FOOD POLICY REPORT LYKKE E. ANDERSEN, CLEMENS BREISINGER, LUIS CARLOS JEMIO, DANIEL MASON-D’CROZ, CLAUDIA RINGLER, RICHARD ROBERTSON, DORTE VERNER, AND MANFRED WIEBELT Prospects for 2050 in Brazil, Mexico, and Peru CLIMATE CHANGE IMPACTS AND HOUSEHOLD RESILIENCE APRIL 2016 Climate Change Impacts and Household Resilience Prospects for 2050 in Brazil, Mexico, and Peru Lykke E. Andersen, Clemens Breisinger, Luis Carlos Jemio, Daniel Mason-D’Croz, Claudia Ringler, Richard Robertson, Dorte Verner, and Manfred Wiebelt A Peer-reviewed Publication International Food Policy Research Institute Washington, DC ABOUT IFPRI The International Food Policy Research Institute (IFPRI), established in 1975, provides research-based policy solutions to sustainably reduce poverty and end hunger and malnutrition. The Institute conducts research, communicates results, optimizes partnerships, and builds capacity to ensure sustainable food production, promote healthy food systems, improve markets and trade, transform agriculture, build resilience, and strengthen institutions and governance. Gender is considered in all of the Institute’s work. IFPRI collaborates with partners around the world, including development implementers, public institutions, the private sector, and farmers’ organizations. Copyright ©2016 International Food Policy Research Institute. All rights reserved. For permission to reprint, contact ifpri-copyright@cgiar.org. Editor: Sandra Yin Design: David Popham Layout: Deirdre Launt and Julia Vivalo Cover photo: Panos/D. Telemans ISBN: 978-0-89629-581-0 DOI: http://dx.doi.org/10.2499/9780896295810 mailto:ifpri-copyright@cgiar.org http://dx.doi.org/10.2499/9780896295810 CONTENTS Tables, Figures, Boxes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . iv Acknowledgments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . vii EXECUTIVE SUMMARY . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . viii Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1 Modeling Suite . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3 Socioeconomic Impacts of Climate Change in Brazil . . . . . . . . . . . . . . . . . . . . . . . 20 Socioeconomic Impacts of Climate Change in Mexico . . . . . . . . . . . . . . . . . . . . . . . 32 Socioeconomic Impacts of Climate Change in Peru . . . . . . . . . . . . . . . . . . . . . . . . . . . 42 Summary, Conclusions, and Proposed Actions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53 Appendix: Supplementary Tables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56 Notes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73 References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 76 About the authors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 84 TABLES Table 1 Brazil: Projected annual crop yield changes (percentage per year), 2000–2050 ....................9 Table 2 Mexico: Projected annual crop yield changes (percentage per year), 2000–2050 ...............10 Table 3 Peru: Projected annual crop yield changes (percentage per year), 2000–2050 ..................... 11 Table 4 Summary of GDP, population, and GDP per capita assumptions for Shared Socioeconomic Pathway Number 2, by region ..............................................................................13 Table 5 Country model characteristics ............................................................................................................. 17 Table 6 Brazil: Agricultural value-added by region and agricultural trade orientation, 2008 .......... 21 Table 7 Brazil: Distribution of households by gender and location (percentage of all households), 2008 .............................................................................................................. 30 Table 8 Brazil: Per capita household income, by household type (reais per month per person), 2008.................................................................................................................. 30 Table 9 Brazil: Household Income Diversification Index, by household type, 2008 .......................... 31 Table 10 Brazil: Probability of being highly vulnerable, by household type (percentage), 2008 .... 31 Table 11 Brazil: Probability of being highly resilient, by household type (percentage), 2008 ........ 31 Table 12 Mexico: Agricultural value-added by region and agricultural trade orientation, 2008 ....33 Table 13 Mexico: Distribution of households by gender and location, 2008 ......................................... 41 Table 14 Mexico: Per capita household income, by household type (pesos per month per person), 2008 ................................................................................................................. 41 Table 15 Mexico: Household Income Diversification Index, by household type, 2008 ..................... 41 Table 16 Mexico: Probability of being highly vulnerable, by household type (percentage), 2008 . 41 Table 17 Mexico: Probability of being highly resilient, by household type (percentage), 2008 ...... 41 Table 18 Peru: Agricultural value-added by region and agricultural commodity, 2008 ....................45 Table 19 Peru: Distribution of households by gender and location (percentage of all households), 2008 .............................................................................................................. 50 Table 20 Peru: Per capita household income, by household type, (nuevos soles per month per person), 2008 ................................................................................................... 50 Table 21 Peru: Household Income Diversification Index, by household type (percent), 2008 ...... 50 Table 22 Peru: Probability of being highly vulnerable, by household type (percent), 2008 ............. 51 Table 23 Peru: Probability of being highly resilient, by household type (percent), 2008 .................. 51 Table 24 Summary of results: Climate-change induced changes in agricultural price ..................... 54 Table A1 Dynamic Computable General Equilibrium model variables and parameters ...................56 Table A2 Full Dynamic Computable General Equilibrium model equations .....................................58 Table A3 Brazil: Economic structure in base year (percent), 2008. ....................................................... 64 iv Table A4 Mexico: Economic structure in base year (percent), 2008 .......................................................66 Table A5 Peru: Economic structure in base year (percent), 2002 .......................................................... 68 Table A6 Brazil: Summary of results ..................................................................................................................69 Table A7 Mexico: Summary of results ...............................................................................................................69 Table A8 Peru: Summary of results ................................................................................................................... 70 Table A9 Sensitivity of macroeconomic results to income elasticity variation (percentage point deviation of annual average growth rates) ................................................................... 71 FIGURES Figure 1 Schematic overview of the modeling suite ........................................................................................4 Figure 2 Comparing CO2 concentration and radiative forcing assumptions ......................................... 5 Figure 3 Projected changes in average annual maximum temperatures (in degrees Celsius) in Brazil, Mexico, and Peru for four downscaled climate scenarios .........................................................6 Figure 4 Projected changes in total annual precipitation in Brazil, Mexico, and Peru for four climate scenarios .................................................................................................................... 7 Figure 5 The Impact System of Models .............................................................................................................12 Figure 6 Global food price scenarios by climate model, 2010–2050 ........................................................14 Figure 7 Agroecological zones of Brazil, Mexico, and Peru ........................................................................15 Figure 9 Brazil: Impacts of global agricultural price changes on net present value of agricultural GDP, by product group ............................................................................................................22 Figure 10 Brazil: Impacts of global agricultural price changes on net present value of agricultural GDP, by region ...........................................................................................................................23 Figure 11 Brazil: Impacts of global agricultural price changes on net present value of GDP, by sector................................................................................................................................................... 24 Figure 12 Brazil: Impacts of global agricultural price changes on household welfare, by income decile .................................................................................................................................................... 24 Figure 13 Brazil: Impacts of local yield changes on net present value of agricultural GDP, by region ...................................................................................................................................................................25 Figure 14 Brazil: Impacts of local yield changes on household welfare .................................................. 26 Figure 15 Brazil: Combined impacts of climate change on net present value of sectoral and total GDP .........................................................................................................................................27 Figure 16 Brazil: Combined impacts of climate change on net present value of agricultural GDP, by region ................................................................................................................................27 Figure 17 Brazil: Combined impacts of climate change on net present value of agricultural GDP, by product group ............................................................................................................... 28 Figure 18 Brazil: Combined impacts on household welfare, by region ...................................................29 Figure 19 Brazil: Combined impacts on household welfare, by income decile .....................................29 v Figure 20 Mexico: Impacts of global agricultural price changes on net present value of agricultural GDP, by region .......................................................................................................................... 34 Figure 21 Mexico: Impacts of global agricultural price changes on net present value of agricultural GDP, by product group ...........................................................................................................35 Figure 22 Mexico: Impacts of global agricultural price changes on net present value of household welfare, by region .........................................................................................................................36 Figure 23 Mexico: Impacts of global agricultural price changes on net present value of household welfare, by household head’s gender ......................................................................................36 Figure 24 Mexico: Impacts of local yield changes on net present value of agricultural GDP, by region ...........................................................................................................................37 Figure 25 Mexico: Impacts of local yield changes on net present value of total GDP, by product group ..........................................................................................................................38 Figure 26 Mexico: Combined impacts of global price changes and local yield changes on net present value of agricultural GDP, by region ...................................................................39 Figure 27 Mexico: Combined impacts of global price changes and local yield changes on net present value of total GDP, by product group ..................................................................39 Figure 28 Mexico: Combined impacts of global price changes and local yield changes on net present value of household welfare, by income decile .................................................. 