d i s c u s s i o n pa p e r n u m B e r 1 august 2010 d e v e l o p m e n t a n d c l i m at e c h a n g e d i s c u s s i o n pa p e r s 1 56659 d e v e l o p m e n t a n d c l i m a t e c h a n g e The Economics of Adaptation to Extreme Weather Events in Developing Countries d i s c u s s i o n pa p e r n u m B e r 1 august 2010 d e v e l o p m e n t a n d c l i m a t e c h a n g e The Economics of Adaptation to Extreme Weather Events in Developing Countries Brian Blankespoor susmita dasgupta Benoit laplante david Wheeler* *Authors' names are in alphabetical order. The authors are respectively Lead Environmental Economist and consultants, World Bank, and Senior Fellow, Center for Global Development. Our thanks to Tim Essam for his assistance with database construction. Thanks also to Sergio Margulis, Urvashi Narain, Gordon Hughes, and Michele de Nevers for useful comments and suggestions. Papers in this series are not formal publications of the World Bank. They are circulated to encourage thought and discussion. The use and citation of this paper should take this into account. The views expressed are those of the authors and should not be attributed to the World Bank. Copies are available from the Environment Department of the World Bank by calling 202-473-3641. © 2010 The International Bank for Reconstruction and Development / THE WORLD BANK 1818 H Street, NW Washington, DC 20433, U.S.A. Telephone: 202-473-1000 Internet: www.worldbank.org/climatechange E-mail: feedback@worldbank.org All rights reserved. August 2010 This volume is a product of the staff of the International Bank for Reconstruction and Development / The World Bank. The findings, interpretations, and conclusions expressed in this volume do not necessarily reflect the views of the Executive Directors of The World Bank or the governments they represent. The World Bank does not guarantee the accuracy of the data included in this work. The boundaries, colors, denominations, and other information shown on any map in this work do not imply any judgement on the part of the World Bank concerning the legal status of any territory or the endorsement or acceptance of such boundaries. RIGHTS AND PERMISSIONS The material in this publication is copyrighted. Copying and/or transmitting portions or all of this work without permission may be a violation of applica- ble law. The International Bank for Reconstruction and Development / The World Bank encourages dissemination of its work and will normally grant permission to reproduce portions of the work promptly. For permission to photocopy or reprint any part of this work, please send a request with complete information to the Copyright Clearance Center Inc., 222 Rosewood Drive, Danvers, MA 01923, USA; telephone 978-750-8400; fax 978-750-4470; Internet: www.copyright.com. Cover photo: Hue, Vietnam. © Shutterstock Images, LLC All dollars are U.S. dollars unless otherwise indicated. taBle oF contents executive summary vii section 1. introduction 1 section 2. losses from extreme Weather since 1960 2 section 3. What lies ahead 4 section 4. development and vulnerability to extreme Weather events 7 section 5. projecting Baseline changes in Female education 11 section 6. climate change, development, and Future vulnerability 12 section 7. estimating the cost of adapting to extreme Weather events 15 7.1 Data on Weather Emergency Preparedness Costs 15 7.2 The Education Alternative 16 7.3 The Cost of Climate Change Neutralization 16 section 8. summary and conclusions 20 references 21 Figures 1 trends in risks from extreme events, 1960­2000 3 2 number affected by droughts in developing countries, 1960­2002 (log scale) 4 3 india--historical and projected data 5 4 extreme Weather losses, 1970­1999: historical and Best practice 10 5 loss risks in india, 2000­50 12 Tables 1 losses From extreme Weather events--developing countries, 1960­99 2 2 catastrophic death tolls From droughts, 1960­2002 3 iv t he econom ics oF ada p tation to e x tr e me Weath er e v e n ts in d ev e lopin g c ou n tr ie s Tables (conTinued) 3a Year 2000 low-income countries through 2050 6 3B Year 2000 lower-middle-income countries through 2050 6 4 Year 2000 low-income and lower-middle-income countries -- changes in life expectancy through 2050 7 5 determinants of net Female enrollment ratios (dependent variable: Flog net enrollment rate) 8 6 Weather risk model: Fixed-effects estimates (dependent variables: [log (variable/population)]) 9 7 historical and simulated "Best practice" Weather-related losses, 1970­2000 9 8 projected net Female primary and secondary enrollment ratios by region 11 9 alternative scenarios for india, 2000­50 13 10 alternative scenarios for developing countries, 2000­50 14 11 disaster preparedness and management data 15 12 climate-neutralizing Female education--students and costs, (republic of congo, nepal, and nicaragua) 17 13 climate-neutralizing Female education--students and costs (developing regions) 18 14 climate-neutralizing Female education (global totals, 2010­50) 19 15 climate-neutralizing Female education, 2002­50 (global and regional costs: selected discount rates [$us Billion]) 20 d e v e l o p m e n t a n d c l i m at e c h a n g e d i s c u s s i o n pa p e r s vii executive summarY Without international assistance, developing countries many additional young women would have to be will adapt to climate change as best they can. Part of educated to neutralize this increased vulnerability? And the cost will be absorbed by households and part by the how much would it cost? public sector. Adaptation costs will themselves be affected by socioeconomic development, which will also Our study relies heavily on fixed-effects estimation of be affected by climate change. Without a better under- risk equations that link losses from floods and droughts standing of these interactions, it will be difficult for during the period 1960­2003 to three basic determi- climate negotiators and donor institutions to determine nants: weather events that increase potential losses, the appropriate levels and modes of adaptation assis- income per capita, and female education. We estimate tance. This paper attempts to contribute by assessing separate equations for (a) the risk of death from a flood, the economics of adaptation to extreme weather events. (b) the risk of being affected by a flood, and (c) the risk We address several questions that are relevant for the of being affected by a drought (the data are too sparse international discussion: How will climate change alter to support estimation for death from droughts). the incidence of these events, and how will their impact be distributed geographically? How will future socio- Our analysis combines the estimated risk equations economic development affect the vulnerability of with projections of economic growth, population affected communities? And, of primary interest to growth, and changes in primary and secondary school- negotiators and donors, how much would it cost to ing. We develop three scenarios: (1) a baseline in neutralize the threat of additional losses in this context? which socioeconomic development continues but the climate does not change; and (2 and 3) two scenarios From a narrow technical perspective, it might be desir- with the same baseline development path but alternative able to address the latter question with a detailed engi- weather paths driven by particularly "wet" and "dry" neering cost analysis of specific disaster prevention GCMs (Global Climate Models). For each GCM measures. However, as we show in the paper, existing scenario, we calculate the associated changes in the risks cross-country information about relevant emergency of death from floods and being affected by floods or preparedness programs is far too sparse to support droughts. Then, choosing the worst-case risk, we calcu- systematic analysis and projection. And in any case, we late the increase in female schooling that would neutral- believe that the effectiveness of such measures is contin- ize this additional risk. We multiply the results by gent on the characteristics of the communities that expenditures per student to estimate the total educa- employ them. We therefore adopt an alternative tional investment required to neutralize the additional approach in this paper, focusing on the role of socioeco- weather risk posed by climate change. nomic development in increasing climate resilience. Drawing on extensive research, our approach highlights Our approach is conservative, in the sense that it is very the importance of female education and empowerment unlikely to underestimate the required investment. in reducing weather-related loss risks. Our cost analysis First, we base our cost assessment on general prepared- asks two key questions: As climate change increases ness via increased education, rather than more narrowly potential vulnerability to extreme weather events, how targeted investment in emergency preparedness. viii t he econom ics oF ada p tation to e x tr e me Weath er e v e n ts in d ev e lopin g c ou n tr ie s Second, we base our cost calculation on worst-case risk Our analysis has generated a set of estimates for scenarios, which require the greatest increase in school- required female schooling and associated costs by GCM ing to neutralize. Third, we incorporate only projected scenario, country and year. Variations in projected increases in vulnerability, not decreases. As an alterna- climatic, socioeconomic and demographic variables are tive, for example, we could perform a net impact analy- more than sufficient to produce wide disparities in sis for a wet-climate scenario that would subtract outcomes by 2050, even among countries within the expected decreased losses from drought from expected same region. At the country and regional levels, neither increased losses from flooding. Fourth, our analysis climate scenario dominates in all cases. The "wet" employs the two GCMs (among approximately twenty) scenario generates higher risk-neutralizing expenditure that generate the wettest and driest scenarios at the on female schooling in somecountries and regions; the global scale. Other GCMs would generate more "dry" scenario is more costly in others. Among regions, moderate intermediate results. Finally, we do not aver- South Asia requires the most expenditure in both age across the two GCMs, which would have the effect climate scenarios, followed by Sub-Saharan Africa and of neutralizing their extreme signals. East Asia, and then more distantly by the other regions. In summary, we believe that our approach is sufficiently At both regional and global levels, we find an impres- conservative to create a strong upward bias in our cost sive scale for the requisite increases in female education estimation. It is certainly possible that the "true" cost of expenditure. By mid-century, neutralizing the impact of adaptation to extreme weather events is lower than our extreme weather events requires educating an additional estimates, but we doubt that it is higher. 