d i s c u s s i o n pa p e r n u m B e r 2 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 56660 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 Costs of Adapting to Climate Change for Infrastructure d i s c u s s i o n pa p e r n u m B e r 2 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 Costs of Adapting to Climate Change for Infrastructure *gordon hughes **paul chinowsky ***Ken strzepek Note: This paper is based upon work that has been commissioned by the World Bank as part of the Economics of Adaptation to Climate Change study. The results reported in the paper are preliminary and subject to revision. The analysis, results, and views expressed in the paper are those of the authors alone and do not represent the position of the World Bank or any of its member countries. * Department of Economics, University of Edinburgh, UK ** Department of Civil , Environmental and Architectural Engineering, University of Colorado, Boulder, CO *** MIT Joint Program on the Science and Policy of Global Change, Cambridge, MA 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: Darjeeling, India. © Shutterstock Images, LLC Note: All dollars are U.S. dollars unless otherwise indicated. taBle oF contents abstract vii section 1. setting the scene 1 section 2. data 5 section 3. climate change 5 section 4. econometric specifications 8 section 5. the effects of climate on demand for infrastructure 13 section 6. calculating the cost of adaptation 30 section 7. estimates of the costs of adaptation 34 section 8. conclusion 43 references 43 Tables 1. correlation matrix of climate variables and historic demographic indicators 7 2. projection equations for electricity generating capacity, fixed telephone lines, and electricity network coverage 13 3. projection equations for municipal and industrial water demand 17 4. projection equations for water and sewer networks 19 5. projection equations for roads 20 6. projection equations for other transport 23 7. projection equations for health 25 8. projection equations for social infrastructure 27 9. projection equations for average household size 31 10. delta-p costs of adaptation by category and country class for 2010­50 (us$ billion per year at 2005 prices, no discounting) 34 iv t h e c os ts oF ad a p tin g to c limate c h an ge For in Fr as tr u c tu r e 11. delta-p costs of adaptation by infrastructure category and World Bank region for 2010­50 (us$ billion per year at 2005 prices, no discounting) 36 12. delta-p costs of adaptation by decade and World Bank region for all infrastructure (us$ billion per year at 2005 prices, no discounting) 38 13. total costs of adaptation by infrastructure category and country class for 2010­50 (us$ billion per year at 2005 prices, no discounting) 40 14. total costs of adaptation by infrastructure category and region for 2010­50 (us$ billion per year at 2005 prices, no discounting) 41 appendix 1. derivation of the climate dose-response relationships 45 references 51 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 aBstract An approach to estimating the costs of adapting to costs. The second component measures the effect of climate change is presented along with results for major climate changes on the long-run demand for infrastruc- components of infrastructure. The analysis separates the ture. The results indicate that the price/cost element is price/cost and quantity effects of climate change. The usually less than 1 percent of baseline costs, while the first component measures how climate change alters the quantity effect may be negative for many countries. cost of a baseline program of infrastructure develop- ment via changes in design standards and operating 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. Setting the Scene benchmark that depends upon factors such as popula- tion and income. This paper presents the results of a global analysis of the costs of adapting infrastructure to climate change In the period from t to t+1, for example from 2010 to over the period from 2010 to 2050. The analysis was 2015, the country will have to invest in order to meet carried out as part of the World Bank's Economics of the efficient level of infrastructure in t+1 and to replace Adaptation to Climate Change study. In this context, infrastructure in situ at date t, which reaches the end of infrastructure has been given a rather broad definition. its useful life during the period. Thus, the total value of It includes the usual types of infrastructure services, investment in infrastructure of type i in country j and including transport (especially roads, rail, and ports), period t is electricity, water and sanitation, and communications.1 In addition, urban and social infrastructure such as I ijt = Cijt [Qijt +1 - Qijt + Rijt ] (1) urban drainage, urban housing, health and educational facilities (both rural and urban), and general public where Cijt is the unit cost of investment and Rijt is the buildings have been included. quantity of existing infrastructure of type i that has to be replaced during the period. The change in the total The basic approach is extremely simple. For any coun- cost of infrastructure investment may be expressed in try j and date t (t = 2010, 2015, ..., 2050), we start from terms of the total differential of (1) with respect to the the assumption that there is some "efficient" level of relevant climate variables that affect either unit costs or provision of infrastructure of type i, which will be efficient levels of provision for infrastructure of type i: denoted by Qijt . The efficient level of infrastructure is that which would be reached if the country had invested I ijt = Cijt [Qijt +1 - Qijt + Rijt ] + (Cijt + Cijt )[ (2) +1 - Qijt + Ri Qijt up to the point at which the marginal benefits of addi- I marginal [Qijt +1 - Q tional infrastructure just cover the ijt = Cijtcosts--bothijt + Rijt ] + (Cijt + Cijt )[ Qijt +1 - Qijt + Rijt ] capital and maintenance--of increasing the stock of infrastructure. It is often argued that developing coun- An equivalent equation may be derived for the costs of tries tend to underinvest in infrastructure and that the operating and maintaining infrastructure. In the extent of underinvestment is particularly large for the discussion that follows, the first part of the right-hand poorest countries (AICD 2009). This is an important side of equation (2) is referred to as the Delta-P development issue, which is not directly related to component of the cost of adaptation, while the second climate change. Hence, the approach attempts to strip part is referred to as the Delta-Q component. These out the effects of country differences in their actual components themselves cover a number of ways in provision of infrastructure by establishing a common which climate change may cause changes in the costs or quantities of providing infrastructure services. 1 Limitations on the availability of comparable data meant that it was not possible to cover gas networks in the study. However, the costs of Delta-P. At the simplest level, changes in temperature, adaptation are likely to be minimal apart from any impacts on the level precipitation, or other climate variables may alter the of demand, which are likely to be similar to the pattern for electricity. 2 t h e c os ts oF ad a p tin g to c limate c h an ge For in Fr as tr u c tu r e direct cost of constructing infrastructure to a standard climate and for a changing climate. A similar specification. For example, seasonal weather variations exercise may be carried out for operating, mainte- can increase the costs of building. However, this is a nance, and replacement costs in order to calculate minor factor. More important is the impact of climate the increment in annualized infrastructure costs as change on the design standards that are applied in order a consequence of climate change. to maintain the quality of infrastructure services provided by a unit of infrastructure, such as a kilometer b. Delta-Q. The quantities of infrastructure assets of paved road or a fixed telephone connection (See required (holding income constant) will change as Canadian Standards Association 2006 for a discussion a consequence of different climatic conditions. of this issue). Again, this has two dimensions. The first is that climate change may change the level or composi- a. Changes in the frequency and/or the severity of tion of demand for energy, transport, and water at storms, flooding, and other extreme weather given levels of income, so we need to calculate the events may compromise the performance of infra- net impact of these changes in terms of capital structure designed to existing standards. Hence, it and operating costs. The second is that climate is common to refer to "climate proofing" invest- change will mean that countries have to invest in ments or ensuring "climate resilience." The study specific additional assets in order to maintain starts from the basis that design standards should specific standards of protection for non-infrastruc- be adjusted so as to deliver the same level of ture activities. performance as would have applied if climate change had not occurred. Thus, if roads or build- The Delta-P dimension of the study is uncontroversial ings are currently constructed to withstand a 1-in- in principle, though more or less difficult in practice. 50 or 1-in-100-year flood or wind storm, then the Various organizations have made broad brush estimates same design standard should apply, but under the of the cost of "climate proofing" existing investment circumstances of a changed frequency or severity programs in developing countries (UNFCCC 2007; of those events. The changes in the unit costs-- McGray et al. 2008). Typically, the analysis starts from Cijt --represent the costs of building infrastruc- a baseline program in investment by time of infrastruc- ture that delivers the same level of performance in ture. Then, an estimate is made of the percentage the face of different climatic stresses. The deriva- increase in unit costs required to ensure that invest- tion of the cost changes, expressed as dose- ments are resilient to climate change. response relationships for different climate stressors, are described in Appendix 1. The dose- One problem with the "climate proofing" approach response functions are applied to estimates of the concerns the investment program to which the cost of average values of climate variables under a climate proofing should be applied. For some sectors or scenario of a stable climate and alternative scenar- countries/regions, it is possible to start from a detailed ios for climate change by country.2 This gives a inventory of infrastructure assets and then to ask what series of cost increases--at constant 2005 prices-- investments will be required to meet future demand for by type of infrastructure, country, and time period. infrastructure services. The best example of this When applied to the baseline projection of infra- approach is a study of the costs of adaptation to climate structure demand, we obtain the Delta-Q cost of change in Alaska (Larsen et al. 2008). However, this adaptation; that is, the difference between the cost type of exercise requires an inventory of infrastructure of the baseline investment program for a stable assets and it does not take account of future investment in infrastructure. 2 Most climate models generate projections for 2° grid squares. For this study, these projections have been downscaled to 0.5° grid squares In the case of developing countries, many institutions and then population-weighted averages of the grid square values have that are concerned with adaptation to climate change been computed for each country. Thus, references to climate variables by country in this paper should be construed as referring to the popu- for infrastructure draw a distinction between (a) the lation-weighted averages of, say, precipitation for the various grid cost of eliminating the "development deficit,"--that is, squares that cover the country. 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 the gap between the infrastructure that a country Relying either upon wish lists or on the envelope of "ought" to have and the infrastructure that it actually what other countries at similar incomes have invested has--and (b) the cost of adapting to climate change on ignores the trade-offs that all governments have to the assumption that the country has an efficient level of make. Even if external assistance is available to fund infrastructure. The former is seen as a development capital projects, it is common experience that lack of problem, while the latter is a climate change problem. funds for operations and maintenance may lead to rapid Even though it is understood that money is fungible, deterioration in the services provided by stocks of infra- the two elements of total investment in infrastructure structure assets. might be financed out of different pots of money. Thus, it may be argued that the analysis should not be The corollary of this distinction between adaptation and based on some notional "efficient" level of infrastructure, the development budget is that the baseline program for but should start from the actual levels and growth of infrastructure investment used in constructing the infrastructure based on decisions that reflect real Delta-Q should not be derived from actual or planned constraints on budgets and the associated priorities. To investment in infrastructure. Instead, it should reflect examine whether the distinction is important in prac- the "efficient" demand for infrastructure.3 This is no tice, the full study has used two sets of baseline projec- simple task. The World Bank has recently completed a tions of demand for infrastructure. The "frontier" detailed assessment of infrastructure investment needs projection is derived by using frontier methods of esti- in 22 African countries, assuming a catch-up from mation to estimate econometric equations that charac- actual to efficient provision over a decade to 2020 terize the envelope of infrastructure demand given (AICD 2009). That exercise involved very substantial exogenous variables such as income, population, urban- work and cannot be extended to all countries in a short ization, etc. This is intended to provide an estimate of period. Instead, the analysis has to be based on an the "efficient" level of infrastructure demand as envis- econometric model that can be used to construct projec- aged in discussions of the development gap. In contrast, tions of the efficient demand for infrastructure up to the "panel" projection uses conventional projections 2050. derived from econometric estimates of the average rela- tionship between infrastructure and the exogenous While the principle of drawing a distinction between variables. the "development deficit" and adaptation to climate change is widely followed in international negotiations, The difference between the Delta-P estimates using the many economists consider that the distinction is either two sets of baseline projections is not as large as some unworkable in practice or simply wrong as a matter of might expect. There is an important reason for this. economic logic. The reason is that most assessments of We find that the relative gap between the frontier and the "efficient" demand for infrastructure ignore the panel projections tends to narrow, because there appears question of resources. A specific country might wish to to be convergence toward standard patterns of infra- have more roads, schools, or hospitals than the stocks structure provision. Further, the income elasticity of that are currently in situ, and the rest of the world demand for infrastructure is generally less than 1 for might agree that this would be a desirable goal. But, the frontier demand equations and is lower than the this is nothing more than a wish list independent of the equivalent income elasticities for the average demand resources that are available. With limited resources equations. For the frontier baselines projection, these some countries may choose to spend their funds on factors lead to a lower level of new investment in infra- providing better roads or more healthcare services. structure, but a higher level of expenditure on replacing and maintaining the initial level of infrastructure. Under the panel baseline projection, lower levels of 3 This paper will refer to the (efficient) demand for infrastructure and spending on replacement and maintenance are offset by will not attempt to address the question of how far the actual stocks of infrastructure are constrained by the supply of infrastructure assets. In higher spending on new investment. Depending on the effect, we assume that (a) we can identify an equation describing the initial development gap and the timing of new invest- long-run demand for infrastructure, and (b) supply constraints are not relevant when projecting the future investment program in calculating ment, it is possible--though not usual--for the cost of adaptation costs. 4 t h e c os ts oF ad a p tin g to c limate c h an ge For in Fr as tr u c tu r e adaptation to be larger for the average baseline than for will influence the nature of investment in roads. There the frontier baseline. are more complex but potentially larger effects operat- ing through the economic geography of urban life, In this paper we will focus exclusively upon the panel industry, and commerce; that is, in the ways in which projections derived from panel data models rather than we organize economic activity in space. Small changes the frontier models. This is consistent with the view may have significant consequences for the level of that the distinction between the development deficit investment in infrastructure. and the adaptation deficit is difficult to draw under the best of circumstances and may not be useful in practical While the principle that climate change may affect the terms. demand for infrastructure seems straightforward, the task of estimating the Delta-Q costs of adaptation is The second, Delta-Q, aspect of this work concerns the much more difficult for two general reasons. impact of changes in climate on the demand for infra- structure. To approach this issue, we have to consider a. Many of the impacts of climate on demand for the mechanisms by which changes in climate may affect infrastructure are long term in nature. This may the demand for infrastructure and how we might iden- not be true for electricity, but any influence of tify these consequences. For example, it is generally climate on the demand for roads will operate via accepted that demand for electricity depends upon the path of economic development over a period climate in general, but it is not so easy to identify the of one, two, or many decades. There are two key climate parameters when estimating the demand for consequences. First, we should not think of the electricity or for electricity-generating capacity. Note Delta-Q component of the costs of adaptation as that even these two variables may be subject to different arising on a regular schedule every five years. The influences because the seasonal or diurnal pattern of calculation merely identifies additions to and electricity demand is strongly influenced by climate.4 subtractions from a liability (or asset) that will Part of the difficulty is that the outcome depends upon materialize in future as economic activity adjusts the relative weights assigned to different factors. An to the changes in climate that are taking place. increase in average temperatures will lead to less Second, in planning for future infrastructure demand for heating in the colder seasons but more development, governments need to consider how demand for cooling in the warmer seasons. The overall climate change may affect the amount and type of direction of change is not easy to predict and is likely to infrastructure that is required if it will influence depend upon the way in which we set up the problem. future patterns of economic activity. Electricity is simple to think about by comparison with b. In practice, there is no way of examining the roads or other transport infrastructure because there is empirical impact of climate on the demand for an intuitive sense of the mechanisms involved in a rela- infrastructure other than through some form of tionship between climate variables and the stock of panel data analysis--pooling data for countries, electricity-generating capacity. But it would be wrong regions, states, or other geographical units over simply to impose the assumption that climate has no time. Inevitably, climate is a cross-sectional vari- effect on the demand for roads. Patently, climate vari- able (since year-to-year variations are weather), ables do affect the structure of economic activity hold- which may easily be confounded with other cross- ing other factors constant --for example, through the section fixed effects. This has prompted various level and composition of agricultural output--and this criticisms of the Ricardian approach to identifying the impact of climate change on agriculture or GDP on the grounds that climate variables are 4 There are also limitations on what one can obtain from climate projec- tions. For example, it is conventional to include degree-days as a cli- acting as a proxy for non-climate factors such as mate variable in equations predicting energy demand because of institutions. Some economists draw the conclu- heating requirements. The number of heating degree-days for a par- ticular location is calculated from the truncated distribution of temper- sion that climate variables should not be used in atures below some threshold--often 18°C--either on an hourly or a this way. We do not accept this view, since it daily basis. 