40 Figure 29 Mexico: Combined impacts of global price changes and local yield changes on net present value of household welfare, by household head’s gender .............................. 40 Figure 30 Peru: Impacts of global agricultural price changes on the net present value of agricultural GDP, by gender of household head .......................................................................... 44 Figure 31 Peru: Impacts of global agricultural price changes on sectoral GDP, by product group ................................................................................................................................................... 44 Figure 32 Peru: Impacts of global agricultural price changes on household welfare, by location ............................................................................................................................................................... 46 Figure 33 Peru: Impacts of local yield changes on agricultural GDP, by region .................................. 46 Figure 34 Peru: Impacts of local yield changes on agricultural GDP, by product group ...................47 Figure 35 Peru: Impacts of local yield changes on household welfare, by location............................. 48 Figure 36 Peru: Combined impacts of climate change on total GDP ..................................................... 48 Figure 37 Peru: Combined impacts of climate change on agricultural GDP, by product .................49 Figure 38 Peru: Combined impacts of climate change on household welfare, by location ...............49 BOXES Box A1 Description of selected equations relevant for model simulations ................................................ 60 Box A2 Country-specific details of Dynamic Computable General Equilibrium models .................... 61 vi ACKNOWLEDGMENTS This report has been produced under a project commissioned by the Office of Evaluation and Oversight (OVE), Inter-American Development Bank (IDB) and was carried out with the International Food Policy Research Institute (IFPRI). The authors thank Joaquim Bento de Souza Ferreira Filho at the University of São Paulo for contributing the Brazil social accounting matrix and are grateful for the research assistance provided by Marcelo Cardona at the Institute for Advanced Development Studies. The authors also thank Teunis van Rheenen of IFPRI for his valuable contributions to this project. The team also thanks Channing Arndt, Alvaro Calzadilla Rivera, John Nash, and David Suarez for their valuable comments and gratefully acknowledges financial support from OVE, which made this report possible. We also thank IFPRI’s Publication Review Committee and two anonymous reviewers for their constructive comments, which helped to further improve the report. The findings, interpretations, and conclusions expressed in this paper are entirely those of the authors. They do not necessarily represent the view of IDB, its executive directors, or its client countries. vii This food policy report is a response to growing concerns about the impacts of climate change on Latin American economies, agriculture, and people. It assesses both local and global effects of changing agricultural yields on the economy, subnational regions, and different household types, including male- and female-headed households in Brazil, Mexico, and Peru. The three countries reflect economic and geographic diversity in Latin America and more than half of the region’s population. MAIN FINDINGS Climate change impacts tend to be relatively small at an economywide level in all three countries. However, sectoral and household-level economic impacts tend to be diverse across countries and subnational levels. They mainly depend on projected changes in agricultural yields, the share of agriculture in regional gross domestic product (GDP), crop-specific international trade balances, net food buyer/seller position, and income diversification of households. As for gender, results from this study suggest that female- headed households may be less vulnerable than male-headed households to the effects of climate change, highlighting the importance of considering women as a source for solutions for building resilience to climate change. Given the relatively small impacts of climate change and the degree of uncertainty associated with them, it is too early to define specific policy recommendations. POLICY IMPLICATIONS All three countries should try to maximize the benefits that may come with higher agricultural world market prices and to minimize the losses from reductions in agricultural yields. EXECUTIVE SUMMARY viii Introduction Climate change affects countries through multiple channels and sectors. The most climate-sensitive sector is agriculture, as rising temperatures and changes in rainfall patterns affect agricultural yields of irrigated and dryland crops (Kang et al. 2009; Mendelsohn and Dinar 2009). Countries that are more dependent on rainfed agriculture, including many in Africa south of the Sahara (SSA), are more vulnerable to a changing climate with projected large losses in their national output (Arndt et al. 2012). But even countries with a larger share of irrigated land, including many Arab countries, are projected to be hit hard by adverse, local impacts of climate change (Wiebelt et al. 2013; Wiebelt et al. 2015). Furthermore, the sum of worldwide adverse climate change effects on agriculture is expected to have strong negative implications for global food supply, trade flows, and commodity prices (Parry et al. 2004; Nelson et al. 2010). Accounting for changing global food prices is therefore an important part of the expected impact at the country level. Depending on the net trade position of a country and the net food producing and consuming status of households, socioeconomic impacts will likely differ. For example, coun- tries that are heavily dependent on food imports, especially those in the Arab world, are particularly hard hit by rising global food prices (Breisinger et al. 2011). Climate change impacts may also differ by gender. The literature suggests that rural women in developing countries are among the most vulnerable groups (Lambrou and Piana 2006; IPCC 2013) because they are responsible for the most climate- sensitive activities, such as water collection and fuelwood collection and, in many instances, agricultural activities (Byrne and Baden 1995; Denton 2009). Within these climate-sensitive activities gendered differences in access to water, land, and resources often exist (Sachs 1996; UN Women Watch 2009; Ringler et al. 2014). A series of previous studies on the potential economic impacts of climate change in Latin America and the Carib- bean (LAC) up to 2100, coordinated by the Economic Commission for Latin America and the Caribbean (CE- PAL), found quite modest climate change impacts on the different economies in Latin America. Results for the study for Mexico indicated that GDP in the scenario with strong climate change was on average only about 0.1 percent 1 lower than the no climate change scenario (Galindo 2009). Similar modest impacts for Mexico were indicated by the econometric estimations presented in Andersen and Verner (2010). The next study in the CEPAL series, from Central America, found adverse impacts that averaged about 0.4 percent of GDP for the rest of the century (CEPAL 2010). The most adverse impacts found in this series were for Bolivia, which averaged up to 4.8 percent of GDP during the rest of the century (BID-CEPAL 2014). None of these studies, however, took into account that climate change in the rest of the world may have a significant effect on well-being in Latin American countries through changes in international commodity prices and trade patterns. Thus, a major contribution of this study is to incorporate these global price effects, compare them with local yield effects, and incorporate both in a general equilibrium framework to visualize how countries are likely to react to these changes. The trend toward urbanization in LAC, as elsewhere, means that fewer people are living off and producing food on the land. Overall in LAC, 15.9 percent of the labor force is directly employed in agriculture, but there is large varia- tion throughout the region: 15.3 percent, 13.4 percent, and 25 percent of the total workforce is employed in agriculture in Brazil, Mexico, and Peru, respectively.1 Hence, climate change is expected to affect fewer households directly through agricultural income change and potentially more households through the indirect effects of climate change- induced fluctuations in food prices. To quantitatively assess these direct and indirect economic effects of climate change, this food policy report uses an innovative model- ing suite to address four specific types of impacts, namely (1) impacts from increased world prices of agricultural products due to global climate change, (2) impacts of local yield changes due to local changes in temperatures and precipitation,2,3 (3) combined impacts following from these two impacts as the economies adjust. The report applies four climate models to illustrate the variability of results for three case-study countries: Brazil, Mexico, and Peru. These three countries together cover more than half of LAC’s total population. The time period covered is the next four decades, to 2050. The report distinguishes itself from previous climate change studies addressing impacts in LAC in three ways. First, it addresses not only the local impacts of chang- ing crop yields due to climate change but also the global impacts from changing crop yields in the rest of the world. Second, it takes into account the indirect effects that these local and global effects on agriculture will have on the rest of the economy through factor reallocation, changes in the cost of intermediate inputs, and final consumption levels. Third, this study includes a gender-differentiated analysis, which has previously been largely absent in the top-down climate change impact literature. The gender dimension was fully included in the analysis of Mexico, as households in the Social Accounting Matrix and computable general equilibrium (CGE) model were not only disaggregated by region, level of education, and level of income, but also by the gender of the heads-of-household. Since gender impacts found in the CGE modeling framework, however, were insignificant for the case of Mexico, we used the framework of Andersen and Cardona (2013) to conduct a household survey-based analysis of vulnerability in all three countries. This complementary analysis did not find any evidence that female-headed households are more vulnerable than male- headed households. Indeed, in all three countries analyzed, female-headed households have both higher and more diversified incomes, which would tend to make them more resilient than male-headed households. 