OECD global long‑run economic scenarios

Annex C. Adding a climate damage channel to the OECD Long-Term Model

C1. Introduction

1. This Annex describes the approach used to add a climate damage channel to the potential output projections in the OECD Long-Term Model (LTM). The approach is grounded in the scientific literature on the likely economic impacts of climate change, relying on consensus estimates and meta-studies wherever possible. Given high uncertainty and evolving research in this field, including at the OECD, the specific parameters used, if not the approach itself, may continue to evolve over time. The chosen approach is designed to be as flexible as feasible, allowing adjustments as research into this issue deepens.

2. The addition of a climate damage channel, together with the extension of the country coverage of the LTM make it possible to produce long-run scenarios that are both more comprehensive, in the sense of covering the near-totality of global output, and more realistic, in the sense of being able to trade off the economic costs of insufficient action to mitigate greenhouse gas (GHG) emissions against those of carbon mitigation itself. The extension of the model horizon from 2060 to 2100 enables fuller account to be taken of the slow dynamics of the carbon cycle and the long-lasting implications of choices regarding the energy transition, and facilitates the use of LTM results as inputs into other models with horizons beyond 2060.

3. The flowchart in Figure A C.1 outlines the approach and the structure of the model and the Annex. Greenhouse gas (GHG) emissions comprise CO₂ emissions from fossil fuel combustion and industrial processes, as described in Annex B of Guillemette and Château (2023[1]), and non-CO₂ emissions (section C2.1), which were not considered in the previous version of the LTM. Aggregate global GHG emissions (section C2.2) then feed into the global carbon cycle and determine the global surface temperature anomaly relative to the pre-industrial era (section C2.3). The global climate damage is a function of this temperature anomaly (section C3.1) and is subsequently allocated to individual countries (section C3.2). As shown in the flowchart, it takes two years (two time steps at the model’s annual frequency) for emissions associated with output in a given year to feed back onto the economy’s productive capacity in the form of climate damages. This is a simplifying assumption with few consequences for the projections given that the atmospheric carbon concentration and downstream variables evolve slowly over time.

Figure A C.1. Schematic view of the approach to climate damages in the LTM

C2. Global carbon cycle and temperature anomaly

4. A global carbon cycle is needed in order to incorporate climate-induced economic damages in the LTM. This first requires completing the coverage of greenhouse gases (GHGs) in the model. Subscripts $i$ and $t$ index countries and time (annual frequency) throughout the Annex.

C2.1. Greenhouse gases other than energy-related carbon emissions

5. In the previous version of the LTM, only CO₂ emissions from fossil fuel combustion and industrial processes (${CO2}_{i,t}$) were projected in the LTM based on projections of output and energy use.1 GHGs other than CO₂ (e.g., methane, nitrous oxide) were ignored. Henceforth, all GHGs are included, except those associated with land use, land use change and forestry (LULUCF). These remain out of the LTM scope for now as they are less straightforward to connect to existing model variables. GHGs other than CO₂ are modelled as a function of the projected evolution of population, living standards and a time trend. They are also assumed to respond to changes in the effective carbon rate with the same sensitivity as energy-related CO₂. The rest of this subsection explains how this approach is implemented. Subscripts $i$ and $t$ (annual frequency) index country and time throughout the Annex.

6. Historical data on total GHG emissions excluding LULUCF, expressed in CO₂-equivalent units (${GHG}_{i,t}$), are sourced from Climate Watch (via Our World in Data).2 Then, a variable capturing all the non-CO₂ gases (${GHG\_EXCO2}_{i,t}$) is calculated as a residual:

${GHG\_EXCO2}_{i,t}={GHG}_{i,t}-{CO2}_{i,t}$

[C1]

7. Next, an equation is estimated relating non-CO₂ emissions per capita to GDP per capita expressed in USD at fixed Purchasing Power Parities (${GDPVD\_CAP}_{i,t}$):

$log\left(\frac{{GHG\_EXCO2}_{i,t}}{{POP}_{i,t}}\right)=\beta \bullet log\left({GDPVD\_CAP}_{i,t}\right)+{\alpha }_i+{\gamma }_t+{\epsilon }_{i,t}$

[C2]

where ${POP}_{i,t}$ is total population, ${\alpha }_i$ and ${\gamma }_t$ are country and time fixed effects and ${\epsilon }_{i,t}$ is an error term. The equation is estimated with panel least squares using data for 139 countries over the period 2000 to 2019, with 2 722 observations. The estimated coefficient $\widehat{\beta }$ indicates an elasticity of non-CO₂ emissions per capita to GDP per capita of 0.31 (p < 0.01). The time fixed effects indicate that emissions intensity across this sample has tended to decline by about 1% per year after accounting for the evolution of GDP per capita.