18 to 23 million young women at a cost of $11 to $14 billion annually. For the period 2000­50 as a whole, At the same time, our approach offers significant both GCM scenarios entail about $280 billion in addi- co-benefits because female education has a much tional expenditure. The present value of these expendi- broader sphere of potential influence than direct invest- tures is substantially reduced by time-discounting, even ment in emergency preparedness. As the development at modest rates, but the basic result stands: In the literature has noted for many years, educating young developing world, neutralizing the impact of worsening women is one of the major determinants (indeed, some weather over the coming decades will require educating would argue, the major determinant) of sustainable a large new cohort of young women at a cost that will development. A disaster-prevention approach that steadily escalate to several billions of dollars annually. focuses on investment in female education therefore has However, it will be enormously worthwhile on other an expected social rate of return on other margins that margins to invest in education for millions of young probably warrants the exercise, even if the expected women who might otherwise be denied its many benefits in reduced disaster vulnerability are overstated benefits. (and in fact, we believe the opposite to be true). d e v e l o p m e n t a n d c l i m at e c h a n g e d i s c u s s i o n pa p e r s 1 1. IntroductIon development in reducing vulnerability to climate shocks. Horwich (2000), Tol and Leek (1999), Burton et al. Without international assistance, developing countries (1993), and Kahn (2005) have focused on the effect of will adapt to climate change as best they can. Part of rising income per capita: As communities get richer, the cost will be absorbed by households and part by the they have greater willingness and ability to pay for public sector. Adaptation costs will themselves be preventive measures. Kahn (2005) finds that the insti- affected by socioeconomic development, which will also tutional improvement that accompanies economic be affected by climate change. Without a better under- development also plays a significant role through standing of these interactions, it will be difficult for enhanced public-sector capability to organize disaster climate negotiators and donor institutions to determine prevention and relief. the appropriate levels and modes of adaptation assis- tance. This paper attempts to assess the economics of Other work focuses on the role of political and human adaptation to extreme weather events. We address development. Albala-Bertrand (1993) identifies politi- several questions that are relevant for the international cal marginalization as a source of vulnerability to natu- discussion: How will climate change alter the incidence ral disasters. Toya and Skidmore (2005) find a of these events, and how will their impact be distributed significant role for education in reducing vulnerability geographically? How will future socioeconomic devel- through better choices, in areas ranging from safe opment affect the vulnerability of affected communi- construction practices to assessment of potential risks. ties? And, of primary interest to negotiators and Recently, Oxfam International (2008) has drawn on donors, how much would it cost to neutralize the threat extensive evidence from South Asia to highlight the of additional losses in this context? particular vulnerability of women, who often suffer far greater losses than men in natural disasters: From a narrow technical perspective, it might be desir- able to address the latter question with a detailed engi- Nature does not dictate that poor people, or women, neering cost analysis of specific disaster prevention should be the first to die. Cyclones do not hand-pick measures. However, as we show in the paper, existing their victims. Yet, history consistently shows that cross-country information about relevant emergency vulnerable groups end up suffering from such events preparedness programs is far too sparse to support disproportionately ... In the 1991 Bangladesh cyclone, systematic analysis and projection. And in any case, we for example, four times more women died than men believe that the effectiveness of such measures is contin- ... Disasters are therefore an issue of unsustainable gent on the characteristics of the communities that and unequal development at all levels ...(Oxfam employ them. We therefore adopt an alternative 2008, p. 1) approach in this paper, focusing on the role of socioeco- nomic development in increasing climate resilience. A logical inference from Oxfam (2008), Albala- Bertrand (1993), and Toya and Skidmore (2005) is that Our analysis builds on empirical work and case studies empowering women through improved education may that have documented the role of socioeconomic be a critical factor in reducing families' vulnerability to 2 t he econom ics oF ada p tation to e x tr e me Weath er e v e n ts in d ev e lopin g c ou n tr ie s weather-related disasters. This would also be consistent with the extensive literature that documents the power- tabLe 1. Losses from extreme Weather ful effect of female education on community-level social events--deveLopIng countrIes, capital and general welfare measures such as life expec- 1960­99 Deaths Number affected tancy (King and Mason 2001). (`000) (`000) Period Floods Droughts Floods Droughts To the best of our knowledge, no empirical research has 1960­69 17.0 1,510.1 34,256 110,000 focused on female education as a potentially critical 1970­79 46.4 319.1 200,000 460,000 determinant of vulnerability to extreme weather events. 1980­89 50.3 556.9 480,000 700,000 Assessing its importance and implications is a core 1990­99 58.5 0.8 1,400,000 190,000 feature of this paper. Drawing on an econometric anal- ysis of panel data, we address two key questions: As total 172.2 2,386.9 2,114,256 1,460,000 climate change increases potential vulnerability to extreme weather events, can expanding female educa- Source: cred (em-dat). tion neutralize this increased vulnerability? If so, how much would it cost? in Brussels.1 For developing countries,2 the CRED data provide a sobering view of weather-related losses since The remainder of the paper is organized as follows. 1960. Table 1 presents the numbers of people killed Section 2 provides a summary of losses from extreme and affected by floods and droughts by decade. Flood- weather events in developing countries during the related deaths rose steadily from 17,000 in the 1960s to period 1960­2006. In Section 3, we review recent over 58,000 in the 1990s. Droughts killed far more projections of climate impacts, economic growth, and people, although their impact was heaviest in the 1960s demographic change. We focus particularly on projec- (1.5 million deaths) and basically absent in the 1990s tions by integrated assessment models that incorporate (800 deaths). Aside from those killed, floods and links between climate change and economic activity. droughts affected huge numbers of people who were Section 4 specifies a set of risk equations for weather- injured, made homeless, or forced to seek emergency related disasters and estimates them by fixed effects. In assistance. Trends for persons affected parallel the Section 5, we develop country-specific projections for trends for deaths: rising for floods, but not for droughts. female education. Section 6 uses our econometric results and education projections to forecast future risks To understand this phenomenon better, it is useful to under alternative assumptions about climate change. In separate losses into their two components: (1) risk of Section 7, we use these projections to estimate the cost loss (total losses/population) and (2) population. The of reducing future weather-related risks through more latter component increased rapidly in developing coun- intensive investment in female education. Section 8 tries, from 2.2 billion in 1961 to 4.7 billion in 2000 summarizes and concludes the paper. (World Bank, World Development Indicators). Figures 2. Losses from extreme 1 To be entered in CRED's EM-DAT database, a natural disaster must Weather sInce 1960 involve at least 10 people reported killed; 100 people reported affect- ed; the declaration of a state of emergency; or a call for international assistance. Recorded deaths include persons confirmed as dead and The most comprehensive data on weather-related losses persons missing and presumed dead. Total affected persons include people suffering from disaster-related physical injuries, trauma, or ill- are maintained by the Centre for Research on the ness requiring medical treatment; people needing immediate assis- tance for shelter; or people requiring other forms of immediate Epidemiology of Disasters (CRED) at the School of assistance, including displaced or evacuated people. Public Health of the Université Catholique de Louvain 2 We define these as countries identified by the World Bank as low income or lower middle income. d e v e l o p m e n t a n d c l i m at e c h a n g e d i s c u