2 CLIMATE CHANGE IMPACTS AND HOUSEHOLD RESILIENCE Modeling Suite because the new conditions or policies have never been observed. The models and results presented in this report rely on structural relationships and explicit causal chains and follow the structural models described and used in a special issue of Review of Development Economics (Arndt et al. 2012) and in Verner and Breisinger (2013). More specifically, the major components of the modeling framework em- ployed in this study are the downscaling and debiasing of global climate scenarios, a crop model, a global agricultural economy model, and countrywide economic models. As shown in Figure 1, the downscaled global climate model scenarios feed precipitation and minimum and maximum daily temperatures into the Decision Support System for Agrotechnology Transfer (DSSAT), which, in turn, gener- ates changes in yields for both rainfed and irrigated crops in the three economies. Yield changes are communicated from DSSAT to both the International Food Policy Research Institute’s (IFPRI’s) International Model for Policy Analysis of Agricultural Commodities and Trade (IMPACT) and to countrywide dynamic computable general equilibrium (DCGE) models. Changes in world food prices derived from IMPACT are communicated to the DCGE model to assess the impacts of climate change on the economic struc- ture and consumption of representative household groups. Only in the case of Mexico are household groups differenti- ated by the gender of the head-of-household in the DCGE model, so we complement the gender analysis by calculat- ing gender-differentiated vulnerability and resilience indica- Projections of changes in precipitation and temperature are typically incorporated into biophysical modeling systems to ultimately assess economic impacts through so-called integrated assessment models (IAM). Tol and Fankhauser (1998) provide an overview of these reduced-form models. While these models have the advantages of being easy to use and of providing a first order estimate of empirical impact, they also have serious disadvantages. In particular, they lump a long causal chain of events into a simplified algebraic relationship. If this causal chain of events naturally evolves through time or is changed deliberately by policy, the only options for capturing these effects is through change in the parameters. Unfortunately, the empirical basis for these changes is often lacking precisely 3 tors using household survey data as suggested by Andersen and Cardona (2013). OVERVIEW OF RECENT CLIMATIC CHANGES IN BRAZIL, MEXICO, AND PERU Knowledge about climate change in LAC has increased significantly over the past 5–10 years as climate observa- tions have become more accessible (IDB 2014). At least three studies have analyzed past temperature and precipita- tion trends in these three countries. Andersen, Román, and Verner (2010) analyzed climate data from May 1948 to March 2008 for 34 high-quality meteorological stations in Brazil and found that of these, 31 stations show significant warming, 3 show no significant change, and none show sig- nificant cooling.4 The authors found that the North region is warming about twice as fast as the South region and the Northeast and Centerwest regions are warming at intermediate rates. In contrast to the results for temperature, the authors found no clear tendencies with respect to precipitation. Using the same data source, Andersen and Verner (2010) ana- lyzed the data from 21 high-quality stations in Mexico during the same period and found more mixed results. Out of the 21 stations, 12 showed a significant positive trend for temperature, 3 showed a signifi- cant negative trend, and the remain- ing 6 showed no significant trend (using a 95 percent confidence level). Since individual stations are subject to idiosyncratic variations, it is necessary to average the results from several stations to get reliable trends for a region. The authors found indications that the central zone of Mexico is warming about three times faster than the coastal zones. A trend analysis reveals no systematic changes in rainfall dur- ing the 1948–2008 period as all stations except one showed no significant trend in monthly precipitation anomalies. Finally, for the case of Peru, Andersen, Suxo, and Verner (2009) analyzed similar data for 24 high-quality stations and found that 15 of these show a significant warming trend, typically by 0.2 to 0.3 ºC per decade, 4 show a signifi- cant negative trend of between –0.1 and –0.2 ºC per decade, and 5 show no significant trend. The authors found that it was not possible to establish any systematic differences between regions. They also concluded that there have been no systematic trends in precipitation in Peru during the past six decades. As we will see in the following section, the ambiguous results about past trends in precipitation carry over to the projections about future changes in precipitation. FIGURE 1 Schematic overview of the modeling suite Source: Authors’ elaboration. Global climate models downscaled and historical meteorological data Decision Support System for Agrotechnology Transfer (DSSAT) Crop Model International Model for Policy Analysis of Agricultural Commodities and Trade (IMPACT) Model Dynamic Computable General Equilibrium (DCGE) Model • Macroeconomic variables • Economic structure (agriculture and other sectors) • Incomes and expenditures of representative household groups Vulnerability and Gender Analysis • Changes in vulnerability • Gender di�erence in vulnerability and impacts 4 CLIMATE CHANGE IMPACTS AND HOUSEHOLD RESILIENCE OVERVIEW OF CLIMATE CHANGE SCENARIOS For this study, five climate change scenarios were used to simulate potential effects of climate change by 2050.5 The perfect mitigation scenario follows historical patterns with regard to temperature and precipitation and their subsequent effects on crop yields, and would be a close proxy of representative concentration pathways (RCP) 2.6, which assumes CO2 concentration levels of just 60 parts per million greater than in 2000. This scenario serves as a counterfactual to allow us to isolate the effects of climate change on agricultural and food systems. The four addi- tional future climates were drawn from results running RCP 8.5 in four earth system models (ESMs), which was used in the Intergovernmental Panel on Climate Change’s (IPCC’s) Fifth Assessment Report, the Inter-Sectoral Impact Model Intercomparison Project (ISIMIP), and the Agriculture Model Intercomparison and Improvement Project.6 RCP 8.5 was selected as it was the concentration pathway that represented the most extreme case of radiative forcing (8.5 watts per square meter by 2100), and because by 2050 both RCP 4.5 and 6.0 are still relatively close to RCP 2.6 with only 8.5 providing a significant divergence (Figure 2).7 The use of a perfect mitigation scenario (similar to RCP 2.6) as well as a more extreme scenario (RCP 8.5) provides a variety of potential climate change effects due to potential greenhouse gas buildup in the atmosphere similar to Nelson et al. (2010) and several recent AgMIP studies (Nelson et al. 2014a; and Nelson et al. 2014b). These ESM results also provide the climatic data needed to run the crop models (see section below). Williamson (1994) observed that ESMs increasing complexity leads to decreasing precision across all variables being simulated. To better handle this uncertainty a multimodel ensemble of four ESMs was used. All formed part of ISIMIP, ensuring a standard reporting of model results. The four ESMs used are the following: • GFDL-ESM2M: designed and maintained by the National Oceanic and Atmospheric Administration’s FIGURE 2 Comparing CO2 concentration and radiative forcing assumptions Source: Downloaded from the RCP Database version 2.0.5 (IIASA 2015). RCP 2.6: van Vuuren et al. (2006); van Vuuren et al. (2007); RCP 4.5: Clarke et al. (2007); Smith and Wigley (2006); Wise et al. (2009); RCP 6.0: Fujino et al. (2006); Hijioka et al. (2008); RCP 8.5: Riahi and Nakicenovic (2007). Note: RCP = representative concentration pathway, radiative forcing = the amount of extra power that is driving the climate (the amount of sunshine com- ing in minus the energy radiated back out). Radiative forcing is measured in terms of power per area: watts (joules of energy per second) passing through the space above a square meter on the Earth's surface. Greenhouse gases act like a blanket, reducing the amount of energy that can be radiated back out into space. The different RCPs represent different ideas on how this blanketing effect might change over time. Their names reflect total radiative forcing assump- tions for the year 2100. They follow different trajectories to get to 2100. The CO2 equivalent concentrations include all forcing agents. 0 200 400 600 800 1000 1200 1400 RCP RCP 2000 2010 2030 2050 2070 2090 0 1 2 3 4 5 6 7 8 9 2000 2010 2030 2050 2070 2090 8.5 6.0 4.5 3.6 8.5 6.0 4.5 3.6 MODELING SUITE 5 Geophysical Fluid Dynamic Laboratory (GFDL) (www .gfdl.noaa.gov/earth-system-model); • HADGEM2-ES: the Hadley Centre’s Global Environ- ment Model, version 2 (www.metoffice.gov .uk/research/modelling-systems/unified-model /climate-models/hadgem2); • IPSL-CM5A-LR: the Institut Pierre Simon Laplace’s ESM (http://icmc.ipsl.fr/index.php/icmc-models /icmc-ipsl-cm5); and • MIROC-ESM: Model for Interdisciplinary Research on Climate, developed by the University of Tokyo, NIES (National Institute for Environmental Studies), and JAMSTEC ( Japan Agency for Marine-Earth Science and Technology) (www.geosci-model-dev-discuss .net/4/1063/2011/gmdd-4-1063-2011.pdf). Figures 3 and 4 show how average annual maximum temperatures and total annual precipitation are projected to change between time slices 1991–2010 and 2041–2060 for the four downscaled ESMs used in this project. From Figure 3 we can see that the climate scenarios in general project higher temperature increases in the interior of each country and lower increases along the coast. MI- ROC is generally the most extreme climate scenario, with temperature increases of up to 6°C in the Amazon basin during the next 50 years. In contrast, the GFDL scenario projects more modest temperature increases, typically of 1–2 °C over 50 years. While all scenarios project temperature increases, al- though of quite different magnitudes for the same extreme RCP, there is much less agreement about precipitation changes. Figure 4 shows, for example, that MIROC projects severe drying in central Brazil (the soybean-producing re- FIGURE 3 Projected changes in average annual maximum temperatures (in degrees Celsius) in Brazil, Mexico, and Peru for four downscaled climate scenarios Source: Authors’ elaboration. Note: GFDL = Earth System Model (ESM) designed and maintained by the National Oceanic and Atmospheric Administration’s Geophysical Fluid Dynamic Laboratory, HADGEM2 = Hadley Centre’s Global Environment Model (version 2), IPSL = Institut Pierre Simon Laplace’s ESM, and MIROC = Model for Interdis- ciplinary Research on Climate. The four climate scenarios depict average annual maximum temperature changes from 2041–2060 minus 1991–2010. HADGEM26 5 4 3 2 1 Brazil Mexico Peru MIROC GFDL IPSL 6 CLIMATE CHANGE IMPACTS AND HOUSEHOLD RESILIENCE http://www.gfdl.noaa.gov/earth-system-model http://www.gfdl.noaa.gov/earth-system-model http://www.metoffice.gov.uk/research/modelling-systems/unified-model/climate-models/hadgem2 http://www.metoffice.gov.uk/research/modelling-systems/unified-model/climate-models/hadgem2 http://www.metoffice.gov.uk/research/modelling-systems/unified-model/climate-models/hadgem2 http://icmc.ipsl.fr/index.php/icmc-models/icmc-ipsl-cm5 http://icmc.ipsl.fr/index.php/icmc-models/icmc-ipsl-cm5 http://www.geosci-model-dev-discuss.net/4/1063/2011/gmdd-4-1063-2011.pdf http://www.geosci-model-dev-discuss.net/4/1063/2011/gmdd-4-1063-2011.pdf gion) while IPSL projects a substantial increase in precipita- tion in the same region. In contrast, IPSL projects severe drying of southern Mexico while GFDL projects a much wetter climate for the same region. IPSL seems to be an outlier in terms of precipitation, although it looks intermediate in terms of temperature increases. By simulating the impacts of all four climate scenarios, a range of possible results, even within the same RCP scenario, can be assessed. For the modeling of impacts of climate change on crop yields, we use not only these average trends but also the intra-annual changes projected in each scenario. DSSAT Crop Model Location-specific effects on crop productivity can be projected using process-based crop simulation models. In turn, the crop models are driven by weather, consistent with conditions indicated by the ESMs, completing the connec- tions between climate conditions at one end and economic outcomes at the other. Crop