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 closes off any possibility of estimating the impact The WDI data has been supplemented with data on of climate change on overall demand for infra- infrastructure availability from a wide variety of sources, structure. Instead, we have carried an extensive including other international organizations (FAO, ITU, econometric analysis of the role of climate vari- WHO, UNICEF, UPU), official country data (espe- ables in modeling the demand for infrastructure. cially census data), and various systematic surveys such The details are technical and take up a large as Demographic and Health Surveys (DHS) and Living amount of space, so that they are reported in a Standards Measurement Surveys (LSMS), which are separate paper (Hughes 2010). The issues and broadly consistent across countries. Even so, the final results are summarized in sections 4 and 5 below. dataset is very patchy in terms of coverage, especially for Our results suggest that the demand for some earlier periods. The panels are unbalanced and there are categories of infrastructure is affected by different many missing values for intermediate years. Thus, it is climate variables with important interactions with not possible to make use of econometric specifications income per capita and urbanization. involving autoregressive or similar errors over time. The results of our econometric analysis suggest that the A further remark concerns the nature of the data relat- absolute magnitude of the Delta-Q component of adap- ing to different types of infrastructure. In a few cases, tation should not be ignored in the long run. On the we have direct measures of the quantity of infrastruc- other hand, the Delta-P component is much more ture assets--for example, kilometers of paved or all predictable as a basis for discussing plans for adapting roads, kilometers of rail track, MW of generating capac- to climate change. For these reasons, our detailed esti- ity. More commonly we have to rely upon measures of mates of the costs of adaptation by 5-year period up to infrastructure output--for example, numbers of house- 2050 concentrate on the Delta-P component, while the holds connected to electricity, water, or sewer systems. Delta-Q estimates are presented as indicative estimates In practice, the efficient levels of infrastructure assets for the whole period. are closely linked to these output or input variables, so we believe that it is reasonable to base our projections on an analysis of these infrastructure indicators. 2. Data The core data used in this study is the World 3. climate change Development Indicators (WDI) database published in 2008 by the World Bank, which provides panel data for Describing the historic climate in a manner that is up to 168 countries and the years 1960 to 2006. The compatible with macroeconomic data is far from year 2005 is treated as the base year for all of our esti- straightforward without any of the complications of mation. Our work relies on the 2008 version of the projecting climate change into the second half of the database. One crucial consequence is that the purchas- 21st century. The literature on the influence of climate ing power parity estimates of GDP per person rely on a on economic variables has tended to rely upon average version of the 2005 ICP baseline due to appear as Penn values of climate variables, primarily temperature, World Tables (PWT) Version 7. These estimates cover measured for the capital city of the country. The classic the period 1980­2007 for a large set of countries. They dataset is the data compiled by NCAR--NOAA's have been extended backwards to 1960 by splicing esti- National Center for Atmospheric Research in Boulder, mates from PWT Version 6.2, which uses the 2000 Colorado--for weather stations around the world iden- ICP baseline. Country gaps have been filled by the tified by their World Meteorological Organization standard approach of using a quadratic equation linking reference code. The difficulty with this dataset is that GDP per person in constant (2000) USD at market there is no consistency across stations in the data that is exchange rates to GDP per person at constant (2005) reported. We have examined average data for capital PPP exchange rates. cities derived from weather stations in or near the capi- tal--including, for example, nearby airports. This is 6 t h e c os ts oF ad a p tin g to c limate c h an ge For in Fr as tr u c tu r e used to obtain an average elevation for the capital city, weighted and inverse population-weighted climate vari- but there are too many missing values to rely upon the ables are shown in Table 1 along with correlations with climate variables for our econometric analysis. historic demographic variables used as instruments for institutional development as discussed in Section 4. The Climate Research Unit at the University of East Anglia has compiled a series of historic weather data for The primary climate variables used in the econometric 0.5 degree grid squares for land areas of the globe. analyses are the two weighted means for (a) annual Summary statistics have been computed for each grid average temperature (computed as the average of cell for monthly average, maximum and minimum monthly average temperatures) in °C; (b) total annual temperatures (in degrees C), and precipitation (in mm) precipitation (computed as the sum of monthly average for the period 1901­2002. The distribution of tempera- precipitation); (c) the temperature range (the average tures is generally accepted as being well-approximated maximum temperature in the hottest month, the aver- by the normal distribution, so it was sufficient to age minimum temperature in the coldest month); and compute the mean and standard deviation for each grid (d) the precipitation range (average precipitation in the cell. For precipitation, the distribution is closer to the wettest month, average precipitation in the driest log-normal, so the mean and standard deviation of month). ln(precipitation+1mm) were calculated in addition to the mean of precipitation.5 One point to note is that annual average temperature measured in degrees C is negative or very small in a Country estimates of the climate variables were number of countries, especially for the inverse popula- constructed using grid cell means for monthly mean, tion-weighted means. Because of the use of the loga- maximum, and minimum temperatures and precipita- rithmic transform, it is necessary either to exclude tion. The primary variables are population-weighted countries with extreme temperatures or to apply some averages using the population in each country in each linear shift to temperatures. The transformation grid cell to weight the grid-cell means, thus reflecting adopted was to add 40°C to all temperatures. This the average exposure for the population of each coun- value reflects the range from the minimum value of the try.6 Alternative sets of country means weighted by (a) monthly minimum temperature (-29.1°C) and the the land areas in each cell, and (b) the inverse of popu- maximum value of the monthly maximum temperature lation in each cell were also constructed. The reason for (+46.9°C). Of course, the shift has no effect on the doing this is linked to the demand for transport and temperature range. other types of hard infrastructure. Consider a country such as Australia. The population is concentrated in The choice and use of climate projections to 2050 and the coastal areas of the continent, while the interior-- beyond is considerably more complex. Global climate with very different climatic conditions--is very thinly models (GCMs) are programmed to produce projec- populated. So the population-weighted averages will tions of different variables for different time periods. reflect the climate on the coast whereas the inverse At a micro scale, there are large differences between the population-weighted averages reflect the climate in the results generated by the various models, so that it is interior, while the area-weighted averages fall in necessary to be very careful about relying upon a single between.7 The correlations between the population- model. The standard deviation of projections for any one grid cell is typically large relative to the mean value of the projected change up to 2050 or even 2100. 5 The shift of +1mm is required because precipitation is zero for many Further, the problem is more serious than simple months at some grid squares, which would generate missing values without the shift. models may suggest. Our econometric models suggest 6 There is one complication. Just over 10 percent of grid cells cover more than one country, but the data only provide the land area of each country in each grid cell plus total population in the grid cell. It is, ed and area-weighted means instead of or in addition to the popula- therefore, necessary to assume that population density is uniform over tion-weighted means improves the performance of our equations. In these grid cells so that population is split between countries in the all of the cases that we have examined, the area-weighted climate vari- same proportion as land area. ables are dominated by the inverse population-weighted (ipop) climate 7 We have tested whether using either the inverse-population-weight- variables. 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 1. correlation matrix of climate variableS anD hiStoric Demographic inDicatorS population-weighted climate inverse population-weighted climate Average Temper- Precipi- Average Temper- Precipi- temper- Precipi- ature tation temper- Precipi- ature tation Birth rate ature tation range range ature tation range range 1950­54 Population-weighted climate average temperature 1.000 precipitation 0.229 1.000 temperature range -0.656 -0.650 1.000 precipitation range 0.600 0.774 -0.618 1.000 Inverse population- weighted climate average temperature 0.830 0.151 -0.630 0.433 1.000 precipitation 0.083 0.811 -0.576 0.525 0.070 temperature range -0.563 -0.612 0.943 -0.526 -0.598 -0.646 1.000 precipitation range 0.373 0.789 -0.630 0.775 0.283 0.885 -0.631 1.000 Historic demographic indicators Birth rate 1950-54 0.728 0.042 -0.373 0.510 0.637 -0.106 -0.293 0.215 1.000 infant mortality 1950-54 0.595 -0.027 -0.201 0.430 0.528 -0.118 -0.165 0.183 0.821 Source: authors' estimates using data for 157 countries with non-missing data for gdp, population, urbanization, and generating capacity in 2005. that the ranges between maximum and minimum For the main scenario analysis in this study, we have monthly temperatures and precipitation are often the used results from the NCAR CCSM-3 and CSIRO-3 primary drivers of infrastructure demand. This means models (abbreviated to NCAR and CSIRO). These that the projections used to calculate the Delta-Q costs have relatively similar changes in the global moisture must be based upon climate scenarios that generate index, but they differ significantly in their patterns of monthly maximum and minimum temperatures as well climate change at the regional and country level. The as average temperatures, which restricts the set of models are part of a larger set of 26 GCMs that have GCMs that can be used. But even more important, the been examined in detail by the MIT Joint Program on variance of the difference between two variables is the the Science and Policy of Global Change. As part of sum of their variances minus their covariance. Under their analysis, the MIT group has down-scaled the most plausible outcomes, this will exceed the variance of climate projections to match the 0.5 degree grid cells each element, so that the uncertainty about climate used for the historic climate data, so population- and ranges will be higher than for climate means.8 area-weighted means were constructed for the countries covered by our study for the NCAR and CSIRO scenarios. 8 This is particularly the case for the precipitation range. Generally, cli- These projections are not sufficient for the Delta-P mate change projections suggest that monthly maximum and mini- analysis, because design standards for certain types of mum temperatures will move roughly in line with average temperatures. That is certainly not the case for precipitation since in infrastructure are driven by extreme values rather than many places it is expected that rainfall patterns will become more monthly average values. However, GCMs are not capa- uneven with zero or even negative covariance between changes for the driest and wettest months. ble of generating reliable estimates of daily maximum/ 8 t h e c os ts oF ad a p tin g to c limate c h an ge For in Fr as tr u c tu r e minimum temperatures, precipitation, or wind speed, so are not relevant when we wish to make projections 40 it is necessary to deal with this requirement in an indi- or more years into the future, since it is neither possible rect manner. We have proceeded as follows: nor desirable to attempt to project how governance or institutions will evolve over that period. a. Use the normal or log-normal distributions of monthly averages of maximum/minimum temper- To the extent that (a) institutional factors influence the ature and monthly precipitation to estimate the current level of infrastructure provision, and (b) there is 99th percentile of monthly maximum temperature, a correlation between institutional development and the 1st percentile of monthly minimum tempera- GDP per person or urbanization, then the impact of ture, and the 99th percentile of maximum monthly institutional development will be (partly) captured by precipitation. the coefficients on GDP per person or urbanization in the reduced form discussed below. This is one reason b. Express these percentiles as a ratio of the maxi- why the elasticities of infrastructure demand with mum/minimum of monthly average maximum/ respect to these variables may be higher when estimated minimum temperatures and the maximum using a sample of all countries than for a sample of monthly precipitation and assume that these ratios high-income countries only. But, equally, there are will remain broadly constant in the future. many other factors that may affect the reduced form elasticities. c. Apply the ratios of the 99th/1st percentiles to the associated monthly extremes for 2050 in order to Quite apart from matters of econometric philosophy, compute the change from extreme values for the the nature of the data available for the purpose of historic climate to extreme values for the climate making projections of future demand for infrastructure scenario in absolute units--degrees C or mm of has an important influence upon the specification of the rainfall. models. There are a very limited number of variables for which independent projections extending to 2050 d. In the case of wind speed, we have estimated the have been constructed and can be used. In addition to elasticity of the 99th percentile of wind speed with the climate variables discussed above, these are total respect to the 99th percentile of precipitation by population, the age structure of the population, urban- fitting extreme value distributions to the historic ization, and growth in income (GDP per capita climate data and used the change in maximum measured at purchasing power parity), plus a number of precipitation to project changes in extreme wind geographical features, which act as country-fixed events. effects.9 The basic approach for the econometric analysis is to develop a reduced form specification of the efficient 4. econometric SpecificationS demand for the services provided by each type of infrastructure--for example, paved roads or railways.10 In considering the specification of the econometric analysis, it has to be remembered that the goal is to 9 The demographic projections are based on the medium fertility pro- generate projections of the average demand for infra- jection in the UN Population Division's 2006 revision, which is linked to structure up to 2050, whether or not these are affected the urbanization projections. The central scenario for growth rates for GDP per person at purchasing power is computed by taking the aver- by climate. We are not trying to examine the factors age of five economic integrated assessment models-- Hope (2003), Nordhaus (2002), Tol (2007), IEA (2008) and EIA (2008). The average that drive the actual amounts of infrastructure assets growth rate for world GDP in real terms is very close to the IPCC A1 supplied today or in the past. The key implication is SRES scenario, but the country growth rates are not based upon the downscaled versions of that scenario since those were constructed that it is not appropriate to include, for example, indica- with a base data of 1990 and the relative country weights are very out tors of governance or institutional development in the of date. The sources of the population and income projections are described in a separate note. analysis. These may be relevant factors explaining 10 There is an extensive literature, much of it originating in the World actual outcomes for individual countries today. But they Bank, on developing econometric models to identify links between 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 We assume that the structural equation defining the Since there are no strong priors on the appropriate efficient demand for infrastructure type i in country j in functional forms for fi{ }, ci{ }, and gi{ }, we can use a period t may be written as: standard flexible functional form to represent the demand equation hi{ } in terms of the explanatory vari- ables. We have adopted a restricted version of the Qijt = f i {Pjt , Y jt , Cijt , X jt , Z ijt ,V jt , t} (3) translog specification for all variables other than popu- lation. Using the notation xj=ln(Xj), the general trans- The variables are defined as follows: log function for infrastructure services may be written as: Pjt is the population of country j in period t;11 Yjt is average income per head for country j in d ijt = a i + b pi p jt + b yi y jt + b xim xmjt + b vir v rjt + g yi y 2 + g (7) period t; d ijt = a i + b pi p jt + b yi y jt jt b x + d cost a i infrastructure type i+ b xim xmjt + + + b virimrjt jt xg yi + jtjyb v rjtx s imnb mjtvxnjt++ y 2imr xmjt vg Cijt is the unit ijt = of + b pi p jt + b yi y jt for p jt + mjt y + ir xim x mjt d ijt = a brpiv y + b yiy jt + gjt xim mjt + r xviry rjtxmjtg+ j+irjt vx 2 2+ f y rjtr i im jt yi jt . = i b mjt d ijt a im+ jt pi p jt + b yiiry jtjt+rjt b xim ximn + im y bxmjt rjt + gjyi ir 2rjt+ g xims mjt xmjt xnjt + f imr xmjt v rjt country j in period + t; r y x + j y v + s xr xnjt + + imr xmjt v v + + mjt jt vir v y y jt rjt x 2imn f mjt jt Xjt is a vector of country characteristics for coun- try j in period t; + r im y jt xmjt + j ir y jt v rjt + s x xnjt + f imr xmjt v rjt imn mjt Zijt is a vector of economic or other variables that affect the demand for infrastructure type i for country j in period t; and In practice, it is often difficult to estimate the full trans- log specification using the more complex econometric Vjt is a vector of climate variables for country j in models, so the approach adopted was to start with the period t. log-linear specification and then test whether the coef- ficients on the quadratic and cross-product terms are We can observe or project values for some of these vari- significant. Because this involves repeated testing of ables, notably P, Y, X, and V (dropping subscripts). For overlapping specifications, we have followed the spirit the other variables we assume that: of the Bonferroni adjustment to test statistics by requir- ing that any coefficients retained in the model are Cijt = ci {Y jt , X j , Z ijt ,V jt , t} (4) significantly different from zero at the 1 percent level using conventional statistical tests.12 and We have noted that including climate variables in equa- tions for the demand for infrastructure may be chal- Z ijt = g i {Y jt , X jt ,V jt , t} . (5) lenged by some economists, especially if one goes on to assume that future demand for infrastructure will be Solving for Zijt and Cijt allows us to write the reduced affected by projected changes in these climate variables. form as The reason for the debate is that climate variables are believed to act as a proxy for institutional and other Qijt = hi {Pjt , Y jt , X jt ,V jt , t} (6) factors that determine actual outcomes, partly as a consequence of historical patterns of development (Acemoglu et al. 2001; Albouy 2008; Dell et al. 2008; Horowitz 2008. For example, attempts have been made infrastructure investment and economic growth and to project future to estimate a relationship linking income per person investment requirements for infrastructure in developing countries-- see Fay and Yepes (2003), Estache et al. (2005), and AICD (2009). 