8. Based on these empirical results, the following projection equation is implemented in the LTM:

$∆log\left({GHG\_EXCO2}_{i,t}\right)=-0.01+∆log\left({POP}_{i,t}\right)+0.31\bullet ∆log\left({GDPVD\_CAP}_{i,t}\right)-0.00369\bullet ∆{ECR}_{i,t}$

[C3]

where ${ECR}_{i,t}$ is the effective carbon rate. In the baseline scenario, ${ECR}_{i,t}$ is assumed to remain constant at its last historical value, implying that non-CO₂ emissions are a function of population, living standards and the estimated global time trend for intensity (-0.01). In the energy transition scenario, ${ECR}_{i,t}$ is calculated as the price which would be necessary to put CO₂ emissions on a path consistent with reaching a net zero target by 2050 (see section 2.6 of Annex B in Guillemette and Château (2023[1])). In the work below, non-CO₂ emissions are assumed to have the same sensitivity to the ${ECR}_{i,t}$ as CO₂, namely a semi-elasticity of -0.00369, as estimated by D’Arcangelo et al. (2022[35]).

9. In the projection period, equation [C1], re-arranged with ${GHG}_{i,t}$ on the left-hand side, is used to add up the projections of CO₂ and non-CO₂ emissions into total GHG emissions in megatonnes of CO₂ equivalent (CO₂e) (${GHG}_{i,t}$).

C2.2. Global aggregation

10. Individual-country GHG emissions are aggregated to the world level ($WLD\_{GHG}_t$) by summing across countries:

$WLD\_{GHG}_t=\sum _i{GHG}_{i,t}$

[C4]

A similar aggregate is calculated for global GHG emissions net of carbon capture and storage ($WLD\_{GHG\_NET}_t$) on the basis of individual country net GHG emissions. As explained in Guillemette and Château (2023[1]), the fraction of global gross emissions captured and stored is assumed to be a logistic function of time, as in Krusell and Smith Jr. (2022[36]), with parameters set exogenously so that it rises slowly at first, then more rapidly and then more slowly again as it nears some upper bound. The current calibration for the baseline scenario assumes that the 1% mark is reached in 2040 and the 50% mark would be reached beyond the projection horizon (in 2120). The energy transition scenario brings forward both waypoints by 15 years.

C2.3. Stock of atmospheric carbon and global temperature anomaly

11. A complete model of the carbon cycle is very complex. The parsimonious approach used here is the five-equation climate-carbon model of the well-known DICE/RICE model (Nordhaus, 1994[37]), which replicates the most advanced Earth system models very well (Folini et al., 2024[38]). The presentation and implementation here draw heavily on Hassler, Krusell and Olovsson (2024[39]):

$∆T_t^S={\sigma }_1\bullet \left(\frac{\eta }{\ln 2}\bullet ln\left(\frac{S_{t-1}}{S_0}\right)-\kappa {\bullet T}_{t-1}^S-{\sigma }_2\bullet \left(T_{t-1}^S-T_{t-1}^O\right)\right)$ $∆T_t^O={\sigma }_3\bullet \left(T_{t-1}^S-T_{t-1}^O\right)$ $∆S_t=-{\sigma }_{12}{\bullet S}_{t-1}+{\sigma }_{21}\bullet S_{t-1}^U+E_{t-1}$ ${∆S}_t^U={\sigma }_{12}{\bullet S}_{t-1}-\left({\sigma }_{21}+{\sigma }_{23}\right)\bullet S_{t-1}^U+{\sigma }_{32}\bullet S_{t-1}^L$ ${∆S}_t^L={\sigma }_{23}\bullet S_{t-1}^U-{\sigma }_{32}\bullet S_{t-1}^L$

[C5] [C6] [C7] [C8] [C9]

12. Equations [C5] to [C9] describe the relation between net GHG emissions in year t-1 ($E_{t-1}$)3 and climate change in year t, as represented by the global mean temperature in the atmosphere at the Earth’s surface ($T_t^S$) and in the oceans ($T_t^O$), both measured as deviations from their respective pre-industrial values (1850-1900).4 This system describes the carbon cycle as having three important characteristics: (i) about half of the emitted CO₂ leaves the atmosphere within a few decades; (ii) between one fifth and one quarter stays for thousands of years; and (iii) the remainder leaves the atmosphere with a half-life of a few hundred years.