models have long been used to assess the possible effects of climate change on agriculture. One of the seminal contributions was Rosenzweig and Parry (1994) which brought together crop models run on weather data for 112 sites modified by global circulation model (GCM) outputs and passed those changes through a basic global economic model. That investigation found negative effects for global yields from climate change in the range of 10–30 percent losses from about 1990 to 2060. Converting to a yearly change, that amounts about 0.2–0.4 percent losses per year, which is comparable to the range found in this study. Many individual studies of particular crops and/or regions have FIGURE 4 Projected changes in total annual precipitation in Brazil, Mexico, and Peru for four climate scenarios Source: Authors’ elaboration. Note: GFDL = Earth System Model (ESM) designed and maintained by the National Oceanic and Atmospheric Administration’s Geophysical Fluid Dynamic Laboratory, HADGEM2 = Hadley Centre’s Global Environment Model (version 2), IPSL = Institut Pierre Simon Laplace’s ESM, and MIROC = Model for Interdis- ciplinary Research on Climate. The four climate scenarios depict total annual precipitation changes from 2041–2060 minus 1991–2010. HADGEM2 400 200 0 –200 –400 Brazil Mexico Peru MIROC GFDL IPSL MODELING SUITE 7 been performed (see White et al. 2011 for a comprehensive survey up to that time). The approach used in this research follows recent global approaches like those described in Rosenzweig et al. (2014) and Nelson et al. (2010). Crop models require a basic set of weather information to adequately represent how the plants grow and respond to their changing environment. The crop models employed here are from the DSSAT family and work on daily high temperature, low temperature, rainfall, and shortwave solar radiation reaching ground level.8 Since all those quantities need to be identified for the crop models to work, we are limited to using climate projections from ESMs that make all the different values available to outside researchers. Comparatively few of the ESM research groups provide the level of detail needed for use in crop modeling. Tempera- ture data typically are the easiest to find and are thought to be the most reliable. However, for agriculture, precipitation (and to a lesser extent, sunshine) is really where the game is played for climate change purposes. As a result, the choice of future climate circumstances is constrained by pragmatic concerns of data availability and usability. The data employed were drawn from four ESMs work- ing under RCP 8.5 conditions as described in the section above: HADGEM2, MIROC, GFDL, and IPSL. The raw ESM outputs cannot be directly input into crop models. The spatial representations are often different between models and in any case are much too coarse for agricultural purposes. Temporally, they operate at scales of minutes or hours while the crop models usually employ a daily time step (and in this case use multiple realizations of generated daily weather based on monthly averages). Hence, several steps are needed to obtain and process the data from different sources into a common spatial grid as well as temporally aggregate the fine time slices into month- ly averages. These tasks were accomplished by colleagues at the Potsdam Climate Institute resulting in maps with global coverage at ½-arc-degree spatial resolution and further aggregated temporally to monthly averages by colleagues at Columbia University. A delta-method approach adjusted the values by putting the changes between the baseline period and future periods on a common baseline. Daily weather realizations were generated by the Simulation of Meteorological variables (SIMMETEO) random weather generator based on the adjusted monthly averages.9 Finally, the crop models within DSSAT used the daily weather as inputs to determine yields under the various conditions. The crop models generate gridded maps of yields. Each crop was modeled for both rainfed and irrigated conditions under perfect mitigation as well as the four future climate situations, meaning 2 × (1 + 4) = 10 maps of yields per crop. One important issue is when each crop is planted. At the purely data level, we do not have full confidence in the accuracy of the target planting month for every location. But, at the substantive level, adjusting the planting month is a significant way for producers to adapt to changing climatic conditions. To deal with these possibilities, the target plant- ing month is used as the center of a three-month window. Each of the three cases are simulated and the highest of the three yields is chosen for each pixel. Additionally, each crop is represented by several varieties since different varieties are appropriate for different locations. So, the final yield maps for each crop and water source reflect both simplistic optimizing behavior regarding planting month and the geographic diversity of varieties. The economic models operate at a more aggregate scale, with scale depending on the particular application. To make the different scales work together, the pixel-level yields were aggregated within the appropriate regional boundaries. An area-weighted average was employed by using maps of physical area allocations by crop from the Spatial Produc- tion Allocation Model as the weighting factors (You et al. 2014; You, Wood, and Wood-Sichra 2006).10 More detailed discussions of the aggregation process and its implications can be found in Mueller and Robertson (2014) and Robert- son et al. (2013). The economic models use these regional yields to adjust the evolution of productivity through time within their frameworks. Please note, however, that these purely biophysical modeling inputs only partially reflect adaptive behavior on the part of producers. Other autono- mous adaptations are reflected through changes in demand and supply as a result of changing food prices under climate change and are partly captured in the economic models. The full extent of autonomous adaptation options is likely slightly underestimated. Tables 1–3 summarize the projected changes in crop yields for the major crops in each of the subregions in the three countries analyzed. These simulations do not include the beneficial effect of CO2 fertilization on crop yields. CO2 8 CLIMATE CHANGE IMPACTS AND HOUSEHOLD RESILIENCE TABLE 1 Brazil: Projected annual crop yield changes (percentage per year), 2000–2050 Region Maize, cassava, and sugarcane Rice Soybeans and cotton Wheat Brazil –0.39 –0.14 –0.27 –0.28 Central tropical subhumid –0.41 0.04 –0.36 –0.28 Eastern semiarid –0.34 –0.21 –0.20 –0.43 Northern tropical humid –0.37 –0.25 –0.20 –0.07 Southern humid –0.21 –0.31 0.06 0.00 Climate scenario: HADGEM2 Brazil –0.46 –0.26 –0.31 –0.56 Central tropical subhumid –0.33 –0.20 –0.26 –0.56 Eastern semiarid –0.50 –0.27 –0.32 –0.88 Northern tropical humid –0.64 –0.30 –0.35 –0.52 Southern humid –0.74 –0.40 –0.22 0.00 Climate scenario: IPSL Brazil –0.57 –0.22 –0.40 –0.36 Central tropical subhumid –0.63 –0.10 –0.59 –0.36 Eastern semiarid –0.35 –0.33 –0.24 –0.50 Northern tropical humid –0.54 –0.28 –0.27 –0.51 Southern humid –0.61 –0.36 0.15 0.00 Climate scenario: MIROC Brazil –0.45 –0.17 –0.32 –0.51 Central tropical subhumid –0.31 0.00 –0.11 –0.51 Eastern semiarid –0.47 –0.22 –0.35 –1.11 Northern tropical humid –0.67 –0.28 –0.47 –0.46 Southern humid –0.65 –0.36 –0.15 0.00 Source: Authors’ calculations based on Decision Support System for Agrotechnology Transfer. Note: GFDL = Earth System Model (ESM) designed and maintained by the National Oceanic and Atmospheric Administration’s Geophysical Fluid Dynamic Laboratory, HADGEM2 = Hadley Centre’s Global Environment Model (version 2), IPSL = Institut Pierre Simon Laplace’s ESM, and MIROC = Model for Interdisciplinary Research on Climate. fertilization was not included in this study for two reasons. First, much uncertainty surrounds the extent to which the potential benefits of CO2 may actually be achieved, as it will depend on soil constraints (Reich and Hobbie 2013; Norby et al. 2010) and future management practices and adaptation (Müller et al. 2010). CO2 fertilization may increase growth (biomass) but diminish nutritional quality in some crops such as cassava (Gleadow et al. 2009), and the lifespan of certain arboreal species (Bugmann and Bigler 2011), which could have major unexpected consequences on arboreal plantation crops like coffee and cacao. In addi- tion to this uncertainty, the purpose of choosing a perfect mitigation scenario and RCP 8.5 was to present an envelope of potential climate results. Applying CO2 fertilization in DSSAT would only decrease the climate possibility space being tested. One of Brazil’s most important agricultural products is soybeans, contributing about 22 percent and 16 percent to regional agricultural GDP in the central tropical subhumid and southern humid regions, respectively. According to the DSSAT model, soybean production in these regions would be adversely affected in all climate scenarios, but especially in the wet IPSL scenario (Table 1). MODELING SUITE 9 In Mexico, maize is the main agricultural product in the west arid, central high arid, and southern humid regions, contributing between 14 and 22 percent of regional agricul- tural GDP. However, maize yields vary significantly across regions in three of the four climate scenarios. Only for the very dry IPSL scenario do we see substantial drops in maize yields in all parts of Mexico (Table 2). In Peru, potatoes contribute more than 10 percent to ag- ricultural GDP in the highlands, and here all models predict very substantial increases in yields due to higher tempera- tures and more precipitation (Table 3). The yield changes modeled for all of these countries are within the diverse range of effects seen across the globe. For example, the annual climate change effect on soybeans rang- es from –0.5 to +0.3 percent, while the whole world average is around –0.2 percent. Brazil, at about –0.3 percent per year, is slightly harder hit. Looking at the most important crops for the other countries in this study, we find that some are in the middle of the pack while others even benefit from climate change. Across the world, the maize situation ranges between –0.8 percent and no change, with –0.5 percent for a typical worldwide value, meaning that Mexico’s roughly –0.3 percent per year is on the better side of typical. The TABLE 2 Mexico: Projected annual crop yield changes (percentage per year), 2000–2050  Region Maize and sugarcane Potatoes Sorghum Beans and alfalfa Wheat Climate scenario: GFDL Mexico –0.22 –0.10 0.01 –0.37 –0.04 West arid –0.58 –0.41 –0.09 –0.06 0.18 North and east arid –0.04 0.11 0.11 –0.46 0.12 Central high arid –0.02 0.08 –0.01 –0.28 –0.24 Southern humid –0.45 –0.13 –0.23 –0.29 –0.40 Climate scenario: HADGEM2 Mexico –0.33 0.08 –0.23 –0.37 –0.16 West arid –0.71 –0.02 –0.30 –0.38 0.08 North and east arid –0.55 0.14 –0.30 –0.37 0.10 Central high arid 0.02 0.17 –0.10 –0.36 –0.39 Southern humid –0.73 –0.03 –0.37 –0.38 –0.38 Climate scenario: IPSL Mexico –0.59 –0.32 –0.09 –0.46 –0.07 West arid –0.78 –0.94 –0.30 –0.12 0.19 North and east arid –0.44 –0.07 0.30 –0.55 0.21 Central high arid –0.38 –0.02 –0.30 –0.36 –0.33 Southern humid –0.87 –0.08 –0.59 –0.38 –0.30 Climate scenario: MIROC Mexico –0.32 –0.12 –0.22 –0.33 –0.04 West arid –0.60 –0.47 –0.20 –0.17 0.15 North and east arid –0.17 0.01 –0.38 –0.33 0.12 Central high arid –0.01 0.10 –0.03 –0.33 –0.22 Southern humid –0.73 –0.09 –0.37 –0.37 –0.33 Source: Authors’ calculations based on Decision Support System for Agrotechnology Transfer. Note: GFDL = Earth System Model (ESM) designed and maintained by the National Oceanic and Atmospheric Administration’s Geophysical Fluid Dynamic Laboratory, HADGEM2 = Hadley Centre’s Global Environment Model (version 2), IPSL = Institut Pierre Simon Laplace’s ESM, and MIROC = Model for Interdisciplinary Research on Climate. 