Sanjay Joshi, Tim Lenton, Jack Oliver, Institute and Faculty of Actuaries
“A development of realistic qualitative and quantitative climate scenarios is required, along with model development to better capture risk drivers, uncertainties and impacts”
Climate-change scenario modelling is now increasingly mainstream in financial services. Financial regulators across the world are mandating that regulated entities carry out climate-scenario modelling and produce TCFD[i] disclosures.
However, using actuarial principles to examine current climate-change scenario analysis limitations and assumptions yields a number of worrying observations, including:
- Many current climate-change scenario models are understating risk, showing benign, or even positive, economic impacts from a hot-house world in which we fail to limit global warming
- There is considerable uncertainty around how much and how quickly we expect the climate to warm – we may have seriously underestimated the pace of climate change
- This translates to uncertainty in carbon budgets, with a real chance that the carbon budgets for 1.5°C of warming could already be zero
- This points to the importance of users of scenario analysis in financial services understanding these limitations and assumptions, and the responsibility to create more realistic models and craft detailed qualitative scenarios
- This underlines the need to race to net zero – with an intentional focus on accelerating a range of positive tipping points in socio-economic systems that we can control.
Many climate-scenario models are severely under-estimating the economic impact of climate change.
Sample TCFD results from UK investors show implausibly benign results for portfolio returns in a hot-house scenario of 3˚°C or more of warming. Some are only slightly less economically damaging than a disorderly or orderly transition. In fact, one institution showed a hot-house scenario to be economically positive compared to transitioning – a result that is patently incorrect and at odds with climate science.
Real-world impacts of climate change, such as the impact of tipping points, sea-level rise and involuntary mass migration do not exist in the models. There are early indicators that we are now approaching some of these tipping points, such as loss of the West Antarctic ice sheet and the Amazon rainforest. They are no longer high-impact, low-likelihood events but increasingly likely as temperatures go past 1.5°C.
Modelled scenarios do not incorporate our experience on climate change. There is considerable uncertainty in key climate-system modelling assumptions, yet no margins are currently included to represent this uncertainty. There is limited consideration of higher warming scenarios.
Damage functions that are used to estimate the economic impacts of climate change exclude many of the risks we expect to fact. Models that are widely used to estimate economic impacts contain a number of simplifying assumptions that do not hold in the real world.
Regulatory scenarios introduce consistency but also the risk of group think, with scenario analysis outcomes being taken too literally and out of context. Investors and regulators assessing financial resilience need to be particularly careful not to place undue reliance on artificially benign model results.
Modelling climate change
Modelling climate change and society’s reaction to it is hugely complex, requiring us to make assumptions about many unknown factors for each scenario we wish to model, including:
- The level of future emissions in each scenario
- How quickly the climate will warm for a given level of emissions
- Whether we cross climate or ecosystem tipping points
- The level of damages we will experience as the climate warms, mitigated by adaptation
- How quickly we will transition as we react to the physical changes we experience
- The pace and scale of the transition in different geographies, economies and sectors
- How to incorporate factors such as land use, technological change and nature
How much will the planet warm?
Climate change is happening more quickly than anticipated, with severe impacts already being felt by millions globally. The Intergovernmental Panel on Climate Change (IPCC) has revised downwards the temperature level at which high risks occur.
We are now at a point where the level of greenhouse gases in the atmosphere is double the pre-industrial level.[ii] This level of emissions is in line with the high emissions scenario RCP8.5, which the IPCC estimates would lead to over 2°C of warming by 2050.
The Earth’s climate takes time to adjust to changes in CO2 concentration. If we stopped all emissions now, what temperature would global warming reach? The IPPC estimates between 2.5°C and 4°C (with a best estimate of 3.0°C)[iii], but there is wide range of uncertainty with an 18% probability that it could be greater than 4.5°C.[iv]
Some scientists assess that Earth-system sensitivity may be double this IPPC range, after allowing for the full impact of reducing ice sheets.
How quickly will it warm?