s s i o n pa p e r s 3 1a­d provide graphical evidence on the risk component. For floods, the trends in risk of being killed or affected tabLe 2. catastrophIc death toLLs from are clearly positive, while no trend is apparent for droughts, 1960­2002 droughts. Figures 1a-d show that the risk series for floods and droughts are very different. Floods exhibit Country Year Deaths roughly continuous behavior through time, while india 1965 500,000 droughts have exhibited huge, rare pulses that have not india 1966 500,000 recurred since the early 1980s. The sources of these india 1967 500,000 pulses are provided by Table 2, which presents the worst ethiopia 1984 300,000 nine cases of drought-related death since 1960. Truly ethiopia 1974 200,000 catastrophic losses of life in droughts in previous sudan 1984 150,000 decades have been limited to four countries: India (1.5 mozambique 1984 100,000 million total), Ethiopia (600,000), Sudan (150,000) and ethiopia 1973 100,000 Mozambique (100,000). somalia 1974 19,000 To summarize, losses from flooding have increased Source: cred (em-dat). markedly since 1960 for two reasons: The risk of loss fIgure 1. trends In rIsks from extreme events, 1960­2000 1a. rIsk of death from fLoodIng 1b. rIsk of beIng affected bY fLoodIng (per mILLIon -- Log scaLe) (per mILLIon -- Log scaLe) 10.00 100,000.00 10,000.00 1.00 1,000.00 0.10 100.00 1960 1970 1980 1990 2000 1960 1970 1980 1990 2000 1c. rIsk of death from drought 1d. rIsk of beIng affected bY drought (per mILLIon -- Log scaLe) (per mILLIon -- Log scaLe) 1,000.00 100,000.00 100.00 10,000.00 10.00 1,000.00 1.00 100.00 0.10 10.00 0.01 1.00 1960 1970 1980 1990 2000 1960 1970 1980 1990 2000 Source: cred (em-dat). 4 t he econom ics oF ada p tation to e x tr e me Weath er e v e n ts in d ev e lopin g c ou n tr ie s has increased significantly, and the population subject to the population forecast (which includes projections of risk has more than doubled. Droughts present a very life expectancies and total fertility rates), we use the different case, with risks dominated by catastrophic UN's medium variant projection (2006 Revision) events in a few countries decades ago, and no clear Because our economic projections incorporate interac- overall trend. Even with rapidly increasing population, tions with projected climate change, they are moderate there is no clear upward trend in persons affected by in their view of future prospects. Overall, the economic drought (Figure 2). Nevertheless, as Table 1 and Figure and demographic projections we employ are relatively 2 show, the number of people affected by droughts close to those in the SRES A2 Scenario.3 continues to be huge. We attempt to bound the set of reasonable expectations about future climate using GCMs with strongly 3. What LIes ahead contrasting predictions. Within the SRES A2 scenario, the GCMs provide a relatively uniform view of future Our approach to impact forecasting incorporates increases in temperature. However, this is not the case projected socioeconomic and demographic trends as for precipitation, which--as we will show in the follow- well as climate change. This requires us to adopt inter- ing section--is closely related to losses from floods and nally consistent assumptions about future changes in droughts. To provide a sense of what is possible from emissions, economies, human development levels, and the current scientific perspective, our projections use populations. The emissions scenario leads to a forecast two GCMs that are the wettest and driest of approxi- of greenhouse gas concentration in the atmosphere, mately 20 available GCMs at the global level. The which is related to changes in global and local climate driest overall is CSIRO's Mk 3.0 model, which was through one of a large number of global circulation transmitted to the IPCC's data collection center in models (GCMs). For the economic forecast, we draw 2005.4 The driest is NCAR's Community Climate on a recent summary of integrated assessment models System Model (CCSM), version 3.0, which was by Hughes (2009), who draws on a critical assessment released to the public in 2004.5 of the IPCC's SRES scenarios by Tol, et al. (2005). Hughes develops a consensus economic projection by To illustrate the implications of the forecasts, Figure 3 taking an average growth rate from five integrated provides historical and projected data for India for assessment models. We use Hughes' constant-dollar income per capita, population, life expectancy, fertility, GDP series, because our econometric risk estimation mean annual rainfall, and maximum (monthly) annual requires the use of data extending back to 1960. For temperature. We include both the CSIRO and NCAR weather projections. It is immediately clear that the India of 2050 bears little resemblance to present-day India, even in our relatively moderate scenario. GDP fIgure 2. number affected bY grows at an annual rate of 4.4 percent, increasing GDP droughts In deveLopIng countrIes, per capita (constant $US 2000) from $450 in 2000 to 1960­2002 (Log scaLe) $4,300 in 2050. Life expectancy increases from 63 to 1,000,000,000 76 years and the total fertility rate declines from 3.25 to 100,000,000 10,000,000 1,000,000 3 IPCC (2000) characterizes A2 as follows: "A very heterogeneous world, 100,000 characterized by self-reliance and preservation of local identities. 10,000 Fertility patterns across regions converge very slowly, which results in high population growth. Economic development is primarily regionally 1,000 oriented and per capita economic growth and technological change 100 are more fragmented and slower than in other scenarios." 10 4 CSIRO is Australia's Commonwealth Scientific and Industrial Research 1 Organization (www.csiro.au/) 1960 1970 1980 1990 2000 5 NCAR is the US National Center for Atmospheric Research. For a detailed description of the CCSM program, see http://www.ccsm.ucar. Source: cred (em-dat). edu/about/. d e v e l o p m e n t a n d c l i m at e c h a n g e d i s c u s s i o n pa p e r s 5 fIgure 3. IndIa -- hIstorIcaL and projected data GDP Per Capita Life Expectancy 5000 80 4000 70 3000 2000 60 1000 0 50 1970 1990 2010 2030 2050 1970 1990 2010 2030 2050 Population Total Fertility Rate 1700 6 1500 5 1300 4 1100 3 900 700 2 500 1 1970 1990 2010 2030 2050 1970 1990 2010 2030 2050 Mean Monthly Maximum Monthly Precipitation (mm.) Temperature (°C) 110 26.5 26.0 100 NCAR NCAR 25.5 CSIRO 90 25.0 24.5 80 CSIRO 24.0 70 23.5 1970 1990 2010 2030 2050 1970 1990 2010 2030 2050 Source: ??? 6 t he econom ics oF ada p tation to e x tr e me Weath er e v e n ts in d ev e lopin g c ou n tr ie s 1.85 (below replacement). In 2050, India is an upper- middle-income country by current World Bank stan- tabLe 3a. Year 2000 LoW-Income dards, closely resembling the Chile of 2000 in income countrIes through 2050 per capita, life expectancy, and fertility. But not, of Lower course, in population: By 2050, India has 1.66 billion Low middle Decade income income Higher Total people in this scenario. 2001­10 56 0 0 56 The climate projections provide a sense of the dispari- 2011­20 49 7 0 56 ties in GCM predictions associated with the IPCC A2 2021­30 42 14 0 56 scenario. From a monthly average of 88 mm in 2000, 2031­40 34 22 0 56 precipitation increases 8 percent by 2050 to 95 mm in 2041­50 29 24 3 56 the NCAR scenario, and decreases by 8 percent to 81 mm in the CSIRO scenario. Thus, the total difference attributable to GCM variation within the same IPCC tabLe 3b. Year 2000 LoWer-mIddLe- scenario is 16 percent--a very large number in this Income countrIes through 2050 context. The two scenarios are much closer on mean monthly temperature. After an increase of 1° C since Lower middle 1970, both predict another increase of 1° C by 2050. Decade income Higher Total 2001­10 38 0 38 This illustration serves to make one point very clearly: 2011­20 26 12 38 In scenario A2, the GCMs concur that India's tempera- 2021­30 11 27 38 ture will rise steadily as greenhouse gases accumulate in 2031­40 7 31 38 the atmosphere. They also predict significant impacts 2041­50 1 37 38 on precipitation, but not consistently, and with different Source: ??? implications for extreme events. A wetter India will experience more floods, while a drier India will have more droughts. But in either case, it is critical to note information for 38 countries identified as lower middle that a changed climate in the full scenario is associated income in 2000. By 2050, only one has not advanced with a changed country. By 2050, India has become a beyond that group. middle-income country, whose per capita income, human resources, and institutional capabilities give it Table 4 provides related information on changes in life much greater resilience in the face of weather changes. expectancy, which are linked to the population projec- tions. Life expectancy information is not available for The Indian case reflects a more general and underap- two of the countries tabulated in Tables 3a and 3b. preciated feature of scenario A2, as well as the other During the present decade, 14 countries still have life SRES scenarios: Anticipated future emissions increases expectancies between 41 and 50 years and 18 have life reflect the expectation that development will also expectancies between 51 and 60 years, while 29 have life continue. By implication, many of the countries expectancies above 70 years. By 2050, the tabulation currently termed low income or lower middle income by has reversed. From 32 countries with life expectancies the World Bank will have departed that status by 2050. between 41 and 60 years in 2000-2010, the number has dropped to 7. And the number of countries with life Tables 3a and 3b, calculated from the Hughes' (2009) expectancies greater than 70 has increased from 29 to projections, provide striking evidence of this shift for 94 64. countries identified as low income and lower middle income by the World Bank in 2000. Table 3a tracks the In summary, even our moderately optimistic projections status of 56 low-income countries through the succeed- entail very large