Climate Change, Agricultural Productivity and Poverty Juliano J. Assunc~ao y Department of Economics, PUC-Rio Flavia Chein z CEDEPLAR/FACE/UFMG Abstract This paper evaluates the impact of climate change on agricultural productivity and poverty. Cross-sectional variation in climate among Brazilian municipios is used to estimate a structural model where geographical attributes determine agricultural productivity and poverty. IPCC predictions based on 20 General Circulation Models (for 2030-2049) are used to simulate the impacts of climate change. Our estimates suggest that the global warming is expected to decrease the agricultural output per hectare in Brazil in 18 JEL Classication: R11; Q1; Q5; Q54; D12. Key words: climate change, poverty, agricultural productivity, Brazil \Warming of the climate system is unequivocal, as is now evident from observations of increases in global average air and ocean temperatures, widespread melting of snow and ice and rising global average sea level." (IPCC - Climate Change 2007, Fourth Assessment Report) 1 Introduction What are the economic consequences of climate change? Is it possible to estimate the long-term economic eects of global warming? What can we say about Latin American countries? This paper studies the impact of climate change on agricultural productivity and poverty in Brazil. This material is based upon work supported by a grant from the World Bank, as a part of the project "Climate Change in Latin America and the Caribbean: Impact and Policy Challenges Adaptation". The opinions, ndings, conclusions, and recommendations are those of the author(s) and not necessarily those of the the World Bank sta. y Address: Rua Marqu^es de S~ao Vicente, 225/F210, Rio de Janeiro, Brazil, 22453-900. Tel.: +55-21-3114-1078; fax: +55-21-3114-1084. E-mail: juliano@econ.puc-rio.br z Bolsista PDJ CNPq. Address: Av. Ant^onio Carlos, 6627, Belo Horizonte, Brazil, 31270-901. E-mail: fchein@cedeplar.ufmg.br
The economic literature on the consequences of climate change is organized in two strands: the production-function approach and the Ricardian approach. The rst one is the most traditional. The consequences of the climate change are estimated from the association between agricultural productivity and climate measures like temperature, rainfall or greenhouse gases levels. This association is specied as a production function. Mendelsohn et al (1994, 2004, 2004b, 2004c) argue that these studies tend to overestimate the climate change damage, since they ignore the capacity of adaptations that farmers can make in response to worse climatic conditions, such as introduction of new crops, migration and occupational mobility. On the other hand, the Ricardian approach assumes that land prices represent the expected present value of all net prots farmers can get from land (Mendelsohn, 1994; Seo and Mendelsohn, 2007b). Empirically, instead of studying yields of specic crops, the Ricardian approach examine how climate in dierent places aects the net rent or the value of farmland. The choice between the two approaches is not trivial. Although the Ricardian approach accounts for a broader range of possible direct and indirect eects of climate change, it is based on the assumption that the land price is determined by the expected present value of future agricultural streams. Especially for Latin American countries, land is used not only as an agricultural input but also as a source of other benets - \as a hedge against ination, as an asset that can be liquidated to smooth consumption in the face of risk, as collateral for access to loans, as a tax shelter, or as a means of laundering illicit funds" [De Janvry, Key and Sadoulet (1997)]. For Berry and Cline (1979), \in countries with poorly developed capital markets, especially those with chronic ination, landowners may nd it attractive to hold land for speculative gain - or merely to accomplish the store of value objective". Assunc~ao (2008) presents evidence compatible with the existence of a non-agricultural component of land demand, showing that land prices in Brazil increased substantially more than rental rates during periods of high macroeconomic instability. Another debate in the literature is the use of climate versus weather information. While most of the papers are based on long-run cross-section variation, Desch^enes and Greenstone (2007) argue that weather variation is orthogonal to unobserved determinants of agricultural prots, what might provide consistency gains. However, this strategy rely largely on temporary shocks and thus do not allow for adaptation mechanisms. Our approach lies somewhere in between the Ricardian and the production function approaches, focusing in cross-section climate variation rather than time-series weather variation. Although we explicitly specify a production function, we allow for adaptation in terms of labor mobility in our framework. The analysis is carried out with a cross-section sample of Brazilian