13. The right-hand side of [C5] has three terms within the bracket. These terms represent the key changes in energy fluxes (flows per unit of area) to and from the atmosphere that drive climate change. The changes use their pre-industrial levels as baseline and are measured in watts per square meter (W/m²). Their sum is called the atmospheric energy balance. If the balance is positive, heat is accumulated, i.e., the atmospheric temperature increases. The change in temperature per period is proportional to the surplus in the energy budget with a proportionality coefficient ${\sigma }_1$.

• The first term within brackets in [C5] captures the greenhouse effect and contains the ratio of the amount of carbon dioxide in the atmosphere in period t-1 ($S_{t-1}$) and the pre-industrial amount ($S_0$).5 This is called CO₂ forcing in the literature. A good approximation to the strength of CO₂ forcing is that it is proportional to the logarithm of the ratio of the current concentration to its pre-industrial value. The parameter $\eta$ is 3.45, implying that a doubling of the CO₂ concentration leads to an increase in the energy budget of 3.45 W/m², ceteris paribus.

• The second term in [C5] captures the fact that as Earth is warmed up, more energy is radiated out into space. The effect is approximated to be linear in $T_t^S$ with a proportionality constant $\kappa$ of 1.06.

• The third term in [C5] represents the cooling effect that arises if the ocean is cooler than the atmosphere. This term is also approximated to be linear in the temperature difference $\left(T_{t-1}^S-T_{t-1}^O\right)$.

14. Equation [C6] describes the dynamics of the ocean temperature. The only mechanism that changes the ocean temperature is the flow of heat between the atmosphere and the ocean, which is proportional to $\left(T_{t-1}^S-T_{t-1}^O\right)$. It enters as in [C5], but with the opposite sign. Since the heat capacity of the oceans is much larger than that of the atmosphere, ${\sigma }_1> {\sigma }_3$.

15. The system [C5]-[C6] is stable. A doubling of the CO₂ concentration implies a steady state where $T_t^S=T_t^O=\frac{\eta }{\kappa }$, here 3¼ °C given the parametrisation. This ratio is called the equilibrium climate sensitivity (ECS). The latest International Panel on Climate Change (IPCC) report gives a likely range of 2½ °C to 4°C and a very likely range of 2°C to 5°C per doubling of the CO₂ concentration (IPCC, 2023[40]).

16. Equations [C7], [C8] and [C9] provide a description of the carbon cycle (or carbon circulation). This is a simple system of three linear difference equations, each describing the change in the size of three reservoirs of carbon (often called carbon sinks). These are the atmosphere ($S_t$), the upper oceans ($S_t^U$) and the lower oceans ($S_t^L$). It is necessary to have these two additional reservoirs in the model because carbon-cycle dynamics are driven by both a relatively rapid flow between the atmosphere, the biosphere, and the surface oceans and a much slower one involving the deep oceans. The dynamics of the reservoirs are given by a linear system where flows are proportional to source reservoirs and emissions.

C3. Economic damages from climate change

17. The link between climate change and economic performance is multi-faceted and complex. Including it within the LTM, given its characteristics, requires taking a simplified approach. As with the rest of the model, the objective is to illustrate broad trends and attempt to provide some quantifications rooted in the relevant science. This requires a method that can be applied to all the economies in the LTM and which makes use of other variables within the model. Since the scientific understanding of potential economic damages due to climate change is far from settled, any projections of such damages are necessarily uncertain. Modelling approaches are also bound to evolve over time as research continues and understanding of the link between climate change and economic performance improves.

18. The approach used in this paper starts from the global surface temperature anomaly projected with the previous section’s global carbon cycle, calculates the global economic damage associated with it using a global damage curve, and then allocates this global damage to individual countries. While the global damage curve is a function of the global surface temperature anomaly only, the relationship it describes should be thought of as a reflection of many different underlying mechanisms, such as changes in weather patterns (humidity, rainfall, the frequency of extreme events, etc), changes in biodiversity, etc. These various impact channels have been shown in the scientific literature to be closely associated with the global temperature anomaly (IPCC, 2023[8]).