changes for developing countries ing decades. By 2050, 27 have moved to lower-middle- during the next four decades. And, more importantly income status or higher. Table 3b provides the same for this analysis, these changes are directly tied to the d e v e l o p m e n t a n d c l i m at e c h a n g e d i s c u s s i o n pa p e r s 7 tabLe 4. Year 2000 LoW-Income and LoWer-mIddLe-Income countrIes -- changes In LIfe expectancY through 2050 Life expectancy at birth Decade 41­50 51­60 61­70 71­80 81­90 Total 2001­10 14 18 31 29 0 92 2011­20 8 18 31 35 0 92 2021­30 4 18 20 50 0 92 2031­40 0 14 22 55 1 92 2041­50 0 7 21 60 4 92 source: ??? emissions scenario that generates the climate change We focus on female education, for reasons that we problem. Countries with much higher incomes and life explained in the introduction. The formal specification expectancies will also have much greater willingness and is as follows for country i in period t: ability to pay for protection from extreme weather events. They will also have more highly educated popu- Lit (1) ln( Rit ) = ln = 0 + 1Git + 2 Eit + 3 Rit + 4Tit + it lations with much greater ability to avoid losses from Pit adverse weather. This turns out to have major conse- quences for our assessment of future risk, as we will see where R = Impact risk (death from floods; affected by in the next section. floods and droughts) L = Total loss (persons killed or affected) P = Population 4. deveLopment and G = GDP per capita E = Female educational enrollment rate vuLnerabILItY to extreme R = Precipitation Weather events T = Temperature = A random error term As we noted in Section 2, extreme floods have killed over 170,000 people since 1960; droughts have killed Event-related losses are drawn from the CRED data- millions; and billions of people have been seriously base; data on population, GDP per capita (in constant affected by extreme weather events. A changing climate $US 2000) and education are drawn from the World may carry even greater risks in the future. But, as we Bank's World Development Indicators. We use noted above, changing development levels will also constant-dollar GDP data to maximize the sample size affect resilience in the face of weather shocks. To deter- for years prior to 1980. Data on net female primary mine the balance between worsening weather and grow- and secondary enrollment rates are limited to the period ing resilience over time, we specify a model of since 1990. To extend the series for panel estimation, weather-related impact risk and estimate it using panel we "backcast" them using panel regressions that relate data for the period 1960­2002. Panel estimation allows enrollment ratios to income per capita, life expectancy, for relatively clear interpretation of results, because it and the total fertility rate (see Section 5). These associ- absorbs many sources of potentially misleading cross- ational regressions are a transformation of the conven- sectional correlation into estimated country effects. At tional fertility equation, in which the total fertility rate the same time, however, the need for lengthy time series is a function of income per capita, life expectancy, and limits the estimation variables to a sparse set. female schooling. Our specification of the risk model incorporates three We specify the education rates as flogs, to ensure that effects: economic development, weather, and education. predictions are restricted to the interval 0­100 percent. 8 t he econom ics oF ada p tation to e x tr e me Weath er e v e n ts in d ev e lopin g c ou n tr ie s For a probability p between 0 and 1, the flog is defined as log [p/(1-p)]. This is an appropriate specification for tabLe 5. determInants of net femaLe the regressions in any case, since it is consistent with enroLLment ratIos (dependent varIabLe: natural lower and upper bounds for net enrollment fLog net enroLLment rate) rates. These are, in effect, first-stage regressions, with (1) (2) the total fertility rate, life expectancy, and per capita primary secondary income playing the role of instruments for second-stage -0.349 -0.555 total Fertility rate estimation of the climate impact regressions. Life (7.35)** (14.87)** expectancy is not significantly affected by deaths from life expectancy 0.068 0.043 (6.04)** (6.51)** floods, which are minuscule by comparison with deaths from other causes. Therefore, use of this variable as a log gdp per capita 0.139 (1.34) 0.893 (10.67)** first-stage instrument should not lead to biased esti- -1.774 -5.213 mates for schooling in the regression for death from constant (1.39) (8.14)** floods. observations 1849 1025 r-squared 0.89 0.98 Table 5 presents the fixed-effects estimation results, which are extremely robust in both regressions for the Note: absolute value of t statistics in parentheses: * significant total fertility rate and life expectancy. Per capita income at 5%; ** significant at 1%. is not a significant determinant of net female primary Source: ???? enrollment, but it has great explanatory power in the secondary enrollment regression. the two sets of econometric results to calculate demands Prior experimentation has indicated that the appropri- for primary and secondary education separately. This ate functional form for equation (1) is log-linear, and introduces something akin to double-counting, thereby that some climate indicators are much more robust than reducing or eliminating the effect of upward bias in the others. In every case, mean rainfall is far more robust regressions themselves. than either maximum or minimum rainfall, as well as other possible transformations (max/min ratio, etc.). The panel estimation results in Table 6 are quite robust No measure of temperature (mean, maximum, mini- for flooding risk, with all variables highly significant mum, max/min ratio, etc.) is significant in any of the and all parameter signs consistent with prior expecta- three risk equations. tions. Flood risk rises significantly as mean precipita- tion rises and falls significantly as per capita income Table 6 presents results for the risk of being killed or rises. The two schooling variables have highly signifi- affected by floods and affected by droughts. Footnote 1 cant impacts of approximately equal magnitude: Flood provides a detailed explanation of the criteria for deter- risk falls as female enrollment increases. The results are mining persons affected. Our instrumental variables much weaker for drought risk, partly because the approach creates high collinearity for primary and smaller sample size reduces degrees of freedom for esti- secondary enrollment ratios, so we estimate separate mation. GDP per capita is insignificant and has a equations for primary and secondary schooling. This perverse sign when it is included in these regressions, so serves our primary objective--inferring the schooling we have excluded it from the estimates. Education and needed to neutralize future climate impacts--but raises rainfall retain the correct signs, although only female the risk of upward bias in the separately estimated primary enrollment is significant at 5 percent. These impact of each schooling variable. Such bias would lead are the maximum likelihood estimates in any case, and to an ultimate underestimate of schooling required to we need to project drought effects, so we retain the two neutralize climate change (the higher the estimated drought equations in Table 6. effect of schooling on risk, the lower the schooling needed to neutralize additional risk from worsening To explore the implications of our results, we compare weather). However, we compensate for this by using actual historical losses with a counterfactual case in d e v e l o p m e n t a n d c l i m at e c h a n g e d i s c u s s i o n pa p e r s 9 tabLe 6. Weather rIsk modeL: fIxed-effects estImates (dependent varIabLes [Log (varIabLe/popuLatIon)]) (1) (2) (3) (4) (5) (6) Floods Floods Floods Floods droughts droughts killed killed affected affected affected affected Female primary -0.018 -0.017 -0.020 enrollment rate (4.67)** (2.70)** (2.05)* Female secondary -0.017 -0.015 -0.011 enrollment rate (5.36)** (2.87)** (1.16) gdp per capita -0.137 -0.122 -0.120 -0.107 ($'000) (6.22)** (5.46)** (2.91)** (2.53)* precipitation 0.010 0.010 0.016 0.016 -0.013 -0.014 (mm.) (3.28)** (3.23)** (3.61)** (3.60)** (1.75) (1.79) -12.833 -13.438 -6.630 -7.297 -0.883 -1.752 constant (28.38)** (36.28)** (9.29)** (12.92)** (1.08) (2.67)** observations 933 929 1051 1047 323 322 countries 120 120 134 134 83 82 r-squared 0.46 0.46 0.41 0.42 0.54 0.53 Note: absolute value of t statistics in parentheses: * significant at 5%; ** significant at 1%. Source: ???? which countries at each World Bank development level are assigned the same female primary enroll- tabLe 7. hIstorIcaL and sImuLated "best ment ratio as the "best practice" country in the practIce" Weather-reLated Losses, same World Bank class in the same year (e.g., all 1970­2000 low-income countries in 1985 are assigned the Best % highest female primary enrollment ratio among Risk category Historical practice Difference Difference low-income countries in 1985). We perform this Flood deaths 153,079 91,541 61,538 40.2 experiment to see how much difference feasible Floods affected (`000) 2,116,243 1,651,065 465,178 22.0 policy changes could have made for extreme droughts 1,342,337 675,797 666,540 49.7 weather risk. As Table 7 and Figure 4 show, our affected (`000) results strongly indicate that more progressive poli- Source: ??? cies would have made an enormous difference. From 1970 to 1999, the CRED database records 153,079 deaths from floods in low-income and lower- We conclude that a huge number of weather-related middle-income countries. In our "best practice" coun- tragedies could have been averted if more developing terfactual, by contrast, flood deaths number 91,541: countries had focused on progressive but feasible female 61,538 fewer people lose their lives. For numbers education policies. Countries that focused on female affected, the estimated differences are very large. From education