10 CLIMATE CHANGE IMPACTS AND HOUSEHOLD RESILIENCE starting point is also important. As already mentioned, to the extent that the relatively coarse climate data can repre- sent the topographic variability in Peru, the high elevations (and concomitant lower temperatures) help to explain that a temperature increase could be beneficial in large parts of the country. The effect of climate change on potatoes range between about –0.6 and +0.2 percent around the world and the overall picture is about –0.2 percent per year. Peru’s +1.0 percent is then at the very high end of benefitting from the changing climate. Overall, the modeled responses for these three countries are not outliers in the global context, al- though they do reflect the diversity of circumstances found on our planet. The IMPACT Global Agricultural Economy Model The country-level DCGE models need guidance on the world prices at national borders. The world prices can be determined using a global model of agriculture, in this case IMPACT. IMPACT was developed at IFPRI in the early 1990s as a partial equilibrium computer simulation model focused on global agriculture. Over time this trade model has been expanded to answer a growing set of ex ante research ques- tions, such as the effects of climate change and water avail- ability on agriculture and food security globally. To respond to this growing demand, IMPACT has been redesigned as a modular network of linked economic, water, and crop models. At the core of this network are the original partial equilibrium trade model and a suite of water models. The trade model is a system of equations offering a methodology for analyzing baseline and alternative sce- narios for global food demand, supply, trade, income, and population. IMPACT simulates agricultural markets in 159 geopolitical regions (Brazil, Egypt, France, and so forth). Within each region supply, demand, and prices for agricul- tural commodities are determined, with all regions linked through trade. World agricultural commodity prices are de- termined annually at levels that clear international markets. To simulate the effects of climate change and the availability of water, a more disaggregated level is required. IMPACT uses the food production unit, which is the intersection of the 159 geopolitical regions with 126 hydrological basins, giving 320 subnational units and allowing for the modeling of water basin management and its effects on agriculture. The integration of a global economic model and a series of water models, allows for the additional modeling of changes in availability of water for irrigation, and the effects that this would have on agricultural productivity. IMPACT simu- lates 62 agricultural commodity markets, which represents the majority of food and cash crops. For more informa- tion about the IMPACT model, please see Rosegrant and IMPACT Development Team (2012) and Robinson et al. (2015). Figure 5 describes the links between the different mod- els that constitute the IMPACT system of models. All mod- els except the climate models (ESMs) are run by IFPRI. The five climate futures were run against the same socioeconomic scenario, which was drawn from IPCC’s Fifth Assessment Report. The socioeconomic scenario was defined as Shared Socioeconomic Pathway Number 2, which is characterized by economic development that more TABLE 3 Peru: Projected annual crop yield changes (percentage per year), 2000–2050 Region Maize Potatoes and root crops Cotton Cereals Climate scenario: GFDL Peru –0.18 0.61 –0.12 –0.03 Coastal –0.41 0.65 –0.12 –0.16 Interior 0.01 0.53 –0.01 0.08 Climate scenario: HADGEM2 Peru –0.27 1.03 –0.16 –0.05 Coastal –0.58 1.15 –0.17 –0.21 Interior –0.02 0.75 0.00 0.10 Climate scenario: IPSL Peru –0.25 1.03 –0.12 –0.06 Coastal –0.52 1.17 –0.12 –0.20 Interior –0.02 0.70 –0.18 0.05 Climate scenario: MIROC Peru –0.25 1.10 –0.13 –0.01 Coastal –0.50 1.27 –0.14 –0.15 Interior –0.05 0.71 –0.10 0.10 Source: Authors’ calculations based on Decision Support System for Agrotech- nology Transfer. Note: GFDL = Earth System Model (ESM) designed and maintained by the National Oceanic and Atmospheric Administration’s Geophysical Fluid Dynamic Laboratory, HADGEM2 = Hadley Centre’s Global Environment Model (version 2), IPSL = Institut Pierre Simon Laplace’s ESM, and MIROC = Model for Interdisci- plinary Research on Climate. MODELING SUITE 11 or less follows historical trends and a medium population growth projection (O’Neill et al. 2014). This socioeconom- ic scenario is analogous to, although not exactly the same as, IPCC’s FourthAssessment Report (AR4) Medium-Medium socioeconomic scenario. Table 4 summarizes the GDP and population growth rate assumptions that define Shared Socioeconomic Pathway Number 2. In general, during the next 40 years, prices of major agricultural products are projected to increase due to the increased demand of a larger and richer world population without equivalent growth in agricultural productivity. In addition to these increases, climate change is expected to affect world prices due to changes in supply. The price indexes in Figure 6 show how world prices for key products are expected to develop under different climate scenarios and a perfect mitigation scenario according to the IMPACT model. Dynamic Computable General Equilibrium (DCGE) Models Climate change affects world prices and local agricultural production (Figure 6), with diverse implications for the three economies analyzed. Moreover, spatial variation in lo- cal climate change impacts within countries means that such effects can vary across subnational regions. We therefore develop economywide models for different agroecological zones (AEZs) (Figure 7 and Table 5) to capture the major linkages between climate change, production, and house- holds. The recursive DCGE models used in this report are constructed to be consistent with neoclassical general equi- librium theory and follow the model described in Diao and FIGURE 5 The Impact System of Models Source: Authors. Note: DSSAT = Decision Support System for Agrotechnology, IMPACT = International Model for Policy Analysis of Agricultural Commodities and Trade. Hydrology— water basin management and stress models Outputs: Yields Production Water demand trends Climate models Crop models (DSSAT) IMPACT global economic model Macroeconomic trends Consumption Trade Harvested area Commodity prices 12 CLIMATE CHANGE IMPACTS AND HOUSEHOLD RESILIENCE Thurlow (2012). Recent applications of this model investi gate climate change impacts for Syria, Tunisia, and Yemen (see Verner and Breisinger 2013; Breisinger et al. 2013; Wiebelt et al. 2015). In the following, we provide a sum- mary of the key model features, with a focus on parameters that are most relevant for tracking climate change impacts. The full set of model equations can be found in Box A1 and a detailed description of the model in Diao and Thurlow (2012). To model the impact of local climate change, yield changes derived from the DSSAT models directly enter the production functions of the DCGE models. More specifi- cally and consistent with the above mentioned literature on the economics of climate change, we use the total factor productivity (TFP) parameter to exogenously impose climate change-induced crop yield changes on the model.11 Reductions in yields tend to then translate into reduced sectoral production, which exerts upward pressure on domestic prices thereby inducing producers to shift their supply toward domestic markets while consumers shift their demand toward foreign markets. To model the impact of global climate change on global food commodity markets, projections from the IMPACT model are introduced as exogenous changes in world market prices. For example, if international prices rise due to climate change, then produc- ers increase supply to international markets to maximize revenues and reduce sales to domestic markets, while consumers lower their demand for imported goods and increase it for domestically produced goods in an attempt to minimize costs of consumption. Besides these direct effects determining the initial domestic output price response in the affected agricultural markets, indirect impacts or economywide effects deter- mine the final resource shifts and final impacts on sectoral income or value added as well as factor income and house- hold income. In the DCGE model, each country faces TABLE 4 Summary of GDP, population, and GDP per capita assumptions for Shared Socioeconomic Pathway Number 2, by region GDP (Billions of USD) Population (millions) Per-capita GDP (USD per 2010 2050 Annual growth (%) 2010 2050 Annual growth (%) 2010 2050 Annual growth (%) Africa and Middle East 6,255 32,593 3.4 1,321 2,508 1.3 4,737 12,996 2.0 East Asia, Southeast Asia, and Oceania 19,277 80,411 2.9 2,216 2,337 0.1 8,699 34,408 2.8 South Asia 4,420 32,574 4.1 1,598 2,296 0.7 2,767 14,184 3.3 Former Soviet Union 2,855 8,984 2.3 279 277 0.0 10,234 32,402 2.3 Europe 14,629 27,784 1.3 537 577 0.1 27,228 48,146 1.1 Latin America and Caribbean 5,899 19,278 2.4 590 746 0.5 10,007 25,852 1.9 North America 14,289 29,929 1.5 344 450 0.5 41,490 66,526 0.9 World 67,624 231,553 2.5 6,884 9,191 0.6 9,823 25,192 1.9 Source: Authors’ compilation. Note: GDP = gross domestic product; Shared Socioeconomic Pathway No. 2 = . For GDP, all local currency converted to US dollars at 2010 prices average purchasing power parity exchange rates. MODELING SUITE 13 FIGURE 6 Global food price scenarios by climate model, 2010–2050 Source: Authors’ elaboration. Note: GFDL = Earth System Model (ESM) designed and maintained by the National Oceanic and Atmospheric Administration’s Geophysical Fluid Dynamic Laboratory, HADGEM2 = Hadley Centre’s Global Environment Model (version 2), IPSL = Institut Pierre Simon Laplace’s ESM, and MIROC = Model for Interdis- ciplinary Research on Climate. 220 180 140 100 2010 2020 2030 Year Wheat Pr ic e in de x 2040 2050 HADGEM2 MIROC GFDL IPSL NoCC 220 180 140 100 2010 2020 2030 Year Sorghum Pr ic e in de x 2040 2050 HADGEM2 MIROC GFDL IPSL NoCC 220 180 140 100 2010 2020 2030 Year Potatoes Pr ic e in de x 2040 2050 HADGEM2 MIROC GFDL IPSL NoCC 220 180 140 100 2010 2020 2030 Year Subtropical and tropical fruits Pr ic e in de x 2040 2050 HADGEM2 MIROC GFDL IPSL NoCC 220 180 140 100 2010 2020 2030 Year Co�ee Pr ic e in de x 2040 2050 HADGEM2 MIROC GFDL IPSL NoCC 220 180 140 100 2010 2020 2030 Year Maize Pr ic e in de x 2040 2050 HADGEM2 MIROC GFDL IPSL NoCC 220 180 140 100 2010 2020 2030 Year Vegetables Pr ic e in de x 2040 2050 HADGEM2 MIROC GFDL IPSL NoCC 220 180 140 100 2010 2020 2030 Year Soybeans Pr ic e in de x 2040 2050 HADGEM2 MIROC GFDL IPSL NoCC HADGEM2 MIROC GFDL IPSL NoCC 220 180 140 100 2010 2020 2030 Year Cotton Pr ic e in de x 2040 2050 HADGEM2 MIROC GFDL IPSL NoCC 220 180 140 100 2010 2020 2030 Year Sugarcane Pr ic e in de x 2040 2050 14 CLIMATE CHANGE IMPACTS AND HOUSEHOLD RESILIENCE FIGURE 7 Agroecological zones of Brazil, Mexico, and Peru Source: Authors’ elaboration. perfectly elastic world demand curves for its exports at fixed world prices assuming that all countries receive the same prices for similar goods and services. However, this “small country assumption” is relaxed by the fact that the results of the IMPACT model take into account world market price variations resulting from producers that are positively or negatively affected by climate change. This is important for countries like Brazil, which is a major exporter of soybeans and coffee. The model distinguishes between various institu- tions, including enterprises, the government, and different household groups. Households and enterprises receive income in payment for the producers’ use of their factors of production. Institutions pay direct taxes and save according to their respective marginal saving propensities. Enterprises pay their remaining incomes to households in the form of dividends. Households use their incomes to consume commodities according to fixed budget shares as derived from a Cobb-Douglas utility function. The government receives revenue from activity taxes, sales taxes, direct taxes, and import tariffs and then makes transfers to households, enterprises, and the rest of the world. The