Climate response time is also uncertain and hard to estimate. Carbon budgets may be smaller than anticipated and risks may develop more quickly. The IPCC indicates that around 320 billion tonnes (Gt) of CO2 can be emitted from the beginning of 2022 to have a 67% chance of staying below 1.5°C. With current emissions running at c.40 billion Gt per annum, this gives us eight years of budget left before we exceed the budget for keeping below 1.5°C of warming.
There is also a one-in-three chance of failure. This means there is a possibility that we have already used up the carbon budget for limiting warming to 1.5°C.
What level of damages?
The loss functions which are commonly used in climate scenarios are based on past data and exclude many of the risks we expect to face. Choice of loss function has a very material impact on results – varying from 10%[v] to 18%[vi] to 63%[vii] [CW1] loss in global GDP by 2100. The Network for Greening the Financial System (NGFS) estimate of 20% (including acute physical losses) should be viewed as a conservative lower limit for the GDP losses we should expect if we do not change course.
Moving forward
A development of realistic qualitative and quantitative climate scenarios is required, along with model development to better capture risk drivers, uncertainties and impacts.
Qualitative scenarios
Visualisations such as flood maps bring home the physical, land use and population movement impacts that climate change may cause. Narratives help decision-makers, companies and investors understand the potential impacts of a hot-house world. For instance, insurance leaders have unequivocally stated that, if climate change raises average temperatures to 4°C above pre-industrial levels, most assets will be uninsurable.
Reverse stress testing
A simple quantitative approach could be to use a reverse stress-testing approach based on a ruin scenario of 100% loss of GDP at a certain temperature limit. With a ruin parameterisation of 6°C, around 30% of GDP could be lost at 3°C of warming, compared with an 80% GDP loss using the 4°C ruin parameterisation. Taking this approach would drive more realistic TCFD results.
Macro-economic model choice
Public reference scenarios, including the NGFS, rely on models referred to as computable general equilibrium models (CGE). CGE models were created by the climate-science community to inform high-level public policy making. Traditionally, they have been used to assess the socio-economic impacts of various climate pathways. The macroeconomic modules of these models had a very different use case from how the financial sector is currently applying them. They have some simplifying neoclassical economics assumptions which generate outputs that do not adequately capture real-world economic dynamics.
Non-equilibrium models, such as the post-Keynesian E3ME model maintained by Cambridge Econometrics, still have limitations but are designed to simulate real-world economic dynamics more accurately. For example, actors are not assumed to be all knowing, perfectly efficient entities but derive behavioural parameters from historical relationships. Also, money can be created by banks through new loans and this investment is not crowded out.[viii]
Climate is a difficult game of global averages and local extremes. We can’t model everything precisely and time is too short to wait for models that are perfect. We need to start by understanding the significant uncertainty in physical climate modelling and avoid that danger that artificially benign results delay vital action to manage climate risks.
[i] Task Force on Climate-related Financial Disclosures
[ii] Hansen et al ‘Global Warming in the Pipeline’, 2022 https://doi.org/10.48550/arXiv.2212.04474 (Noting paper is not yet peer reviewed)
[iii] The Intergovernmental Panel on Climate Change (IPCC) AR6 Synthesis Report, 2023, p.68
[iv] Kemp et al ‘Climate Endgame: Exploring catastrophic climate change scenarios’, 2022 https://www.pnas.org/doi/10.1073/pnas.2108146119
[v] Piontek et al ‘Integrated perspective on translating biophysical to economic impacts of climate change’, 2021 https://doi.org/10.1038/s41558-021-01065-y (Kalkuhl & Wenz, 2020)
[vi] NGFS ngfs_climate_scenarios_for_central_banks_and_supervisors_.pdf.pdf, 2022
[vii] Piontek et al, Integrated perspective on translating biophysical to economic impacts of climate change, 2021 https://doi.org/10.1038/s41558-021-01065-y (Burke & Tanutama, 2015)
[viii] Bowdrey & Hidi Under the bonnet: different economic engines that drive climate change scenario models, The Actuary, 2022
[CW1]I was not entirely sure from the paper whether the 10% was from the Burke & Tanutama 2015 or 2019 research paper