suffered far fewer losses from extreme 1960 to 1999, the CRED database indicates that 2.12 weather events than less-progressive countries with billion people in developing countries were affected by equivalent income and weather conditions. It seems floods. In the counterfactual best practice case, this falls reasonable to assert that what has been true in the past by 465 million to 1.65 billion. For droughts, the will also be true in the future. Given the significance of number affected falls by about 667 million--from 1.34 income and female education in determining vulnerabil- billion to 676 million. ity to extreme weather events, we would expect 10 the e conom ics o F a dap tation to e xtr e me Weath er e ve n ts in d ev e lopin g c ou n tr ie s fIgure 4. extreme Weather Losses, 1970­1999: hIstorIcaL and best practIce Source: ??? d e v e l o p m e n t a n d c l i m at e c h a n g e d i s c u s s i o n pa p e r s 11 tabLe 8. projected net femaLe prImarY and secondarY enroLLment ratIos bY regIon East Asia and Latin America Middle East and Region Sub-Saharan Africa Pacific Islands and Caribbean North Africa South Asia Year P S P S P S P S P S 2000 54.9 19.7 95.0 80.5 90.8 61.3 80.5 56.0 69.5 42.0 2010 68.7 28.5 96.9 86.5 93.3 73.1 89.1 67.4 81.3 60.5 2020 78.0 43.8 97.8 89.9 94.9 81.1 92.9 77.6 85.8 73.5 2030 86.0 59.4 98.2 93.0 96.0 85.8 94.9 84.7 88.8 81.9 2040 90.8 70.9 98.6 94.8 96.9 89.6 96.1 89.3 90.7 87.2 2050 93.5 78.0 98.8 96.2 97.4 92.1 96.9 92.1 92.4 90.8 Source: ??? countries' future resilience to increase with economic the results with country projections of GDP per capita, growth and improvements in education. life expectancy, and the total fertility rate to predict future net female primary and secondary enrollment rates. We combine these rates with appropriate UN medium-variant female population cohort data to calcu- 5. projectIng baseLIne changes late the actual number of females enrolled in primary and secondary school. Finally, we total enrolled females In femaLe educatIon and relevant cohort females by region for primary and secondary schooling, and form the ratios to project The panel estimation results in Table 5 provide a regional primary and secondary enrollment ratios. reasonable basis for projecting the future paths of female primary and secondary education in each coun- Although our economic and demographic projections try, given our exogenous projections of income and are "moderate" by current standards, they nevertheless population. Our projections for life expectancy and the entail continued rapid progress in female education. By total fertility rate are taken directly from the UN's 2050, Sub-Saharan Africa increases its net female medium-variant population forecast. Given the paths primary enrollment rate from 54.9 to 93.5, and its net of the three variables (life expectancy, total fertility rate, female secondary enrollment rate from 19.7 to 78.0. income per capita), we use the fixed-effects results South Asia also makes rapid progress, moving its female reported in Table 5 to plot the paths of future net primary and secondary enrollment rates from 69.5 to female primary and secondary enrollment rates. Here it 92.4 and 42.0 to 90.8, respectively. East Asia/Pacific, is worth repeating that we estimate both equations in Latin American/Caribbean, and Middle East/North Table 5 using flog transformations on net enrollment Africa also move upward, but proportionately less rates, which insure that projections are bounded in the because they start from higher bases. While educa- range 0­100. tional progress is quite noteworthy in this scenario, it still falls well short of the Millennium Development To illustrate the implications of our approach, Table 8 Goals. For example, Sub-Saharan Africa only presents projected future schooling rates by region. We approaches the MDG for female primary education by compute these rates in several steps. First, we estimate 2050. the panel regressions, incorporating subregional dummies as well as country effects.6 Then we combine 6 We include fixed effects for 25 subregions. 12 the e conom ics o F a dap tation to e xtr e me Weath er e ve n ts in d ev e lopin g c ou n tr ie s 6. cLImate change, deveLopment, have a significant impact on climate vulnerability. To assess the relevant magnitudes, we develop a detailed and future vuLnerabILItY illustration for India that incorporates variations in the GCMs and development conditions. Now we turn our attention to simulating the future. Our approach is identical to the counterfactual We focus on computed risks, or incidence probabilities, approach in Section 4, except that all of the right-hand because they provide clear insight into the impact of variables are projected for this exercise. To get a clear model variables on projected future losses. In each case, sense of the stakes, we introduce several variants: We the relevant probability is multiplied by population to estimate impacts with and without climate change, for provide an overall loss estimate. India's population is both GCM climate projections, and with and without projected to continue growing to about 1.66 billion improvements in income and female education. We people by 2050. Since the population is growing, even introduce the latter variant to highlight the stark differ- constant risk (measured as a loss probability) will trans- ence in results when only future emissions are counted, late to more losses when it is multiplied by the growing not the economic and human development that accom- population. pany those emissions. As we have noted in Section 3, forty years hence many of today's low- and lower- In Figure 5, our historical baseline for India is set at middle-income countries will have experienced major income per capita and life expectancy in 2000, and growth in income and female education. Our panel mean precipitation during the period 1995­2000. The results in Table 6 suggest that the latter changes will associated annual loss probabilities are .62 per million fIgure 5. Loss rIsks In IndIa, 2000­50 Source: ??? d e v e l o p m e n t a n d c l i m at e c h a n g e d i s c u s s i o n pa p e r s 13 for being killed by a flood; 0.0039 for being affected by India (at 2000 income per capita and educational a flood; and .0509 for being affected by a drought. enrollment rates) would yield 44,038 expected deaths From this baseline, we forecast the impact of from flooding. In the static case, this rises to 45,458 for GCM-projected changes in mean precipitation, while the NCAR climate change scenario and falls to 41,325 holding income and life expectancy at their 2000 levels. for CSIRO. The results, labeled "Static" in Figure 5, show the magnitudes of the projected impacts, as well as their In comparison to these relatively small mortality effects, directions. NCAR is the wettest global scenario, and the number of people affected by floods is larger by one this is reflected in the Indian projections. In the static order of magnitude and people affected by droughts by NCAR case, the risk of death from flooding rises from two orders of magnitude. For floods, the baseline prob- 0.62 in 2000 to 0.66 in 2050. Conversely, CSIRO, ability of being affected (in the static-India case with no which produces the driest scenario, projects a drop in climate change) is .0039. This rises to .0043 for NCAR mean precipitation and an associated fall in flood- climate change and falls to .0035 for CSIRO. These related death risk: from 0.62 in 2000 to 0.57 in 2050. are much larger fractions than the death risks, and they translate to large absolute numbers. Projected people Thus, holding economic and social development affected annually rises from 4.0 million to 6.5 million in constant at 2000 levels, precipitation variations across the baseline case (constant risk, population growth), GCMs include a range of about .09 per million in flood rises to 7.2 million for NCAR (in a static India), and death risk. In all cases, the deviation from current falls to 5.7 million for CSIRO. Thus, the range of death risk is sufficiently small to be dominated by impacts is between 700,000 more people and 800,000 population growth in the assessment of losses. Table 9 fewer people affected by floods. The associated totals presents the relevant projections. In the "Historical" and differences in Table 9 are quite large: A fifty-year case, projected annual deaths increase from 626 to 1,023 increase from the baseline of 15 million for NCAR (the risk of death remains constant, while population (293.6 vs. 278.6 million), and a decrease of 27.7 million continues growing). Projected annual deaths from for NCAR. The numbers are larger by another order of flooding rise to 1,092 for NCAR, the wettest scenario, magnitude for drought, but in the opposite direction. and fall to 950 for CSIRO, the driest. Clearly, the For annual numbers affected by drought, the baseline differences are quite small in absolute terms: By 2050, case rises from 51.6 million in 2000 to 84.4 million in climate is responsible for 69 additional flood-related 2050. The numbers rise less rapidly in the NCAR deaths per year in the NCAR scenario, and 73 fewer (wet) case for a static India, to 77.1 million and more deaths in CSIRO. For the entire fifty-year period, rapidly for CSIRO (dry), to 93.4 million. Translated to continuation of the historical climate pattern in a static fifty-year totals, the differences are huge: 337 million tabLe 9. aLternatIve scenarIos for IndIa, 2000­50 Affected by floods Affected by droughts Year Scenario Flood deaths (Million) (Million) csiro ncar historical csiro ncar historical csiro ncar historical 2000 626.1 626.1 626.1 4.0 4.0 4.0 51.6 51.6 51.6 2050 static 950.4 1,092.3 1,023.4 5.7 7.2 6.5 93.4 77.1 84.4 2050 devel 461.0 461.0 1,023.4 2.8 3.6 6.5 75.7 62.5 84.4 totals 2000­50 static 41,325.2 45,457.6 44,038.4 250.9 293.6 278.6 3,967.4 3,477.1 3,630.2 2000­50 devel 27,768.1 30,269.3 44,038.4 176.7 204.1 278.6 3,362.6 2,956.4 3,630.2 Note: Flood deaths in 2000, 2050 (static), and 2050 (devel) are annual. Source: ??? 