government also purchases commodities (actually remuneration for the provision of public goods) in the form of government consumption expenditures, and the government saves the remaining income (with recurrent budget deficits repre- senting negative savings). All savings from households, enterprises, government, and the rest of the world (foreign savings) are collected in a savings pool from which invest- ment is financed. The model includes three macroeconomic accounts: government balance, a current account, and a savings- investment account. To balance the macro accounts, it is necessary to specify a set of macro-closure rules, which provide a mechanism through which balance is achieved. A savings-driven investment macro closure is assumed such that investment is endogenously determined by the sum of private, public, and foreign savings. Private savings are assumed to be fixed proportions of net enterprise and household income. In the government account, the fiscal balance and therefore public savings are endogenous, with government demand fixed and all tax rates held constant, so that government savings or dis-savings depend on the level of economic activity. Finally, for the current account, both Brazil Mexico Peru West arid North and east arid Central high arid Southern humid Coast Inland Northern tropical humid Central tropical subhumid Southern humid Eastern semiarid MODELING SUITE 15 foreign savings in foreign currency terms and the nominal exchange rate are assumed to be fixed, while the domestic consumer price index adjusts to reach overall equilibrium. Several labor and capital categories in the model are dif- ferentiated according to country characteristics (Table 5). All types of labor are assumed to be fully employed and mo- bile across sectors. Capital is assumed to be sector specific within periods but mobile over time. New capital from past investment is allocated to sectors according to profit rate differentials under a “putty-clay” specification. This means that once capital stocks have been invested, it is difficult to transfer them to other uses. In agriculture, cultivated land in each AEZ is assumed to be mobile and can be reallocated across crops in response to shocks.12 Thus, changes in crop production could result from changes in yields, intensifi- cation, and land use patterns. In the Peruvian model, all factors, including labor types and capital, are intersectorally mobile but interregionally immobile (Table 5). Long-run sectoral factor productivity growth is speci- fied exogenously. Within the DCGE model, the decisions of consumers, producers, and investors change in response to changes in economic conditions driven by different sets of climate outcomes and market outcomes. The model allows a degree of endogenous adaptation within periods, with changes in labor and land allocation across sectors and crops in response to shocks. The DCGE models for Brazil, Mexico, and Peru are specifically built to capture the economic and distributional effects of climate change in these three Latin American countries. Since global and local climate change affects dif- ferent crops differently, the model captures both the sectoral and the spatial heterogeneity of crop production and its linkages to other sectors such as food processing, manufac- turing, and services. Wiebelt et al. (2015) provides a more detailed descrip- tion of the essential features of the DCGE model and a discussion about how structural features and crucial param- eters affect the direct and indirect effects of world market price changes and agricultural yield changes on resource allocation as well as functional and socioeconomic income distribution for the case of Tunisia. GENDER AND RESILIENCE At first glance, climate change and all its associated ef- fects, such as temperature increases, changes in rainfall, sea level increase, and glacier retreat, do not seem to affect women and men differently. However, there are complex and dynamic links between gender relations and climate change (Terry 2009). The literature generally finds that rural women in developing countries are among the most vulnerable groups (for example, Lambrou and Piana 2006) because they often are responsible for the most climate- sensitive activities, such as water and fuelwood collection and sometimes agriculture (Byrne and Baden 1995; Denton 2009; Scheurlen 2015); although men often are mostly responsible for agricultural activities. Moreover, there are gendered differences in access to water, land, and resources (Sachs 1996; UN Women Watch 2009; Ringler et al. 2014). One of the reasons women are left with the more climate- sensitive activities is gendered differences in labor market access (Buechler 2009) and mobility (UN Women Watch 2009). Research on mobility and migration patterns suggests that in times of disaster and stress, such as those that might arise from climate change, men tend more often to migrate from rural areas than women and girls. The latter tend to stay behind. The resulting increased work burdens, specifi- cally related to agriculture, may make it difficult for women and girls to continue their existing income-generating activities, let alone take on additional work (Denton 2009). The fact that women and girls are often responsible for most of the unpaid care tasks around the household also means their lives are directly affected by the changes brought about by climate change. They often have to walk farther to find increasingly scarce food, fuel, and water as well as care for family members who are susceptible to the health risks linked to climate change. For example, climate change is expected to cause more extreme precipitation pat- terns, with more droughts and more floods (IPCC 2013), and in both cases it is typically women who have to work harder to obtain water during droughts and deal with the in- creased disease burden caused by floods (Denton 2009). As a result, women and girls find themselves with less time for education, income-generating activities, and participation in community decisionmaking processes, further entrenching unequal gender relations (Skinner 2011). There also may be 16 CLIMATE CHANGE IMPACTS AND HOUSEHOLD RESILIENCE gendered differences in spending patterns. Households that spend a large share of their income on food may be particu- larly vulnerable to food price increases brought about by climate change (FAO 2011). However, gender roles and relations are highly context specific and therefore must be studied and addressed based on local contexts (Verner 2012). To complement the simulations by a direct vulnerability analysis of household survey data from the three countries, we use the vulnerability indicators proposed by a recent Inter-American Development Bank study (Andersen et al. 2014). This report argues that although vulnerability is a complex concept, it can be quantified and analyzed at the household level using just two main indicators: (1) per capita household income and (2) household income diver- sification. The most vulnerable households are those that simultaneously have low levels of income and low levels of diversification because any shock could threaten their entire income base. In contrast, households with high and well- diversified incomes will be much better able to withstand any adverse shock. Andersen and Cardona (2013) develop a simple typology of vulnerability based on these two indicators. A household that has a per capita income below the national poverty line and a diversification index (DI) less than 0.5 is classified as highly vulnerable, while a household that has a per capita income above the poverty line and a DI greater than 0.5 is classified as highly resilient (Figure 8). Andersen and Cardona (2013) argue that since diversi- fication is the opposite of income concentration; a simple and logical way of constructing a DI is simply 1 minus the widely used Herfindahl-Hirschman Index of Concentration: DI = 1 –∑i=1 pi (1) where N is the total number of income sources and pi represents the income proportion of the ith income source. The value of the index is 0 when there is complete special- ization (100 percent of total household income comes from one source only) and approaches 1 as the number of income sources increases and no single source dominates household incomes. The advantage of using the DI instead of just the number of livelihood sources is that the DI is not very sensitive to the grouping of small income sources with bigger ones. For example, if a household had three sources, contributing 90 percent, 9 percent, and 1 percent, respectively, the DI would be 0.1818. If we lump together the last two sources, the index changes only marginally to 0.1800. This is a reduction of less than 1 percent in the index, whereas the reduction in number of livelihood sources would be 33 percent. This property of robustness to alternative classifications is impor- tant as we will necessarily have to make some assumptions about how to classify and group different income sources together (Andersen and Cardona 2013). In principle, one should define sources so that there is very low correlation TABLE 5 Country model characteristics Country Sectors and commodities Agroecological zones (AEZs) Factors Households Brazil 37 sectors, 19 nonagriculture, 18 agriculture, all differentiated by 4 AEZs 4 AEZs: northern tropical humid (4.1% of total agricultural GDP), central tropical subhumid (34.3%), southern humid (47.5%), eastern semiarid (14.1%) 10 labor types by wage level, mobile across sectors and AEZs; sector- specific capital; 4 land by AEZ 40 household types by AEZ and income deciles Mexico 39 sectors, 20 nonagriculture, 19 agriculture differentiated by 4 AEZs 4 AEZs: west arid (36.2%), north and east arid (18.6%), central high arid (21.8%), southern humid (23.4%) 6 labor types by skill category and male/female, mobile across sectors and AEZs; sector-specific capital; 4 land by AEZ 80 household types by AEZ, gender, and income deciles Peru 36 sectors, 21 nonagriculture, 15 agriculture, all differentiated by 2 AEZs 2 AEZs: Coast (31.8%) and Inland (68.2%) 3 labor types by skill category, mobile across sectors; sector-specific capital; 2 land by AEZ 4 household types, rural and urban by AEZ Source: Country social accounting matrixes. Note: AEZ = agroecological zone, GDP = gross domestic product. N 2 MODELING SUITE 17 across states of nature. Thus, if both the husband and the wife are engaged in subsistence agriculture, that would count as only one income source because adverse climatic or market conditions would affect both in a similar way. If they also had some cattle, they would count as an additional income source, as cattle and agricultural productivity are not strongly correlated. Indeed, cattle are often used as an insurance mechanism in Latin America. In practice, the exact classification of sources will depend on the amount of detail available in the household surveys of each country. Thus, while the index can be compared across groups within the same country, it is more difficult to compare across countries. Since these two indicators can be calculated for ev- ery household in each of the countries analyzed, we can compare the probability of falling into the highly vulnerable corner for any group of interest, including female- and male- headed households. The methodology, however, does not permit the analysis of intrahousehold gender differences in vulnerability. This is a potential weakness, as there may well be intrahousehold differences that the household-level analysis cannot capture (IADS 2014). There are multiple gender-specific vulnerabil- ities related to differing familial and community obligations and life cycle events (for example, childbirth, childcare, lower social status, lower access to and control over assets, and mobility constraints), and these shape opportunities for men, women, and children to build resilience and constrain the coping strategies they employ to manage risk (Holmes and Jones 2013). INTERTEMPORAL DISCOUNT FACTOR AND REPORTING OF RESULTS To summarize the results of annual climate change impacts over the next 40 years, we calculate the net present value (NPV) of the impacts and compare them to the NPV of the relevant variable in the perfect mitigation scenario. This will give us an average measure of impact during the 40 years, but the impacts will typically be smaller than average in the beginning of the period, when the climate has not yet changed much. Impacts will be larger than average by the end, when the full climate changes illustrated in Figures 3 and 4 have occurred. In order to calculate the NPV figures of all relevant variables, we apply a very low inter- temporal discount rate of 0.5 percent. Using a deliberately low discount rate is justified based on a worst-case hypothesis. The findings presented in this report show that the climate-change effects are small using a low discount rate. They would be even smaller if a higher discount rate were used. In other words, the overall conclusion of this report holds even if other (higher or lower) discount rates were used. Three indicators are used to assess the impacts of climate change on total GDP, agricultural GDP, and households’ welfare. The three indicators are the following: FIGURE 8 Four main vulnerability types Cuto�: National poverty line C: Poor but resilient A: Highly vulnerable B: Highly but resilient A: Rich but vulnerable Cuto�: DI = 0.5 Per capita household income D iv er si �c at io n in de x Source: Andersen and Cardona (2013). Note: DI = Diversification Index. 