C3.1. Global damage function

19. The LTM’s initial global damage curve is based on the meta-analysis of Howard and Sterner (2017[9]). At the outset, it is important to be clear that the choice of this specific damage function does not represent an assessment that it is the ‘correct’ or ‘best’ one. Given huge uncertainty about the shape of the true function, it is simply a convenient specification for scenario-generation purposes and also a flexible one, as its sole parameter can be adjusted up or down to generate various scenarios. It also represents a rough median within the wide range of functions that have been proposed in the literature, as illustrated below.

20. Global climate damages, measured as the fraction of global output lost in year t relative to a no-warming scenario (${wld\_damage}_t$), is a function of the global surface temperature anomaly in that year ($T_t^S$) from [C5]:

${wld\_damage}_t=0.010038\bullet {\left(T_t^S\right)}^2$

[C10]

The second-degree equation implies that damages rise at an increasing rate as the temperature anomaly increases (Figure A C.2). The parameter 0.010038 is from the preferred specification of Howard and Sterner (2017[9]) accounting for non-catastrophic and catastrophic climate damages. The individual studies underlying their meta-results consider a wide range of potential economic impacts of global warming, such as declining agricultural productivity, disease spread, lower labour productivity due to heat stress, property damage, reconstruction costs associated with extreme weather events, etc. The damage projection from [C10] should therefore be thought of as encompassing that same range of impacts. Conversely, the individual studies underlying the meta-results typically omit the social and political impacts of climate change, including migration and conflict, so the projection from [C10] should also be thought of as omitting these potentially important impact channels.

Figure A C.2. Global climate damage as a function of the global surface temperature anomaly

Per cent of global GDP

Source: Howard, P. and T. Sterner (2017), “Few and not so far between: A meta-analysis of climate damage estimates”, Environmental and Resource Economics, Vol. 68, pp. 197-225.

21. The calibration of the five-equation global carbon cycle in the previous section starts from a surface temperature anomaly of 1.078°C in 2015. Taking account of measured carbon emissions since then, the system [C5]-[C9] projects a surface temperature anomaly of 1.235°C in 2024.6 According to the global damage function in [C10], this temperature anomaly implies that global output is already 1.6% below what it would be in the absence of climate change, more than the share of Spain in global GDP.7 Going further up along the curve, the specification and parametrisation imply a reduction of global GDP of about 4% for an increase of 2°C in global temperature above pre-industrial levels, rising to 9% of global GDP for a 3°C increase.

22. The predicted damage at 3°C of warming represents a middle ground between damage functions based on structural modelling, which typically predict global output losses between zero and 5% at this warming level, and those based on statistical modelling, which predict output losses between 10% and 30% (IPCC, 2023[41]). The predicted damage at 3°C of warming is close to that predicted by Kahn et al. (2021[42]) as well as Waidelich et al. (2024[43]). At 3.7°C of warming, as predicted for 2099 in the Representative Carbon Pathway (RCP) 8.5 (van Vuuren et al., 2011[44]), the climate damage is 14% of global GDP, slightly above the 7-12% loss predicted by Nath, Ramey and Klenow (2024[7]), whose approach combines both statistical and structural elements. This suggests that the simplified global damage curve chosen for the LTM provides a plausible median estimate within the wide spectrum of findings in the scientific literature, as intended (see Figure 11 in the main part of the paper). Moreover, the modelling approach is such that alternative global damage estimates can be readily incorporated to provide alternative illustrative scenarios.

23. Another way to assess the global damage curve and compare to other studies is to calculate the implied social cost of carbon (SCC), which is the present value of future output losses associated with emitting one additional tonne of CO₂ today. This value depends on factors besides the damage curve itself, such as the relative sizes of economies, the projection horizon and the discount rate. Based on the BAU1 scenario described in the main part of the paper, a 2100 horizon and a conventional 2% discount rate, the LTM gives a SCC of USD 253 per tonne of CO₂ (Table A C.1). This is much higher than typical SCC estimates from pre-2020 studies, which tended to be lower than USD 100, but close to more recent estimates based on the same 2% discount rate, including a comprehensive synthesis of hundreds of estimates from various studies, which finds a mean SCC of USD 284 (Moore et al., 2024[4]). Some recent studies suggest the SCC could be much higher, however, with recent estimates exceeding USD 1000 (Bilal and Känzig, 2024[11]). Again, this comparison suggests that the functional forms and parameters chosen for the initial LTM scenario yield climate damages that are well within the range of contemporaneous estimates.

