The Sectoral and Regional Economic Consequences of Climate Change to 2060

Abstract

This paper presents a new detailed global quantitative assessment of the economic consequences of climate change (i.e. climate damages) to 2060. The analysis is based on an assessment of a wide range of impacts: changes in crop yields, loss of land and capital due to sea level rise, changes in fisheries catches, capital damages from hurricanes, labour productivity changes and changes in health care expenditures from diseases and heat stress, changes in tourism flows, and changes in energy demand for cooling and heating. A multi-region, multi-sector dynamic computable general equilibrium model is used to link different impacts until 2060 directly to specific drivers of economic growth, including labour productivity, capital stocks and land supply, as well as assess the indirect effects these impacts have on the rest of the economy, and on the economies of other countries. It uses a novel production function approach to identify which aspects of economic activity are directly affected by climate change. The model results show that damages are projected to rise twice as fast as global economic activity; global annual Gross Domestic Product losses are projected to be 1.0–3.3% by 2060. Of the impacts that are modelled, impacts on labour productivity and agriculture are projected to have the largest negative economic consequences. Damages from sea level rise grow most rapidly after the middle of the century. Damages to energy and tourism are very small from a global perspective, as benefits in some regions balance damages in others. Climate-induced damages from hurricanes may have significant effects on local communities, but the macroeconomic consequences are projected to be very small. Net economic consequences are projected to be especially large in Africa and Asia, where the regional economies are vulnerable to a range of different climate impacts. For some countries in higher latitudes, economic benefits can arise from gains in tourism, energy and health. The global assessment also shows that countries that are relatively less affected by climate change may reap trade gains.

Appendices

Appendix 1: Brief Description of the ENV-Linkages Model

ENV-Linkages is a global multi-sectoral, multi-regional model that links economic activities to energy and environmental issues; Chateau et al. (2014) provide an in-depth description of the model.¹⁵ The model is calibrated for the period 2010–2060 using the macroeconomic trends consistent with the OECD Economic Outlook No 93 (2013). This Annex summarises the main components of the model: household demand, production, foreign trade, environmental aspects, factor supply, market equilibrium, government and closure rules and dynamic behaviour.

1.1 Household and Other Final Demand

Household consumption demand is the result of a static maximization behaviour which is formally implemented as an “Extended Linear Expenditure System”. In each region, a representative consumer, who takes the consumer-prices of commodities k ($P_{k}^{c})$ as given, optimally allocates disposable income (Y) among the full set of consumption commodities ($C_{k})$ and real savings (S). Following Lluch (1973) savings are considered to be a standard good and therefore do not rely on forward-looking behaviour by the consumer. The resulting demand equation of this optimization process is shown in Eq. (1), with parameters: $\theta _{k,}$ the maximum levels of per capita consumption for each good k; $\mu$, the marginal propensity to consume good k;  and Pop, total population:

$$\begin{aligned} C_k= & {} Pop\times \theta _k +\frac{\mu _k }{P_k^c }\times \left( {Y-Pop\times \sum _k {P_k^c \times \theta _k } } \right) , \quad \hbox { for all goods k and regions}\qquad \end{aligned}$$

(1)

$$\begin{aligned} S= & {} Y-\sum _k {P_k^c \times C_k},\quad \hbox { for all regions} \end{aligned}$$

(2)

Households derive income from supplying factors of production (labour, capital, land) to firms, subject to taxation. There are also direct transfers between households and governments, reflecting the net value of various transactions, such as income taxes and social security benefits.

Two more final demand categories are specified in the model. The Investment sector and the Government, they both consume a bundle of final goods, with a CES specification, reflecting a simplified version of household demand (excluding savings and excluding the necessary consumption bundle). Then, the total demand of a good in the economy is equal to the consumer final demand plus the intermediary demands from firms plus the government and investment expenditures of this good.

1.2 Production

Firms in all sectors minimise the cost of producing the goods and services that are demanded by consumers and other producers (domestic and foreign), taking producer prices as well as input prices as given. Production is represented by constant returns to scale technology. Moreover for each good or service, output is produced by different production streams, differentiated by capital vintage (old and new). Capital that is implemented contemporaneously is new—thus investment has an effect on current-period capital, but then it becomes ‘old’ capital (added to the existing stock) in the subsequent period.

Each production stream has an identical production structure, but with different technological parameters and substitution elasticities. It is based on a nested Constant Elasticity of Substitution (CES) production functions in a branching hierarchy. Moreover while the nested structure is replicated for the production of each commodity—with the notable exception of energy and agricultural sectors—, the parameterisation of the CES functions differs across sectors and regions. Energy and agricultural sectors use the same basic structure, but with adapted hierarchy of inputs to reflect sector-specific circumstances (see below).

Fig. 13

Nested structure of production of non-energy and non-agricultural sectors

Figure 13 illustrates the typical nesting of the model’s non-energy and non-agricultural sectors, where each node represents a constant elasticity of substitution (CES) production function, with the respective elasticity of substitution $\upsigma$. This gives marginal costs and represents the different substitution (and complementarity) relations across the various inputs in each sector. Each sector uses intermediate inputs—including energy inputs—and primary factors (labour and capital). Agricultural sectors also need land input while in some sectors, primary factors also include a sector-specific natural resource factor, e.g. minerals in mining sectors.

The top-level production nest considers final output (xp) as a composite commodity combining process emissions and the production of the sector net of these emissions. In sectors that do not emit such gases, the corresponding emission rate is set equal to zero. For greenhouse gas (GHG) emissions, the values of the substitution elasticities $\upsigma ^{\mathrm{top}}$ are calibrated such as to fit to marginal abatement curves available in the literature on alternative technology options (see for instance US EPA 2006); for air pollutants, these values equal zero.

On the right-hand side of the tree, the second-level nest considers the gross output of each sector (net of process emissions) as a combination of aggregate intermediate demands and a value-added bundle, including energy (VA). For sake of readability of this annex only the detailed equations about this part of the firm optimization program will be shown (optimal programs at the other nest levels would be very similar). The firms’ optimal allocation program between the aggregate intermediate bundle (ND) and the value-added bundle consists in choosing optimal demand for the two bundles, taking as given their prices, in order to minimize the cost of producing a given level $x_{i,v}$ of gross output of sector i (net of GHGs) using capital of vintage v. Using the dual representation of the top-level production nest expressed in prices, the first-order conditions of the firm’s profit maximization objective then give following optimal demands, with $\alpha$ the corresponding CES share parameter of the two bundles and tfp the total factor productivity in sector i:

$$\begin{aligned} ND_i= & {} \sum _v {\alpha _{i,v}^{nd} \times \left( {\frac{P_{i,v} }{P_i^{nd} }} \right) ^{\sigma _{i,v}^p}} \nonumber \\&\times \,\left( {tfp} \right) ^{\sigma _{i,v}^p-1}\times x_{i,v},\quad \hbox { for all regions and sectors i} \end{aligned}$$

