Extreme weather events are bringing devastating consequences for an increasing number of people and countries, so quantifying their economic impact is highly relevant. The effect of changes in climate-related variables on economic output is typically estimated using a ‘damage function’. The methodologies and damage estimates relating to these functions vary a great deal, with each approach having its advantages and limitations. This first in a two-column series explores the concept of damage functions, highlighting the persisting uncertainties and the importance of continued dialogue to refine our understanding of climate-induced economic losses.
Climate change is the foremost global challenge of our time, posing widespread and potentially very severe risks to societies worldwide. In 2023, the global average temperature reached yet another record high, rising to almost 1.5°C above its pre-industrial level (Copernicus 2024), while the summer of 2024 was the world’s hottest on record. This signals how our climate is rapidly evolving, bringing about changes that go well beyond mere increases in temperature. A wide variety of risks can indeed be observed, which may lead to substantial economic losses. Examples of such risks include capital asset damages caused by floods, reduced labour productivity due to elevated temperatures, and crop failures produced by severe drought.
The far-reaching consequences of climate change merit the attention of financial institutions, including central banks and supervisory authorities, which are seeking to learn more about the risks to the financial sector (Landau and Brunnermeier 2020, Löyttyniemi 2021, Hartmann et al. 2022, Hiebert 2024). The Network for Greening the Financial System has been a prominent contributor to the understanding of such risks by developing climate scenarios, which are extensively used in climate-related financial risk assessments.
The corresponding analyses generally require estimates of the economic losses that are incurred when physical climate risks materialise.The Network has thus included loss projections in its scenarios by building on a particular strand of literature aimed at quantifying economic losses (typically expressed as GDP impacts) for a given set of climate change conditions (typically warming levels), widely known as damage functions (Bilal and Rossi-Hansberg 2023). Economist William Nordhaus pioneered the integration of estimates of economic damage caused by climate change into economic modelling three decades ago (Nordhaus 1991, 1994, Gillingham 2018). Since then, many alternative methodologies and estimates have been proposed.
Dissecting climate damage functions
Climate damage functions, in their broadest sense, refer to any type of research that quantifies the impact of climate (or weather) on economic variables. Consequently, some of these studies focus on a single hazard, sector, transmission channel, or region. One subset of this literature examines the global aggregate impact of climate change on economic output. These aggregate estimates are especially relevant for policymakers and for the financial sector, as they can provide insight into future economic developments across various climate pathways.
The differences among damage estimates stem from both assumptions and methodologies. We can distinguish five distinct calibration methods (with an econometric approach being most common over recent years):
- Enumeration approach. This type of study aims to list all possible channels through which climate change damages the economy, quantifying the impacts at various warming levels and summing up the separate estimates to obtain an aggregate. However, these studies are often criticised as being incomplete and omitting significant impacts (Revesz et al. 2014).
- Econometric estimates. This approach – especially the use of panel regressions – has received increasing attention in recent literature. While offering interesting empirical findings, it is often criticised for capturing short-term weather dynamics rather than long-term climate change (Dell et al. 2014, Tol 2024).
- Computable general equilibrium (CGE) models. These models operate in a very similar manner to the enumeration approach. They start by identifying multiple climate shocks, such as crop-yield declines and human health impacts, and use these as inputs for a CGE model. This method allows more complex dynamics and interactions between shocks to be captured. Like the enumeration approach, CGE-based methods suffer from the omission of important climate change impacts (Howard and Sterner 2017).
- Expert surveys. In this approach, damage estimates are constructed based on the expectations of many experts (usually climate scientists and economists). While they combine multiple perspectives, the subjectivity of this approach is a concern (Oppenheimer et al. 2016).
- Meta-studies. These studies combine estimates from previous studies to produce a central estimate across literature and methodologies. However, they integrate the latest insights with a delay, which makes them susceptible to being built on obsolete findings in a rapidly evolving field.
What we don’t yet know about: linearity, growth or level effects, and obtaining an exhaustive measure of climate risk and its economic impact
In addition to the variety of calibration methods, researchers have divided opinions about assumptions and findings.
Linearity
First, there is uncertainty surrounding the (non-)linear form of damage functions. How climate damage will evolve at increasing warming levels is not well known due to the limited availability of historical data. For instance, a linear relationship could be assumed to exist between global warming and output losses. Based on this assumption, damages incurred by going from 1°C to 2°C of global warming are equal to those incurred by going from 2°C to 3°C. However, many researchers argue that the impacts of climate change are likely to have non-linear dynamics, such as tipping points (e.g. the melting of the Greenland and West Antarctic ice sheets) after certain warming thresholds have been crossed, which could result in catastrophic, irreversible damages. This would imply that losses incurred at high warming levels are much higher than those currently being observed.