11 For some types of infrastructure, total population may be replaced by population in each age group; i.e., the number of children (ages 0 to 14), the number of elderly (ages 65+). The country-fixed effects include 12 In fact almost all of the coefficients are significantly different from country size and the proportions of land area that are desert, arid, zero at the 0.1 percent level. The exceptions to this procedure relate to semiarid, steep, or very steep using standard FAO land classifications. linear terms in exogenous variables when one or more of the quadratic In addition, we have used the proportion of land that has no significant terms is significant at the 0.1 percent level. In such cases the linear soil constraints for agriculture. term is retained, since it may be important for scaling the predictions. 10 t h e c os ts oF ad a p tin g to c limate c h an ge For in Fr as tr u c tu r e and average temperature as a basis for measuring the quadratic terms or cross-products with other impact of climate change at highly aggregated level. explanatory variables. Consistently, one or both of Indeed, any simple correlation of these variables appears the variables have coefficients that are significantly to show that a higher average temperature (usually for different from zero at the 95 percent or 99 percent the capital city of the country) is associated with lower levels. For this reason, the variables are included average income per person. But even this relationship is in all of the models discussed below. So, it must complicated by the role of natural resource endowments. be remembered that--even without further Acemoglu et al. (2001) suggest that, in part, tempera- controls for the possible role of climate as an ture is serving as an instrument for institutional devel- instrument for institutional development--the opment, so they include historical mortality rates in analysis starts from a point that matches the state their analysis on the grounds that this is an alterna- of the art in the current literature. tive--and better--proxy for institutional development. b. The role of climate as an instrument for institu- The strategy adopted for our analysis relies on a number tional development is a geographical argument-- of alternative ways of dealing with this problem. that is, it is about the geography of regional development--as much as it is about climate per a. The Acemoglu et al (AJR) study used colonial se. Thus, the natural approach-- again made (mostly 18th century) mortality as an instrument difficult in the past by data limitations--is to for institutional development and found that this consider the use of spatial econometrics in which had a very significant coefficient in their equations spatially weighted values of variables are used as for recent economic growth. However, estimates instruments for institutional and other factors. of colonial mortality are not available for more The standard model of spatial interaction (or than one-half of the countries in our sample and, autocorrelation) is: in any case, there is considerable controversy about the reliability of the estimates that have been used. Instead, we have used an alternative set of instru- yi = a + j Wj i ij y j + b xi + ei (8) mental variables. The UN's population statistics include a variety of demographic variables for the where the matrix W is a matrix of weights capturing early 1950s for almost all countries. These the spatial influence of location j on location i, is provide good instruments because they are closely called the spatial autocorrelation coefficient, and is the correlated with the historical endowment of both error term whose distribution depends upon the model institutions and infrastructure, but demographic specification. The inverse-distance model has been used changes over the past 50 years mean they are less for this analysis for which the elements of W are associated with current patterns.13 Two instru- proportional to the reciprocal of the distance between ments have been used--the crude birth rate and the population centroids for countries i and j up to a infant mortality. These two were chosen because maximum of 2,500 km.14 The W matrix is normalized they capture the highest proportion of the cross- so that the row sums are equal to 1. The equations are country variation of the demographic variables estimated using panel GMM with spatially weighted examined. Reflecting their special role, these vari- values of population, GDP per person, urbanization, ables were included on their own without and country size as instruments. The details of the analysis are given in a separate paper, but the overall 13 The actual variable used in the AJR study is ln (settler mortality). For 63 countries in their samples (excluding Bahamas), the correlations between ln (settler mortality) and our historic demographic variables 14 The distance band is chosen to ensure that all countries have at least are 0.46 for ln (crude birth rate), 0.67 for ln (infant mortality), and -0.69 three "neighbors" within the band. This is a particular concern for for ln (life expectancy). The correlations with AJR's proximate indicator large/isolated countries or territories such as Australia, Brazil, Canada, of institutions (average protection against expropriation risk 1985­95) and Papua New Guinea. Reducing the distance band to 2,000 km are -0.58 for ln (settler mortality), -0.57 for ln (crude birth rate), -0.69 for would mean that seven countries or territories have only one "neigh- ln (infant mortality), and 0.65 for ln (life expectancy). Hence, our histor- bor" within the band, while reducing it to 1,500 km excludes Australia, ic demographic indicators should provide better instruments for insti- Mongolia, Papua New Guinea, and Timor-Leste as having no "neigh- tutional influences than AJR's use of settler mortality. bors." 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 conclusion is that (a) the spatial interactions are consis- problem is that governance ratings change over tently insignificant, and (b) including them does not time, whereas the climate variables are constant. alter the role of the climate variables in our equations. To get around this, we have computed country averages for the years for which data is available c. Setting aside the spatial argument, the central and constructed the correlation matrices for both econometric contention of the argument that population-weighted and inverse population- climate variables do not reflect the role of climate weighted climate variables. Population-weighted per se is that some or all of these variables are mean temperature and precipitation range would correlated with the error terms in the regression. be the best instruments, as they have simple corre- This is a classic econometric problem that may be lations of -0.41 to -0.49 for the main WGI gover- caused by omitted variables, measurement errors, nance variables--notably government or other factors. The standard solution is to treat effectiveness, regulatory quality, and rule of law. the suspect climate variables as endogenous and Population-weighted precipitation and tempera- look for instruments that are correlated with the ture range have very low correlations with the climate variables but not with the error term--see, governance variables, and the correlations for the for example, Cameron and Trivedi (2006, chapter inverse population-weighted variables are signifi- 4) or Baum (2005, chapter 8). It is not easy to cantly worse than for the population-weighted find suitable instruments for all of the climate variables. With squared correlations of 0.2 or less, variables, especially as a group, since physical char- climate variables would have high standard errors acteristics of countries are included in the infra- if they were acting as instruments for gover- structure demand equations. We have investigated nance--roughly 5 times the true standard errors a range of potential instruments, such as the abso- for the governance variables--which is difficult to lute value of latitude (the best instrument for reconcile with the relatively high t-ratios actually temperature); internal renewable water resources; obtained. Further, including governance variables numbers of bird, mammal, and plant species per sq in the tests reported below may reduce or increase km; percentage covered by water and snow/ice; the F-values for the joint tests on the sets of and spatially weighted physical characteristics for climate variables, but it does not alter the infer- neighboring countries. These instruments ence. Overall, our results provide little support for perform reasonably well for mean temperature and this interpretation. temperature range (both population-weighted and inverse population-weighted) on their own. In It is not possible to prove a negative. Our analysis these cases, the use of instrumental variables does cannot demonstrate conclusively that the coefficients on not alter our conclusions. The variables turn out our climate variables reflect the effects of climate per se to be weak instruments for total precipitation and rather than the indirect influence of other, non-climate, precipitation range on their own or for all climate factors. Nonetheless, we would argue that the cumula- variables together, but no one has seriously tive weight of evidence is strong enough to shift the proposed that either total precipitation or precipi- burden of proof. A key point is that the influence of tation range act as proxies for other influences on climate in our infrastructure equations rarely depends infrastructure demand. Finally, the analysis using upon a single climate variable on its own, whereas argu- instrumental variables consistently fails to reject ments about the role of climate as a proxy or instrument the hypothesis that the climate variables--either for other factors focus almost exclusively on mean individually or as a group--can be treated as exog- temperature. There is even less reason to believe that enous; that is, that the correlation between the inverse population-weighted climate variables act in this climate variables and the error term is zero. way, since by definition these reflect climate patterns in areas where people do not live and have not lived in d. A final possibility is that climate variables act as instruments for governance variables. One 12 t h e c os ts oF ad a p tin g to c limate c h an ge For in Fr as tr u c tu r e large numbers.15 Hence, after extensive and careful significant fraction of observations are censored from investigation, we conclude that the evidence supports above with the upper limit equal to logit (0.999). the view that climate does and will have a significant influence on future demand for infrastructure. In In addition to climate variables, the explanatory vari- reaching this conclusion, we have been particularly strict ables in the base models are: when considering the inclusion of mean temperature (population-weighted and inverse population-weighted) · Log of population in our projection models, so there has been a bias in favor of omitting these variables unless there was unam- · Logs of GDP per person at 2005 PPP, country size, biguous support for retaining them. On that basis, we and urban population as percentage of total popula- believe it is reasonable to estimate the Delta-Q compo- tion plus quadratic terms in these variables nent of adaptation over the full period from 2010 to · Log of a cross-country building cost index with the 2050 on the basis of the demand projections generated U.S.=1.0 by our equations. · Logs of the proportions of land area that are desert, The primary investigation of alternative specifications is arid, semi-arid, steep, very steep, and have no soil carried out using pooled OLS with Driscoll-Kraay stan- constraints for agriculture--obtained from FAO's dard errors, which allow for a general pattern of spatial Terrastat database dependence between countries (Driscoll and Kraay · Logs of the birth rate and infant mortality for 1998; Hoechle 2007).16 In the case of the proportions 1950-54 of the population covered by electricity, water, and sewer networks, the dependent variable is the logit of the rele- · Dummy variables for World Bank regions. vant shares in order to translate values between 0 and 1 to the entire real line. It is necessary to censor values The last two groups of variables are retained in all that are reported as either 0 or 1 in order to avoid models. Tests for the inclusion of non-climate and degeneracy. Thus, the minimum and maximum values climate variables are performed separately. At the first correspond to shares of 0.001 and 0.999, as the shares stage, the non-climate variables are tested for signifi- are reported to the nearest 0.1 of a percentage point. A cance in a model containing the seven climate vari- panel tobit model has been used to estimate the ables--pop and ipop variants other than temperature demand equations for coverage rates for which a range. After dropping non-climate variables that do not have significant coefficients, tests on the hypotheses that the coefficients for (a) the population-weighted 15 The absolute values of the correlation coefficients between the logs climate variables, (b) the inverse population-weighted of similarly weighted climate variables are less than 0.66 across our sample of countries, with the sole exception of total precipitation and climate variables, and (c) all climate variables are all precipitation range (see Table 1). Both temperature and precipitation equal to zero are carried out. If one or more of these are negatively correlated with temperature range. The correlation coefficients between population-weighted and inverse population- hypotheses are rejected, the set of climate variables weighted variables range from 0.78 to 0.83, with the exception of tem- included in the model is reduced by first dropping perature range, for which the value is 0.94. In view of this last correlation, we have excluded the inverse-population weighted tem- either the pop or the ipop variants and then those vari- perature range from the analysis. ables within each category that do not have significant 16 Driscoll-Kraay standard errors are robust to panel heteroscedasticity coefficients. Finally, interactions with GDP and urban- and temporal autocorrelation as well as spatial interdependence. The estimation is carried out using Hoechle's xtscc procedure in Stata, ization are tested for the climate variables that have which generalizes the Driscoll-Kraay estimator to allow for unbalanced panels. There is an important feature of the Driscoll-Kraay/Hoechle been retained. procedure that needs to be kept in mind. The method relies upon the derivation of a robust covariance matrix for a sequence of cross-sec- tional averages. The panels used for our analysis of some categories of Finally, we have used total, urban, or rural population infrastructure are very unbalanced and do not span continuous periods weights (as appropriate) in estimating equations for of time. Nonetheless, cross-sectional averages can be calculated for more than 25 years. The sample of countries in each cross-sectional which the dependent variable is the log or logit of an average differs, but this is consistent with the way in which the covari- infrastructure indicator per person or per household; ance estimator is specified. Thus, even though the Driscoll-Kraay anal- ysis relies upon asymptotics as T, the nature of our data is for example, municipal industrial water use per person, consistent with its basic requirements. 