14 t he econom ics oF ada p tation to e x tr e me Weath er e v e n ts in d ev e lopin g c ou n tr ie s more people affected for CSIRO, and 153 million fewer probabilities of being killed or affected plunge so for NCAR. sharply that they dominate rising population in the calculation of total losses. The result is many fewer All of the cases discussed above have elements in deaths and people affected by floods in 2050, although common: In an India that experiences no change in there is still a climate affect at a much lower level. A income and life expectancy, population growth alone rapid fall is also evident for risk in the NCAR scenario (with constant loss risk) ensures that losses from for drought, although much less so for CSIRO. extreme weather events increase substantially, even if there is no climate change. The projected range of When we translate these risks into total losses, the climate changes will alter the forecast, making it lower results are quite striking. The India of 2050 has annual in some cases and higher in others. But in the static- flooding deaths of about 461 for NCAR and CSIRO, India case, all the scenarios project greater future losses, vs. 1,023 in the baseline. It has 3.6 million people and climate effects that are smaller proportionally than affected by floods in NCAR and 2.8 million in CSIRO, anticipated effects from population change. vs. 6.5 million in the baseline. In the case of droughts, 62.5 million are affected in NCAR and 75.7 million in Of course, all of the projections above are unrealistic, CSIRO, vs. 84.4 million in the baseline. because they assume a static India that bears no resem- blance to the India in the climate change forecasts. As We draw two conclusions from these results. First, it we have seen in a previous section, that India is quite still makes sense to discuss financial support for adapta- close to present-day Chile in income and education tion to climate change in this context, because risks and levels by 2050. Despite their unreality, we have losses can still be greater with climate change than included the static-India forecasts because we believe without it. But second, and perhaps more important, that they reflect the implicit assumptions in many our results strongly indicate that the India of 2050 will current climate-impact analyses. suffer fewer losses from extreme weather after four more decades of climate change than present-day India For an instructive contrast, we now turn to projections suffers. that also utilize the fixed-effects estimates for equation (1), but incorporate our income and education projec- For global perspective, we have included the same tions for India. Although we will review the numbers projection comparisons for all developing countries in in some detail, the basic results are made graphically Table 10. Although the magnitudes are larger, they clear by the "Development" scenarios in Figure 5. For replicate the patterns that we have just discussed. The flooding, in both NCAR and CSIRO scenarios, the developing world of 2050 may well suffer more losses tabLe 10. aLternatIve scenarIos for deveLopIng countrIes, 2000­50 Affected by floods Affected by droughts Year Scenario Flood deaths (Million) (Million) csiro ncar historical csiro ncar historical csiro ncar historical 2000 5,520 5,520 5,520 16.4 16.4 16.4 142.7 142.7 142.7 2050 static 10,861 11,018 10,871 24.6 27.1 25.5 262.2 241.6 250.5 2050 devel 3,425 3,464 10,871 8.3 9.3 25.5 184.8 169.2 250.5 totals 2000­50 static 419,849 427,755 423,185 1,054 1,140 1,102 10,690 10,116 10,304 2000­50 devel 231,330 234,889 423,185 645 693 1,102 8,516 8,055 10,304 Note: Flood deaths in 2000, 2050 (static), and 2050 (devel) are annual. Source: ??? d e v e l o p m e n t a n d c l i m at e c h a n g e d i s c u s s i o n pa p e r s 15 with climate change than without it (the impact cost of measures for emergency preparedness. A study depends on whether the wet or dry scenario dominates), of this type would be aided considerably by country- but the available evidence makes it very likely that it specific cost information for measures targeted on will suffer far fewer losses than presently in either case. floods or droughts. However, the representative infor- And our results for female education in equation (1) mation in Table 11 illustrates why we have not been reinforce a fundamental point: If we are really inter- able to employ such information. Its entries have been ested in reducing losses from climate events, assistance extracted from country reports by the Asian Disaster for greater resiliency now can make a huge difference. Reduction Center. The reports generally focus on summary information rather than specific information for emergency preparedness by type of disaster (e.g., 7. estImatIng the cost of floods, droughts). In the case of Japan, for example, much of the $34 billion expenditure is clearly for earth- adaptIng to extreme Weather quake-related measures. The listed funds for events Bangladesh are more than twice China's and four times Indonesia's, and they include both emergency food assistance and disaster management. Much of the 7.1 data on Weather Indonesian fund undoubtedly relates to geologic disas- em erg encY p re pa re dne s s ters (earthquakes, volcanic eruptions, tsunamis), as well co sts as weather-related disasters. Systematic work on the cost of adaptation to extreme Most reports do not provide time series information, weather events has been hindered by scanty data on the nor do they go beyond reporting for single funds or tabLe 11. dIsaster preparedness and management data Annual equivalent Country Agency Year (Million $US) armenia emergency management administration 2006 7.0 Bangladesh Food and disaster management Budget annual 500.0 china various agencies 2005 217.7 indonesia contingency budget for disaster response annual 125.8 india calamity relief Fund 2000-2005 5.1 Japan Budget for disaster risk reduction annual 34,000.0 Kazakhstan For debris flows 1999 200.0 republic of Korea national emergency management agency annual 300.0 thailand department of disaster prevention and mitigation 2003 25.6 thailand department of disaster prevention and mitigation 2006 63.9 thailand department of disaster prevention and mitigation 2005 46.0 thailand department of disaster prevention and mitigation 2004 32.4 mongolia total Budget 2006 12.5 malaysia disaster relief Fund annual 15.5 nepal emergency Fund 2006 0.015 pakistan ten-Year-perspective development plan 2001-2011 18.8 philippines national calamity Fund 2005 12.8 russian Federation Fund for prevention and elimination of emergency situations 2003 687.4 tajikistan activities for disaster management annual 5.5 Source: asian disaster reduction center, country reports. http://www.adrc.asia/ 16 t he econom ics oF ada p tation to e x tr e me Weath er e v e n ts in d ev e lopin g c ou n tr ie s national-level agencies.7 Local expenditures are not the 50-year sum of positive Ei for each of the three included. risks and, in keeping with our conservative approach, we choose the risk for which the sum is largest. In summary, the available data are far too spotty, non- specific, non-standardized, and temporally limited to permit estimation of cost functions that could be used 7 . 3 t he co s t o F cl i mat e for projection. In addition, they cover relatively few cha ng e ne ut ra l i z at i o n countries. There is simply no way to construct a reasonable cost analysis from such information. Now we are ready to compute the cost of neutralizing the risk impact of more extreme weather events via increased schooling for young women. Once we have 7.2 th e e ducation a lte rnat i v e chosen the appropriate Ei for each schooling level in each country, computing the associated incremental cost Since direct cost measures cannot be derived from the involves two steps. First, we obtain the number of new available data, we turn to an indirect approach. As the students by multiplying Ei (as a percent) by the panel results in Table 6 show, improvements in female number of females in the appropriate age cohort.9 education are powerfully associated with reductions in Then we multiply by projected expenditure per pupil. disaster risks once changes in weather and income are To compute projected unit expenditures, we have drawn accounted for. In this section, we exploit this relation- on the World Bank's World Development Indicators to ship to address the adaptation cost question indirectly. estimate panel regressions for primary and secondary expenditures per student as a proportion of GDP per Our approach applies straightforward algebra to equa- capita. After extensive experimentation with available tion (1). Given an anticipated change in precipitation, and plausible right-hand variables (e.g., per capita we calculate the increase in education that will be just income, size of student population, time trend), we find sufficient to restore the risk level prior to the precipita- significance only for country, subregional, and regional tion change. With subscripts B for the baseline case fixed effects. We use all three sets of fixed effects to get and N for the risk-neutralizing case, we impose the the most accurate estimates for countries that have no following constraint on the relevant elements of equa- unit expenditure data in the WDI. Then we apply the tion (1) (the others cancel because they remain country estimates to projected per capita income to unchanged): obtain predicted expenditures per primary and second- ary student by year. We combine these with the calcu- (2) 2 EiN + 3 RiN = 2 EiB + 3 RiB lated numbers of primary and secondary students required for "climate change neutralization" to obtain This yields the change in the educational enrollment our estimate of the public cost. rate that will neutralize the change in risk introduced by deviation of rainfall from the baseline case: Table 12 provides a representative set of results for three countries: Republic of Congo, Nepal, and Nicaragua. 