municipios. Brazil is an interesting case-study for our purposes for many reasons. First, it has a large territory with substantial variation of agroclimatic conditions. Not only the long-run climate indicators are quite dierent across regions, but also the IPCC forecasts for climate change in the country vary accordingly. Second, the agriculture sector is very active and diversied. Third, an important fraction of the population is poor. Forth, we have municipio-level data to perform the study, which gives us almost 5,000 observations for the estimation. The paper is divided in two parts. In the rst part, we investigate the eect of climate change on agricultural productivity, with an explicit use of a production function. Input and crop choices are the only form 2
of adaptation available for households in this rst analysis. The econometric specication is derived from a structural model of agricultural production with heterogeneity in labor skills. We estimate the model and simulate it with the IPCC projections on temperature and rainfall. According to the IPCC, Brazil will experience an increase of 1.43 degrees Celsius in the average temperature and a reduction of 1.44 percent in rainfall in the period from 2030 to 2049. Our simulations suggest it will reduce the agricultural productivity in 18%, with substantial heterogeneity in the dierent parts of the territury. The impacts on the municipios range from -40% to almos +15%. The Southern region of the country is expected to gain while the Northern and Northeastern regions are expected to suer with the global warming. The second part of the paper studies the consequences in terms of poverty that arise from the variation in agricultural output which is related with climate change. Again, we use a structural model to guide the empirical analysis. We show that dierent forms of adaptation can be considered in the estimation procedure, through sample adjustments. Each model is estimated and simulated based on the IPCC forecasts. We start with a sample of rural households, assuming there is no labor mobility and that households have limited capabilities to mitigate the impacts of the climate change. For this sample, our estimates suggest the poverty is expected to increase by 3.2 percentage points with the global warming. We then allow for labor mobility considering poverty measures computed from two other samples: (i) total (urban and rural) poverty rate; (ii) migrationadjusted poverty rate. Our model interprets the use of the total poverty rate as a means of taking into account the reaction of households in terms of sector and occupational changes. For the case of the migration-adjusted poverty rate, the interpretation is that households are allowed to mitigate adverse eects of climate change through migration. In both cases, the model is particularly helpful in specifying the conditions under which wage eects from the labor mobility can be ignored. When we allow for labor mobility, the average impact of climate change on poverty reduces to 2.0 percentage points. Similarly to what happens with the agricultural productivity, our results are quite dierent along the territory. The Northern region is expected to increase poverty in 6.2 percentage points, while poverty is expected to reduce in 0.9 percentage point in the Southern region. The impacts at the municipio-level range from -3.9 to 7.2 percentage points. 2 Data We combine data from dierent sources. All variables and respective sources are depicted in table 1. [Table 1 - Data Description] The information about climate and soil types comes from the Climate Research Unit- University of East Anglia (CRU-UEA) and EMBRAPA, respectively. The information on temperature and precipitation comprises a historical average of the period from 1961 to 1990. The geographical coordinates (longitude, latitude and altitude) of each municipio are obtained from the Brazilian Census Bureau (IBGE). 3
The agricultural output and agricultural output per hectare are averaged for the years 1997 to 2006, considering data from the Annual Municipal Agricultural Survey (PAM) collected by the IBGE. We decided to consider averages in order to smooth out idiosyncratic shocks and thus characterize the agricultural productivity of each municipio in a longer term. For poverty, we consider three measures obtained from the Demographic Census of 2000 collected by the IBGE. The rst one is the proportion of individuals in rural areas that live in households with per capita income less than one dollar per day. The second poverty measure is similar, but for all individuals in each municipio (both rural and urban households). The third measure is a poverty rate adjusted for migration. To the population of each municipio, we add all individuals that migrate to other cities and exclude everyone who comes from other places in the previous 5 years. Finally, the future scenarios are comprised by temperature and rainfall forecasts computed by Intergovernmental Panel on Climate Change (IPCC), a scientic organization set up by the World Meteorological Organization (WMO) and by the United Nations Environment Programme (UNEP). The forecasts are based on 20 General Circulation Models (GCM) and cover the period from 2030 to 2049. The appendix describe all the models and the scenario considered. 