Table A C.1. Global social cost of carbon in the LTM

USD at 2021 prices and Purchasing Power Parity exchange rates

Assumed discount rate	Social cost of carbon
1%	372
2%	253
3%	179
4%	132
5%	102

Note: The estimates are based on the present discounted value of lost global output to the end of the 2100 projection horizon.

24. The single parameter in the global climate damage function [C10] can be adjusted up or down to generate various scenarios with more or less damages. Although the functional form remains the same and cannot perfectly match damage curves with other functional forms, in practice, changing that single parameter is often sufficient to closely mimic the implied global damage at 2°C and 3°C from other damage curves, which is the range most relevant to the scenarios that are normally considered with a 2100 horizon. For instance, replacing the parameter in [C10] by 0.048 very closely approximates the damage predicted by Bilal and Känzig (2024[11]) at these warming levels, which is at the high end of the range in the literature (Figure A C.3).

25. A potential limitation of the approach is that the damage function is assumed to be smooth. While this assumption is conventional in existing climate-economy models and simplifies computation, the true relationship between the climate and the economy is almost certainly not smooth. Instead, the global damage curve likely features climate ‘tipping points’ where the relationship could simply break down, reflecting fundamental qualitative shifts in the climate, such as icecaps melting, the Gulf Stream stopping or El Niño turning from a temporary phenomenon into a permanent one (Lenton et al., 2019[45]). Where along the temperature continuum these discontinuities might occur remains highly uncertain, however.

26. Keeping with the design philosophy of the LTM and computational feasibility, climate tipping points would also have to be included in a reduced-form manner if they are to be considered. The results of Dietz et al. (2021[46]) would be one way of doing so within the current framework. They survey the literature and derive a reduced-form function which expresses the additional climate damage due to tipping points as a function of the mean surface temperature predicted without tipping points. This works out to 0.9% of GDP in additional damage at 2°C of warming and 1.3% at 3°C of warming. Raising the single parameter in the global damage function [C10] (from 0.010038 to 0.0118) is sufficient to fairly closely approximate the additional damage implied by the Dietz et al. (2021[46]) results. Thus, an illustrative tipping points adjustment could be incorporated in future scenarios, with appropriate caveats. A more detailed accounting of tipping points in the LTM would however require significant additional model development and research.

C3.2. Allocating global damages to individual countries

27. A second step is to allocate the projected global damage to individual countries.8 The chosen attribution rule is designed to meet three broad objectives: 1) it should be fairly simple and applicable to all individual countries and regions in the LTM in a consistent manner; 2) individual-country damages should aggregate up to the global damage function in [C10]; and 3) the cross-country distribution of relative damages should be broadly consistent with the scientific literature on country-specific vulnerabilities to global warming.

28. A recent meta-analysis of the economic impacts of climate change finds that much of the cross-country variance in economic damages can effectively be summarised by two dimensions, climate (meaning today’s, without considering future changes) and income (Tol, 2024[47]). All else equal, warm‑climate and low-income countries tend to be more vulnerable to, and less ready for, global warming than cold-climate and high-income countries. A combination of these two dimensions can thus serve as a proxy measure of climate damage sensitivity. Again, this reduced-form approach summarises several underlying structural mechanisms that are correlated with damages in larger climate-economy models. For instance, warm-climate countries tend to have a higher share of their economic activity in sectors, such as agriculture, that are directly exposed to climate change. Similarly, the income variable captures many underlying mechanisms through which high-income countries are better able to adapt to climate change than low-income ones (e.g. availability of technologies, the ability to pay for them, government competence in raising funds and delivering projects, etc).