(3)

$$\begin{aligned} VA_{i,v}= & {} \alpha _{i,v}^{va} \times \left( {\frac{P_{i,v} }{P_{i,v}^{va} }} \right) ^{{\sigma _{i,v}^p }}\times \left( {tfp} \right) ^{{\sigma _{i,v}^p }-1}\nonumber \\&\times \, x_{i,v}, \quad \quad \hbox { for all regions, sectors i and the two vintages} \end{aligned}$$

(4)

$$\begin{aligned} P_{i,v}= & {} \frac{1}{A_i }\times \left[ \alpha _{i,v}^{nd} \left( {P_i^{nd} } \right) ^{1-\sigma _{i,v}^p }+\,\alpha _{i,v}^{va} \left( {P_{i,v}^{VA} } \right) ^{1-\sigma _{i,v}^p} \right] ^{(1/(1-\sigma _{i,v}^p ))} \nonumber \\&\hbox {for all regions, sectors i and the two vintages} \end{aligned}$$

(5)

On the right part of the Fig. 13, the next nest describes the bundle VA as a combination of labour demand and a capital-energy bundle KTE. Similar sub-components also exist in the KTE bundle: a composite capital-natural resources (when one exists) and aggregate energy bundle (XEP). The energy bundle XEP is of particular interest for analysis of climate change issues. Energy is assumed to be first a composite of a “fossil fuels” bundle (NELY) and electricity demand. In turn, the NELY is a composite of coal demand and the bundle “Liquid fossil fuels” (OLG). At the lowest nest, OLG bundle consists of a mix of crude oil, refined oil products and natural gas demands.

On the left part of Fig. 13 the firm’s optimization program describes the optimal allocation of the aggregate intermediate demand bundle ND between intermediary demands for all goods and services (energy excluded) for all sectors and regions. Then these intermediary demands are split between domestic and imported commodities.

The nodes in Fig. 13 do not apply to agricultural and energy sectors. Indeed the nesting of the production function for the agricultural sectors is further re-arranged to reflect substitution between intensification (e.g. more fertiliser use) and extensification (more land use) of crop production; or between intensive and extensive livestock production. The structure of electricity production assumes that a representative electricity producer maximizes its profit by using the five available technologies to generate electricity: fossil-fuel based, hydro and geothermal, nuclear, solar and wind, and renewable combustibles (incl. biomass) and waste. A CES specification is used with a large value for the elasticity of substitution to reflect that the different technologies produce very close substitute products. Fossil-fuel based electricity production follows the generic production structure described in Fig. 13. Production of each of the non-fossil electricity technologies has a structure similar to that of the other sectors, except for a top nesting combining a sector-specific natural resource production factor on the one hand, and all other inputs on the other. This specification aims at controlling the penetration of these electricity technologies over time.

The final set of firms’ demand categories is given by demand for primary factors $x^{fp},$ comprising labour, land, capital (for both vintages) and a sector-specific ‘natural resource’ (determined at different nodes of Fig. 13). In this case an additional efficiency parameter $\uplambda ^{\mathrm{fp}}$ enters into the equation and plays an important role in the dynamic calibration and in supply shocks:

$$\begin{aligned} \begin{array}{ll} x_i^{fp} =&{}\alpha _i^{fp} \times \left( {\frac{P_{i,v}^{bd} }{P_i^{pf} }} \right) ^{^{\sigma _i^{bd} }}\times \left( {\lambda _i^{fp} } \right) ^{^{\sigma _i^{bd} }-1}\times bd_i, \hbox { where bd represents the `parent bundle'}\\ &{} \hbox { and takes different forms according to the primary factor considered}\\ &{}\hbox { (example bd = VA when fp is labour; see Fig.}~13) \end{array} \end{aligned}$$

(6)

1.3 Agricultural Land Supply and Land Allocation Across Agriculutural Land Uses

Land mobility and extensification of the area used for production are introduced in two steps. First, land supply curves are used to govern the introduction of currently unmanaged land. This requires information on the potential land supply by region, which are obtained from the MAGNET model (LEI 2014). The impact of the land supply curve in the model is to reduce the land supply elasticity over time as land use gets closer to the potential land use.

Secondly, a multi-level transformation function captures a hierarchical mobility of land use across sectors. This transformation function describes the ease with which different land uses can be transformed into other uses. The characterization of the nested transformation function across agricultural uses of land is based on the OECD PEM model (OECD 2011). Under this specification the total agricultural land (tland) is firstly allocated between “rice”, “other crops”, “vegetables”, “fibre plants” and a Field Crops and Pasture bundle (FCP), in turn this FCP land bundle is decomposed into land demands for “livestock”, “sugar plant” and a composite Cereals, Oilseeds and Protein crops bundle (COP); at last this COP bundle is decomposed into demand for between “wheat”, “other grains” and “oilseed”. The corresponding equations for the optimization of the land allocation of tland t are shown below (equations for the other land bundles are symmetric):

$$\begin{aligned} land_{cr}= & {} \alpha _{cr}^{land} \times \left( {\frac{P_{cr}^{land} }{P^{tland} }} \right) ^{{\omega t^{tl} }}\times tland, \hbox { for land type } \textit{cr} \nonumber \\= & {} \hbox { ``rice'', ``other crops'', ``vegetables'', ``fiber''} \end{aligned}$$

(12)

$$\begin{aligned} COP= & {} \alpha _{COP}^{land} \times \left( {\frac{P^{COP}}{P^{tland} }} \right) ^{\omega ^{tl}}\times tland, \qquad \qquad \quad \hbox { for all regions} \end{aligned}$$

(13)

$$\begin{aligned} P^{tland}= & {} \left[ {\sum _{cr} \alpha _{cr}^{land} \left( {P_{cr}^{land}} \right) ^{1+\omega ^{tl}}+\alpha _{COP} \left( {P^{COP}} \right) ^{1+\omega ^{tl}}} \right] ^{(1/(1+\omega ^{tl}))}, \quad \quad \quad \hbox {for all regions}\nonumber \\ \end{aligned}$$

(14)

1.4 Foreign Trade

International trade is based on a set of regional bilateral flows. The model adopts the Armington specification, assuming that domestic and imported products are not perfectly substitutable. Moreover, imports are also imperfectly substitutable between regions of origin. Therefore, in each region, total import demand for each good is allocated across trading partners according to the relative export prices of the originating regions. The Armington specification is implemented using two CES nests (see the bottom left part of Fig. 13). At the top nest, domestic agents choose the optimal combination of the domestic good ($x^{d})$ and aggregate import good ($x^{m})$, where $P^{a}$ is the price of Armington good i, $P^{m}$ the price of aggregate import of i and $P^{p}$ the production price (e.g. the price of domestically produced good):

$$\begin{aligned} x^{m}_i= & {} \beta _i^m \times \left( {\frac{P_i^a }{P_i^{^{m}} }} \right) ^{\sigma _i^m }\times x_i^a, \quad \hbox { for all commodities i and regions} \end{aligned}$$

(7)

$$\begin{aligned} x^{d}_i= & {} \beta _i^d \times \left( {\frac{P_i^a }{P_i^p }} \right) ^{\sigma _i^m }\times x_i^a, \quad \hbox { for all commodities i and regions} \end{aligned}$$