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 average household size, or the percentages of house- the inverse population-weighted climate variables. The holds connected to electricity, water, or sewer networks. rejection of the hypothesis of zero coefficients is partic- In all cases, the weights are normalized to sum to the ularly strong for the inverse-population weighted number of observations used for the analysis. climate variables; this is reinforced by the higher values of the t-ratios for these coefficients. The only climate variable with a coefficient that is not significantly different from zero is population-weighted mean 5. the effectS of climate on temperature. On the other hand, both temperature range and inverse population-weighted mean tempera- DemanD for infraStructure ture have coefficients that are highly significant. The signs of the coefficients differ, but these variables are Electricity generating capacity. Model (1) in Table 2 negatively correlated (see Table 1) so that warmer coun- shows that the tests for the joint significance of the tries tend to have less generating capacity, holding other climate variables reject the hypothesis of zero coeffi- factors constant. cients decisively for both the population-weighted and table 2. projection equationS for electricity generating capacity, fixeD tele- phone lineS, anD electricity network coverage Logit(Urban Logit(Rural Ln(Generating capacity) Ln(Fixed telephone lines) electricity coverage) electricity coverage) Variables (1) (2) (3) (4) (5) (6) (7) (8) ln(population) 0.954*** 0.959*** 1.088*** 1.081*** 0.729** 0.641** 1.882*** 1.750*** (0.028) (0.021) (0.033) (0.030) (0.222) (0.217) (0.150) (0.152) ln(gdp per per- 0.975*** 0.926*** 0.618*** 0.608*** 3.706*** 3.703*** 5.909*** 6.496*** son) (0.141) (0.129) (0.024) (0.016) (0.562) (0.556) (0.587) (0.642) ln(country size) 1.111*** 1.034*** -0.159*** -0.150*** 4.454*** 5.208*** (0.137) (0.117) (0.026) (0.026) (0.630) (0.697) ln(% urban) 2.936*** 4.248*** 2.084*** 1.340 -10.59*** -10.43*** -7.703*** 8.478 (0.563) (0.711) (0.474) (1.593) (3.141) (3.122) (1.554) (4.349) ln(% urban) 0.324*** 0.309*** squared (0.050) (0.044) ln(gdp per per- -0.115*** -0.108*** -0.621*** -0.714*** son) * ln(country (0.014) (0.012) (0.070) (0.078) size) ln(gdp per per- -0.239*** -0.226*** -0.203** -0.191** 1.683*** 1.660*** 0.627** 0.776*** son) * ln(% urban) (0.056) (0.060) (0.066) (0.062) (0.428) (0.427) (0.211) (0.218) ln(country size) * 0.131*** 0.106*** 1.001*** 1.091*** ln(% urban) (0.027) (0.026) (0.136) (0.140) (continued) 14 t h e c os ts oF ad a p tin g to c limate c h an ge For in Fr as tr u c tu r e table 2. projection equationS for electricity generating capacity, fixeD tele- phone lineS, anD electricity network coverage (continued) Logit(Urban Logit(Rural Ln(Generating capacity) Ln(Fixed telephone lines) electricity coverage) electricity coverage) Variables (1) (2) (3) (4) (5) (6) (7) (8) ln(Building cost) -1.878*** -1.712*** 21.44** 18.21** 17.94*** 14.30*** (0.289) (0.284) (6.636) (5.967) (3.502) (3.348) ln(% desert) 0.0276*** 0.0278*** 0.0250*** 0.0351*** (0.004) (0.005) (0.005) (0.007) ln(% semi-arid) -0.0378*** -0.0297** 0.0296*** 0.0307*** (0.011) (0.011) (0.004) (0.004) ln(% steep land) -0.107*** -0.111*** -1.509*** -0.321 (0.008) (0.012) (0.413) (0.350) ln(% very steep 0.0947*** 0.0792*** 0.0647*** 0.0647*** land) (0.008) (0.007) (0.004) (0.004) ln(% no soil con- -0.0522*** -0.0414*** 0.0365*** 0.0346*** -0.805*** -1.146*** straint) (0.010) (0.008) (0.006) (0.007) (0.167) (0.142) ln(temperature - -0.837 -2.158*** -2.620*** -5.626 -7.115* pop) (0.480) (0.202) (0.376) (6.088) (3.084) ln(precipitation - 0.395*** 0.152*** -0.001 -3.175** -3.884*** -1.541* pop) (0.069) (0.040) (0.085) (1.224) (1.098) (0.767) ln(temp range - 0.313** 0.386*** -0.250** 0.144* -6.061** -3.735** -4.629*** pop) (0.103) (0.074) (0.090) (0.067) (1.866) (1.228) (1.119) ln(precip range - -0.431*** -0.272*** -0.169** -0.0449** 3.574** 3.799*** 2.094** pop) 1.388*** (0.055) (0.052) (0.065) (0.017) (1.342) (1.055) (0.756) (0.310) ln(temperature - -1.057*** -1.447*** -0.388*** 0.305*** -0.203 -10.53*** ipop) -13.65*** (0.137) (0.125) (0.111) (0.068) (2.396) (1.418) (1.946) ln(precipitation - -0.272*** -0.269*** -0.149 -1.771 -1.270** ipop) (0.045) (0.036) (0.098) (0.997) (0.490) ln(precip range - 0.479*** 0.468*** 0.240** 0.0542* 1.018 1.012* ipop) (0.055) (0.048) (0.077) (0.025) (1.001) (0.509) ln(% urban) * -1.006** ln(temperature - (0.363) pop) 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 table 2. projection equationS for electricity generating capacity, fixeD tele- phone lineS, anD electricity network coverage (continued) Logit(Urban Logit(Rural Ln(Generating capacity) Ln(Fixed telephone lines) electricity coverage) electricity coverage) Variables (1) (2) (3) (4) (5) (6) (7) (8) ln(% urban) * -0.388*** ln(precipitation - (0.050) pop) ln(% urban) * 0.328*** ln(temp range - (0.070) pop) ln(% urban) * 0.270*** 0.139*** ln(precip range - (0.066) (0.031) pop) ln(% urban) * 0.862*** -4.295*** ln(temperature - (0.102) (1.151) ipop) ln(% urban) * -0.0749*** ln(precip (0.009) range - ipop) model pols pols pols pols tobit tobit tobit tobit observations 6027 6027 5130 5130 906 906 853 853 number of coun- 165 165 186 186 130 130 127 127 tries r-squared 0.923 0.924 0.938 0.939 log-likelihood -250.6 -253.2 -436.6 -438.4 dF 26 27 25 28 19 15 24 20 no of censored 716 716 661 661 obs p-value for all cli- mate variables = 0.000 0.000 0.008 0.000 0 p-value for pop climate variables 0.000 0.000 0.005 0.000 =0 p-value for ipop climate variables 0.000 0.000 0.167 0.000 =0 Note: standard errors are shown in brackets underneath the relevant coefficients with *** p < 0.001, ** p < 0.01, * p < 0.05. in addi- tion to the variables shown, all of the equations include the following explanatory variables: ln(birthrate 1950), ln(infant mortality 1950) and dummy variables for World Bank regions. Source: authors' estimates. 16 t h e c os ts oF ad a p tin g to c limate c h an ge For in Fr as tr u c tu r e Overall, however, temperature and temperature range inclusion in the equation for urban coverage. For rural are less important influences on the amount of generat- coverage, population-weighted precipitation range and ing capacity than precipitation and precipitation range inverse population-weighted temperature--on its own together with their interactions with urbanization. and interacted with urbanization--are the key climate There are various mechanisms by which precipitation variables. The chi-square statistic for the test of zero may affect installed capacity. One factor is the role of influence of the temperature variables is 59.7, so that hydro power in total electricity supply, since utilization these cannot be excluded. factors tend to be lower for hydro plants. Another is the role of pumped irrigation systems and similar influ- Water use. The dependent variables for water use are ences on patterns of electricity demand in countries the logs of water abstractions per person for municipal with high interseasonal variations in rainfall. Even and industrial use, which are derived from FAO data. though the absolute values of the coefficients for precip- This includes water that is lost in treatment and in itation and precipitation range are smaller than the water supply networks. Models (1) and (2) in Table 3 equivalent coefficients for temperature, these variables summarize the results of the econometric analysis for have an important effect in the calculation of the municipal water use per person. In this case, the tests Delta-Q changes because the distributions of changes for the joint significance of the climate variables reject in total precipitation and precipitation range are much the hypothesis of zero coefficients decisively for the more dispersed and much larger relative to their historic population-weighted variables, but not for the inverse values than are the equivalent distributions for population-weighted variables. The best specification temperature. includes population-weighted precipitation and precipi- tation range. Another point to note is the quadratic in Fixed telephone lines. The projection equations for fixed GDP per person. The results seem to be intuitively telephone lines are reported as Models (3) & (4) in reasonable, reflecting rainfall patterns where people live Table 2. Again, the hypotheses of zero coefficients for and the effect of changes in GDP on water use. The climate variables are decisively rejected, particularly for quadratic terms in GDP per person imply that water the population-weighted variables, with mean tempera- consumption per person reaches a peak at an income of ture, temperature range, and precipitation range all about $12,000 per capita in 2005 PPP, and falls gradu- having significant coefficients. There are strong interac- ally as countries get richer beyond this point. tions with urbanization, so that the impact of climate change on demand for telephones varies markedly both Models (3) and (4) in Table 3 summarize the results for within and across country classes. industrial water use per person. In this case, the tests reject the hypotheses that the population-weighted and/ Electricity network coverage. Models (5) through (8) in or inverse population-weighted climate variables have Table 2 show the estimated equations for the logits of zero coefficients. The detailed investigation identifies electricity coverage for urban and rural households population-weighted temperature range and precipita- weighted by the relevant populations in 2005.17 Panel tion range plus inverse population-weighted precipita- tobit models are used with an upper censoring value tion and precipitation range as having significant corresponding to a coverage of 99.9 percent. Since the coefficients. There are significant interactions between majority of observations are censored, the number of the inverse-population weighted climate variables and exogenous variables is reduced in each equation by GDP per person with urbanization. Use of water in much more than for electricity generating capacity. industry is a derived demand, so the influence of Nonetheless, population-weighted precipitation, precipi- climate variables operates through the scale and location tation range, and temperature range clearly warrant of food processing and similar resource-based industries. Hence, it is climate conditions in rural and thinly popu- lated areas that have a significant influence. 17 For the purpose of projecting the total numbers of connections, it is necessary to allow for non-household connections. We have assumed that the total numbers of electricity connections are 108 percent of the Water and sewer connections. Table 4 summarizes the numbers of households connected to the network. This multiplier results for coverage rates of piped water supply and reflects the typical ratio for upper-middle and high-income countries. 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 3. projection equationS for municipal anD inDuStrial water DemanD Ln(Municipal water use per person) Ln(Industrial water use per person) Variables (1) (2) (3) (4) ln(gdp per person) 2.159** 2.000** 3.455** 2.953* (0.679) (0.632) (1.073) (1.197) ln(country size) ln(% urban) 0.530*** 0.559*** (0.071) (0.067) ln(gdp per person) squared -0.115** -0.105** -0.191** -0.219*** (0.039) (0.036) (0.064) (0.064) ln(Building cost) -2.477*** -2.342* (0.662) (0.904) ln(% steep land) 0.970*** 0.943*** (0.124) (0.100) ln(% very steep land) 0.152*** 0.156*** -0.225*** -0.183** (0.023) (0.032) (0.054) (0.059) ln(% no soil constraint) -0.265*** -0.156*** (0.040) (0.027) ln(temperature - pop) 0.923* -0.027 (0.391) (1.219) ln(precipitation - pop) -0.150 -0.306*** 0.456 (0.173) (0.087) (0.354) ln(temp range - pop) 0.459** 2.091*** 2.003*** (0.163) (0.294) (0.221) ln(precip range - pop) 0.205 0.367*** -0.819* -0.594*** (0.175) (0.103) (0.323) (0.127) ln(temperature - ipop) 0.079 -0.842 (0.239) (0.592) ln(precipitation - ipop) 0.102 -0.512* -5.318*** (0.081) (0.240) (0.836) ln(precip range - ipop) -0.054 0.902** 5.682*** (0.103) (0.287) (0.931) ln(gdp per person) * 0.577*** ln(precipitation - ipop) (0.099) ln(gdp per person) * -0.577*** ln(precip range - ipop) (0.107) model pols pols pols pols observations 368 368 337 337 number of countries 161 161 158 158 (continued) 18 t h e c os ts oF ad a p tin g to c limate c h an ge For in Fr as tr u c tu r e table 3. projection equationS for municipal anD inDuStrial water DemanD (continued) Ln(Municipal water use per person) Ln(Industrial water use per person) Variables (1) (2) (3) (4) r-squared 0.980 0.979 0.954 0.955 log-likelihood dF 19 14 19 18 no of censored obs p-value for all climate variables = 0 0.000 0.000 p-value for pop climate 0.000 0.000 variables = 0 p-value for ipop climate 0.087 0.000 variables = 0 Note: standard errors are shown in brackets underneath the relevant coefficients with *** p < 0.001, ** p < 0.01, * p < 0.05. in addi- tion to the variables shown, all of the equations include the following explanatory variables: ln(birthrate 1950), ln(infant mortality 1950) and dummy variables for World Bank regions. Source: authors' estimates. sewer networks in urban and rural areas. Models (1) to For the purpose of costing wastewater treatment, we (6) are based upon panel tobit estimation, allowing for have assumed that the BOD/COD concentration and the upper censoring of countries with reported coverage other characteristics of sewage handled by wastewater of 99.9 percent or higher. In general, population- treatment plants correspond to typical values for munic- weighted climate variables have a significant influence ipal wastewater. This implies that industries will be on coverage rates in urban areas, while inverse popula- expected to process wastewater with high concentra- tion-weighted climate variables are more important in tions of industrial pollutants. Further, it is assumed that rural areas. The only exception is rural water supply, for wastewater treatment plants are scaled to process 80 which both sets of climate variables are significant. percent of the volume of water treated by water treat- Interactions with GDP per person and urbanization are ment plants, allowing for network losses and wastewater not significant. Since coverage rates for piped water that is not discharged to sewers. supply are close to or equal to 99.9 percent in high income countries, changes in climate variables will not Roads. Table 5 shows equations for the total length of have any effect on costs of adaptation in many coun- roads (both paved and unpaved) and for the logit of the tries. However, changes in average temperature--and share of paved roads in total road length, weighted by precipitation for rural households--may affect the total road length in the latter case. The key climate numbers of households connected to collective sewer variables affecting the length of roads are temperature systems.18 and precipitation range--both population-weighted and inverse population-weighted--plus population-weighted temperature range. There are strong interactions with 18 It should be emphasized that this is not a matter of whether house- holds have access to some form of adequate sanitation. The depen- GDP per person for temperature and precipitation dent variable is the proportion of households that are connected to community sewers, rather than relying upon septic tanks or equivalent individual arrangements. Community sewers are more expensive to construct and the wastewater that is collected must be treated, so tions by assuming that the total numbers of water supply and sewer costs of adaptation arise from shifts to or away from reliance on com- connection are 10 percent higher than the numbers of household con- munity sewers. Again, we have allowed for non-household connec- nections, based on typical ratios for middle-income countries. 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 table 4. projection equationS for water anD Sewer networkS Logit(Urban water Logit(Rural water Logit(Urban sewer Logit(Rural water coverage) coverage) coverage) coverage) Variables (1) (2) (3) (4) (5) (6) (7) (8) ln(population) 0.353** 0.300* 0.513*** 0.488*** 0.357** 0.278* 0.765*** 0.783*** (0.127) (0.118) (0.113) (0.113) (0.119) (0.117) (0.149) (0.230) ln(gdp per person) 0.889*** 0.901*** 0.430 0.647 2.576*** 2.629*** 1.405*** 1.407*** (0.150) (0.148) (0.826) (0.824) (0.348) (0.350) (0.346) (0.335) ln(country size) -0.580*** -0.539*** 1.327*** 1.439*** 2.035*** 2.161*** -0.636*** -0.635*** (0.112) (0.095) (0.318) (0.314) (0.407) (0.405) (0.102) (0.175) ln(% urban) -3.744*** -3.797*** 1.388*** 1.371*** 1.247*** 1.226*** (0.995) (0.982) (0.230) (0.229) (0.339) (0.268) ln(gdp per person) 0.157** 0.149** squared (0.053) (0.053) ln(gdp per person) * -0.282*** -0.293*** -0.276*** -0.285*** ln(country size) (0.035) (0.034) (0.041) (0.042) ln(gdp per person) * 0.451*** 0.462*** ln(% urban) (0.131) (0.130) ln(Building cost) -7.790** -9.089*** 13.91** 14.16* (2.878) (2.607) (4.162) (5.822) ln(% desert) -0.188* -0.201*** (0.080) (0.051) ln(% arid land) -0.421*** -0.295*** -0.254*** -0.266*** (0.082) (0.073) (0.053) (0.072) ln(% semi-arid land) 0.141* 0.162* 0.375*** 0.449*** (0.064) (0.063) (0.075) (0.074) ln(% no soil constraint) -0.282* -0.422*** (0.114) (0.091) ln(temperature - pop) -8.469*** -8.000*** -0.351 -5.950** -7.603*** 0.281 (2.303) (1.352) (2.406) (2.128) (1.452) (3.102) ln(precipitation - pop) -0.262 -1.690** -1.319*** 0.128 0.149 (0.530) (0.570) (0.214) (0.529) (0.625) ln(temp range - pop) -0.693 -1.793* -1.498** -0.119 -0.023 (0.742) (0.767) (0.484) (0.747) (0.325) ln(precip range - pop) -1.184* -1.500*** 0.870 0.288 -0.413 (0.517) (0.238) (0.593) (0.531) (0.823) ln(temperature - ipop) -0.761 -6.472*** -6.385*** -1.098 -3.842*** -3.745*** (1.033) (0.947) (0.863) (0.889) (0.864) (0.485) ln(precipitation - ipop) -0.017 0.131 -0.283 -0.981*** -0.855*** (0.375) (0.381) (0.356) (0.190) (0.085) (continued) 20 t h e c os ts oF ad a p tin g to c limate c h an ge For in Fr as tr u c tu r e table 4. projection equationS for water anD Sewer networkS (continued) Logit(Urban water Logit(Rural water Logit(Urban sewer Logit(Rural water coverage) coverage) coverage) coverage) Variables (1) (2) (3) (4) (5) (6) (7) (8) ln(precip range - ipop) -0.084 -0.486 -0.429 0.255 (0.416) (0.425) (0.430) (0.269) model tobit tobit tobit tobit tobit tobit pols pols observations 582 582 547 547 318 318 272 272 number of countries 157 157 155 155 140 140 124 124 r-squared 0.901 0.901 log-likelihood -452.1 -452.9 -461.5 -464.7 -327.0 -335.1 dF 21 16 21 17 21 15 20 15 no of censored obs 94 94 36 36 10 10 p-value for all climate 0.000 0.000 0.000 0.000 variables = 0 p-value for pop climate 0.000 0.000 0.024 0.021 variables = 0 p-value for ipop climate 0.854 0.001 0.002 0.000 variables = 0 Note: standard errors are shown in brackets underneath the relevant coefficients with *** p < 0.001, ** p < 0.01, * p < 0.05. in addi- tion to the variables shown, all of the equations include the following explanatory variables: ln(birthrate 1950), ln(infant mortality 1950) and dummy variables for World Bank regions. Source: authors' estimates. table 5. projection equationS for roaDS ln(total road length) logit(share of paved roads) variables (1) (2) (3) (4) ln(population) 0.584*** 0.590*** 0.599*** 0.668*** (0.004) (0.005) (0.057) (0.064) ln(gdp per person) -0.042 2.070*** 1.980*** 8.010*** (0.034) (0.098) (0.094) (0.781) ln(country size) -0.0931* -0.007 3.645*** 0.081 (0.045) (0.029) (0.473) (0.300) ln(% urban) 0.395*** 0.786*** -2.207*** -0.845 (0.069) (0.074) (0.389) (0.520) ln(country size) squared 0.0166*** 0.0175*** -0.108*** -0.0668*** (0.002) (0.001) (0.019) (0.013) ln(% urban) squared -0.155*** -0.0581*** (0.015) (0.010) (continued) 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 table 5. projection equationS for roaDS (continued) ln(total road length) logit(share of paved roads) variables (1) (2) (3) (4) ln(gdp per person) * 0.0331*** 0.0217*** -0.316*** 0.010 ln(country size) (0.003) (0.002) (0.023) (0.023) ln(gdp per person) * -0.105*** -0.199*** ln(% urban) (0.006) (0.012) ln(country size) * 0.434*** 0.036 ln(% urban) (0.053) (0.037) ln(Building cost) -11.84*** -9.597*** (0.778) (0.527) ln(% desert) 0.0347*** 0.0546*** -0.157*** -0.232*** (0.009) (0.006) (0.025) (0.033) ln(% arid) 0.396*** 0.371*** (0.041) (0.065) ln(% semi-arid) -0.0492*** -0.0503*** (0.005) (0.004) ln(% steep land) 0.100*** 0.0660*** (0.013) (0.011) ln(% very steep land) -0.0245*** -0.0111*** 0.196*** 0.174*** (0.004) (0.002) (0.040) (0.045) ln(% no soil constraint) 0.0279*** 0.0402*** (0.006) (0.005) ln(temperature - pop) -1.267*** 1.589*** 2.041 (0.190) (0.424) (1.121) ln(precipitation - pop) 0.017 0.262 (0.030) (0.242) ln(temp range - pop) -0.108** -0.208*** -2.074*** 16.46*** (0.038) (0.042) (0.106) (1.595) ln(precip range - pop) -0.170** -0.154*** -2.099*** 0.663* (0.055) (0.028) (0.319) (0.315) ln(temperature - ipop) 0.593*** 0.684*** -2.213*** -1.829*** (0.141) (0.145) (0.530) (0.256) ln(precipitation - ipop) 0.039 -0.646*** (0.069) (0.176) ln(precip range - ipop) 0.156* 1.334*** 0.976*** (0.060) (0.070) (0.071) ln(% urban) * 0.110*** ln(precip range - pop) (0.011) ln(gdp per person) * -0.373*** ln(temperature - pop) (0.025) (continued) 22 t h e c os ts oF ad a p tin g to c limate c h an ge For in Fr as tr u c tu r e table 5. projection equationS for roaDS (continued) ln(total road length) logit(share of paved roads) variables (1) (2) (3) (4) ln(gdp per person) * -2.150*** ln(temp range - pop) (0.192) ln(gdp per person) * -0.190*** ln(precip range - pop) (0.038) ln(gdp per person) * -0.127*** ln(precip range - ipop) (0.008) model pols pols pols pols observations 2040 2040 1790 1790 number of countries 182 182 179 179 r-squared 0.922 0.926 0.816 0.822 log-likelihood dF 27 28 25 23 no of censored obs p-value for all climate variables = 0 0.000 0.000 