29. In line with this evidence, an individual country’s climate damage, expressed as a fraction of lost output (${damage}_{i,t}$), is modelled as a function of the global surface temperature anomaly ($T_t^S$), log GDP per capita in 2015 expressed in USD at purchasing power parity (PPP) exchange rates (${ycap}_i$), and population-weighted average annual temperature in 2015 ($avg{temp}_i$), as a proxy for climate, according to the following equation:

${damage}_{i,t}=\left(0.010038+\frac{{\theta }_i}{{2.5}^2}\right)\bullet {\left(T_t^S\right)}^2$ ${\mathrm{w}\mathrm{h}\mathrm{e}\mathrm{r}\mathrm{e} \theta }_i=-0.0182\bullet \left({ycap}_i-wld\_ycap\right)+0.0037\bullet (avg{temp}_i-wld\_avgtemp)$

[C11]

This equation, original to the LTM, combines the global damage curve discussed previously from Howard and Sterner (2017[9]) with the results of Tol (2024[47]). The parameter 0.010038 is from the former (and the same as in [C10]) while the parameters -0.0182 and 0.0037 are from the latter. ${\theta }_i$ is a country-specific and time-invariant index measuring a country’s relative sensitivity to climate damage, with zero corresponding to the same sensitivity as the global outcome (Figure A C.3). The fixed parameters $wld\_ycap$ and $wld\_avgtemp$ are, respectively, log global GDP per capita and global average annual temperature in 2015, both aggregated using 2015 GDP weights in USD at PPPs. They ensure that aggregating ${\theta }_i$ with these same 2015 GDP weights yields zero, and thus that aggregating ${damage}_{i,t}$ along the country dimension yields [C10]. The value ${2.5}^2$ ensures that wealth and climate effects on damages match the values estimated by Tol (2024[47]) for 2.5°C of global warming.

30. Despite its simplicity, the climate damage sensitivity indicator [C11] captures most of the relevant variation across countries in more complex measures of the expected impacts of climate change. For instance, it has a correlation of 0.87 with the latest ND-GAIN country index (Notre Dame Global Adaptation Initiative, 2024[48]).9 This more comprehensive index combines a country’s vulnerability to climate change (based on 36 indicators) with its readiness to confront it (based on 9 indicators) into an overall measure of a country’s vulnerability to climate disruption.

31. The specification in [C11] is convenient for the LTM as it allows the climate damage sensitivity of the global economy to be changed without changing the relative sensitivities of individual countries, as one might desire when interested in global outcomes. It also allows easy integration of a different set of individual-country relative sensitivities (such as the ND-GAIN index mentioned previously) without changing the overall global damage curve. Alternatively, any set of country-level climate damage results for a given level of warming can be transformed into a set of ${\theta }_i$ by expressing country-specific damages as deviations from their global aggregate.

32. Analytical work is ongoing in the OECD Economics Department to produce a finer-grained analysis of a country’s sensitivity to global warming using a richer set of variables and country interlinkages (Box A C.1). The approach chosen for the LTM readily allows the relative sensitivity indices to be aligned at a future date with the results from the broader modelling exercises. Consideration of how these relative sensitivities might evolve over time, in response to changing levels of development and policy actions, could also be given in future model development.

Figure A C.3. Index of relative sensitivity to climate damage (${\mathit{\theta }}_{\mathit{i}}$)

Lower is less sensitive and zero is equal to the global outcome

Note: Values correspond to ${\theta }_i$ in [C11]. OTHERAFRIC, OTHERASIA and OTHERLATIN are regions with together 55 countries that are not modelled individually.

Box A C.1. Measuring the macroeconomic costs of climate change: top-down and bottom-up approaches

Economic assessments of climate change rely on two main primary methodologies: top-down and bottom-up approaches.

Top-down approaches often take a macroeconomic perspective, relying on aggregate data to assess how climate-related damages influence overall economic performance. These models typically begin with global, cross-country, or regional analyses and derive country-specific estimates by applying the fitted values of regressions to local conditions. For instance, Bilal and Känzig (2024[11]) analyse the impact of global temperature on economic output using cross-country data, with the results used to estimate structural damage functions. Future OECD work builds on global estimates from Kotz, Levermann and Wenz (2024[49]) to project future macroeconomic damages across different countries (Murtin, Wenz and Ziegler, forthcoming[50]). Top-down models can capture systemic effects and are more feasible when micro-level data is scarce, but face challenges in fully accounting for country-specific policies and local variations, both of which are likely to enhance the accuracy of climate impact assessments.

Bottom-up approaches analyse climate impacts at the micro level, using detailed data from households, firms, or sectors. These methods attempt to capture the underlying mechanisms through which climate change affects the economy. For example, Hsiang et al. (2017[51]) combine sector-specific empirical analysis with downscaled climate projections to derive economic damages in the United States. Previous OECD work uses a detailed multi-sectoral, multi-regional dynamic general equilibrium modelling framework (the OECD ENV-Linkages model) to assess economic damages from specific climate hazards through particular economic channels (OECD, 2015[52]). These methods offer granular insights and support tailored policy interventions, but can be geographically limited, and may struggle to capture the full effects of macroeconomic, cross-sectoral, and spatial interactions, limiting their applicability.