(8)

$$\begin{aligned} P_i^a \times x_i^a= & {} P_i^p \times x_i^d +P_i^m \times x_i^m, \quad \hbox { for all commodities i and regions} \end{aligned}$$

(9)

At the second nest, domestic agents optimally allocate demand for the aggregate import good across the range of the different trading partners r, and this determines the full range of world bilateral trade flows for all goods i between all origins r and destination r, with $P^{IM}$ being the domestic import price of a particular good or service from region of origin r (CIF price).

$$\begin{aligned} WTF_{r{\prime },i}= & {} \beta _{r{\prime },i}^w \times \left( {\frac{P_i^{m} }{P_{r{\prime },i}^{^{IM}} }} \right) ^{\sigma _i^w }\nonumber \\&\times x^{m}_i, \quad \quad \hbox { for all commodities I, all origins r' and all destinations r} \end{aligned}$$

(10)

and the price of aggregate import

$$\begin{aligned} P_i^m =\left[ {\sum _r {\beta _{i,r}^W} P_{i,r}^{{IM}^{1-\sigma _{i,r}^w}}} \right] ^{(1/(1-\sigma _i^w ))}, \quad \quad \hbox {for all commodities } i \hbox { and all destinations } r\nonumber \\ \end{aligned}$$

(11)

For sake of simplicity, it is assumed that firms will not differentiate between domestic and foreign buyers (so net-export prices and domestic prices are identical), therefore the following equations hold (with xmarg the quantity of international trade and transport services that capture difference between origin and destination port):

$$\begin{aligned} xp_i =x_i^d +\sum \nolimits _{r{\prime }} {WTF_{r{\prime },i} } +x_i^{m\arg }, \quad \quad \hbox { for all commodities } i \hbox { and regions } r \end{aligned}$$

(12)

1.5 Market Equilibria

Once a sector’s optimal combination of inputs is determined from relative prices, sectoral output prices are calculated assuming competitive supply (zero-profit) conditions. Market goods equilibria imply that, on the one side, the total production of any good or service is equal to the demand addressed to domestic producers plus exports; and, on the other side, the total demand is allocated between the demands (both final and intermediary) addressed to domestic producers and import demand. Similarly, equilibrium is ensured on all factor markets.

1.6 Government and Model Closure

The government in each region collects various kinds of taxes in order to finance government expenditures. Assuming fixed public savings (or deficits), the government budget is balanced through the adjustment of the income tax on consumer income. In each period, investment net-of-economic depreciation is equal to the sum of government savings, consumer savings and net capital flows from abroad.

Each region runs a current-account surplus (or deficit), which is fixed (in terms of the model numéraire). Closure on the international side of each economy is achieved by having, as a counterpart of these imbalances, a net outflow (or inflow) of capital, which is subtracted from (added to) the domestic flow of saving. These net capital flows are exogenous. In each period, the model equates investment to saving (which is equal to the sum of saving by households, the net budget position of the government and foreign capital flows). Hence, given exogenous sequences for government and foreign savings, this implies that investment is ultimately driven by household savings.

1.7 Dynamic Behaviour

The ENV-Linkages model has a simple recursive-dynamic structure as agents are assumed to be myopic and to base their decisions on static expectations concerning prices and quantities. Dynamics in the model originate from two endogenous sources: (i) accumulation of productive capital and (ii) the putty/semi-putty specification of technology. The dynamics also depend on exogenous drivers like population growth, autonomous energy efficiency or sector specific autonomous labour efficiency improvements, et cetera.

At the aggregate level, the basic capital accumulation function equates the current capital stock (kstock) to the depreciated stock inherited from the previous period (at depreciation rate $\delta )$ plus net investment (I):

$$\begin{aligned} kstock=(1-\delta )\times kstock(-1)+I, \quad \quad \hbox { for all regions } r \end{aligned}$$

(13)

Where the new investment determined residually as the result of investment-saving equilibrium, with $P^{I}$ le price of investment, $S^{g}$ nominal government savings, and $S^{f}$ net foreign savings:

$$\begin{aligned} P^{I}\times I=S+S^{g}+P\times S^{f}+\delta \times P^{I}\times kstock, \quad \hbox { for all regions } r \hbox { but one } \end{aligned}$$

(14)

Differences in sectoral rates of return determine the allocation of investment across sectors. The model features two vintages of capital, but investment adds only to new capital. Sectors with higher investment, therefore, are more able to adapt to changes than sectors with low levels of investment. Indeed, declining sectors whose old capital is less productive can to a limited extent sell existing capital stock to other firms, which they can use after incurring some adjustment costs.

1.8 Emissions

Emissions from combustion of energy, including $\hbox {CO}_{2}$ and a range of air pollutants, are directly linked to the use of different fuels in production. Other GHG emissions are linked to output in a way similar to Hyman et al. (2002). The following non-$\hbox {CO}_{2}$ emission sources are considered: i) methane from rice cultivation, livestock production (enteric fermentation and manure management), fugitive methane emissions from coal mining, crude oil extraction, natural gas and services (landfills and water sewage); ii) nitrous oxide from crops (nitrogenous fertilizers), livestock (manure management), chemicals (non-combustion industrial processes) and services (landfills); iii) industrial gases ($\hbox {SF}_{6}$, PFCs and HFCs) from chemicals industry (foams, adipic acid, solvents), aluminium, magnesium and semi-conductors production. Over time, there is, however, some relative decoupling of emissions from the underlying economic activity through autonomous technical progress, implying that emissions grow less rapidly than economic activity.

Emissions of air pollutants not linked to combustion are assumed to be in fixed proportion to output of the sector. Emissions of key air pollutants - sulphur dioxide ($\hbox {SO}_{2})$, nitrogen oxides ($\hbox {NO}_{\mathrm{x}})$, black carbon (BC), organic carbon (OC), carbon monoxide (CO), volatile organic components (VOCs) and ammonia ($\hbox {NH}_{3})$—have been linked to the projected sectoral production activities in ENV-Linkages. As mentioned above, emission coefficients from combustion processes in industrial sectors, transport and residential and commercial energy demand are linked to the inputs of fossil fuels. Other emission coefficients are instead linked directly to output (e.g. different agricultural goods, cement, metals and waste). Other sources of emissions (e.g. from biofuels) are considered but accounted for separately as it is not possible to link them to specific economic activities in ENV-Linkages.

Appendix 2: Parameter Values and Model Calibration

1.1 Calibration Procedure

The process of calibrating the ENV-Linkages model is broken down into three stages. First, a number of parameters are calibrated, for given elasticity values, to represent the data for a representative historical year (using 2004 rather than the more recent 2007 data as 2007 was less representative) as an initial economic equilibrium. This process is referred to as the static calibration. However, the information available on the values of these parameters is insufficient for the model simulation to be able to reproduce base-year data values. Given the modelling choices made with regard to the representation of both behaviour and structural technical relationships, some model parameters must be calculated to fit to the data for the initial year. As a general rule, the parameters used to do this are those whose impact on the outcomes in terms of variation rates remains limited (scale parameters) or parameters for which there are no widely accepted empirical estimates (CES share coefficients).