p-value for pop climate variables = 0 0.000 0.000 p-value for ipop climate variables = 0 0.000 0.000 Note: standard errors are shown in brackets underneath the relevant coefficients with *** p < 0.001, ** p < 0.01, * p < 0.05. in addi- tion to the variables shown, all of the equations include the following explanatory variables: ln(birthrate 1950), ln(infant mortality 1950) and dummy variables for World Bank regions. Source: authors' estimates. range and with urbanization for precipitation range. GDP per person. In particular, higher temperatures in These climate variables are directly linked to the cost of rural areas--that is, inverse population-weighted building and maintaining roads--temperature is partic- temperature--lead to a lower share of paved roads, ularly important for paved roads subject to heavy use, which is exactly what one would expect in view of the while temperature and precipitation ranges affect the higher costs of construction and maintenance for rural capital and maintenance costs of both paved and paved roads associated with higher temperatures. unpaved roads. The fact that both population-weighted and inverse-population weighted climate variables are Other transport. Table 6 shows the projection equations significant reflects the impact of climate on all types of for rail track length, aircraft movements, and container roads--rural, urban, and national. It is likely that traffic handled by ports. The last two are indicators climate may also play a role through the structure of the used in estimating investments in airports and sea/river economy--for example, the nature and role of agricul- ports. With one exception, the tests reject the hypothe- tural production--and through geographical patterns of sis of zero coefficients for all climate variables decisively. economic development. The exception is for inverse population-weighted climate variables in the rail equation. The primary The share of paved roads in total road length is influ- climate influences are: enced by the same variables and their interactions with 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 23 table 6. projection equationS for other tranSport Ln(Rail track length) Ln(Aircraft movements) Ln(Container traffic) Variables (1) (2) (3) (4) (5) (6) ln(population) 0.484*** 0.472*** 0.540*** 0.540*** 0.649*** 0.649*** (0.043) (0.041) (0.035) (0.032) (0.034) (0.023) ln(gdp per person) 0.259*** 0.971 0.710*** 0.692*** 0.005 -2.830*** (0.053) (0.741) (0.065) (0.066) (0.074) (0.748) ln(country size) 0.425*** 0.405*** -0.184*** -0.167*** -2.751*** -3.495*** (0.076) (0.049) (0.030) (0.030) (0.195) (0.363) ln(% urban) -0.315 -0.215 -0.376** 2.202*** 3.989*** -0.071 (0.172) (0.146) (0.143) (0.571) (0.389) (1.116) ln(country size) squared 0.0337*** 0.0325*** 0.0299*** 0.0278*** (0.003) (0.003) (0.003) (0.003) ln(% urban) squared -0.112* 1.090*** 1.563*** (0.047) (0.168) (0.171) ln(gdp per person) * 0.222*** 0.302*** ln(country size) (0.015) (0.034) ln(country size) * -0.507*** -0.513*** ln(% urban) (0.021) (0.026) ln(Building cost) -3.062*** -2.936*** (0.349) (0.339) ln(% desert) 0.0862*** 0.0728*** 0.266*** 0.223*** (0.021) (0.020) (0.009) (0.015) ln(% arid) -0.333*** -0.363*** (0.012) (0.011) ln(% semi-arid) 0.0404*** 0.0710*** (0.006) (0.004) ln(% steep land) -0.132*** -0.144*** (0.029) (0.032) ln(% very steep land) 0.0743*** 0.0819*** (0.010) (0.011) ln(temperature - pop) -2.235*** 3.225* -0.466 0.712* (0.660) (1.288) (0.334) (0.307) ln(precipitation - pop) 0.362*** -1.378*** -0.396*** -0.351*** 0.591*** 0.909*** (0.035) (0.153) (0.102) (0.080) (0.140) (0.124) ln(temp range - pop) 0.939*** -0.797 -1.038*** -1.400*** 0.232 (0.183) (0.512) (0.131) (0.111) (0.170) ln(precip range - pop) -0.175 0.373*** 0.313*** 0.035 (0.123) (0.053) (0.044) (0.079) ln(temperature - ipop) 0.814 -1.005*** -1.084*** -0.906*** -6.590*** (0.745) (0.114) (0.091) (0.093) (1.305) (continued) 24 t h e c os ts oF ad a p tin g to c limate c h an ge For in Fr as tr u c tu r e table 6. projection equationS for other tranSport (continued) Ln(Rail track length) Ln(Aircraft movements) Ln(Container traffic) Variables (1) (2) (3) (4) (5) (6) ln(precipitation - ipop) 0.149 0.326*** 0.026 0.429*** 0.261*** (0.133) (0.046) (0.072) (0.075) (0.051) ln(precip range - ipop) -0.217 -0.183*** 0.060 -0.457*** -0.326*** (0.174) (0.041) (0.059) (0.034) (0.058) ln(% urban) * 0.707*** ln(precipitation - pop) (0.163) ln(% urban) * -0.508*** ln(temp range - pop) (0.074) ln(% urban) * -0.354*** ln(precipitation - ipop) (0.089) ln(% urban) * 0.324*** ln(precip range - ipop) (0.058) ln(gdp per person) * -0.580*** ln(temperature - pop) (0.156) ln(gdp per person) * 0.167*** ln(precipitation - pop) (0.014) ln(gdp per person) * 0.159** ln(temp range - pop) (0.051) ln(gdp per person) * 0.612*** ln(temperature - ipop) (0.139) model pols pols pols pols pols pols observations 1969 1969 5040 5040 407 407 number of countries 133 133 175 175 69 69 r-squared 0.741 0.740 0.831 0.833 0.805 0.822 log-likelihood dF 19 18 23 24 25 24 no of censored obs p-value for all climate variables = 0 0.000 0.000 0.000 p-value for pop climate variables = 0 0.000 0.000 0.000 p-value for ipop climate variables = 0 0.040 0.000 0.000 Note: standard errors are shown in brackets underneath the relevant coefficients with *** p < 0.001, ** p < 0.01, * p < 0.05. in addi- tion to the variables shown, all of the equations include the following explanatory variables: ln(birthrate 1950), ln(infant mortality 1950) and dummy variables for World Bank regions. Source: authors' estimates. 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 25 a. Rail length--temperature, precipitation, and movement will be affected by factors such as the temperature range, both on their own and inter- amount and distribution of tourism (both internal and acted with GDP per person external), the dispersion and nature of natural-resource based industries, and the availability of alternative b. Aircraft movements--all climate variables other methods of transport. than population-weighted temperature plus inter- actions with urbanization Health care. Our analysis of adaptation costs relies upon two health care inputs--the numbers of hospital c. Container traffic--both precipitation variables, beds and physicians --as indicators used in assessing inverse population-weighted temperature, and the baseline cost of health infrastructure (hospitals and precipitation plus interactions with urbanization clinics) and the impact of climate change. The projec- and GDP per person. tion equations are shown in Models (1) through (4) of Table 7. In addition, Models (5) and (6) report equa- In these cases, there is no easy explanation for the tions for one important indicator of health outcomes-- results since it is clear that there are multiple influences the log of the infant mortality rate. on the indicators. For example, the number of aircraft table 7. projection equationS for health Ln(No of hospital beds) Ln(No of doctors) Ln(Infant mortality rate) Variables (1) (2) (3) (4) (5) (6) ln(population 0-14) 0.283** 0.323*** -0.520*** -0.558** 1.423*** 1.444*** (0.099) (0.060) (0.135) (0.187) (0.116) (0.129) ln(population 15-64) 0.881*** 0.736*** 1.300*** 1.196*** -0.982*** -0.964*** (0.131) (0.064) (0.120) (0.182) (0.132) (0.143) ln(population 65+) -0.224*** -0.128*** 0.275*** 0.391*** -0.475*** -0.508*** (0.032) (0.023) (0.045) (0.034) (0.035) (0.033) ln(gdp per person) 1.721*** -2.680*** 0.246** -4.951*** 0.928* 0.947* (0.327) (0.605) (0.080) (0.589) (0.390) (0.384) ln(country size) -0.155*** -0.0984*** 0.364*** 0.147** 0.105*** 0.109*** (0.021) (0.021) (0.106) (0.046) (0.031) (0.031) ln(% urban) -0.094 1.077*** 1.301*** -0.165*** 1.738** (0.072) (0.211) (0.162) (0.031) (0.584) ln(gdp per person) squared -0.100*** -0.0853*** -0.0661** -0.0661** (0.019) (0.010) (0.023) (0.023) ln(country size) squared 0.0173*** 0.0124*** 0.0111*** 0.0116*** (0.002) (0.001) (0.001) (0.001) ln(gdp per person) * -0.0454*** -0.0185*** -0.0172*** -0.0186*** ln(country size) (0.011) (0.004) (0.004) (0.004) ln(gdp per person) * -0.118*** -0.130*** ln(% urban) (0.027) (0.024) ln(country size) * 0.0966*** 0.0642*** ln(% urban) (0.015) (0.010) (continued) 26 t h e c os ts oF ad a p tin g to c limate c h an ge For in Fr as tr u c tu r e table 7. projection equationS for health (continued) Ln(No of hospital beds) Ln(No of doctors) Ln(Infant mortality rate) Variables (1) (2) (3) (4) (5) (6) ln(% arid) 0.0627*** 0.0462*** (0.011) (0.007) ln(% very steep land) 0.0306*** 0.0244*** (0.005) (0.006) ln(temperature - pop) -1.275*** -10.13*** -1.512*** -7.442*** 0.954*** 0.345 (0.084) (1.416) (0.173) (1.432) (0.142) (0.216) ln(precipitation - pop) -0.275*** -0.357*** -0.244*** 0.054 (0.079) (0.040) (0.031) (0.047) ln(temp range - pop) 0.011 -0.037 0.263*** 0.203*** (0.102) (0.044) (0.052) (0.053) ln(precip range - pop) 0.168* 0.0722* -0.102* (0.080) (0.035) (0.049) ln(temperature - ipop) 0.102 -0.276* -5.061*** -0.262*** -0.208*** (0.110) (0.113) (1.114) (0.049) (0.047) ln(precipitation - ipop) 0.164* 0.0756*** -0.0824* 0.0911** 0.0622** (0.079) (0.021) (0.039) (0.034) (0.020) ln(precip range - ipop) -0.039 0.104** -0.028 (0.058) (0.037) (0.044) ln(% urban) * -0.463** ln(temperature - pop) (0.139) ln(gdp per person) * 1.022*** 0.709*** ln(temperature - pop) (0.159) (0.162) ln(gdp per person) * 0.536*** ln(temperature - ipop) (0.125) model pols pols pols pols pols pols observations 1852 1852 2650 2650 2486 2486 number of countries 177 177 180 180 177 177 r-squared 0.936 0.939 0.950 0.955 0.917 0.917 log-likelihood dF 22 17 25 23 24 22 no of censored obs p-value for all climate vari- 0.000 0.000 0.000 ables = 0 p-value for pop climate vari- 0.000 0.000 0.000 ables = 0 p-value for ipop climate vari- 0.000 0.000 0.000 ables = 0 Note: standard errors are shown in brackets underneath the relevant coefficients with *** p < 0.001, ** p < 0.01, * p < 0.05. in addi- tion to the variables shown, all of the equations include the following explanatory variables: ln(birthrate 1950), ln(infant mortality 1950) and dummy variables for World Bank regions. Source: authors' estimates. 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 27 Again, the hypothesis that climate variables have no for population-weighted temperature. How this effect on either health inputs or health outcomes is works out country-by-country depends upon the consistently rejected at very high confidence levels. temperature distribution across heavily and thinly populated areas. In this case, the coefficient on a. Hospital beds--Population-weighted temperature precipitation is negative and is linked to the popu- on its own and interacted with GDP per person lation-weighted variable. plus inverse population-weighted precipitation are the key variables in this case. The overall coeffi- c. Infant-mortality--This is influenced by tempera- cient (elasticity) on mean temperature increases-- ture, including an interaction with urbanization from -3.07 for a low-income country with a GDP plus temperature range and inverse population- per person of $1,000, to -0.72 for a middle- weighted precipitation. The signs of the coeffi- income country with a GDP of $10,000 per cients on temperature can be misinterpreted. person, and to +0.70 for a high-income country Assuming that temperature increases (or with a GDP of $40,000 per person. Thus, it is decreases) uniformly throughout a country, the net easy to be misled by simple assumptions about coefficient on temperature is +0.88 for a country how climate "ought" to affect investment in with an urbanization rate of 20 percent, but +0.24 healthcare facilities that are based on experience in for a country with an urbanization rate of 80 a narrow range of countries. The coefficient on percent. Hence, an increase in mean temperature precipitation is positive but quite small. It is is likely to increase infant mortality, but by more possible that this reflects an increased need for in low-income countries with low levels of urban- dispersed hospital facilities when communications ization than in middle- and high-income coun- are subject to disruption caused by high rainfall. tries with higher levels of urbanization. These results conform with a priori expectations. In b. Doctors --The results for the number of doctors addition, a higher temperature range and higher are similar to those for hospital beds, but the precipitation in rural areas tend to increase infant influence of temperature is divided between popu- mortality, both of which seem reasonable. lation and inverse population-weight variables. This seems reasonable since hospitals are invari- Social infrastructure. The number of teachers is used as ably located in urban areas, whereas doctors may the indicator for investment in schools, while the be more dispersed, though this is not the case in number of post offices is used as one indicator for the poorest countries. Again, the interactions with municipal infrastructure. The equations are shown in GDP per person mean that the overall coefficients Table 8. As one would expect, one cannot reject the switch from negative to positive at a GDP per hypothesis that the inverse population-weighted climate person of $12,600 for inverse population-weighted variables have no effect on the number of post offices, temperature, and at a GDP per person of $36,200 which are concentrated in areas of greater population table 8. projection equationS for Social infraStructure Ln(No of teachers) Ln(No of post offices) Variables (1) (2) (3) (4) ln(population 0-14) 0.360*** 0.397*** 0.276*** 0.366*** (0.062) (0.074) (0.065) (0.060) ln(population 15-64) 0.670*** 0.544*** 0.363*** 0.048 (0.115) (0.132) (0.099) (0.091) (continued) 28 t h e c os ts oF ad a p tin g to c limate c h an ge For in Fr as tr u c tu r e table 8. projection equationS for Social infraStructure (continued) Ln(No of teachers) Ln(No of post offices) Variables (1) (2) (3) (4) ln(population 65+) -0.059 0.029 0.231*** 0.445*** (0.042) (0.048) (0.032) (0.030) ln(population) ln(gdp per person) 0.0878*** 0.0851*** 1.977*** -1.124** (0.022) (0.020) (0.155) (0.403) ln(country size) -0.188*** -0.111*** -0.0338*** -0.018 (0.011) (0.014) (0.010) (0.011) ln(% urban) -0.404*** -3.032*** -2.066*** -2.178*** (0.064) (0.429) (0.097) (0.078) ln(gdp per person) squared -0.107*** -0.109*** (0.009) (0.009) ln(country size) squared 0.00308*** 0.00191* 0.0139*** 0.0133*** (0.001) (0.001) (0.001) (0.001) ln(% urban) squared -0.222*** -0.185*** -0.641*** -0.701*** (0.023) (0.025) (0.031) (0.027) ln(gdp per person) * 0.0170*** 0.0114*** ln(country size) (0.001) (0.002) ln(country size) * 0.0861*** 0.0729*** ln(% urban) (0.011) (0.011) ln(% desert) 0.0318*** 0.013 (0.007) (0.007) ln(% arid) -0.121*** -0.0989*** (0.006) (0.005) ln(% semi-arid) 0.0423*** 0.0489*** (0.006) (0.007) ln(% steep land) -0.0549*** -0.0488*** (0.012) (0.013) ln(% very steep land) 0.0167*** 0.00934*** 0.0827*** 0.0667*** (0.003) (0.003) (0.006) (0.010) ln(% no soil constraint) 0.0532*** 0.0314*** (0.003) (0.007) ln(temperature - pop) -0.923*** -0.694*** -2.167*** -10.35*** (0.071) (0.084) (0.119) (0.859) ln(precipitation - pop) -0.130*** -0.289*** -0.173** 0.650*** (0.027) (0.017) (0.059) (0.084) ln(temp range - pop) -0.277*** -0.217*** -0.518*** -0.537*** (0.017) (0.026) (0.066) (0.047) (continued) 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 29 table 8. projection equationS for Social infraStructure (continued) Ln(No of teachers) Ln(No of post offices) Variables (1) (2) (3) (4) ln(precip range - pop) -0.125*** 0.0871*** -0.058 (0.014) (0.014) (0.052) ln(temperature - ipop) 0.222*** 0.751*** -0.110 (0.046) (0.056) (0.178) ln(precipitation - ipop) 0.041 -0.0824* (0.028) (0.038) ln(precip range - ipop) -0.022 0.068 (0.031) (0.046) ln(% urban) * -0.444*** ln(precipitation - pop) (0.043) ln(% urban) * 0.485*** ln(precip range - pop) (0.036) ln(% urban) * 0.825*** ln(temperature - ipop) (0.045) ln(gdp per person) * 0.931*** ln(temperature - pop) (0.112) ln(gdp per person) * -0.100*** ln(precipitation - pop) (0.007) model pols pols pols pols observations 950 950 3251 3251 number of countries 167 167 173 173 r-squared 0.979 0.982 0.909 0.911 log-likelihood dF 25 26 29 27 no of censored obs p-value for all climate variables = 0 0.000 0.000 p-value for pop climate variables = 0 0.000 0.000 p-value for ipop climate variables = 0 0.000 0.065 Note: standard errors are shown in brackets underneath the relevant coefficients with *** p < 0.001, ** p < 0.01, * p < 0.05. in addi- tion to the variables shown, all of the equations include the following explanatory variables: ln(birthrate 1950), ln(infant mortality 1950) and dummy variables for World Bank regions. Source: authors' estimates. 30 t h e c os ts oF ad a p tin g to c limate c h an ge For in Fr as tr u c tu r e density. Otherwise the results show a mixture of Step 1--Construct baseline projections of infrastructure climate interactions with urbanization for the number investment. The projection equations discussed in the of teachers and with GDP per person for post offices. previous section are used to construct baseline projec- Focusing again on mean temperature in the equation tions of the efficient stock of infrastructure assets for for the number of teachers, the overall coefficients for a periods from 2010 to 2050 under the assumption of no uniform increase in temperature are -0.55 for an urban- climate change. The projections of physical infrastruc- ization rate of 20 percent and -0.02 for an urbanization ture demand are based upon standard assumptions rate of 80 percent, so the effect of climate on teachers is about income and population growth, population struc- much larger in low-income and rural countries. This is ture, and urbanization. The value of new investment consistent with the well-established difficulty of equip- required for infrastructure type i for country j in period ping and staffing rural schools. Of course, low-income t is obtained by multiplying Qijt = Qijt+1 - Qijt by countries today are likely to be much more urban in Cij, the unit cost of infrastructure type i in country j at 2050, so that the cumulative impact of an increase in 2005 prices. The unit costs have been compiled from a temperature on the number of teachers will be small large variety of World Bank and other sources. A stan- even in these cases. dardized construction cost index has been used to allow for broad cross-country differences in construction Household size. In several cases, the amount of infra- costs, but allowances are also made for location (urban structure is linked to the projected number of house- or rural) and other special factors. In addition to new holds, so it is necessary to rely upon equations that investment, we have estimated the amount of invest- project the average household size in urban and rural ment that would be required to replace infrastructure areas. These are shown in Table 9. It seems that assets that reach the end of their economic life. There climate does affect average household sizes. The is no realistic way of modeling the age structure of primary mechanism is that higher temperatures are assets in situ at the beginning of the analysis. associated with larger average sizes for both urban and Implementing a full vintage model of infrastructure is rural households. There is also a significant but quanti- not sensible given the uncertainty about other parame- tatively small impact of precipitation on rural household ters in the model. Hence, we have adopted a continu- size. ous depreciation assumption--that is, in period t the required replacement investment is (5/Li)*Qijt where Li is the typical economic life of infrastructure of type i. 6. calculating the coSt of Step 2--Add alternative climate scenarios. The data used for the baseline projections is supplemented with aDaptation projections of the climate variables taken from the climate scenarios that are being used for the whole The calculation of the cost of adaptation involves a EACC study. These are constructed as deltas at differ- number of steps. The description that follows focuses ent dates with respect to the no-climate-change base- on investment or capital costs. A similar process is line derived from calculations of monthly average, required to estimate changes in the costs of operation maximum, and minimum temperatures and precipita- and maintenance, both for the baseline level of infra- tion. To avoid instability in the projections arising from structure and for changes in infrastructure resulting path-dependency and other effects, the climate variables from changes in climate conditions.19 for 2010 are 20-year averages centered on 2010. These are computed for 2010, 2030, ... and then interpolated to give the projections for the 5-year periods. 