18 CLIMATE CHANGE IMPACTS AND HOUSEHOLD RESILIENCE Accumulated Absolute Value of Losses Indicator 1 was calculated following three steps. First, simulation outputs of the model present figures for all three variables expressed in constant-value local currency for all scenarios. These figures were converted into constant-value US dollars by dividing the whole series by the exchange rate in the base year. Second, aggregated net present values of the variables were calculated for each of the scenarios by adding up the string of discounted values of the variable for the whole time horizon of the simulation. Finally, the indicator was calculated as the deviations of the aggregated discounted values for each of the scenarios from those cor- responding to the no climate change scenario. I1 = Σ VJi – Σ Vncci , where I1 = indicator 1, VJi = value of variable V in climate change scenario j and in year i, and Vncci = value of variable V in no climate change sce- nario and in year i. Percentage of Accumulated GDP in No Climate Change Scenario Indicator 2 was calculated by dividing indicator I1 by the aggregated string of discounted values of the variable in the no climate change scenario: I2 = (Σ VJi – Σ Vncci) / Σ Vncci . Percentage of GDP in Base Year Indicator 3 was calculated by dividing indicator I1 by the value of the variable in the base year (2009). I3 = (Σ VJi – Σ Vncci) / VBY , where VBY = value of the variable in the base year. When using these indicators, it is important to bear in mind several caveats to avoid potential misinterpretations: 1. Indicator 1 does not provide a clear idea of the size of the impacts, since we cannot tell by just looking at this indi- cator whether the impacts are large or small. It is always important to compare it with reference to the value of an- other known variable. This shortcoming can be solved by expressing this indicator in per capita terms. The positive aspect of this indicator is that its meaning can be easily understood. However, this indicator is very sensitive to the choice of discount rate. 2. Indicator 2 does provide a reference to assess whether the effects are large or small as it is expressed as a percent- age. It thus provides a more precise idea of the size of the effects as it compares an aggregated string of discounted GDP values of impacts over the long term with an ag- gregated string of discounted GDP values over the same time horizon. This indicator is less sensitive to the choice of discount rate, because the string values of both, the nu- merator and denominator of the indicator, are discounted using the same rate. However, the result of comparing two NPVs is path-dependent. Since GDP is assumed to increase at a fixed rate, and the impacts follow a different path over time, concentrated at the far end of the period near 2050, the choice of discount rate does influence the indicator value. However, the sensitivity of the indicator to the choice of discount rate tends to be smaller, which leads the impacts of climate change to be smaller. That is the case presented in this report. 3. Indicator 3, like indicator 2, provides a reference to assess the magnitude of the impacts. However, this indicator tends to provide an artificially enlarged magnitude of the impacts since it compares an aggregated string of discounted values of impacts over the long term with the value of the variable in one single year. Therefore, while the meaning of this indicator can be easily interpreted and understood, the indicator tends to be very sensitive to the choice of discount rate. Tables A6 through A8 provide an overview of results for each country and each indicator. Throughout most of the country sections in the main text, we discuss the results using indicator 2. MODELING SUITE 19 Socioeconomic Impacts of Climate Change in Brazil Brazil is one of the world’s largest producers of agricultural products and a leading exporter of soy and coffee. It can therefore expect important impacts from both global price changes and local yield changes. Previous studies on the impacts of climate change on agriculture in Brazil have found potentially severe impacts, with farms losing on average 23 percent of their land values by 2060 in the worst climate scenario (Seo and Mendelsohn 2008). Seo and Mendelsohn’s (2008) study on the economic costs and opportunities of climate change in Brazil estimates that except for sugarcane, all crops are adversely affected by a reduction in low-risk production areas, especially soybeans (by 30 to 34 percent), maize (by 15 percent), and coffee (by 17 to 18 percent). Crop yields would fall in particular for staple crops in the northeast region (Margulis and Burle Schmidt Dubeux 2011). However, although absolute agricultural production is large in Brazil, agriculture contributes only about 5.5 percent to total GDP (Table 6), so the effect on the overall economy is still bound to be relatively small. Moreover, most of Brazil’s production and consumption of agricultural goods is directed toward domestic markets. Direct impacts of global price changes will therefore be concentrated on a few export-oriented agricultural subsectors (such as soy- beans, coffee, and maize) and import-substituting subsec- tors (wheat), together making up 30 percent of agricultural value-added, while all other agricultural sectors largely pro- duce for the domestic market and therefore are not directly affected by world market price changes.13 Yet domestic agricultural subsectors and forestry compete with directly affected subsectors for agricultural land as well as other pri- mary and secondary inputs such as labor and capital and are therefore likely to be impacted. In addition, nonagricultural sectors also compete with agriculture for labor, capital, and intermediate inputs. Since Brazil is located in the tropics and spans the equator, the climate is already hotter than ideal for almost all crops. So, as indicated in Table 1, further temperature increases are expected to have a negative effect on crop yields almost everywhere. Again the direct impact is concentrated on a few agricultural subsectors (maize, wheat, rice, soybeans, cassava, cotton, and sugarcane) making up 40 percent of agricultural value-added. It is important to note that rice, cassava, cotton and sugarcane are exclusively sold in domestic markets. Moreover, these agricultural 20 goods do not face competition from abroad; they are purely nontradable. For such crops with weak or no links to inter- national markets, productivity declines induced by climate change may be more than offset by rising domestic prices. In this case, farmers may have incentives to shift resources toward rather than away from some strongly affected crops. In this section we will first analyze the effects of changes in global agricultural prices due to global climate change, and then we will analyze the effects of changes in crop yields due to local climate change. After presenting the combined effect of the two types of impacts, we will present a gender- differentiated analysis of vulnerability. IMPACTS OF GLOBAL AGRICULTURAL COMMODITY PRICE CHANGES ON THE BRAZILIAN ECONOMY AND HOUSEHOLDS Brazil’s demand for agricultural goods is strongly biased toward domestic markets, with imported goods making up just 3.6 percent of total absorption of agricultural goods and 1.9 percent of the total import bill (Table 6). Wheat is the TABLE 6 Brazil: Agricultural value-added by region and agricultural trade orientation, 2008 Northern tropical humid Central tropical subhumid Southern humid Eastern semiarid All regions Agricultural goods Billions of reais Percent Billions of reais Percent Billions of reais Percent Billions of reais Percent Billions of reais Percent EXP- shr EXP- OUTshr IMP- shr IMP- DEM- shr Maize 0.18 3.23 4.27 9.06 6.55 10.05 0.93 4.80 11.93 8.69 0.59 10.44 0.07 1.56 Wheat 0.08 0.17 1.74 2.67 1.82 1.33 0.08 10.56 1.00 61.58 Rice 0.18 3.23 0.85 1.80 3.01 4.62 0.19 0.98 4.23 3.08 Soybeans 0.09 1.62 10.10 21.43 10.49 16.10 1.39 7.18 22.07 16.08 3.83 38.46 Cassava 0.55 9.87 0.72 1.53 0.76 1.17 0.57 2.94 2.60 1.89 Tobacco 2.78 4.27 0.05 0.26 2.83 2.06 0.02 1.72 Citrus fruits 0.03 0.54 0.28 0.59 2.94 4.51 0.26 1.34 3.51 2.56 0.03 1.95 Cotton 1.55 3.29 0.16 0.25 0.79 4.08 2.50 1.82 Coffee 4.63 9.82 0.84 1.29 0.34 1.76 5.81 4.23 1.61 64.47 Sugarcane 0.06 1.08 1.82 3.86 6.67 10.24 1.52 7.85 10.07 7.34 Other crops 1.04 18.67 5.52 11.71 8.45 12.97 6.68 34.50 21.69 15.81 0.36 3.57 0.50 5.81 Live cattle 1.48 26.57 7.72 16.38 5.87 9.01 2.14 11.05 17.21 12.54 0.20 2.21 Poultry 0.21 3.77 1.13 2.40 4.83 7.41 1.07 5.53 7.24 5.28 Live milk cattle 0.27 4.85 4.29 9.10 3.25 4.99 1.21 6.25 9.02 6.57 Pigs 0.12 2.15 0.88 1.87 1.62 2.49 0.39 2.01 3.01 2.19 Eggs 0.10 1.80 0.89 1.89 1.53 2.35 0.63 3.25 3.15 2.30 0.04 1.77 Forestry 0.84 15.08 2.14 4.54 2.97 4.56 0.69 3.56 6.64 4.84 0.05 1.67 0.31 10.22 Fishery 0.42 7.54 0.27 0.57 0.69 1.06 0.51 2.63 1.89 1.38 0.01 1.00 0.06 6.46 Total 5.57 100.00 47.14 100.00 65.15 100.00 19.36 100.00 137.22 100.00 6.9 10.4 1.9 3.6 Percent 4.06 34.35 47.48 14.11 100.00 Source: Social accounting matrix for Brazil. Note: EXPshr = Commodity share of total export revenues; EXP-OUTshr = Share of exports in commodity demand, IMPshr = Commodity share of total import expen- ditures; IMP-DEMshr = Share of imports in commodity supply, GDP = gross domestic product. Goods with EXP-OUTshr or IMP-DEMshr of less than 1 percent are treated as nonexport or nonimport goods. Values in red are directly affected by world market price changes. Values highlighted in gray are directly affected by yield changes. Total GDP at factor cost = 2.5 trillion reais. Agriculture’s share of GDP = 5.5 percent. SOCIOECONOMIC IMPACTS OF CLIMATE CHANGE IN BRAzIL 21 only agricultural commodity with a sizeable import inten- sity of 62 percent, yet wheat production does not exceed 2 percent of total agricultural value-added in any of the AEZs and the world market price of wheat is not much affected by global climate change (Figure 5).14 Thus, the impact of global climate change in Brazil is felt largely on the supply side of agricultural markets. Although overall agricultural export intensity is rather low, with only 10 percent of do- mestic agricultural production being exported, producers of maize, wheat, soybeans, and coffee, which contribute 30 percent to total agricultural value-added, all export a size- able share of their production, ranging from 10 percent for maize and wheat to 39 percent and 65 percent for soybeans and coffee, respectively. Improvements in agricultural international terms of trade caused