The second step of the calibration starts from the base year data and updates all parameters to a reference year for the model baseline (currently 2010) by simulating the model dynamically to match historical trends over the period 2005–2010. For 2010 the SAMs of the model are re-calculated to ensure that all economic values are defined in 2010 USD; 2010 is then the base year for the ENV-Linkages simulations.

Third, the baseline projection for the model horizon 2011–2060 reproduces various OECD macro-economic projections as well as other sectoral projections.

The sectoral and regional aggregation of the model, as used in the analysis for this paper, are given in Tables 2 and 3, respectively.

Table 2 Sectoral aggregation of ENV-Linkages

Table 3 Regional aggregation of ENV-Linkages

1.2 Starting Year Data and Main Elasticities

The core of the static calibration is formed by the set of comprehensive input–output tables that describe how economic activities in the different sectors are linked to each other, and linked to economic activity in other regions. This set of mutually consistent accounting matrices is based upon the Social Accounting Matrices (SAMs) of the Global Trade Analysis Project (GTAP) database (Narayanan et al. 2012). For the model version described here, the GTAP version 8.2 is used for the base year 2004.

Many key parameters are set on the basis of information drawn from various empirical studies and data sources (elasticities of substitution, income elasticities of demand, supply elasticities of natural resources, etc.). Table 4 reports some key production elasticities used in the current version of the model. Use of these parameters is illustrated in Fig. 13 in “Appendix 1”, as well as by the associated equations.

Following the putty/semi-putty technology specification, Table 4 shows that the substitution possibilities among factors, inputs and production bundles are assumed to be higher with new vintage capital than with old vintage capital.

Table 4 Key parameter values

In addition to these parameters, the income elasticities of household demand for the starting year are taken from the GTAP database.

1.3 Dynamic Calibration and Projections Used

Ideally, an informed choice of prospective trends in exogenous variables would produce a set of acceptable scenarios. The approach followed here considers only one single set of drivers to provide a central baseline projection while recognising that alternative sets may potentially generate somewhat different simulation results. The central baseline projection allows calculating the values of a number of parameters over time (such as energy efficiency gains), in order to reproduce the evolution of the drivers (such as energy demands; for more details see below). Alternative baseline scenarios could be built by varying specific model assumptions or by adopting a different storyline on major trends. Exploration of such alternatives is not considered here.

1.4 Macroeconomic and Socio-Economic Trends

The first step of the baseline projection is to derive a set of macro-economic variables that are reproduced in the model by adjusting key parameters. Macro-economic projections are performed using two long run macroeconomic growth models in combination. First, the growth scenarios are resulting from simulations of the BLT model hosted at the OECD Economics Department (Johansson et al. 2013). This model covers only 42 countries: OECD and G20 countries up to 2060. For the current publication the scenario used is identical to the “Long-term baseline projections” presented in the OECD Economic Outlook No 93 (OECD 2013). Secondly, the ENV-Growth model, hosted at the OECD Environment Directorate, is used to complete these projections for countries not covered by the BLT model. Together, macroeconomic projections are provided for almost 180 countries. The ENV-Growth model could in practice run up to 2100 as for example used in the GDP projections of the Shared Socio-Economic Pathways (Dellink et al. 2017).

Both models adopt the same philosophy: they project country by country a gradual process of convergence towards a balanced growth path along the lines of an augmented Solow growth model (like Barro and Sala-i-Martin 2004). These models adopt as common principle a so-called conditional convergence hypothesis: country income levels (e.g. GDP per capita) will converge towards those of most developed economies based on convergence hypotheses for all the key drivers of per capita economic growth. But the long run potential differs across economies as their drivers start from different values and the dynamics to their long run values are conditioned by different parameter values; furthermore, for some drivers in some countries, the convergence process is not completed before the end of the century.

The main difference between the two models is that the BLT model projects the dynamics of transition of economies from 2010 to their long run potential (e.g. a balanced growth path) in detail, while for the ENV-growth model the short and medium transition to the long run potential are derived with simple convergence equations, which requires much less data. Nevertheless both models adopt the same drivers of the long run potential GDP per capita. Each driver’s calculations are based on real income comparisons expressed in purchasing power parity (PPP) and not in market exchange rates. Technically, the dynamic global projection is obtained by running ENV-Growth taking variables for the 42 countries from OECD (2013). Once this is done data are aggregated across countries to recover the aggregation of the ENV-Linkages model (Table 3).

The baseline simulation of ENV-Linkages, in the absence of climate change and local air pollution damages, uses these projections to calibrate future macro-economic variables over the period 2010–2060. The baseline is characterised by the continuation of current policies, incl. energy and environmental policies. Table 5 details some of the main adjustments done.

Table 5 Macro-economic calibration

1.5 Sectoral Trends and Other Adjustment of Parameters Over the Baseline

Next, the baseline construction features specific sectoral projections for the energy and agricultural sectors. Energy system projections are calibrated on the “2013 World Energy Outlook” trends (IEA 2013b) while agricultural and food projections from IFPRI’s IMPACT V.3. model (Rosegrant 2012) are used to calibrated agriculture system.The main calibration choices are summarized in Table 6.

Table 6 Sectoral calibration

In models like ENV-Linkages, which use the Armington specification to represent international trade flows, countries face downward sloping demand for their exports. Therefore, a fast-growing country would typically experience a decline in its relative factor prices, implying a depreciation of its real exchange rate, ceteris paribus (abstracting from the offsetting Balassa–Samuelson effect). This appears inconsistent with past history, which shows that imports from fast-growing countries have typically increased through the creation of new products rather than through price reductions (see in particular Krugman 1989). In order to capture this historical feature in a simplified manner, the baseline projection further assumes a gradual exogenous increase in the share of non-OECD countries in overall imports of OECD countries.

In addition, the increase in global competition is accompanied by growth in the use of services in production, in line with the argument advanced in OECD (2005). This is simulated by adjusting dynamically the input–output structure so as to increase the weight of services in the composition of the bundle of intermediate goods, for non-agricultural and non-fossil fuels sectors.

Finally, the parameters relative to household demand need to be recalibrated dynamically in the baseline simulation. Household preferences in ENV-Linkages include a minimum subsistence level of demand for each good that makes the utility function non-homothetic. However, when using the model over a rather long projection horizon, household income can increase quite substantially and, if the minimum subsistence demands are not adjusted, the income elasticity of demand for all goods converges towards unity. This problem is offset by adjusting the subsistence parameters in the baseline scenario for each period in order to reproduce a desired set of income elasticities for future periods.

Moreover, in the baseline simulation, income elasticities of demand are evolving over time, assuming a conditional convergence of household preferences (e.g. income elasticities of demand for non-energy and non-agricultural goods) of the non-OECD countries to the OECD standard, based on relative income per capita.