19 The analysis is formulated in terms of periods that are referred to by the first year in the period--that is,2010­14 is shortened to 2010. No attempt is made to allow for within-period changes in variables. Some Step 3--Project infrastructure quantities under the alterna- of the demographic variables (urbanization and population age struc- ture) used in the projection equations are based on period averages. tive climate scenarios. This is similar to the projection of For other variables, such as income and total population, the added baseline infrastructure quantities in Step 1, but using complexity of using period averages outweighs the benefits because the main projection equations are frontier models and may overstate the climate variables for the alternative climate the levels of infrastructure required to meet demand over relatively scenarios. short periods. 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 31 table 9. projection equationS for average houSeholD Size ln(urban household size) ln(rural household size) variables (1) (2) (3) (4) ln(population 0-14) 0.295*** 0.353*** 0.285*** 0.304*** (0.051) (0.044) (0.047) (0.031) ln(population 15-64) -0.247 -0.357** -0.208 -0.258* (0.132) (0.120) (0.113) (0.109) ln(population 65+) -0.119 -0.073 -0.120 -0.089 (0.099) (0.094) (0.084) (0.096) ln(gdp per person) -0.038 -0.028 0.521*** 0.373** (0.028) (0.027) (0.150) (0.112) ln(country size) 0.0254* 0.0303* 0.0897*** 0.0798** (0.011) (0.012) (0.022) (0.025) ln(% urban) 0.109** 0.119** 0.017 0.044 (0.034) (0.036) (0.026) (0.023) ln(gdp per person) squared -0.0292** -0.0200* (0.009) (0.008) ln(gdp per person) * -0.0112*** -0.0107*** ln(country size) (0.002) (0.003) ln(% desert) -0.0196*** -0.0212*** (0.003) (0.003) ln(% arid) 0.0152*** 0.0195*** (0.004) (0.004) ln(% semi-arid) 0.0217*** 0.0248*** (0.003) (0.004) ln(temperature - pop) 0.859*** 0.777*** 0.844*** 0.718*** (0.140) (0.138) (0.178) (0.111) ln(precipitation - pop) -0.118* -0.118* -0.119*** (0.047) (0.053) (0.025) ln(temp range - pop) 0.054 0.156 (0.079) (0.080) ln(precip range - pop) 0.109 0.039 (0.066) (0.064) ln(temperature - ipop) -0.166** -0.177*** 0.057 (0.063) (0.047) (0.043) ln(precipitation - ipop) 0.0785*** 0.0795*** 0.0296** (0.016) (0.022) (0.010) ln(precip range - ipop) -0.0827*** -0.0496** (0.019) (0.018) model pols pols pols pols (continued) 32 t h e c os ts oF ad a p tin g to c limate c h an ge For in Fr as tr u c tu r e table 9. projection equationS for average houSeholD Size (continued) observations 322 322 254 254 number of countries 126 126 112 112 r-squared 0.991 0.991 0.996 0.996 log-likelihood dF 20 15 25 21 no of censored obs p-value for all climate variables = 0 0.000 0.000 p-value for pop climate variables = 0 0.000 0.000 p-value for ipop climate variables = 0 0.000 0.001 Note: standard errors are shown in brackets underneath the relevant coefficients with *** p < 0.001, ** p < 0.01, * p < 0.05. in addi- tion to the variables shown, all of the equations include the following explanatory variables: ln(birthrate 1950), ln(infant mortality 1950) and dummy variables for World Bank regions. Source: authors' estimates. Step 4--Apply the dose-response relationship to estimate b. Variant 2 assumes that the asset is designed to changes in unit costs for alternative climate scenarios. We withstand the worst conditions that it might be calculate the changes in unit costs for infrastructure exposed to over its life--that is: type i in country j for period t, Cijt, using the climate change deltas for the alternative climate scenarios and the dose-response relationships discussed in Appendix Cijt = d [max(V jt ,...,V j ,t + Li )]Cij (10) 1. There is a complication that has to be considered. This concerns the question of whether the design stan- on the assumption that the severity of storms dards used for infrastructure are--or should increases monotonically with the relevant climate be--forward looking. Normal engineering practice does variable(s) V. not take account of changes in underlying climate conditions. Thus, in designing for a 100-year storm, the The difficulty with Variant 2 is that it implies that the engineer looks at the characteristics of the 100-year asset is significantly overdesigned for most of its work- storm on the basis of evidence of storms up to the ing life because it will only be exposed to the most current date. Clearly, this does not allow for changes in severe weather conditions at the very end of its life. In the severity of the 100-year storm that might be economic terms, Variant 2 is not the optimal solution expected to occur over the life of the asset. There are and it would be sensible to design for the 100-year two possible approaches that can be adopted. storm consistent with the expected climate at some earlier date. There is no general solution, since the a. Variant 1 assumes that the dose-response adjust- optimal period to look ahead depends upon both the ment to unit costs is calculated using current expected increase in the severity of storms over the climate conditions--that is: future and the shape of the dose-response relationship. For consistency with the analysis of coastal protection, we have modified Variant 2 to look ahead for a fixed Cijt = d [V jt ]Cij (9) period of 50 years. where d[ ] is the dose-response relationship. 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 33 Step 5--Estimate the change in total investment costs for pro rata with maximum monthly precipitation. the baseline projections. This yields the Delta-P estimates of the cost of adaptation for each climate scenario with d. Changes in temperature affect the rate at which two variants corresponding to the alternatives at Step 4 oxygen levels recover in rivers to which the efflu- above. ent is discharged from waste water treatment plants. Thus, a higher level of BOD removal is Step 6--Estimate the change in investment costs due to the required to maintain the quality of receiving difference between the baseline infrastructure quantities and waters. This implies higher consumption of elec- the alternative climate scenario quantities. This yields the tricity or use of chemicals at treatment plants. Delta-Q estimates of the cost of adaptation for each The increase in O&M costs is linked to the climate scenario. increase in average temperatures and is incorpo- rated in our estimates of the cost of adaptation. Step 7--Special adjustments. We have incorporated some special factors in the calculation of the costs of adapting These steps are followed in deriving the estimates of to climate change that could not be represented by the Cijt used in calculating the Delta-P costs of adapta- general dose-response relationships. These are: tion in the first part of equation (2): a. For electricity generation, we have taken account of the decrease in the operating efficiency of exist- I jt [1] = Cijt [Qijt +1 - Qijt + Rijt ] (11) i ing thermal power plants as the ambient tempera- ture increases. The effect is documented in the with the Qijt, etc. given by the baseline projections of literature for ambient temperatures above 15°C, infrastructure investment. The Delta-Q costs of adapta- though it is possible to design new power plants tion are defined by: to include absorption chillers to bring the ambient temperature of the air entering turbines down to 15°C for a relatively minor penalty on operating I jt [2] = (Cijt + Cijt )[Qijt +1 - Qijt + R(12) ijt ] i costs. I jt [2] = (Cijt + Cijt )[Qijt +1 - Qijt + Rijt ] i b. Another special factor for electricity generation concerns the efficiency and feasibility of water in which Qijt are obtained from the changes in the cooling as temperatures increase, because of limits baseline investments associated with the alternative on the temperature rise that can be permitted in climate scenarios. the receiving waters. Dry cooling can be adopted either in parallel with wet cooling or as an alterna- Equation (12) yields engineering estimates of the tive in particularly hot or dry locations. The Delta-Q costs, which reflect an assumption that coun- model assumes that an increasing proportion of tries will respond to climate change by building more or power plants will rely upon dry cooling as average less infrastructure. However, it should be noted that temperatures rise. more cost-effective options may be available. In another paper, we examine one of these options in more c. The operating costs of water treatment plants may detail for the water sector (Hughes, Chinowsky, and increase as a result of climate change. Primary Strzepek 2010). We show that the welfare cost of using attention has focused on the amount of chemicals water abstraction fees to limit increases in demand for used for flocculation if the levels of turbidity and water may be lower than the cost of building additional suspended solids in raw water rise. This is likely capacity for water and wastewater treatment. Our to be associated with changes in levels of peak results demonstrate that this economic approach can flow in rivers from which water is abstracted, so reduce the cost of adaptation in the water sector by a the model allows for cost of chemicals to increase substantial amount relative to the engineering approach of building more infrastructure assets in response to an 34 t h e c os ts oF ad a p tin g to c limate c h an ge For in Fr as tr u c tu r e increase in demand for water. 7. eStimateS of the coStS of aDaptation There is an obvious instrument--water abstraction fees--available in the water sector. Similar policies Our estimates of the costs of adaptation for electricity could be followed for some other types of infrastruc- and water services are shown in Tables 10 to 14. To ture--for example, energy and transport. As a conse- facilitate comparisons all figures in the tables are quence, the estimates of the Delta-Q costs of adaptation presented as average costs per year at 2005 prices over will tend to overstate the economic costs of adaptation the relevant period--that is, for 2010­50 as a whole or in countries that face an increase in demand for infra- for each decade--with no discounting. Figures are structure as a consequence of climate change. Since the rounded to the nearest $1 billion per year to avoid any effect is one-sided--that is, an economic approach can impression of spurious accuracy. As a consequence, reduce costs when the demand for infrastructure sums of the separate numbers may differ from the rele- increases, but would not be required when the demand vant totals due to rounding. The Delta-P increases in for infrastructure decreases--it is safe to conclude that investment, O&M, and total costs for the two climate the engineering estimates of the Delta-Q costs of adap- scenarios are shown by infrastructure category and tation presented in the next section represent an upper country class in Table 10. The baseline costs without bound on the costs of following a cost-effective strategy any climate change are shown as a point of reference. of adaptation. In all cases, the costs of adaptation are substantially table 10. Delta-p coStS of aDaptation by category anD country claSS for 2010­50 (uS$ billion per year at 2005 prices, no discounting) Lower middle Upper middle NCAR scenario Cost type Low income income income High income Total 1. power & telephones capital cost 0 0 0 1 2 o&m cost 0 0 0 0 0 total cost 0 1 0 1 2 Baseline cost 132 173 92 304 701 2. Water & sewers capital cost 0 0 0 0 0 o&m cost 0 0 0 0 1 total cost 0 0 0 0 1 Baseline cost 119 154 95 193 562 3. roads capital cost 3 2 1 6 11 o&m cost 0 0 0 0 0 total cost 3 2 1 7 12 Baseline cost 67 56 60 215 398 4. other transport capital cost 0 0 1 0 1 o&m cost 0 0 3 1 5 total cost 0 0 4 1 6 Baseline cost 8 18 86 31 142 5. health & schools capital cost 0 1 0 1 2 o&m cost 0 0 0 0 0 total cost 0 1 0 1 2 Baseline cost 36 121 92 302 551 (continued) 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 35 table 10. Delta-p coStS of aDaptation by category anD country claSS for 2010­50 (uS$ billion per year at 2005 prices, no discounting) (continued) Lower middle Upper middle NCAR scenario Cost type Low income income income High income Total 6. urban infrastructure capital cost 8 6 2 5 20 o&m cost 0 0 0 0 0 total cost 8 6 2 5 20 Baseline cost 287 219 209 841 1,555 total capital cost 11 9 4 14 37 o&m cost 0 0 4 2 6 total cost 11 9 8 15 43 Baseline cost 649 740 634 1,887 3,910 CSIRO scenario 1. power & telephones capital cost 0 0 0 1 1 o&m cost 0 0 0 0 0 total cost 0 0 0 1 2 Baseline cost 132 173 92 304 701 2. Water & sewers capital cost 0 0 0 0 0 o&m cost 0 0 0 0 1 total cost 0 0 0 0 1 Baseline cost 119 154 95 193 562 3. roads capital cost 1 1 0 5 6 o&m cost 0 0 0 0 0 total cost 1 1 0 5 7 Baseline cost 67 56 60 215 398 4. other transport capital cost 0 0 0 0 1 o&m cost 0 0 2 1 3 total cost 0 0 2 1 4 Baseline cost 8 18 86 31 142 5. health & schools capital cost 0 0 0 1 2 o&m cost 0 0 0 0 0 total cost 0 0 0 1 2 Baseline cost 36 121 92 302 551 6. urban infrastructure capital cost 4 2 1 4 11 o&m cost 0 0 0 0 0 total cost 4 2 1 4 11 Baseline cost 287 219 209 841 1,555 total capital cost 5 4 2 10 21 o&m cost 0 0 2 1 4 total cost 5 4 4 11 25 Baseline cost 649 740 634 1,887 3,910 Source: authors' estimates 36 t h e c os ts oF ad a p tin g to c limate c h an ge For in Fr as tr u c tu r e higher for the NCAR scenario than for the CSIRO the low-income countries, the costs of adaptation are scenario, so we will focus on the NCAR figures. The highest for East Asia (EAP) and for South Asia (SAS), total Delta-P cost of adaptation over 40 years is about 1 reflecting their populations and aggregate income. The percent of the baseline cost for all countries. The ratio costs for Europe and Central Asia (ECA) are higher of adaptation costs to baseline costs is highest for low- than might have been anticipated, but this reflects the income countries at about 1.7 percent and is lowest for initial level of infrastructure leading to relatively high high-income countries. O&M costs. Sub-Saharan Africa (SSA) has the high- est ratio of adaptation costs to baseline costs at 2.3 Table 11 shows the same information for developing percent. Broken down by infrastructure category and countries disaggregated by World Bank region. Outside region, the heaviest burden of adaptation is for other table 11. Delta-p coStS of aDaptation by infraStructure category anD worlD bank region for 2010­50 (uS$ billion per year at 2005 prices, no discounting) NCAR scenario Cost type EAP ECA LCA MNA SAS SSA Total capital 1. power & telephones 0 0 0 0 0 0 1 cost o&m cost 0 0 0 0 0 0 0 total cost 0 0 0 0 0 0 1 Baseline 137 75 41 24 79 41 397 cost capital 2. Water & sewers 0 0 0 0 0 0 0 cost o&m cost 0 0 0 0 0 0 0 total cost 0 0 0 0 0 0 1 Baseline 115 71 54 25 81 22 368 cost capital 3. roads 1 0 1 0 2 1 5 cost o&m cost 0 0 0 0 0 0 0 total cost 1 0 1 0 2 1 5 Baseline 36 37 31 13 44 23 183 cost capital 4. other transport 0 1 0 0 0 0 1 cost o&m cost 0 3 0 0 0 0 4 total cost 0 4 0 0 0 0 5 Baseline 16 80 6 2 4 4 111 cost capital 5. health & schools 1 0 0 0 0 0 1 cost o&m cost 0 0 0 0 0 0 0 total cost 1 0 0 0 0 0 1 Baseline 93 49 52 20 25 9 249 cost capital 6. urban infrastructure 5 1 2 0 5 2 15 cost o&m cost 0 0 0 0 0 0 0 (continued) 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 37 table 11. Delta-p coStS of aDaptation by infraStructure category anD worlD bank region for 2010­50 (uS$ billion per year at 2005 prices, no discounting) (continued) NCAR scenario Cost type EAP ECA LCA MNA SAS SSA Total total cost 5 1 2 0 5 2 15 Baseline 163 159 78 32 252 31 714 cost capital total 8 2 3 1 8 3 24 cost o&m cost 0 3 0 0 0 0 4 total cost 8 5 3 1 8 3 28 Baseline 560 470 262 116 485 130 2,023 cost CSIRO scenario capital 1. power & telephones 0 0 0 0 0 0 1 cost o&m cost 0 0 0 0 0 0 0 total cost 0 0 0 0 0 0 1 Baseline 137 75 41 24 79 41 397 cost capital 2. Water & sewers 0 0 0 0 0 0 0 cost o&m cost 0 0 0 0 0 0 0 total cost 0 0 0 0 0 0 0 Baseline 115 71 54 25 81 22 368 cost capital 3. roads 0 0 0 0 1 0 2 cost o&m cost 0 0 0 0 0 0 0 total cost 0 0 0 0 1 0 2 Baseline 36 37 31 13 44 23 183 cost capital 4. other transport 0 0 0 0 0 0 0 cost o&m cost 0 2 0 0 0 0 2 total cost 0 2 0 0 0 0 3 Baseline 16 80 6 2 4 4 111 cost capital 5. health & schools 0 0 0 0 0 0 1 cost o&m cost 0 0 0 0 0 0 0 total cost 0 0 0 0 0 0 1 Baseline 93 49 52 20 25 9 249 cost capital 6. urban infrastructure 2 1 1 0 3 1 7 cost o&m cost 0 0 0 0 0 0 0 total cost 2 1 1 0 3 1 7 (continued) 38 t h e c os ts oF ad a p tin g to c limate c h an ge For in Fr as tr u c tu r e table 11. Delta-p coStS of aDaptation by infraStructure category anD worlD bank region for 2010­50 (uS$ billion per year at 2005 prices, no discounting) (continued) CSIRO scenario NCAR scenario Cost type EAP ECA LCA MNA SAS SSA Total Baseline 163 159 78 32 252 31 714 cost capital total 3 1 1 0 4 1 11 cost o&m cost 0 2 0 0 0 0 3 total cost 3 3 1 1 4 1 14 Baseline 560 470 262 116 485 130 2,023 cost Source: authors' estimates. transport in the ECA region, largely because of the Tables 13 and 14 give details of the costs of adaptation high level of O&M costs. This is followed by roads in by infrastructure category and country class or region South Asia, but in both cases the cost of adaptation is when the definition of the cost of adaptation is little more than 5 percent of baseline costs. extended to include both the Delta-P and the Delta-Q components in the analysis. Recall that the Delta-Q Table 12 shows the breakdown of the Delta-P costs of costs are driven by the increase or decrease in the demand for infrastructure associated with the projected adaptation for all infrastructure by decade. The relative changes in climate. Table 13 shows that the Delta-Q cost of adaptation increases gradually from about 1 changes are negative for the world as a whole in both percent of baseline costs for 2010­19 to about 1.6 scenarios. This means that total expenditure on infra- percent for 2040­49. One component of this increase is structure will fall as a consequence of climate change, the rise in O&M costs in the ECA region, which has though more investment may be required in some coun- already been highlighted, but even for all regions other tries and some sectors. However, the fall in total expen- than ECA there is an increase from about 1.2 percent diture is most important for high-income countries, so of baseline costs in the first decade to 1.6 percent in the that the overall scale of the Delta-Q adjustments for final decade. developing countries is similar to that of the Delta-P table 12. Delta-p coStS of aDaptation by DecaDe anD worlD bank region for all infraStructure (uS$ billion per year at 2005 prices, no discounting) NCAR scenario Cost type EAP ECA LCA MNA SAS SSA Total 2010-19 capital cost 5 2 1 1 4 1 13 o&m cost 0 0 0 0 0 0 1 total cost 5 2 1 1 4 1 14 Baseline cost 417 395 197 77 273 79 1,438 2020-29 capital cost 7 2 2 1 6 2 21 o&m cost 0 2 0 0 0 0 3 total cost 7 5 2 1 7 2 24 Baseline cost 505 452 238 101 396 109 1,801 (continued) 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 39 table 12. Delta-p coStS of aDaptation by DecaDe anD worlD bank region for all infraStructure (uS$ billion per year at 2005 prices, no discounting) (continued) NCAR scenario Cost type EAP ECA LCA MNA SAS SSA Total 2030-39 capital cost 9 2 3 1 9 3 27 o&m cost 0 5 0 0 0 0 6 total cost 9 7 3 1 9 3 33 Baseline cost 608 497 283 127 550 146 2,213 2040-49 capital cost 11 2 4 1 11 5 34 o&m cost 0 6 0 0 0 0 7 total cost 11 8 4 1 12 5 41 Baseline cost 710 538 330 156 719 187 2,641 CSIRO scenario 2010-19 capital cost 3 1 1 0 1 0 6 o&m cost 0 0 0 0 0 0 1 total cost 3 1 1 0 1 1 7 Baseline cost 417 395 197 77 273 79 1,438 2020-29 capital cost 3 1 1 0 2 1 7 o&m cost 0 1 0 0 0 0 2 total cost 3 2 1 1 2 1 9 Baseline cost 505 452 238 101 396 109 1,801 2030-39 capital cost 3 1 1 0 4 1 12 o&m cost 0 3 0 0 0 0 4 total cost 4 4 1 1 4 1 15 Baseline cost 608 497 283 127 550 146 2,213 2040-49 capital cost 4 2 1 1 8 2 19 o&m cost 0 3 0 0 0 0 4 total cost 4 6 2 1 8 2 23 Baseline cost 710 538 330 156 719 187 2,641 Source: authors' estimates. costs. This is illustrated in the breakdown of adaptation The striking feature of the results--taking account of costs by World Bank region in Table 14, which shows both Delta-P and Delta-Q costs of adaptation--is how that the sum of Delta-P and Delta-Q costs of adapta- small the overall costs of adaptation are relative to the tion is close to zero for all developing countries. The baseline costs. The impact of climate change is far from net costs of adaptation per year over the full period vary evenly distributed, but even in the worst-affected from a negative cost (that is, a saving) of $7 billion per region--MNA--the net cost is little more than 2 year for East Asia to a positive cost of $2 billion per percent of baseline expenditures. Thus, in practice the year for the Middle East and North Africa (MNA). cost of adaptation for infrastructure is well within all of the margins of error inherent in this type of exercise. 