3 For ease of interpretation, we present results at ten-year (3) Ei = EiN - EiB = - [ R - RiB ] (2<0, 3>0) 2 iN intervals, beginning in 2010. The results highlight the For each education level, we calculate Ei for persons global diversity that interacts with the GCM killed by floods, persons affected by floods, and persons affected by droughts. Adopting a conservative losses from droughts) But in the case of floods, such an exercise would approach, we only consider positive Ei.8 We compute require assigning relative weights to being killed and being affected (injured, rendered homeless, or requiring temporary assistance). It would also require the assignment of relative weights to the affects of floods and droughts. Rather than adopt arbitrary weights, we take the 7 Our thanks to Tim Essam for his help in gathering this information. conservative approach and use the greatest change in enrollment ratio across the three risk categories. 8 In a wetter climate regime, the number of people killed and affected by floods will rise, but the number affected by droughts will fall. The 9 Our primary-school cohort is young women in age group 5­9, plus half converse is true for a dryer regime. A complete accounting would of young women 10­14. Our secondary cohort is half the age group therefore involve calculation of net impacts (losses from flooding vs. 10­14, plus the age group 15­19. d e v e l o p m e n t a n d c l i m at e c h a n g e d i s c u s s i o n pa p e r s 17 tabLe 12. cLImate-neutraLIzIng femaLe educatIon--students and costs, (repubLIc of congo, nepaL, and nIcaragua) CSIRO NCAR CSIRO new NCAR new Cost per Cost per CSIRO NCAR new primary secondary new primary secondary primary secondary total cost total cost students students students students student student Year ($'000) ($'000) (`000) (`000) (`000) (`000) ($US) ($US) Republic of Congo 2010 1,399 0 4.7 4.1 0.0 0.0 97 233 2020 4,341 0 10.5 9.8 0.0 0.0 128 307 2030 11,868 4,729 21.9 20.9 8.7 8.3 164 395 2040 24,179 16,055 32.9 32.4 21.8 21.5 218 524 2050 23,501 23,501 24.8 24.7 24.8 24.7 280 672 Nepal 2010 5,935 0 93.6 88.4 0.0 0.0 32 33 2020 15,059 0 182.5 183.7 0.0 0.0 40 42 2030 27,796 500 274.9 262.4 4.9 4.7 51 53 2040 27,237 9,382 198.8 204.6 68.5 70.5 66 69 2050 26,494 14,319 151.4 154.8 81.8 83.7 85 88 Nicaragua 2010 0 5,889 0.0 0.0 35.9 69.4 89 39 2020 0 12,343 0.0 0.0 37.0 155.3 118 51 2030 1,788 12,488 6.1 11.8 27.7 116.9 159 69 2040 9,425 12,911 21.6 46.0 21.6 81.4 226 99 2050 3,695 13,549 6.4 12.4 17.8 58.3 313 137 Source: ??? projections. Overall costs are higher for CSIRO in costs of $218 and $524, yields a total expenditure of Congo and Nepal, and for NCAR in Nicaragua; we $24.2 million. Projected short-term moderation of focus on the higher-cost scenario for each country. climate impact after 2040 reduces the numbers in 2050. Here it is useful to recall that these are adjustments from a baseline in which the countries continue their In Nepal, the number of needed additional students in socioeconomic development. Weather-impact risks the CSIRO (higher-cost) scenario is far greater than for decline as income and life expectancy increase. The the Congo and the unit costs substantially lower. When numbers in Table 12 reflect the deviations from this combined, the two factors yield climate-neutralizing baseline, which assumes no climate change. costs that increase from $5.9 million in 2010 to $27.2 million in 2040, then fall to $26.5 million in 2050. The In Congo, neutralizing the effect of climate change in NCAR scenario is more potentially damaging for the CSIRO scenario requires 4,700. additional females Nicaragua than CSIRO. Neutralizing the impact of in primary school in 2010, along with 4,100 additional NCAR-projected change requires the addition of female students in secondary school. The associated 35,900 young women to primary schooling and 69,400 annual schooling costs for primary and secondary to secondary schooling in 2010. The total cost is $5.9 students are $97 and $233, respectively. When these are million in 2010, increasing to $13.5 million in 2050.10 applied to the schooling increments, the result is an additional expenditure of $1.4 million in 2010. The numbers increase steadily through 2040. In that year, 10 Like many countries in Latin America, Nicaragua is reported by the the addition to schooling is 32,900 primary students World Development Indicators as spending more per capita on primary students than on secondary students. The results for Nicaragua in and 32,400 secondary students which, at projected unit Table 12 reflect this disparity. 18 t he econom ics oF ada p tation to e x tr e me Weath er e v e n ts in d ev e lopin g c ou n tr ie s Table 13 summarizes our results at the regional level. similar. For both GCM scenarios, the requisite annual Here the scale of effort needed for climate neutraliza- expenditure rises from about $200 million in 2010 to tion becomes apparent. In Sub-Saharan Africa, the over $2 billion in 2050. By the latter date, climate overall impacts of CSIRO and NCAR are roughly neutralization in CSIRO requires 1.7 million additional tabLe 13. cLImate-neutraLIzIng femaLe educatIon--students and costs (deveLopIng regIons) CSIRO CSIRO NCAR NCAR CSIRO NCAR new new new new total total primary secondary primary secondary cost cost students students students students Year ($'000) ($'000) (`000) (`000) (`000) (`000) Sub-Saharan Africa 2010 179,036 211,757 800 880 981 974 2020 562,746 672,806 1,847 2,258 2,422 2,508 2030 1,001,736 1,044,923 2,117 2,843 3,038 4,262 2040 1,680,756 1,623,282 2,038 3,311 3,383 6,481 2050 2,294,642 2,203,969 1,708 3,488 2,967 7,053 South Asia 2010 528,691 266,339 2,264 2,603 990 1,020 2020 1,567,993 784,403 4,354 5,277 1,961 2,024 2030 2,983,226 1,771,123 5,129 7,143 2,960 3,139 2040 4,101,156 3,502,443 4,470 6,040 3,752 4,056 2050 5,446,895 4,853,930 4,277 5,357 3,539 3,681 East Asia and Pacific Islands 2010 339,783 397,749 1,100 1,237 872 1,276 2020 883,470 1,019,321 1,423 2,580 1,561 1,984 2030 1,352,575 1,312,094 990 2,681 1,130 1,635 2040 2,104,735 1,723,739 768 3,367 820 1,554 2050 795,767 2,636,736 241 1,315 780 2,307 Eastern Europe and Central Asia 2010 112,192 150,593 200 246 214 313 2020 339,834 479,968 419 436 459 561 2030 417,795 727,301 301 359 375 561 2040 476,751 868,335 216 265 311 429 2050 524,959 1,554,609 148 156 345 447 Middle East and North Africa 2010 60,080 32,367 114 147 46 93 2020 171,062 95,127 269 318 101 205 2030 327,340 226,553 255 339 135 259 2040 573,419 443,764 232 322 181 312 2050 891,987 958,291 219 323 244 372 Latin America and the Caribbean 2010 396,497 513,496 465 597 619 959 2020 1,016,825 1,266,047 555 1,486 929 1,928 2030 617,302 1,187,125 412 833 700 1,687 2040 218,728 1,370,353 200 644 526 1,445 Source: ??? d e v e l o p m e n t a n d c l i m at e c h a n g e d i s c u s s i o n pa p e r s 19 primary school students and 3.5 million additional secondary tabLe 14. cLImate-neutraLIzIng femaLe educatIon students. In the NCAR case, (gLobaL totaLs, 2010 ­50) these numbers rise to 3.0 million CSIRO CSIRO NCAR NCAR CSIRO NCAR new new new new primary students and 7.1 million total total primary secondary primary secondary secondary students. The cost cost students students students students Year ($million) ($million) (`000) (`000) (`000) (`000) Sub-Saharan case points to 2010 1,616,279 1,572,299 4,943 5,710 3,721 4,634 another feature of geographic 2020 4,541,929 4,317,672 8,867 12,355 7,433 9,209 diversity that has implications for 2030 6,699,975 6,269,119 9,203 14,199 8,339 11,542 the cost of climate neutralization. 2040 9,155,545 9,531,917 7,923 13,948 8,973 14,278 Climate changes in the two 2050 10,900,000 13,600,000 6,959 11,341 8,282 14,892 scenarios have different geographic distributions. In the Source: ??? African case, by happenstance, the countries most adversely billion). In both scenarios, the implications for climate- affected in NCAR have significantly lower unit school- neutralizing female education are massive. By 2050, ing costs than the countries with the greatest effects in neutralizing CSIRO requires 7 million additional young CSIRO. As a result, climate-neutralizing expenditure is women in primary school and 11.3 million in secondary slightly lower in NCAR, even though the number of school. The corresponding numbers for NCAR are 8.3 additional students is substantially higher. million and 14.9 million. South Asia is also not far from cost parity in the two Table 15 provides a final summary by totaling annual climate scenarios. Although the expenditure difference expenditures for the period 2002­5011 at varying is large in 2010--$529 million in CSIRO vs. $266 discount rates. Overall, at a 0 discount rate, impact- million in NCAR--by 2050 the numbers are propor- neutralizing expenditures for additional female educa- tionally much closer ($5.4 billion and $4.9 billion, tion are $279.4 billion for CSIRO and $288.1 billion respectively). Other regions exhibit disparities, but with for NCAR. The totals fall sharply as the discount rate different patterns. East Asia and the Pacific Islands are increases. At 7 percent, present values in 2002 are dominated by China, whose rising prosperity generates $40.2 billion for CSIRO and $39.5 billion for NCAR. steadily increasing schooling costs. Costs in the Among the world's regions, it is both clear and unsur- CSIRO scenario dominate until 2040, when a projected prising that the largest climate-neutralizing expendi- climate shift significantly moderates climate stress tures are in the areas whose low incomes and schooling during the same period that it increases in NCAR. The rates are associated with higher climate impact risks. result is a reversal for CSIRO, as regional expenditures South Asia has the greatest expenditure in both climate fall from $2.1 billion in 2040 to $796 million in 2050, scenarios, with CSIRO much more costly than NCAR while NCAR expenditures continue expanding, from ($121.2 billion vs. $89.7 billion for a zero discount rate). $1.7 billion to $2.6 billion. In the remaining three Sub-Saharan Africa and East Asia/Pacific Islands are in regions, NCAR dominates expenditures in varying the next