Appendix 3: Background Information on the Modelled Impacts of Climate Change

1.1 Agriculture

The climate change impacts on agriculture that are modelled in ENV-Linkages involve sector- and regional-specific changes in crop yields for each of the 8 crop sectors (see “Appendix 2” for the sectoral disaggregation of the ENV-Linkages model). The input data on crop yield changes (physical production per hectare) are those shared by the modelling teams involved in the Agricultural Model Intercomparison Project Ag-MIP (Rosenzweig et al. 2013; Nelson et al. 2014; Von Lampe et al. 2014). This project contains the most robust global assessment of agricultural impacts from climate change published to date. Although impacts on grasslands follow very similar patterns as impacts on crop land, the AgMIP project has not provided information on how grasslands are affected, so impacts on livestock are excluded from the modelling analysis. From the available scenarios shared in the AGMIP project, the central projection uses the HadGEM model, for the specification of the climate system variables, coupled with the DSSAT crop model (Hoogenboom et al. 2012; Jones 2003). The specification of regional climate impacts coming from this model combination was then used as input for the International Food Policy Research Institute’s IMPACT model (Rosegrant 2012) to calculate the exogenous yield shocks from changes in crop growth and water stress by water basin. These shocks were then aggregated to the ENV-Linkages model regions. A pathway between 2010 and 2050 was produced by proportionally changing the effect of climate change on the yield growth rate such that in 2050, the yield shocks correspond to the AgMIP projection for the 2050s; this delivers a non-linear impact on yield levels.

One specific scenario was chosen for the central projection, using projections of regional climate changes from the HadGEM model and using the DSSAT crop model to identify the agronomic impacts. In the AgMIP project, alternative choices are available for both the climate system model (using IPSL instead of HadGEM) and the crop model (using LPJmL instead of DSSAT), which are used in Sect. 4.6 to illustrate the uncertainties. In line with AgMIP, the estimated yield shocks used in the central projection do not consider the carbon fertilization effects on vegetation as these are deemed too uncertain, although Rosenzweig et al. (2013) do identify it as “a crucial area of research”. Again, this is subjected to sensitivity analysis in Sect. 4.6.

Since the IMPACT model does not contain a full production function while ENV-Linkages does, the crop yield shocks have been translated into specific elements in the production functions of the ENV-Linkages agricultural sectors. The yield shocks are implemented in the model as a combination of the productivity of the land resource in agricultural production, and the total factor productivity of the agricultural sectors.¹⁶

Climate change affects crop yields heterogeneously in different world regions. Further, the effects are also not the same for different crops. Figure 14 illustrates changes in crop yields in 2050 for the central projection using the HadGEM climate model in combination with the DSSAT crop model and exclude a $\hbox {CO}_{2}$ fertilisation effect; the uncertainty related to the choice of climate and crop model, and the effect of $\hbox {CO}_{2}$ fertilisation, is further investigated in Sect. 4. While the maps illustrate impacts for 2050, the impacts are not constant over time. They follow a non-linear trend that is extended from 2050 to 2060 using the increase from the previous decade. Impacts for other crops can deviate substantially from the impacts for rice and wheat; they are not reproduced here, but are described in detail in Nelson et al. (2014). Note that these impacts refer to potential shocks: in the CGE model, farmers have options to change their production process and adapt to these shocks and will do so in order to minimise their costs, i.e. market-driven adaptation is endogenously handled inside the economic modelling framework. The modelling framework excludes the possibility to increase the size of irrigated agricultural land. In regions with low water stress levels, this adaptation option can be an important part of the response to climate change (Ignaciuk and Mason-D’Croz 2014), but is excluded here as markets forces alone are usually insufficient to achieve large-scale expansion of irrigated areas (Ignaciuk 2015).

Fig. 14

Source: IMPACT model, based on the AgMIP study

Impacts of climate change on crop yields in the central projection.

Changes in yields of paddy rice by 2050 are strongest in tropical areas, including Central American and Mexico, Saharan African countries, some parts of the Middle East and a large part of South and South-East Asian countries. Some regions have large positive impacts on paddy rice yields. In particular, the highest gains will take place in the Southern parts of Latin America, and particularly in Chile, in large parts of Africa, including Morocco, South Africa and other Sub-Saharan African countries, and in parts of Easter Europe and continental Asia. Such heterogeneity in impacts suggests that climate change will largely change trade patterns in widely traded commodities such as rice.

Changes in yields of wheat by 2050 are somehow less differentiated, as most regions are negatively affected. The most severe negative impacts take place in Mexico, Western and Eastern Africa, some Southern African countries, Middle East, South and South East Asia, and some Western European regions, such as Belgium, the Netherlands and Germany. While these are the most affected regions, negative impacts are widely spread and also affect most of Europe, continental Asia and North America. Some regions are positively affected by climate change. These include regions with cold climates such as Canada, Russia and Scandinavian countries, most of Central America, Argentina, some countries in Eastern Europe and continental Asia, and a few African countries.

While the data from Nelson et al. (2014) do not provide projections of effect of higher carbon concentrations on crop growth (the $\hbox {CO}_{2}$ fertilisation effect), the underlying crop model information as reported in Rosenzweig et al. (2013) and synthesised in the Geoshare tool¹⁷ is used in the sensitivity analysis in Sect. 4.6. For the default combination of the DSSAT cropmodel and HadGEM climate model, projections on the $\hbox {CO}_{2}$ fertilisation effect are available for rice, wheat, maize and soybeans. A wider set of crops, including also rapeseed, millet, sugarcane and sugar beets, is covered by the combination of the LPJmL cropmodel and HadGEM climate model. $\hbox {CO}_{2}$ fertilisation effects are not available for fruits and vegetables (including potatoes), and for the plant fibres sector (which includes cotton). For the alternative climate model IPSL, $\hbox {CO}_{2}$ fertilisation effects are only provided for non-irrigated lands. For missing data, the assumption is made that there is no $\hbox {CO}_{2}$ fertilisation effect, although the literature suggests that for some crops not included here (not least potatoes) the effect may be quite strong (Leakey 2009).

As an illustration of the projection data that is used in the simulations, Table 7 presents the yield shocks for 2050 for the Sub-Saharan Africa region (excl. South Africa). The table shows that the impacts of climate change on crop yields varies widely between crops, and the model choice can even change the sign of the yield effect, as shown in the table for sugarcane / sugar beets. The table also illustrates that the effect of $\hbox {CO}_{2}$ fertilisation is quite strong and positive and can limit some of the major negative consequences in agriculture.