40 t h e c os ts oF ad a p tin g to c limate c h an ge For in Fr as tr u c tu r e table 13. total coStS of aDaptation by infraStructure category anD country claSS for 2010­50 (uS$ billion per year at 2005 prices, no discounting) Lower middle Upper middle NCAR scenario Cost type Low income High income Total income income 1. power & telephones delta-p 0 1 0 1 2 delta-Q -4 -2 -2 -20 -29 delta-p+delta-Q -4 -2 -2 -19 -27 Baseline cost 132 173 92 304 701 2. Water & sewers delta-p 0 0 0 0 1 delta-Q -5 1 1 -1 -4 delta-p+delta-Q -5 1 1 -1 -3 Baseline cost 119 154 95 193 562 3. roads delta-p 3 2 1 7 12 delta-Q 0 -2 -3 -23 -27 delta-p+delta-Q 3 0 -2 -17 -16 Baseline cost 67 56 60 215 398 4. other transport delta-p 0 0 4 1 6 delta-Q 0 0 -4 -1 -6 delta-p+delta-Q 0 0 0 0 0 Baseline cost 8 18 86 31 142 5. health & schools delta-p 0 1 0 1 2 delta-Q 0 -1 1 3 2 delta-p+delta-Q 0 0 1 4 4 Baseline cost 36 121 92 302 551 6. urban infrastructure delta-p 8 6 2 5 20 delta-Q -3 -5 -3 -10 -21 delta-p+delta-Q 5 0 -1 -5 -1 Baseline cost 287 219 209 841 1,555 total delta-p 11 9 8 15 43 delta-Q -13 -10 -10 -53 -86 delta-p+delta-Q -1 -1 -2 -38 -43 Baseline cost 649 740 634 1,887 3,910 CSIRO scenario 1. power & telephones delta-p 0 0 0 1 2 delta-Q -4 1 -1 -4 -8 delta-p+delta-Q -4 2 0 -4 -6 Baseline cost 132 173 92 304 701 2. Water & sewers delta-p 0 0 0 0 1 delta-Q -2 1 1 -8 -8 delta-p+delta-Q -2 1 1 -7 -7 Baseline cost 119 154 95 193 562 3. roads delta-p 1 1 0 5 7 (continued) 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 41 table 13. total coStS of aDaptation by infraStructure category anD country claSS for 2010­50 (uS$ billion per year at 2005 prices, no discounting) (continued) CSIRO scenario Lower middle Upper middle NCAR scenario Cost type Low income High income Total income income delta-Q -2 -2 -4 -23 -30 delta-p+delta-Q -1 -1 -4 -18 -24 Baseline cost 67 56 60 215 398 4. other transport delta-p 0 0 2 1 4 delta-Q 0 0 -2 -1 -3 delta-p+delta-Q 0 0 0 1 0 Baseline cost 8 18 86 31 142 5. health & schools delta-p 0 0 0 1 2 delta-Q -1 -2 0 1 -2 delta-p+delta-Q -1 -1 0 1 0 Baseline cost 36 121 92 302 551 6. urban infrastructure delta-p 4 2 1 4 11 delta-Q -5 -5 -3 -10 -24 delta-p+delta-Q -1 -3 -3 -7 -13 Baseline cost 287 219 209 841 1,555 total delta-p 5 4 4 11 25 delta-Q -14 -7 -9 -45 -75 delta-p+delta-Q -8 -3 -5 -34 -50 Baseline cost 649 740 634 1,887 3,910 Source: authors' estimates. table 14. total coStS of aDaptation by infraStructure category anD region for 2010­50 (uS$ billion per year at 2005 prices, no discounting) NCAR scenario Cost type EAP ECA LCA MNA SAS SSA Total 1. power & telephones delta-p 0 0 0 0 0 0 1 delta-Q -5 -1 1 2 -3 -1 -9 delta-p+delta-Q -5 -1 1 2 -3 -1 -7 Baseline cost 137 75 41 24 79 41 397 2. Water & sewers delta-p 0 0 0 0 0 0 1 delta-Q -3 4 0 1 -4 -1 -3 delta-p+delta-Q -3 4 0 1 -4 -1 -3 Baseline cost 115 71 54 25 81 22 368 3. roads delta-p 1 0 1 0 2 1 5 delta-Q -1 -1 -1 0 0 0 -4 delta-p+delta-Q 0 -1 0 0 2 1 1 Baseline cost 36 37 31 13 44 23 183 (continued) 42 t h e c os ts oF ad a p tin g to c limate c h an ge For in Fr as tr u c tu r e table 14. total coStS of aDaptation by infraStructure category anD region for 2010­50 (uS$ billion per year at 2005 prices, no discounting) (continued) NCAR scenario Cost type EAP ECA LCA MNA SAS SSA Total 4. other transport delta-p 0 4 0 0 0 0 5 delta-Q 0 -4 0 0 0 0 -5 delta-p+delta-Q 0 0 0 0 0 0 0 Baseline cost 16 80 6 2 4 4 111 5. health & schools delta-p 1 0 0 0 0 0 1 delta-Q -1 1 0 0 0 0 -1 delta-p+delta-Q 0 1 0 0 0 0 1 Baseline cost 93 49 52 20 25 9 249 6. urban infrastructure delta-p 5 1 2 0 5 2 15 delta-Q -4 -2 -1 -1 -2 0 -11 delta-p+delta-Q 1 -2 0 -1 3 1 4 Baseline cost 163 159 78 32 252 31 714 total delta-p 8 5 3 1 8 3 28 delta-Q -15 -5 -2 1 -10 -2 -33 delta-p+delta-Q -7 0 1 2 -2 1 -5 Baseline cost 560 470 262 116 485 130 2,023 CSIRO scenario 1. power & telephones delta-p 0 0 0 0 0 0 1 delta-Q -1 0 0 1 -3 -2 -4 delta-p+delta-Q -1 1 0 1 -3 -2 -3 Baseline cost 137 75 41 24 79 41 397 2. Water & sewers delta-p 0 0 0 0 0 0 0 delta-Q -1 2 0 1 -2 0 0 delta-p+delta-Q -1 2 0 1 -2 0 0 Baseline cost 115 71 54 25 81 22 368 3. roads delta-p 0 0 0 0 1 0 2 delta-Q -1 -2 -2 -1 -1 -1 -8 delta-p+delta-Q -1 -2 -2 0 0 0 -6 Baseline cost 36 37 31 13 44 23 183 4. other transport delta-p 0 2 0 0 0 0 3 delta-Q 0 -2 0 0 0 0 -3 delta-p+delta-Q 0 0 0 0 0 0 0 Baseline cost 16 80 6 2 4 4 111 5. health & schools delta-p 0 0 0 0 0 0 1 delta-Q -1 0 0 -1 -1 0 -2 delta-p+delta-Q -1 0 0 0 0 0 -1 Baseline cost 93 49 52 20 25 9 249 6. urban infrastructure delta-p 2 1 1 0 3 1 7 (continued) 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 43 table 14. total coStS of aDaptation by infraStructure category anD region for 2010­50 (uS$ billion per year at 2005 prices, no discounting) (continued) NCAR scenario CSIRO scenario Cost type EAP ECA LCA MNA SAS SSA Total delta-Q -4 -3 -1 -1 -4 -1 -13 delta-p+delta-Q -2 -2 -1 -1 -1 0 -6 Baseline cost 163 159 78 32 252 31 714 total delta-p 3 3 1 1 4 1 14 delta-Q -8 -5 -3 0 -10 -3 -29 delta-p+delta-Q -5 -2 -1 1 -7 -2 -16 Baseline cost 560 470 262 116 485 130 2,023 Source: authors' estimates. 8. concluSion Delta-Q effects may be positive or negative--increasing or decreasing the costs of adaptation--in different The work reported in this paper represents the most countries. Summed by region, the Delta-Q totals are extensive and careful effort that has been made to esti- negative in all regions except MNA. The results of our mate the costs of adapting to climate change in the econometric analysis do not dictate that climate change infrastructure sector at a global level. Our primary will have the effect of reducing demand for generating conclusion is that the cost of adapting to climate capacity or roads. The equations contain complex inter- change, given the baseline level of infrastructure provi- actions between income and various climate variables sion, is no more than 1­2 percent of the total cost of --not merely temperature--with both population- providing that infrastructure. While there are differ- weighted and inverse population-weighted variants. It ences across regions and sectors, the pattern is clear and does not seem plausible that these effects are merely unambiguous--the cost of adaptation is small in rela- capturing the influence of one or more omitted vari- tion to other factors that may influence the future costs ables. Hence, estimates of the costs of adaptation that of infrastructure. We accept that we may have omitted ignore the potential impact of climate change on the or underestimated some of the costs of adaptation. On demand side may give a rather partial view of the over- the other hand, we have consistently tried to err on the all picture. generous side--increasing our estimates of probable costs when there is reasonable doubt. Further, it can be shown that an economic rather than an engineering approach to adaptation when climate change increases referenceS the demand for infrastructure will reduce the Delta-Q costs by a substantial amount in some cases. Thus, in Acemoglu, D., S. Johnson, and J.A. Robinson. 2001. our view it is extremely unlikely that revised estimates "The Colonial Origins of Comparative Development." will alter our conclusion about the relative magnitude of American Economic Review 91:1369­1401. the costs of adaptation. AICD. 2009. Africa's Infrastructure: A Time for The second conclusion of our study is that the impact Transformation. Washington, DC: World Bank. of climate change on the overall demand for infrastruc- ture may be more important than the increase in the Albouy, D.Y. 2008. "The Colonial Origins of cost of providing the baseline level of provision. These Comparative Development: An Investigation of the 44 t h e c os ts oF ad a p tin g to c limate c h an ge For in Fr as tr u c tu r e Settler Mortality Data." National Bureau of Economic Horowitz, J.K. 2008. "The Income-Temperature Research Working Paper. Cambridge, MA: NBER. Relationship in A Cross-Section of Countries and its Implications for Predicting the Effects of Global Baum, C.F. 2005. An Introduction to Modern Warming." Working Paper. College Park, MD: Econometrics using Stata. College Station, TX: Stata Department of Agricultural and Resource Economics. Press. Hughes, G.A. 2010. "The Econometrics of Climate Cameron, A.C., and P.K. Trivedi. 2006. Change." Department of Economics Working Paper. Microeconometrics. Cambridge, UK: Cambridge Edinburgh: University of Edinburgh. University Press. Hughes, G.A, P. Chinowsky, and K. Strzepek. 2010. Canadian Standards Association. 2006. The Role of "The Costs of Adaptation to Climate Change for Standards in Adapting Canada's Infrastructure to the Water Infrastructure in OECD Countries." Department Impacts of Climate Change. Toronto: Canadian Standards of Economics Working Paper. Edinburgh: University of Association. Edinburgh. Dell, M., B.F. Jones, and B.A. Olken. 2008. "Climate IEA. 2008. World Energy Outlook 2008. Paris: Shocks and Economic Growth: Evidence from the Last International Energy Agency. Half Century." National Bureau of Economic Research Working Paper. Cambridge, MA: NBER. Larsen, P.H., S. Goldsmith, O. Smith, M.L.Wilson, K. Strzepek, P. Chinowsky, and B. Saylor. 2008. Driscoll, J.C., and A.C. Kraay. 1998. "Consistent "Estimating future costs for Alaska public infrastructure Covariance Matrix Estimation with Spatially at risk from climate change." Global Environment Dependent Panel Data." Review of Economics and Change 18(3):442­457. Statistics 80: 549­560. McGray, H., A. Hammill, and R. Bradley, eds. 2008. EIA. 2008. International Energy Outlook 2008. Weathering the Storm: Options for Framing Adaptation Washington, DC: Energy Information Agency. and Development. Washington, DC: World Resources Institute. Estache, A., B. Speciale, and D. Veredas. 2005. "How much does infrastructure matter to growth in Nordhaus, W.D. 2002." Modelling induced innovation Sub-Saharan Africa?" Working Paper. Brussels: in climate-change policy." In A. Grubler, N. EeCARES, Universite Libre de Bruxelles. Nakicenovic and W.D. Nordhaus (eds), Modeling Induced Innovation in Climate Change Policy. Fay, M., and T. Yepes. 2003. "Investing in Infrastructure: Washington, DC; Resources for the Future Press. What is needed from 2000 to 2010?" Policy Research Working Paper. Washington, DC: World Bank. Tol, R.S.J. 2007. On the optimal control of carbon dioxide emissions: an application of FUND. Amsterdam: Hoechle, D. 2007."Robust standard errors for panel institute for environmental studies, vrije regression with cross-section dependence."The Stata universiteit. Journal 7:281­312. UNFCCC. 2007. Climate Change: Impacts, Vulnerabilities Hope, C. 2003. The Marginal Impacts of CO2, CH4 and and Adaptation in Developing Countries. UNFCCC SF6 Emissions. Judge Institute of Management Working Secretariat. Bonn: UNFCCC. Paper No. 10/2003. Cambridge, UK: University of Cambridge. 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 45 appenDix 1. Derivation of 1. Estimation of Dose-Response Values for the climate DoSe-reSponSe Construction Costs relationShipS To generate dose-response values for infrastructure construction costs, we employed two general Paul Chinowsky approaches. The first estimates dose-response values (Department of Civil, Environmental and Architec- based on the cost associated with the change in the tural Engineering, University of Colorado) typical building code update, while the second more directly estimates the incremental costs of climate stres- Jason Price & Jim Neumann sors and design changes. We use the building code (Industrial Economics Inc, Cambridge, Mass) approach to generate dose-response values for paved roads, buildings, and transmission towers and the latter for bridges and unpaved roads. The dose-response relationship between climate change and the cost of building and maintaining infrastructure Our assessment of dose-response values for infrastruc- is a central component of the World Bank's assessment ture construction costs assumes perfect foresight with of infrastructure adaptation costs. The magnitude of respect to climate change. Therefore, these dose- the dose-response relationship is likely to vary both by response values represent the relationship between infrastructure type and by country. Variation in this infrastructure construction costs at the time of relationship by infrastructure type reflects, among other construction and the changes in climate projected factors, differences in the materials with which different during the infrastructure's lifespan. types of infrastructure are constructed and the ways in which different types of infrastructure are used; for A. Building Code Methodology example, buildings often provide heating and cooling. In addition, variation in the dose-response relationship The building code methodology is based on the premise by country reflects inter-country variation in labor and that a major update of design standards results in a 0.8 materials costs as well as terrain; for example, varying percent increase in construction costs (FEMA 1998). degrees of flat versus mountainous terrain. The readily available data suggest that such code updates would occur with every 10 centimeter (cm) The data and methods supporting the World Bank's increase in precipitation for paved roads and buildings; assessment of dose-response values by infrastructure therefore, we express the precipitation dose-response type and country are outlined in the sections below. relationship for these specific types of infrastructure as This information is presented separately for infrastruc- follows: ture construction costs and infrastructure maintenance costs. Exhibits 1 and 2 describe the specific dose- response relationships analyzed. We note that the dose- (1) C P , BPR = 0.8%(B BPR ) response values estimated for both construction costs where and maintenance costs are based on the cost of building and maintaining infrastructure in the United States. To CP,BPR = change in building and paved road construc- develop dose-response values specific to individual tion costs associated with a 10 cm change in annual developing countries, we scaled the U.S.-based cost esti- precipitation mates using an inter-country construction cost index published by Compass International Consultants Inc. BBPR = base construction costs for buildings and paved (2009). The country-specific values that make up this roads index represent average construction costs for each country relative to costs in the United States. Based on published construction cost information, we assume base construction costs of $185 per square foot for medical buildings as a base for public facilities 46 t h e c os ts oF ad a p tin g to c limate c h an ge For in Fr as tr u c tu r e exhibit 1 -- DoSe-reSponSe DeScriptionS for conStruction coStS Precipitation Dose-Response Temperature Dose-Response Wind Dose-Response Bridges change in construction costs per not estimated. impact likely to not estimated. impact likely to bridge per 1 foot increase in bridge be minimal. be minimal. height. Paved Roads change in costs of constructing a km change in cost of constructing not estimated. impact likely to of paved road per 10 cm change in a km of paved road per step- be minimal. annual precipitation projected during wise increase in the maximum lifespan relative to baseline climate. of monthly maximum tempera- dose-response represents change in ture values projected during costs for every 10 cm increment. lifespan relative to baseline cli- mate. the first increase occurs after a 1 degree celsius change in maximum tempera- ture. every other step occurs 3 degrees celsius beyond that. Unpaved Roads change in construction costs per km not estimated. impact likely to not estimated. impact likely to per 1% change in the maximum of the be minimal. be minimal. monthly maximum precipitation values projected during lifespan relative to baseline climate. Transmission Poles not estimated. impact likely to be not estimated. impact likely to percent change in costs per 15 minimal. be minimal. mph (~24 kmh) increase in the maximum of the monthly maxi- mum wind speeds projected during lifespan, relative to base- line climate. Buildings change in costs per square foot, per change in costs per square not estimated. impact likely to 10 cm change in annual precipitation foot, per 0.5 degree change be minimal. projected during lifespan. celsius in annual average tem- perature during lifespan, rela- tive to baseline climate. exhibit 2 -- DoSe-reSponSe DeScriptionS for maintenance coStS Precipitation Temperature Paved Roads - Existing change in annual maintenance costs per change in annual maintenance costs per km per 1 degree km per 10 cm change in annual rainfall change celsius in maximum of monthly maximum temper- projected during lifespan relative to base- ature projected during lifespan. line climate. Paved Roads - Newly paved roads constructed after 2010 would have no maintenance impact if designed for changes in cli- Constructed mate expected during their lifetime. Unpaved Roads change in annual maintenance costs per not estimated. impact likely to be minimal. 1% change in maximum of monthly maxi- mum precipitation projected during lifespan. Railroads not estimated. impact likely to be mini- change in annual maintenance costs per km per 1 degree mal. celsius change in maximum of monthly maximum temper- ature projected during lifespan. Buildings - Existing change in annual maintenance costs per change in annual maintenance costs per square foot per square foot per 10 cm change in annual 1 degree change celsius in annual average temperature rainfall projected during lifespan. projected during lifespan. Buildings - Newly Buildings constructed after 2010 would have no maintenance impact if designed for changes in climate Constructed expected during their lifetime. 