Recognising the strengths and limitations of these different approaches, ongoing work at the OECD aims to develop a micro-based, bottom-up approach to estimating the macroeconomic costs of climate change: the Climate and Adaptation Spatial General Equilibrium Model (CASGEM – OECD). This quantitative spatial general equilibrium model plans to integrate in-house micro-level estimates of the impact of climate on local economic fundamentals, for example Costa et al. (2024[53]), to assess how macroeconomic impacts of climate change vary across different regions and economic agents. This macroeconomic and spatial approach is intended to help develop economic scenarios that balance local insights with global coherence, bridging micro and macro perspectives, and ensuring more effective policy responses.

33. Setting damages at the country level based on [C11] is not equivalent to allocating fixed shares of the global damage to individual countries. A fixed-shares approach would ignore the effect of changing relative country sizes across different scenarios. This can be understood by considering a hypothetical scenario whereby a small country, say Switzerland, undertakes structural reforms that raise its potential output over some baseline scenario path substantially. Allocating a fixed share of the global damage to it would imply lower damages as a share of its own GDP than in the baseline scenario. This is because Switzerland is small, so the additional GHG emissions associated with higher output make a trivial difference to the global climate damage, which is dwarfed by the increase in its GDP. On the contrary, the individual damage curve [C11] ensures that the projected damage would be the same as in the baseline scenario, save for the trivial feedback coming from the global temperature effect. Switzerland’s share of global damages would be slightly higher than in the baseline scenario, an effect coming from its slightly larger share of global output, with every other country having slightly smaller shares of the global damage via the same relative country-size effect.

C3.3. Implementation in the LTM

While [C11] refers to a loss of potential output, in practice this loss is allocated to the trend labour efficiency component of potential output (see Figure A C.1 above and Box 1 in the main part of the paper). Given the Cobb-Douglas functional form of the production function used in the LTM and the assumed 0.67 labour income share, two-thirds of climate damages thus occur through lower trend labour efficiency. Moreover, as the LTM targets a fixed capital-to-output ratio over the long run, a decline in trend labour efficiency implies a proportional decline in the level of capital over time, all else equal. Thus, the remaining third of climate damages comes through lower capital accumulation. Combining the direct impact on trend labour efficiency and the indirect impact on capital accumulation, the total loss of potential output works out to ${damage}_{i,t}$ given by [C11].

34. By construction, climate damages are assumed to affect only the productivity of labour but not trend employment. In reality, premature mortality due to heat stress or diseases as well as changing international migration may be important sources of damages for some countries. Making population projections endogenous to climate developments within the LTM is a possible avenue for future model development.

C3.4. The impact of changing country weights on the global damage estimate

35. Global damage curves such as the one shown in Figure A C.2 are typically derived from some type of comparative static analysis, whereby country weights in global production remain as they are today and only the global temperature is assumed to change. The reverse experiment can be conducted in the LTM by keeping the global temperature anomaly constant – and therefore individual-country climate damages constant as a share of their own GDP – but calculating the influence of changing country weights in global output on the global damage aggregate. This is done by setting $T_t^S$ equal to 2°C in [C11], letting individual-country GDP weights change over the projection period according to LTM projections and aggregating global damages as usual. Due to heterogeneity in projected growth rates and in climate damages – low-income countries tend to have higher damages and tend to grow faster than rich ones – global climate damages rise over time even with a fixed temperature anomaly. This effect increases climate damages by about 0.18 percentage points of global output per decade (Figure A C.4).

Figure A C.4. Global environmental damages for 2°C of warming due to changing country weights

Per cent of global output lost

Note: The line shows estimated global environmental damages in per cent of global output for a fixed global surface temperature anomaly of 2°C. The aggregation of damages is done using GDP in USD at 2021 purchasing power parity exchange rates, including projections from the LTM’s current baseline scenario. The projected increase in damages through 2100 is due purely to changing country weights over time.