Table 7 Impacts of climate change on yields in Sub-Saharan Africa (percentage change in yields in 2050 relative to current climate)

For the fisheries sector, the damages reflect projected changes in global fish catch potential caused by climate change. This is modelled in ENV-Linkages as a change in the natural resource stock available to fishing sectors, which approximates the impacts of climate change for fish stocks and the resulting effects for the output of the fisheries sector. Acknowledging that the empirical basis for estimating the impacts on the fisheries sector is very small and uncertainties on projections are very large, the input data used for the modelling is based on one of the most comprehensive assessments, the EU’s SESAME project, which in turn uses results from Cheung et al. (2010). This study applies an empirical model (Cheung et al. 2008) that predicts maximum catch potential as dependent upon primary production and distribution. It considers a range of 1066 species of exploited fish and invertebrates. Future projected changes in species distribution are simulated by using a model (Cheung et al. 2008, 2010) that starts with identifying species’ preference for environmental conditions and then links them to the expected carrying capacity. The environmental conditions considered include seawater temperature, salinity, distance from sea-ice and habitat types, but the assessment excludes any effects related to ocean acidification. The model assumes that carrying capacity varies positively with habitat suitability of each spatial cell. Finally, the related change in total catch potential is determined aggregating spatially and across species.

The input data for the fisheries sector in the ENV-Linkages model is the percentage change in fish catch with respect to 2000 as described above. The most negatively affected regions by 2060 are North Africa (− 27%) and Indonesia (− 26%). Some European regions, the Middle East, Chile and several countries in South East Asia have impacts ranging from − 10 to − 15%. Smaller negative impacts also take place in China, Korea, Brazil and other Latin American countries, Mexico, and some European countries. In some countries fish catches actually increase. The highest increases will occur in Russia ($+$ 25%) and in the five major European economies ($+$ 23%). Small positive impacts are seen in the United States, Canada, Oceania and the Caspian region. Other world regions (India, other developing countries in Asia, South Africa and the rest of Africa) are basically unaffected.

1.2 Coastal Zones

Coastal land losses due to sea level rise are included in the ENV-Linkages model as changes in the availability of land as well as damages to physical capital. Both modifications concern land and capital stock variables by region in the model. As information on capital losses are not readily available, in line with Bosello et al. (2012), land and capital stock changes are approximated by assuming that changes in capital supply match land losses as a percentage change from baseline.

Estimates of coastal land lost to sea level rise are based on the DIVA model outputs (Vafeidis et al. 2008) as used in the European Union’s (EU) FP7 ClimateCost project (Brown et al. 2011) and generated with the HadGEM model. DIVA is a sector model designed to address the vulnerability of coastal areas to sea level rise and other ocean- and river-related events, such as storm surges, changes in river morphology and altered tidal regimes. The model is based on a world database of natural system and socioeconomic factors for world coastal areas reported with spatial details. Changes in natural and socio-economic conditions of possible future scenarios are implemented through a set of impact and adaptation algorithms. Impacts are then assessed both in terms of physical losses (i.e. sq. km of land lost) and economic costs (i.e. value of land lost and adaptation costs).

The regions that are most affected by sea level rise are those in South and South East Asia, with highest impacts in India, and other developing countries in the region. The projected land and capital losses expressed as percentage of total regional agricultural land area in 2060 with respect to the year 2000 are respectively − 0.63% for India and − 0.86% for the Other Developing Asia region of ENV-Linkages. Other countries in the region are also affected but to a smaller extent. Some impacts are also felt in North America, with Canada, Mexico and the United States being affected. Canada has the highest loss in land (and capital) in this region (− 0.47% in 2060 with respect to 2000). Smaller impacts occur in Middle East (− 0.35%) and in Europe, where the highest impacts are felt in the aggregate non-OECD Europe region (− 0.37%), which includes, among other countries, Israel, Norway and Turkey. Other world regions, such as Africa, South America and continental European regions are on balance hardly affected by sea level rise.

1.3 Extreme Events

There are many types of extreme events and they affect the economy in different ways. However, given the uncertainties involved in the frequency and damages caused by these events and the difficulties in attributing such events to climate change, the available data on how the economy will be affected is still scarce. Recently, the assessment by Mendelsohn et al. (2012) has provided some quantitative assessment and projections on damages from hurricanes that can be used as input in an economic framework. Mendelsohn et al. (2012) stress that the regional damages are quite sensitive to the climate model that is used to project future climate conditions, and projections based on the HadGEM model are not available. Hence, the analysis of the economic consequences was realised in ENV-Linkages on a multi-model average.¹⁸ This may of course dampen some of the more severe consequences projected by individual models.

Mendelsohn et al. (2012) find that climate change is predicted to increase the frequency of high-intensity storms in selected ocean basins as the century progresses, although this depends on the climate model used. These climate-induced damages are included in ENV-Linkages as reductions in regional capital stocks from tropical cyclones. Due to lack of information, this is assumed to affect all sectors (which is in line with the normal CGE model assumption that new investments in capital are fully malleable across the economy).

Mendelsohn et al. (2012) also find that the current annual global damage from tropical cyclones is USD 26 billion, which is equivalent to 0.04% of global GDP, and roughly double (in absolute amount) by the end of the century under current climate conditions, i.e. due to changes in socioeconomic conditions.¹⁹ However, these damages are projected to double again by the end of the century due to climate change. Most additional climate-induced damages are predicted to take place in North America, East Asia and the Caribbean–Central American region, where the United States, Japan and China will be most affected.

1.4 Health

Within the health impact category, the ENV-Linkages model covers both climate-related illnesses and effects related to heat stress. Impacts on human health linked with climate-related diseases are expressed by changes in mortality and morbidity, following Bosello et al. (2012) and Bosello and Parrado (2014).²⁰ The illnesses considered are vector-borne diseases (malaria, schistosomiasis and dengue), diarrhoea, cardiovascular and respiratory diseases. The assessment for cardiovascular diseases includes both cold and heat stress. Within the production function approach, the modelling technique for these health impacts is to translate the results of the empirical literature into changes in labour productivity and demand for health services (Bosello et al. 2006)—explicitly excluding the welfare (or “disutility”) impacts of premature deaths from climate change. While there are other factors that are affected by these illnesses, labour productivity is the most suitable variable to capture the effects that climate-related diseases have on the economy.

Estimates of the change in mortality due to vector-borne diseases are taken from Tol (2002), which are based on modelling studies (Martens et al. 1995, 1997; Martin and Lefebvre 1995; Morita et al. 1994) as well as on mortality and morbidity figures from the World Health Organization’s Global Burden of Disease data (Murray and Lopez 1996).²¹ These studies suggest that the relationship between climate change and malaria is linear. This relationship is also applied to schistosomiasis and dengue fever. To account for changes in vulnerability possibly induced by improvement in living standards, Tol (2002) applies a relationship between per capita income and disease incidence (Tol and Dowlatabadi 2001). This relationship is used to assess the impacts for the CIRCLE baseline by using the projected per capita regional income growth of the ENV-Linkages model.

For diarrhoea, an estimated equation describes how increased temperatures increase both mortality and morbidity, while negative income elasticities imply lower impacts with rising income, wi5th mortality declining more rapidly than morbidity (Link and Tol 2004). For premature deaths due to cardiovascular and respiratory diseases, data are based on a meta-analysis performed in 17 countries (Martens 1998). Tol (2002) extrapolates these findings to all other countries, using the current climate as the main predictor. Cardiovascular (for both cold and heat stress) and respiratory mortality (for heat stress only) are assumed to only affect urban population.