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 47 (DCD 2007) and $621,000 per kilometer (km) for projected, we assume a 0.8 increase in construction costs paved roads, the latter of which represents the average due to a design standard update. cost per km of constructing a 2-lane collector road in rural areas (FDOT 2009a).20 The readily available data suggests several relationships will have no impact or minimal impact in these catego- The code update methodology that we employed for ries as follows: 1) no impact from wind on paved roads temperature effects is similar to the approach outlined or buildings, 2) no impact from temperature on trans- in Equation 1 for precipitation. Unlike the code mission poles, and finally 3) no impact from precipita- update approach for precipitation, we do not apply the tion on transmission poles. full 0.8 percent cost increase of a code update to each incremental change in temperature. Instead, we scale B. Example of Building Code Methodology the 0.8 percent value to reflect the portion of construc- tion costs likely to be associated with temperature Two examples are presented here to illustrate the appli- effects. Based on data published by Whitestone cation of the building code methodology to new Research (2008), we assume that 28 percent of the costs construction, a building example for precipitation and a associated with a code update for buildings are related paved road example for temperature. For the former, to HVAC equipment affected by temperature. assume that a new hospital is to be built in a location Similarly, research into the effects of temperature on that has a base precipitation level of 100 cm per year. It roads provides a guideline of 36 percent of the costs for is projected that due to climate change, the location will a code update for roads is temperature-related (Miradi have a 15 cm increase during the 40-year anticipated 2004). Based on these values, we assume a 0.22 percent lifespan of the building. Given the 10cm threshold for increase in building construction costs for each incre- a building code update, the design of the structure mental change in temperature and a 0.29 percent would anticipate the precipitation increase and the asso- increase in paved road construction costs for such ciated building code update. Essentially, the building changes. Based on professional judgment and the will be overbuilt for Year 0 to anticipate the need later design parameters for HVAC systems, which are typi- in the lifespan to accommodate the increased precipita- cally based on the number of degree days per year tion. The cost of this overbuild will be the cost of one (NOAA 2009), we assume that the 0.22 percent value is code update for the 10 cm increase, or 0.8 percent of applied to building costs for each 0.5 degree Celsius the base construction costs. increase in average annual temperature. For paved roads, we apply the 0.29 percent increase as a step func- In the context of temperature, consider the example of tion, with the first increase occurring after a 1 degree paved road construction. Using the 36 percent relative Celsius increase in temperature and later increases impact discussed above, the standard 0.8 percent cost occurring with each 3-degree increase in temperature. increase for a code update is modified by this percent- This reflects the need for new pavement binders with age resulting in a modified value of 0.29 percent of base every 3-degree increase in temperature and a change in construction costs. However, to apply this to new practice for a 1-degree change as an initial safety factor construction, the guidelines for pavement design are (Blacklidge Emulsions, Inc. 2009). brought into the equation. Specifically, temperature increases require new pavement binders every 3 degrees We also apply the building code methodology to trans- Celsius (Blacklidge Emulsions, Inc. 2009). Therefore, mission line towers, but instead of precipitation and for a new construction scenario where the maximum temperature, wind is the climate stressor of concern. temperature will increase 2 degrees over the 20-year For every 15 mile per hour (~24 km per hour) increase lifespan of the road, a cost increase of 0.29 percent of in the maximum of the maximum monthly wind speeds base construction costs is applied after the first 1 degree to account for an initial safety factor built into the design. Since the increase does not total an additional 3 20 Both of these base cost values represent the costs of construction in degrees, the total increase from the temperature impact the United States. We developed values specific to other countries based on an inter-country construction cost index published by is 0.29 percent of base construction costs. Compass International Consultants Inc. (2009), as indicated above. 48 t h e c os ts oF ad a p tin g to c limate c h an ge For in Fr as tr u c tu r e C. Direct Response Methodology most significant design changes associated with an increase in clearance would involve changes to bridge- For bridges and unpaved roads, we use a more direct deck support structures, which account for approxi- approach for estimating the cost impact of changes in mately 50 percent of bridge construction costs (Kinsella climate stressors. Under this approach, we directly and McGuire 2005). In addition, based on the standard relate changes in infrastructure construction costs to 16-foot clearance for bridges on highways (FHWA specific changes in climate or infrastructure design 2009), a one-foot increase in bridge clearance would requirements. In general terms, this approach is represent a 6.25 percent increase. Assuming that the summarized by Equation 2. increase in costs for bridge foundations would be proportional to the change in clearance, we assume that construction costs for the bridge support structures (2) CURBT = M × BURBT would increase by 6.25 percent with each 1-foot where CURB = change in construction costs for increase in clearance. Because support structures repre- bridges and unpaved roads associated with a unit sent approximately 50 percent of bridge construction change in climate stress or design requirements costs, we assume that the total construction costs for a bridge would increase by approximately 3.13 percent M = cost multiplier (50 percent x 6.25 percent) with each one-foot increase in clearance. BURB = base construction costs for bridges and unpaved roads The base cost of a bridge is likely to vary significantly due to differences in the number of lanes per bridge and Implementation of the approach represented by bridge length. For the purposes of this analysis, we use Equation 2 is somewhat different for unpaved roads the costs of a 2-lane bridge spanning 100 feet. than it is for bridges. For unpaved roads, we express the Assuming an average lane width of 12 feet, this trans- dose-response relationship represented by Equation 2 as lates to a bridge deck with an area of approximately the change in construction costs associated with a 1 2,400 square feet. Based on a unit cost of $220 per percent change in maximum monthly precipitation. square foot (FDOT 2009b), we estimate that the total Research findings have demonstrated that 80 percent of base construction costs for a bridge are approximately degradation of unpaved roads can be attributed to $528,000. Applying the 3.13 percent value derived precipitation (Ramos-Scharron and MacDonald 2007). above to this estimate, we assume an increase of The remaining 20 percent is attributed to factors such $16,500 in bridge construction costs for each one-foot as tonnage of traffic and traffic rates. Given this 80 increase in bridge clearance.21 percent attribution to precipitation, we assume that the base construction costs for unpaved roads increase by 80 The readily available data suggest no impact or minimal percent of the total percentage increase in maximum impact will originate from wind or temperature monthly precipitation; that is, a 0.8 percent increase in increases for new construction of bridges or unpaved costs for each 1 percent increase in maximum precipita- roads. tion. For example, if the maximum monthly precipita- tion increases by 10 percent in a given location, then 80 percent of that increase is used (8 percent) as the 2. Estimation of Dose-Response Values for increase in base construction costs. In addition, we Maintenance Costs further assume a base construction cost of $13,000 per Similar to our development of dose-response values for km for unpaved roads, based on published cost data infrastructure construction costs, we employed two basic (Cerlanek et al. 2006). The readily available data methodologies to generate dose-response values relating suggest no relationship between temperature and the cost of building unpaved roads. 21 This value is based on U.S. construction cost data. We developed val- For bridges, we estimate the climate-related change in ues specific to other countries based on an inter-country construction cost index published by Compass International Consultants Inc. (2009), costs per one-foot increase in bridge clearance. The as indicated above. 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 49 changes in climate stressors to changes in infrastructure change in climate stress, scaled for the stressor's effect maintenance costs. The first approach is based on on maintenance costs, as shown in Equation 4. infrastructure lifespan decrements that could potentially result from climate change if maintenance practices S remain unchanged following changes in climate stress. (4) LERB = (SMT ) We use this methodology to develop dose-response BaseS values for existing paved roads and buildings.22 Newly where LERB = Potential percent change in lifespan for constructed paved roads and buildings are assumed to existing paved roads and buildings associated with a not be affected by climate stressors because forward- unit change in climate stress looking design allows these structures to accommodate future climate changes at the time of construction. For S = Change in climate stress (i.e., precipitation or railroads and unpaved roads (both existing and newly temperature) constructed), we use a separate methodology similar to the direct dose-response approach outlined above for BaseS = Base level of climate stress with no climate bridge and unpaved road construction costs. change A. Avoided Lifespan Decrement Methodology SMT = Percent of existing paved road or building To assess the relationship between climate stressors and maintenance costs associated with a given climate stres- maintenance costs for existing paved roads and build- sor (i.e., precipitation or temperature) ings, we use an approach based on the cost of prevent- ing the reduction in lifespan that may result from As indicated in Equation 4, the potential change in changes in climate-related stress. As indicated by lifespan is dependent on the change in climate stress. Equation 3, implementation of this approach involves For precipitation effects, we assume a potential reduc- two basic steps: (1) estimating the lifespan decrement tion in lifespan for existing paved roads and buildings that would result from a unit change in climate stress for every 10 cm increase in annual rainfall. For temper- and (2) estimating the costs of avoiding this reduction ature, we assume a potential lifespan reduction with in lifespan. every 1 degree Celsius change in temperature (average annual temperature for existing buildings and maximum (3) M TERB = (LERB)(CERB) annual temperature for existing paved roads). where MTERB = Change in maintenance costs for Equation 4 also illustrates that our estimate of the existing paved roads and buildings associated with a potential reduction in lifespan associated with a given unit change in climate stress change in climate stress reflects the contribution of that stressor to baseline maintenance costs (i.e., variable LERB = Potential percent change in lifespan for exist- SMT). For buildings, we assume that changes in ing paved roads and buildings associated with a unit precipitation associated with climate change will affect change in climate stress roofing and external enclosures and changes in temper- ature will affect HVAC systems. Because roofing and external enclosures represent 15 percent of building CERB = Cost of preventing a given lifespan decrement maintenance costs (Whitestone Research 2008), we for existing paved roads and buildings assumed that precipitation contributes 15 percent to a building's maintenance costs. Similarly, because HVAC To estimate the reduction in lifespan that could result represents 28 percent of building maintenance costs from an incremental change in climate stress (LERB), (Whitestone Research 2008), we assume that tempera- we assume that such a reduction is equal to the percent ture effects are responsible for 28 percent of a building's maintenance costs. We also identified similar data for paved roads suggesting that precipitation-related main- 22 By existing roads and buildings, we mean those roads and buildings in service as of 2010, the first year in the time horizon of this analysis. tenance represents 4 percent of maintenance costs and 50 t h e c os ts oF ad a p tin g to c limate c h an ge For in Fr as tr u c tu r e that temperature-related maintenance represents 36 road construction costs. More specifically, we estimate percent (Miradi 2004). the change in railroad and unpaved road maintenance costs associated with a unit change in climate stress as a After estimating the potential reduction in lifespan fixed percentage of baseline construction costs (for rail- associated with a given climate stressor, we estimate the roads) or maintenance costs (for unpaved roads), as costs of avoiding this reduction in lifespan. To estimate illustrated by Exhibit 5. these costs, we assume that the change in maintenance costs would be approximately equal to the product of (5) MTURR = M × BURR (1) the potential percent reduction in lifespan (LERB) and (2) the base construction costs of the asset. where MTURR = Change in maintenance costs for Therefore, if we project a 10 percent potential reduction unpaved roads and railroads associated with a unit in lifespan, we estimate the change in maintenance costs change in climate stress as 10 percent of base construction costs. As indicated above, we estimate base construction costs of $185 per M = Cost multiplier square foot for buildings and $621,000 per km for paved roads in the U.S.23 BURR = Baseline maintenance (for unpaved roads) costs or construction costs (for railroads) B. Example of Avoided Lifespan Decrement Approach Similar to the direct response methodology for As an example of the avoided lifespan methodology, construction costs, implementation of this approach for consider a country with baseline annual precipitation of maintenance costs also varies by infrastructure type. 80 cm without climate change. For such a country, a For railroads, we express the relationship described by 10-cm increase in annual precipitation would represent Equation 5 as the change in maintenance costs associ- a 12.5 percent increase in precipitation. Because precip- ated with a 1 degree Celsius change in the maximum of itation accounts for approximately 15 percent of build- the maximum monthly temperature projections for an ing-related maintenance costs, we would assume a 1.9 area. Based on research on the effect of heat stress on percent potential reduction in building lifespan (12.5% rails and the associated costs, we estimate that for every x 15%). If baseline building construction costs in this 1 degree increase in maximum temperature, railroad country are approximately $175 per square foot, we maintenance costs increase by 0.14 percent of railroad would estimate an increase in maintenance costs of construction costs (DRPT 2008). Therefore, assuming approximately $3.30 per square foot for every 10 cm construction costs of approximately $404,000 per km increase in annual precipitation. If the country were to (in the U.S) (Railroad 2009; Vickers 1992), we estimate experience a 15-cm increase in annual precipitation, we that railroad maintenance costs would increase by $565 would still assume a $3.30 per square foot increase for every 1 degree increase in maximum temperature. because the 15-cm increase includes just one 10-cm incremental change. However, if we were to project a For unpaved roads, we express the dose-response rela- 21-cm increase, we would assume an increase of $6.60 tionship represented by Equation 5 as the change in per square foot. maintenance costs associated with a 1 percent change in maximum monthly precipitation. As indicated above, research has demonstrated that 80 percent of unpaved C. Direct Response Methodology road degradation can be attributed to precipitation, To estimate dose-response values for railroad and while the remaining 20 percent is due to traffic rates unpaved road maintenance costs, we follow an approach and other factors (Ramos-Scharron and MacDonald similar to that outlined above for bridge and unpaved 2007). Given this 80 percent attribution to precipita- tion, we assume that maintenance costs increase by 0.8 percent with every 1 percent increase in the maximum 23 As indicated above, we developed values specific to other countries of the maximum monthly precipitation values projected based on these U.S. values and an inter-country construction cost index published by Compass International Consultants Inc. (2009), as indicat- for any given year. Published data indicates that the ed above. 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 51 baseline cost of maintaining an unpaved road is approx- Washington, DC: Federal Emergency Management imately $930 per km (Cerlanek et al. 2006). Therefore, Agency. for every 1 percent increase in maximum temperature, we assume a maintenance cost increase of $7.45 per km. FHWA. 2009. "Right of Passage: Controversy Over Vertical Clearance on the Interstate System." Federal The readily available data suggest climate stressors will Highway Administration. http://www.fhwa.dot.gov/ have no impact or minimal impact in these categories as infrastructure/50vertical.cfm, viewed May 31, 2009. follows: 1) no impact from temperature on unpaved roads, 2) no impact from precipitation on railroads. Kinsella, Y., and F. McGuire. 2005. "Climate Change Impacts on the State Highway Network: A Moving Target." Proceedings of the NIZHT Conference, Christchurch, New Zealand, November 2005. referenceS Miradi, M. 2004. "Artificial neural network (ANN) Blacklidge Emulsions, Inc. 2009. "SHRP Performance models for prediction and analysis of ravelling severity Graded Asphalt Binders." http://www.blacklidgeemul- and material composition properties." In M. sions.com/shrp.htm, viewed May 31, 2009. Mohammadian, ed. CIMCA 2004. Gold Coast, Australia: CIMCA. Cerlanek , W. D., C.M Zeigler, and S.E. Torres. 2006. Maintenance of Paved and Unpaved Roads in Alachua NOAA. 2009. "Heating and Cooling Degree Day County. Alachua, FL: Department of Public Works. Data." NOAA Satellite and Image Information Service. http://www.ncdc.noaa.gov/oa/documentlibrary/ Compass International Consultants, Inc. 2009. Global hcs/hcs.html#51overview. Construction Costs Yearbook. Morrisville, PA: Compass International. Railroad. 2009. "Railroad Construction Costs." Henry County Library. http://tacnet.missouri.org/history/rail- DCD. 2007. "The DCD Cost Guide for Medical roads/rrcosts.html#1995, Clinton, MO, viewed May 31, Buildings." DCD Construction Magazine, March/April 2009. Issue. Ramos-Scharron, C. E., andL.H. MacDonald. 2007. DRPT. 2008. Heat Order Issues Technical Memorandum. "Runoff and suspended sediment yields from an Richmond, VA: Virginia Department of Rail and unpaved road segment, St. John, U.S. Virgin Public Transportation. Islands."Hydrological Processes 21(1): 35­50. FDOT. 2009a. "Generic Cost Per Mile Models." Vickers, R.A. 1992. Cost Effective Maintenance of Florida Department of Transportation. htp://ftp.dot. Railway Tracks. London: Institute of Civil Engineers. state.fl.us/LTS/CO/Estimates/CPM/summary.pdf, viewed May 31, 2009. Whitestone Research. 2008. Building Maintenance and Repair Cost Reference 2008-2009. Santa Barbara, CA: FDOT. 2009b. "Bridge Costs." Florida Department of Whitestone Research. Transportation. http://www.fhwa.dot.gov/ infrastructure/50vertical.cfm, viewed May 31, 2009. FEMA. 1998. "Promoting the Adoption and Enforcement of Seismic Building Codes." FEMA 313. 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