rank, with rough balance across the two scenar- degrees. ios--around $46 billion for Sub-Saharan Africa and $55 billion for East Asia/Pacific. In the next rank, both Table 14 summarizes the annual results, which tell a Eastern Europe/Central Asia and Latin America/ story of impressive magnitude. Overall, annual expen- Caribbean have NCAR expenditures about twice as ditures are remarkably close for CSIRO and NCAR high as CSIRO expenditures. The pattern reverses for until 2040. Climate-neutralizing educational expendi- Middle East/North Africa, where expenditures are ture is about $1.6 billion for both in 2010. By 2040, $15.7 billion for CSIRO and $12.9 billion for NCAR. NCAR is slightly ahead ($9.5 billion vs. $9.2 billion). Projected short-run climate shifts and a host of other factors shift the balance by 2050, and NCAR finishes 11 A few missing value problems prevent generation of fully comparable numbers for 2000 and 2001. well ahead of CSIRO ($13.6 billion annually, vs. $10.9 20 t he econom ics oF ada p tation to e x tr e me Weath er e v e n ts in d ev e lopin g c ou n tr ie s tabLe 15. cLImate-neutraLIzIng femaLe educatIon, 2002­50 (gLobaL and regIonaL costs: seLected dIscount rates) ($us bILLIon) Discount Global Global Sub-Saharan East Asia and Eastern Europe Rate (%) Total Total Africa Pacific Islands and Central Asia csiro ncar csiro ncar csiro ncar csiro ncar 0 279.4 288.1 46.4 47.0 53.8 57.7 15.8 30.2 3 110.3 111.1 17.1 17.8 21.7 23.2 6.6 11.7 5 64.5 64.1 9.5 10.1 12.8 13.8 4.0 6.8 7 40.2 39.5 5.7 6.1 8.0 8.8 2.6 4.2 Discount Latin America and Middle East and Rate (%) Caribbean North Africa South Asia csiro ncar csiro ncar csiro ncar 0 26.5 50.6 15.7 12.9 121.2 89.7 3 13.0 22.3 5.7 4.4 46.1 31.7 5 8.8 14.0 3.2 2.3 26.2 17.1 7 6.2 9.3 1.9 1.3 15.8 9.8 Source: ??? 8. summarY and concLusIons asks two key questions: As climate change increases potential vulnerability to extreme weather events, how In this paper, we have addressed several questions that many additional young women would have to be are relevant for the international discussion of adapta- educated to neutralize this increased vulnerability? And tion to climate change: How will climate change alter how much would it cost? the incidence of these events, and how will their impact be distributed geographically? How will future socio- Our study relies heavily on fixed-effects estimation of economic development affect the vulnerability of risk equations that link losses from floods and droughts affected communities? And, of primary interest to during the period 1960­2003 to three basic determi- negotiators and donors, how much would it cost to nants: weather events that increase potential losses, neutralize the threat of additional losses in this context? income per capita, and female education. We estimate separate equations for the risk of death from a flood, From a narrow technical perspective, it might be desir- the risk of being affected by a flood, and the risk of able to address the latter question with a detailed engi- being affected by a drought (the data are too sparse to neering cost analysis of specific disaster prevention support estimation for death from droughts). measures. However, as we show in the paper, existing cross-country information about relevant emergency Our analysis combines the estimated risk equations preparedness programs is far too sparse to support with projections of economic growth and population systematic analysis and projection. And in any case, we change, along with accompanying changes in primary believe that the effectiveness of such measures is contin- and secondary schooling. We develop three scenarios: gent on the characteristics of the communities that A baseline in which socioeconomic development employ them. We therefore adopt an alternative continues but the climate does not change, and two approach in this paper, focusing on the role of socioeco- scenarios with the same baseline development path but nomic development in increasing climate resilience. alternative weather paths driven by particularly "wet" Drawing on extensive research, our approach highlights and "dry" GCMs. For each GCM scenario, we calculate the importance of female education and empowerment the associated changes in the risks of death from floods in reducing weather-related loss risks. Our cost analysis and being affected by floods or droughts. Then, d e v e l o p m e n t a n d c l i m at e c h a n g e d i s c u s s i o n pa p e r s 21 choosing the worst-case risk, we calculate the increase scenario, country, and year. Variations in projected in female schooling that would neutralize this addi- climatic, socioeconomic, and demographic variables are tional risk. We multiply the results by expenditures per more than sufficient to produce wide disparities in student to estimate the total educational investment outcomes by 2050, even among countries within the required to neutralize the additional weather risk posed same region. At the country and regional levels, neither by climate change. climate scenario dominates in all cases. The "wet" scenario generates higher risk-neutralizing expenditure Our approach is conservative, in the sense that it is very on female schooling in some countries and regions; the unlikely to underestimate the required investment. "dry" scenario is more costly in others. Among regions, First, we base our cost assessment on general prepared- South Asia requires the most expenditure in both ness via increased education, rather than more narrowly climate scenarios, followed by Sub-Saharan Africa and targeted investment in emergency preparedness. East Asia, and then more distantly by the other regions. Second, we base our cost calculation on worst-case risk scenarios, which require the greatest increase in school- At both regional and global levels, we find an impres- ing to neutralize. Third, we incorporate only projected sive scale for the requisite increases in female education increases in vulnerability, not decreases. As an alterna- expenditure. By mid-century, neutralizing the impact of tive, for example, we could perform a net impact analy- extreme weather events requires educating an additional sis for a wet climate scenario that would subtract 18 to 23 million young women at a cost of $11 to $14 expected decreased losses from drought from expected billion annually. For the period 2000­50 as a whole, increased losses from flooding. Fourth, our analysis both GCM scenarios entail about $280 billion in addi- employs the two GCMs (among approximately twenty) tional expenditure. The present value of these expendi- that generate the wettest and driest scenarios at the tures is substantially reduced by time-discounting, even global scale. Other GCMs would generate more at modest rates, but the basic result stands: In the moderate intermediate results. Finally, we do not aver- developing world, neutralizing the impact of worsening age across the two GCMs, which would have the effect weather over the coming decades will require educating of neutralizing their extreme signals. a large new cohort of young women at a cost that will steadily escalate to several billions of dollars annually. In summary, we believe that our approach is sufficiently However, it will be enormously worthwhile on other conservative to create a strong upward bias in our cost margins to invest in education for millions of young estimation. It is certainly possible that the "true" cost of women, who might otherwise be denied its many adaptation to extreme weather events is lower than our benefits. estimates, but we very much doubt that it is higher. At the same time, our approach offers significant references co-benefits because female education has a much broader sphere of potential influence than direct invest- Albala-Bertrand, J. 1993. Political Economy of Large ment in emergency preparedness. As the development Natural Disasters. New York: Oxford University Press. literature has noted for many years, educating young women is one of the major determinants (indeed, some Burton, K., R. Kates, and G. White. 1993. The would argue, the major determinant) of sustainable Environment as Hazard. 2nd Edition. New York: development. A disaster-prevention approach that Guilford Press. focuses on investment in female education therefore has an expected social rate of return on other margins that Horwich, G. 2000. "Economic Lessons From the Kobe probably warrants the exercise, even if the expected Earthquake." Economic Development and Cultural benefits in reduced disaster vulnerability are overstated Change 48: 521­542. (and in fact, we believe the opposite to be true). Hughes, Gordon. 2009. The Economics of Adaptation to Our analysis has generated a set of estimates for Climate Change: Global Estimates for Infrastructure. required female schooling and associated costs by GCM Washington, DC: World Bank. 22 t he econom ics oF ada p tation to e x tr e me Weath er e v e n ts in d ev e lopin g c ou n tr ie s IPCC. 2000. Tol, Richard S.J., Brian O'Neill, and Detlef P. van Vuuren. 2005. A Critical Assessment of the IPCC SRES Kahn, Matthew. 2005. "The Death Toll From Natural Scenarios. Hamburg: Research Unit, Sustainability and Disasters: The Role of Income, Geography, and Global Change, Hamburg University. Institutions." The Review of Economics and Statistics 87(2) 271­284. Tol, R., and F. Leek. 1999. "Economic Analysis of Natural Disasters." In T. Downing, A. Oisthoorn, and King, Elizabeth, and Andrew Mason. 2001. Engendering R. Tol, eds. Climate Change and Risk. London: Rutledge Development. World Bank Policy Research Report. Washington, DC: World Bank. Toya, H., and M. Skidmore. 2005. "Economic develop- ment and the impacts of natural disasters." Economics Oxfam International. 2008. Rethinking Disasters: Why Letters 94 (1): 20­25. Death and Destruction is not Nature's Fault but Human Failure. New Delhi: South Asia Regional Centre, Oxfam (India) Trust. The World Bank Group 1818 H Street, NW Washington, D.C. 20433 USA Tel: 202-473-1000 Fax: 202-477-6391 Internet: www.worldbank.org/climatechange