The resulting changes in labour productivity from climate-induced diseases, which have been summarised in Bosello et al. (2012), are used as input in ENV-Linkages. By 2060, the highest negative effects take place in Africa and the Middle East (− 0.6% for South Africa, − 0.5% for North Africa and the Middle East and − 0.4% for other African countries). Smaller impacts take place in Brazil, Mexico and in developing countries in Asia (− 0.3%), as well as in Indonesia, the United States, South-East Asia and most of Latin America (− 0.2%). Some regions are projected to have positive impacts on labour productivity from climate-induced diseases, the highest being in Russia ($+$ 0.5%), Canada ($+$ 0.4%) and China ($+$ 0.2%). In other regions the impacts are either very small or inexistent.

Changes in health care expenditures for climate-related diseases are also taken from Bosello et al. (2012). The costs of vector borne diseases are based on Chima et al. (2003), who report the expenditure on prevention and treatment costs per person per month. Their data suggest the following relationships

$$\begin{aligned} P= & {} \mathop {0.1406}\limits _{(0.3103)} + \mathop {0.0026Y}\limits _{(0.0008)} \end{aligned}$$

(1)

$$\begin{aligned} T= & {} \mathop {-\,0.4646}\limits _{(0.8217)} + \mathop {0.0053Y}\limits _{(0.0018)} \end{aligned}$$

(2)

where P is monthly prevention costs ($/capita), T is monthly treatment costs ($/cap) and Y is income per capita ($/cap). We scale this up with the increase in mortality.

Changes in health expenditure are small as percentage of GDP.²² In 2060, they are projected to be highest in the developing countries in Asia (0.5%), in Brazil and in the Middle East and North Africa region (0.3%). Additional demands for health services are very small in other regions. Interestingly, they are negative in Canada and in large EU economies, such as Germany and France (− 0.1%), where reduced cardiovascular disease expenditures dominate.

Occupationalheat stress is modelled as having an impact on labour productivity. This builds on Kjellstrom (2009), who identify a link between the global temperature, heat and humidity, and work ability for different types of activities (agriculture, industry and services). Ideally one would combine the sectoral reductions in work ability to the regional temperature increases to identify labour productivity losses. Unfortunately, there is insufficient data to do so. However, data derived from Roson and Van der Mensbrugghe (2012) translates the underlying regional climate profiles into labour productivity losses as a function of global average temperature increase. For the European regions, data from Ciscar et al. (2014) are adopted.

Until now, most assessments (Eboli et al. 2010; Ciscar et al. 2014) have first aggregated the various productivity losses across sectors and then apply these averages to all economic sectors. Recent research (e.g. Graff Zivin and Neidell 2014; Somanathan et al. 2014) has highlighted that productivity of sectors where workers are mostly outdoor (i.e. heat-exposed industries) are on balance much more affected by increased heat. For indoor activities, in turn, the most severe consequences can be avoided by increased air conditioning, which is at least partially captured under the impacts of changed energy demand (see e.g. Somanathan et al. 2014). Hence, the impacts are assumed to be limited to heat-exposed sectors. In line with Ciscar et al. (2014), labour productivity losses are concentrated in the agricultural, forestry, fisheries, and construction sectors, and exclude most manufacturing and services sectors (see “Appendix 2” for a full list of sectors in ENV-Linkages).

The highest impacts on labour productivity caused by heat stress in 2060 take place in regions with relative large proportions of outdoor workers and warm climates. The most severely affected regions, with productivity losses between 3 and 5% for outdoor activities for a one degree temperature increase are in non-OECD, non-EU European countries, Latin America (incl. Brazil and Chile), Mexico, China, Other Developing Asia and South Africa. Most OECD countries, including USA, Japan and the OECD EU countries have much smaller effects, of less than 1%.²³

The health impacts of climate change have economic consequences that go beyond market costs. These costs, such as the costs of premature deaths, cannot be accounted for in the ENV-Linkages model. However, they can be evaluated using WTP techniques and, for premature deaths, the Value of a Statistical Life.

1.5 Energy Demand

Residential energy demand has been projected to change due to climate change. As discussed in e.g. IEA (2015), the energy sector is heavily influenced by mitigation policies. But even in the absence of mitigation policies will there be impacts on energy demand and supply. Changes in households’ demand for oil, gas, coal and electricity from less energy consumption for heating and more for cooling, have been captured directly in the model as a change in consumer demand for the output of these energy services. Changing residential energy demand in response to climate change is derived from the IEA, which provides data on space heating and cooling by carrier until 2050 under its Current Policies Scenario (IEA 2013b). Data until 2060 was extrapolated using trends in demand from 2040 to 2050. The IEA derives its projections from its World Energy Model, which is a large-scale partial-equilibrium model designed to replicate the functioning of energy markets over the medium- to long-term. It determines future energy supply and demand for different energy carriers (supply: coal, oil, natural gas, biomass; demand: coal, oil, natural gas, nuclear, hydro, bioenergy, and other renewables) according to trends in energy prices, $\hbox {CO}_{2}$ prices, technologies, and socioeconomic drivers. The baseline trends without climate change are characterised by increased heating and cooling demand for most economies, driven to a large extent by higher incomes; the baseline also projects a strong trend of electrification, which affects especially heating energy demand. Demand for space heating and cooling under climate change is affected by factors including the anticipated change in heating and cooling degree days due to climate change (IEA 2013b, 2014).

Overall, global energy demand for space cooling is projected to grow by roughly 250% between 2010 and 2060 under no climate change and by 330% if climate change is taken into account. Increases in demand for heating are much lower, with a projected 42% rise until 2060 without climate change and a 16%-increase in the climate change scenario, but start from a much larger base level. Non-OECD countries drive most of the increase in demand both for heating and cooling, and particularly so in heating. By 2060, household demand for cooling is projected to be 27% of the total demand for space heating and cooling purposes as compared to 9% in 2010.

Climate change-induced shifts in the demand for electricity until 2060 can reflect both an increased demand for cooling purposes, i.e. air conditioning, as well as decreased demand for electric space heating as a response to higher average temperatures. Globally, total annual electricity demand is projected to remain largely unaffected by climate change by 2060, with increases in demand for cooling during summers balancing decreases in demand for heating during winters. Of the 25 regions in ENV-Linkages, about half are projected to increase their total demand for electricity due to greater need for cooling, including the EU7 ($+$ 11.4%), Chile ($+$ 8.4%) and non-OECD EU countries ($+$ 7.3%). Korea (− 6.7%), non-EU OECD Europe (− 3.7%), and the Caspian region (− 3.1%) are part of the other half of the regions in which decreased consumption of electricity for heating will more than offset increases in demand for cooling (given the strong trend towards electrification of heating in the baseline without climate change). In the vast majority of the regions, positive or negative variations in demand for cooling and heating under climate change are projected to stay below 3% relative to the baseline total electricity demand of households.