C3.5. Additional illustrative results

36. In the BAU1 baseline scenario presented in the main part of the paper, the assumption is that the energy transition continues along recent historical trends. This business-as-usual path implies that global temperature keeps rising and the surface anomaly $\left(T_t^S\right)$ reaches 2½ °C by the end of the century (Figure A C.5, Panel A). Economic damages in the median country rise from about 2% today to 8.4% of output by 2100 (Figure A C.5, Panel B). The global damage aggregate is slightly lower than that of the median country in the early part of the projection period, reflecting the fact that many of today’s largest economies are relatively wealthy and located in colder climates. The range of damage outcomes widens over time. The least affected country (Iceland) shows little damage, whereas the most affected country (Niger) loses 17% of output by the end of the century. The regional dispersion of damages aligns well with the findings of Nath, Ramey and Klenow (2024[7]), who also project little damage in rich European countries in 2099 in a no-abatement scenario, while the Middle East and Africa are the most affected regions with substantial damages (Figure A C.6).

Figure A C.5. Temperature anomaly and economic damages in baseline scenario

Note: The shaded area in panel B shows the min-max range of damages across all countries in the LTM.

Figure A C.6. Economic damages in 2100 in BAU1 scenario

Per cent of output lost

Note: Greyed out areas are missing from the LTM coverage.

Notes

← 1.  See sections 2.2 and 2.3 of Annex B in Guillemette and Château (2023[1]) for a description of the approach for the original set of countries in the LTM. Energy consumption and CO₂ emissions from energy use for the additional countries/regions now incorporated in the LTM are projected in the same way, that is, by first making assumptions about the evolution of the energy mix consistent with the scenario being considered, then projecting total energy supply and then CO₂ emissions from energy use. While industrial process emissions are not strictly a function of energy use, they are closely associated with it and are projected on this basis for simplicity.

← 2.  GHG emissions other than CO₂ are converted into CO₂-equivalent (CO₂e) using a metric known as the Global Warming Potential (GWP). The GWP is a measure of how much heat a GHG traps in the atmosphere up to a specific time horizon, relative to CO₂. It is calculated over a specific time interval, commonly 20, 100 or 500 years. The data source used here (Climate Watch) uses a 100-year time interval, which seems appropriate given the 75-year projection period of the LTM.

← 3.  $E_t$ is expressed in gigatonnes of carbon (GtC). This variable is obtained by applying a conversion factor to aggregate global net emissions from [C4], which is expressed in megatonnes of CO₂-equivalent.

← 4.  The parameters and initial values are from Hassler, Krusell and Olovsson (2024[39]). The parameters are ${\sigma }_1=0.137$, ${\sigma }_2=0.73$, ${\sigma }_3=0.00689$, $\eta =3.45$, $\kappa =1.06$, ${\sigma }_{12}=0.053$, ${\sigma }_{21}=0.0536$, ${\sigma }_{23}=0.0042$ and ${\sigma }_{32}=0.001422$. The initial values are $T_{2015}^S=1.078$, $T_{2015}^O=0.3132$, $S_{2015}=850$, $S_{2015}^U=765$ and $S_{2015}^L=1799$.

← 5.  For calibration and comparison with other models and projections, it can also be useful to express the atmospheric carbon content in CO₂ parts per million (PPM) as opposed to gigatonnes of carbon (GtC). For this purpose, a conversion factor (2.13 GtC per PPM of CO₂) can be applied to $S_t$. The pre-industrial era CO₂ concentration is conventionally taken to be 280 PPM and the Global Carbon Budget estimates this concentration at 422.5 PPM in 2024.

← 6.  The latest measurements peg the current temperature anomaly at a higher level, for instance 1.46°C above the pre-industrial average (1850-1900) in 2024 according to NOAA National Centers for Environmental Information (2025[71]). However, the temperature concepts used in modelling applications typically refer to smoothed series and it is often unclear what type of smoothing is applied and over what period.

← 7.  This estimate is much higher than economic losses from extreme weather events, estimated at about USD 160 billion annually from 2010 to 2019 and rising quickly (OECD, 2024[66]). There is no inconsistency, however, as the full economic damage from climate change would include many other effects besides extreme weather events, for instance reduced labour productivity due to heat.

← 8.  An alternative approach would be to first downscale the global temperature result to a country-level resolution and make country-level damages a function of country-level temperature. Results from geophysical models could be used to establish the downscaling algorithm. This approach would be more complicated but could potentially better reflect non-linearities between global and local outcomes.

← 9.  Across the overlapping set of 134 countries and based on ND-GAIN scores for the latest available year (2022).