by increasing global prices for these com- modities, which result from global climate change (Figure 5), would have a favorable effect on agricultural GDP in Brazil (Figure 9). As higher world market prices for these agricultural goods provide an incentive to domestic pro- ducers to expand production and restructure their supply from domestic toward foreign markets, export expansion would boost agricultural GDP. This effect can be especially observed in the central tropical subhumid region that has specialized in large-scale soybean, coffee, and maize produc- tion (Figure 10). In contrast, the northern tropical humid region, with its production of cassava, other crops, live cattle, and wood for domestic markets and wood processing, is largely unaffect- ed by global price changes. Finally, agricultural production in the southern humid and to a lesser extent in the eastern semiarid regions benefit from higher world market prices for maize and soybeans, which together make up 16 percent and 12 percent of agricultural value added, respectively (Table 6). Overall and driven by world market price changes, the NPV of agricultural GDP in Brazil is expected to increase by between 1.5 to 3.6 percent compared to a perfect mitigation scenario, depending on the climate scenario (Figures 9 and 10). This advantage for the agricultural sector, however, turns into a disadvantage for other sectors, including some agricultural subsectors, since several of these—especially the industrial sector—use agricultural inputs and will suffer from more expensive inputs. Cost push as a result FIGURE 9 Brazil: Impacts of global agricultural price changes on net present value of agricultural GDP, by product group Source: Authors’ elaboration based on dynamic computable general equilibrium simulation result Note: _G = Global scenario, GFDL = Earth System Model (ESM) designed and maintained by the National Oceanic and Atmospheric Administration’s Geo- physical Fluid Dynamic Laboratory, HADGEM2 = Hadley Centre’s Global Environment Model (version 2), IPSL = Institut Pierre Simon Laplace’s ESM, and MIROC = Model for Interdisciplinary Research on Climate. –2 0 2 4 6 8 10 12 14 16 Ch an ge c om pa re d to pe rf ec t m iti ga tio n (% ) Agricultural sector HADGEM2_G Climate model MIROC_G GFDL_G IPSL_G Cereals Export cro ps Oth er c ro ps Livesto ck Forestr y & �sh ing To tal a gric ultu re 22 CLIMATE CHANGE IMPACTS AND HOUSEHOLD RESILIENCE of backward linkages is most pronounced in food process- ing, where intermediate input costs make up more than 80 percent of total production costs.15 In addition, since the export-oriented agricultural sector will become more profitable due to higher output prices, over the long run this sector will attract more investment, labor, and capital at the expense of the remaining sectors. Moreover, the strong expansion of export-oriented soybeans and coffee produc- tion and import-substituting wheat production will attract land from other agricultural uses, especially the forestry sector, thereby potentially leading to increased deforestation (Figure 9). Finally, export expansion and import substitu- tion in agriculture will lead to a balance-of-trade surplus. Eliminating the surplus will require an appreciation of the real exchange rate—for example, an increase in the rela- tive price of tradable to nontradable goods—that leads to additional resource shifts, now from tradables to nontrad- ables producing sectors. Since the agricultural sector is small compared to the nonagricultural sectors, the increase of agricultural real income of about 2.5 percent on average across the four climate scenarios is outweighed by average real income losses of 0.3 percent in industry, leaving overall real GDP practically unaffected by global climate change (Figure 11).16 While total real GDP is unaffected by global climate change and increasing world market prices for agricultural goods, the terms-of-trade improvement leads to higher do- mestic producer prices and higher wages, profits, and land rental rates, thereby raising domestic factor incomes and absorption including household consumption. Although all types of production factors benefit from higher factor returns, the increases in factor remuneration are largest for land, capital, and unskilled labor, which are all used inten- sively in the production of soybeans and coffee. The positive terms-of-trade effect and the resulting increases in wages, profits, and land rentals translate into a small improvement of welfare for all households—from the poorest to the richest (Figure 12).17 The total increase in household welfare is between 0.15 and 0.36 percent, depending on the climate change scenario, and is larger for the poorest households (the two lowest-income deciles, which receive a larger than average share of their functional FIGURE 10 Brazil: Impacts of global agricultural price changes on net present value of agricultural GDP, by region Source: Authors’ elaboration based on dynamic computable general equilibrium simulation results Note: _G = Global scenario, GFDL = Earth System Model (ESM) designed and maintained by the National Oceanic and Atmospheric Administration’s Geo- physical Fluid Dynamic Laboratory, HADGEM2 = Hadley Centre’s Global Environment Model (version 2), IPSL = Institut Pierre Simon Laplace’s ESM, and MIROC = Model for Interdisciplinary Research on Climate. 0 1 2 3 4 5 6 Northern tropical humid Central tropical subhumid Southern humid Eastern semiarid All regions Ch an ge c om pa re d to pe rf ec t m iti ga tio n (% ) Agricultural region HADGEM2_G MIROC_G GFDL_G IPSL_G Climate model SOCIOECONOMIC IMPACTS OF CLIMATE CHANGE IN BRAzIL 23 FIGURE 11 Brazil: Impacts of global agricultural price changes on net present value of GDP, by sector Source: Authors’ elaboration based on dynamic computable general equilibrium simulation results. Note: _G = Global scenario, GFDL = Earth System Model (ESM) designed and maintained by the National Oceanic and Atmospheric Administration’s Geo- physical Fluid Dynamic Laboratory, HADGEM2 = Hadley Centre’s Global Environment Model (version 2), IPSL = Institut Pierre Simon Laplace’s ESM, and MIROC = Model for Interdisciplinary Research on Climate. –1.0 –0.5 0.0 0.5 1.0 1.5 2.0 2.5 3.0 3.5 4.0 Agriculture Industry Services Total GDP Ch an ge c om pa re d to pe rf ec t m iti ga tio n (% ) Sector HADGEM2_G MIROC_G GFDL_G IPSL_G Climate model FIGURE 12 Brazil: Impacts of global agricultural price changes on household welfare, by income decile Source: Authors’ elaboration based on dynamic computable general equilibrium simulation result. Note: HHD01 to HHD10 = First to tenth income decile; _G = Global scenario, GFDL = Earth System Model (ESM) designed and maintained by the National Oceanic and Atmospheric Administration’s Geophysical Fluid Dynamic Laboratory, HADGEM2 = Hadley Centre’s Global Environment Model (version 2), IPSL = Institut Pierre Simon Laplace’s ESM, and MIROC = Model for Interdisciplinary Research on Climate. 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 HADGEM2_G MIROC_G GFDL_G IPSL_G HHD01 HHD02 HHD03 HHD04 HHD05 HHD06 HHD07 HHD08 HHD09 HHD10 Total Ch an ge c om pa re d to pe rf ec t m iti ga tio n (% ) Climate model 24 CLIMATE CHANGE IMPACTS AND HOUSEHOLD RESILIENCE income from the relatively scarce factor—land, and the sup- ply of low-skilled labor—which are both used intensively in the export-oriented agricultural sectors. IMPACTS OF LOCAL CROP YIELD CHANGES ON THE BRAZILIAN ECONOMY AND HOUSEHOLDS While the primary impact of changes in world market prices is limited to the exported part of domestic production of a few agricultural goods in Brazil, changes in local crop yields affect more goods (Table 6) and total production of these goods. As shown in Table 1, the impacts of climate change on Brazilian crop yields are expected to be overwhelmingly negative. This will imply reductions in agricultural GDP in all regions, but especially in the central tropical subhumid region specializing in large-scale soybean and maize pro- duction. In this region, the negative effect is on the order of 1 to 6 percent. In contrast, crops that are affected by local climate change make up a small share of agricultural pro- duction in the northern tropical humid region. Moreover, cassava—which makes up 50 percent of affected crops in this region—is a pure nontradable. Thus, increasing prices almost compensate for output losses resulting from lower yields. Overall, the fall in agricultural GDP due to the local yield changes is on the order of 1.5 percent in the GFDL scenario and up to 3.0 percent in the HADGEM2 scenario (Figure 13). All households, both net producers and net consumers across all regions, suffer from the reduction in agricultural yields due to local climate change—both as a result of lower factor income and higher prices for consumption goods— but the poorer deciles suffer relatively more than the richer deciles (Figure 14). Again, poorer households are more affected—now from local climate change—because these receive a larger share of their functional income from land rents, which will fall more strongly than profit and wage rates since land is exclusively used in agriculture (and for- estry), while labor and capital are also used in other sectors that are not directly affected by local climate change. On the other hand, income losses in rice, cassava, and sugarcane FIGURE 13 Brazil: Impacts of local yield changes on net present value of agricultural GDP, by region Source: Authors’ elaboration based on dynamic computable general equilibrium simulation results. Note: _L = Local scenario, GFDL = Earth System Model (ESM) designed and maintained by the National Oceanic and Atmospheric Administration’s Geo- physical Fluid Dynamic Laboratory, HADGEM2 = Hadley Centre’s Global Environment Model (version 2), IPSL = Institut Pierre Simon Laplace’s ESM, and MIROC = Model for Interdisciplinary Research on Climate. –6 –5 –4 –3 –2 –1 0 HADGEM2_L MIROC_L GFDL_L IPSL_L Northern tropical humid Central tropical subhumid Southern humid Eastern semiarid All regions Ch an ge c om pa re d to pe rf ec t m iti ga tio n (% ) Agricultural region Climate model SOCIOECONOMIC IMPACTS OF CLIMATE CHANGE IN BRAzIL 25 production resulting from lower yields are partly compen- sated for by higher domestic prices for these nontradable goods. Yet these agricultural commodities make up a low share of total agricultural production. On average during the 40-year simulation period, the size of the drop in household welfare due to this effect is between 0.25 percent and 0.50 percent. COMBINED IMPACTS OF GLOBAL AND LOCAL CLIMATE CHANGE ON THE BRAZILIAN ECONOMY The global and local effects of climate change jointly show that (1) local effects dominate global effects, and (2) indirect effects on other sectors dominate direct effects on agriculture. Although macroeconomic growth on average does not differ much from the case of perfect mitigation, the sec- toral contribution to growth is slightly changed in favor of agriculture due to the global effects of climate change, which render the production of various agricultural com- modities more profitable in three of the four climate change scenarios. The exception is the HADGEM2 scenario, where the combination of lower world market price increases and higher yield losses hampers agricultural growth. Yet the indirect impact of climate change on nonagricultural sectors is even worse, thus still favoring agriculture at least relatively (Figure 15) Results for the agricultural sector differ noticeably between climate scenarios, across both regions and agri- cultural products (Figures 16 and 17). Agricultural income rises under the combined MIROC, GFDL, and IPSL scenarios with increasing speed over time (not shown in the figures) but falls in the HADGEM2 scenario. As shown in the previous section, the impacts of local climate change in isolation have negative implications for agricultural in- comes. The overall fall in yields due to the local impacts of climate change translates into higher domestic agricultural prices induci