As global temperatures increase with climate change, the IEA projections suggest that climate change will lead to reduced household demand for all major space heating fuels, i.e. gas, oil and coal, relative to the baseline. In total household demand for gas, change in gas-based space heating is projected to fall by about 7%; oil-based space heating by 1% (out of total household demand for oil); and coal-fired heating by 17% (out of total household demand for coal). If aggregated, the demand for gas, oil, and coal for heating purposes is projected to decline by approximately 3% of total household demand for these fuels (or decline by 2% out of total energy demand for heating and cooling if electricity is included). At the regional level, changes can be more pronounced: the demand of gas, oil and coal in total household demand for these fuels is going to decrease most by 2060 in other EU countries (− 19.8%), OCE (− 17.6%), EU7 (− 13.5%), and Chile (− 12.5%). The other regions will experience falling demand for these fuels by less than 10% of household fuel demand (with 12 regions reducing demand by less than 2.5%).

While the consequences of climate change on space cooling and heating demand are captured in the model, the impacts of climate change on energy supply (such as interruptions in the availability of cooling water for thermal electricity generation due to heat extremes or droughts, the change in water availability for hydro-electricity, etc.) are not included in the analysis. The direct impacts of climate change on energy supply are not included in the analysis, but the CGE model does capture the endogenous effects on energy markets induced by the impacts on demand, and cause changes in energy prices and supply. As with other demand impacts, any endogenous shifts in demand between carriers, as a result of changing prices and income levels, are also fully captured in the model.

1.6 Tourism Demand

Changes in tourism flows reflect projected changes in tourist destinations due to changes in climate. For instance, projected decreases in snow cover in the Alps in the future might lead tourists to go skiing in other regions. These changes in the regional demand for tourism services are derived from simulations based on the Hamburg Tourism Model (Bigano et al. 2007); it thus contains only one projection and alternative models may provide different projections, especially at the regional scale. However, this approach has been amply used in EU research projects and in previous applications in CGE models (Berrittella et al. 2006 and Bigano et al. 2008). The Hamburg Tourism Model is an econometric simulation model that projects domestic and international tourism by country. The share of domestic tourists in total tourism depends on the climate in the home country and on per capita income. Climate change is represented in this simple approximation by annual mean temperature. A number of other variables, such as country size, are included in the estimation, but these factors are held constant in the simulation. International tourists are allocated to all other countries on the basis of a general attractiveness index, climate, per capita income in the destination countries, and the distance between origin and destination. Total tourism expenditure is then calculated multiplying the number of tourists times an estimated value of the average individual expenditure.

The model is based on three econometrically estimated equations. One equation for arrivals (Eq. 3) and two for departures (Eqs. 4 and 5) define the core of the model. In these equations the variables are, respectively:

A:

Total arrivals per year

G:

Land area (km2)

T:

Annual average temperature (C$^{\circ }$)

C:

Length of coastline (km)

Y:

Per capita income

D:

Total departures per year

P:

Population (in thousands)

B:

The number of countries with shared land borders

H:

Total domestic tourist trips per year

D:

The destination country

O:

The origin country

Arrivals are defined by:

$$\begin{aligned} \ln A_d= & {} \mathop {5.97}\limits _{0.97} +\mathop {2.05}\limits _{0.96} \cdot 10^{-7}G_d +\mathop {0.22}\limits _{0.07} T_d -\mathop {7.91}\limits _{2.21} \cdot 10^{-3}T_d^2 +\mathop {7.15}\limits _{3.03} \cdot 10^{-5}C_d +\mathop {0.80}\limits _{0.09} \ln Y_d \nonumber \\ N= & {} 139;R_{adj}^2 =0.54 \end{aligned}$$

(3)

Departures are determined following a two-step procedure: firstly the number of tourists “generated” by a country is defined, then it is subdivided between domestic tourists and those who travel abroad. The number of tourists that a country generates depends on the size of the population and on average income. The share of domestic tourists in total tourism depends on the climate in the home country and on per capita income. Missing observations are completed using two regressions. Total tourist numbers, D+H, where H is the number of domestic tourists are interpolated using:

$$\begin{aligned}&\ln \frac{D_o +H_o}{P_o}=\mathop {-1.67}\limits _{0.83} +\mathop {0.93}\limits _{0.10} \ln Y_o \nonumber \\&N=63;R_{adj}^2 =0.60 \end{aligned}$$

(4)

The ratio of domestic to total holidays was interpolated using

$$\begin{aligned} \ln \frac{H_o }{D_o +H_o }= & {} \mathop {-3.75}\limits _{1.19} +\mathop {0.83}\limits _{0.42} \cdot 10^{-1}\ln G_o +\mathop {0.93}\limits _{0.30} \cdot 10^{-1}\ln C_o \nonumber \\&+\mathop {0.16}\limits _{0.32} \cdot 10^{-1}T_o -\mathop {0.29}\limits _{1.11} \cdot 10^{-3}T_o^2 \nonumber \\&+\left( {\mathop {0.16}\limits _{0.12} -\mathop {4.43}\limits _{1.24} \cdot 10^{-7}Y_o } \right) \ln Y_o \nonumber \\ N= & {} 63;R_{adj}^2 =0.36 \end{aligned}$$

(5)

The model is calibrated to 1995 data. Climate is proxied by the annual mean temperature. A number of other variables such as country size, were included in the estimation, but these factors are held constant in the simulation. International tourists are allocated to all other countries on the basis of a general attractiveness index, climate, per capita income in the destination countries, and the distance between origin and destination. Again, other explanatory variables were included in the regression for reasons of estimation efficiency, but these are held constant in the simulations. The number of international tourists to a country is the sum of international tourists from the other 206 countries. Total tourism expenditure is then calculated multiplying the number of tourists times an estimated value of the average individual expenditure.

In the ENV-Linkages model, climate damages from tourism have been modelled by modifying the quality of tourism services in different regions. In contrast to Bosello et al. (2012), changes in demand for tourism are not forced changes in household expenditures, but they are rather induced by a change in the quality of the service provided. Such quality changes are represented in the model as a change in total factor productivity of the tourism services sector. Changes in tourism expenditures are largely a shift between countries, plus a shift from tourism to other commodities in domestic consumption, as changes in domestic tourism flows do not affect economy-wide expenditures. On balance, global expenditures on tourism are decreasing, implying a net negative impact on the global economy. The main reason for the reduction of global tourism expenditures is that the average quality of tourism services goes down, and to some extent consumers respond to the increase price of tourism to shift towards other consumption categories.

The regional effects are crucial for tourism, as some countries are negatively and others positively affected. The countries with the highest gains in tourism expenditure by 2060, expressed as percentage change with respect to the baseline scenario, are Canada ($+$ 92%), Russia ($+$ 66%) and the United States ($+$ 21%). Smaller positive impacts (of around $+$ 10%) also occur in Chile and Japan. The largest negative impacts instead take place in Latin America, excluding Chile and Brazil (− 27%), Mexico (− 25%), as well as Africa, excluding South Africa, China, South-East Asia and developing countries in Asia (with impacts around − 20%). Smaller negative impacts take place in South Africa (− 14%), Indonesia (− 13%) the EU OECD regions (− 9%) and the Caspian region (− 7%). Other countries have much smaller impacts.