In early studies, such as those presented by Nordhaus, the relationship was assumed to be quadratic, leading to relatively modest damages at very high warming levels (e.g. the damage function estimates made by Nordhaus [2017] project losses of around 8.5% at a 6°C warming level). Over time, researchers have proposed the use of higher-order polynomials, or even exponential relationships (Weitzman 2012), which generate similar results at current warming levels, but much higher – or even catastrophic – damages at exceptionally high warming levels (Figure 1).
Figure 1 Loss projections across various damage function forms
Notes: Illustration of four fictional damage curves displaying how the form of a damage function affects loss projections. Since higher warming levels have not yet been observed, the corresponding loss trajectories are assumption-based.
Source: Authors’ elaboration.
Growth or level effects
Second, there is ongoing debate around whether climate change shocks have level or growth effects on economic output (Newell et al. 2021). If pure growth effects are assumed, a permanent change in climate causes a perpetual alteration in the growth rate of an economy. In the case of pure level effects, a climate shock only results in a one-off decrease in economic output. Afterwards, the economy reverts to its pre-shock growth rate. The chosen assumption greatly affects loss projections over long time horizons (Figure 2). 5 For instance, Burke et al. (2015), who found growth effects, project much higher damages than those foreseen by assuming only instantaneous level effects, such as Kalkuhl and Wenz (2020).
Recent studies have tried to identify a middle ground between both approaches, for instance, by assuming that losses affect output – rather than growth – over multiple years, but not in perpetuity. 6 For example, Kotz et al. (2024) extended on the premise of level effects by incorporating lagged effects on top of the instantaneous ones. As such, they could account for losses up to ten years after the initial shock but without assuming a permanent change in the growth rate. A similar conclusion was obtained by Kahn et al. (2021) using a different approach.
Figure 2 Level vs growth effects following a one-off permanent climate shock
Notes: The illustration shows how losses due to a permanent change in climate evolve differently when assuming level or growth effects of a climate shock on economic output. Assuming level effects, GDP reverts to its pre-shock growth rate after the shock. Assuming growth effects, the growth rate is permanently altered.
Source: Authors’ elaboration.
Obtaining an exhaustive measure of climate risk and its economic impact
Third, there are diverging approaches with respect to the selection of climate-related variables to measure climate risk. Climate change is most prominently associated with increasing temperatures, which are used by most damage functions to calibrate losses. However, relying solely on warming levels could result in incomplete damage estimates. Therefore, some studies seek to incorporate other climatic variables, such as precipitation changes or temperature variability (Kotz et al. 2024).
The geospatial specification of climatic variables is another aspect on which there is no consensus in the literature. While many damage functions rely on local temperature shocks, there is some evidence that this approach may lead to an underestimation of losses. For instance, Bilal and Känzig (2024) argue in favour of using global temperature shocks, which are said to more comprehensively represent climate change risks due to a stronger correlation with extreme weather events such as droughts and cyclones.
Moreover, there are some climate-related risks that are excluded from most damage functions, such as climate-induced socioeconomic risks (e.g. migration, armed conflict, mortality). 7 Recent advancements in dynamic spatial integrated assessment models (S-IAMs) allow accounting for spatial frictions like migration barriers, costs related to trade, or physical infrastructure to obtain a more accurate evaluation of the economic impacts of climate change. For more, see Desmet and Rossi-Hansberg (2024). Lastly, most damage functions do not explicitly account for adaptation, which could result in a smoother impact on economic output (see e.g. Barreca et al. 2016 for a discussion on adaptation effects).
Conclusion
Quantifying the potential economic losses caused by unabated climate change is crucial for effective planning and decision-making. It is also highly relevant for financial stability assessments. Economists and climate scientists have made great advances in recent years to improve their methods for estimating these losses. While an unbiased view of the economic impact of climate change is urgently needed, results vary greatly depending on the methodology and assumptions applied. An open dialogue remains essential to improving our understanding of climate change and the damages it may cause.
The Network for Greening the Financial System is an active supporter of these discussions. A follow-up column will take stock of this literature in terms of quantitative impacts and discuss the integration of damage functions into Network climate scenarios.
Source : VOXeu