1 Descriptive statistics of the data are reported in table 2. [Table 2 - Summary Statistics] The temperature in Brazilian municipios ranges from 14 C to 28 C in our sample. Figure 1 shows the kernel density estimation of temperature. We can observe a high concentration of temperatures around 25 C. The highest average temperature levels appear at municipios in the North and Northeast regions, the less developed regions in the country (Figure 2). [Figure 1 - Distribution of Temperature] [Figure 2 - Average Temperature across Brazilian Municipios] The rainfall levels, as shown in table 2, ranges approximately from 29 to 280 millimeters per month. Figure 3 depicts the distribution of the municipio average rainfall levels, and gure 4 shows its geographical distribution. The Northeast region presents a large number of municipios with low levels of precipitation. [Figure 3 - Distribution of Rainfall] [Figure 4 - Average Rainfall across Brazilian Municipios] 1 We are thankful to Michael Westphal for kindly provide us with these data. For more information about GCM models, access http://www-pcmdi.llnl.gov/ipcc/standard output.html. 4
Data from IPCC is depicted in gures 5 to 8. Figures 5 and 7 show that the distribution of the expected changes in both temperature and rainfall is approximately symmetric. Figures 6 and 8, on the other hand, shows many clusters with very dierent forecasts. Temperature is expected to increase more intensely in the North and Central-West regions. Rainfall predictions suggest an increase in precipitation in the South region and Amaz^onia, and a reduction in precipitation in the Central and Northeast regions. [Figure 5 - Distribution of Predicted Changes in Temperature] [Figure 6 - Predicted Changes in Temperature across Brazilian Municipios] [Figure 7 - Distribution of Predicted Changes in Rainfall] [Figure 8 - Predicted Changes in Rainfall across Brazilian Municipios] 3 Climate Change and Agricultural Productivity 3.1 Theoretical Framework Consider the following Cobb-Douglas aggregate production function for each municipio i: Y i = A i T 1 i K i L i exp (" i ) ; i = 1; :::; M; (1) where Y i represents the total output which price is normalized to 1; T i is the available area; K i and L i represent the amount of nonlabor and labor input; A i is a technological factor; and " i is an error term accounting for idiosyncratic determinants of the output such as climatic shocks. We also assume that labor is heterogeneous - agricultural workers have dierent skills. A worker of type 2 [0; 1] has productivity represented by (), where 0 > 0 and (0) = 1. Then, the total labor input employed in municipio i is given by: L i = Z 1 0 i () L i () d; where L i () is the number of workers of type employed in the production. Consider now a competitive environment with no externality. For any arbitrary plot size T i, we assume farmers in municipio i maximize the expected prot given the observed climate conditions C i : max E A i T 1 i K i ;L i K i Z 1 0 Z! 1 i () L i () d exp (" i ) r i K i w i () L i () djc i : 0 (2) 5
The rst-order conditions for K i E E Y i L i C i Y i Ki! and L i (), 2 [0; 1] are given by:! C i = r i ; (3) i () = w i () ; for all 2 [0; 1] : (4) The system (3)-(4) shows that the expected marginal revenue is equal to the marginal cost of each input. Moreover, equation (4) implies that w i() = w i( 0 ) for all, i () i ( 0 ) 0 2 [0; 1]. Thus, we have that w i () = i () w i (0) : (5) Equation (5) shows that the wage schedule is completely determined by the baseline wage w i (0) and the productivity function i (), because workers of dierent abilities are substitutes. The optimal amount of nonlabor and labor inputs is given by: K i = T i A i L i = Z 1 0 r 1 i w i (0) 1 E(exp(" i )jc i ) i () L i () d = T i! 1 1 ; (6)! 1 A i ri w i (0) 1 1 1 E(exp(" i )jc i ) : (7) Finally, the agricultural output for the municipio with land endowment of T i is given by:! 1 Yi A i = T i ri w i (0) 1 E(exp("i )jc i ) + 1 exp ("i ) : (8) In addition to the agricultural sector, we also assume there is a subsistence activity in the economy which provides a very low income w, which is typically below the poverty and indigence levels of income. This subsistence activity does not require any special skill and, therefore, is available to everyone in the economy. As a consequence, it establishes a lower bound for the baseline wage rate, i.e., w i (0) w. This structure is commonly used in the occupational choice literature. 2 We now adopt two assumptions to drive our empirical analysis. Assumption A1: There is perfect capital mobility across municipios: r i i = 1; :::; M. = r for all Assumption A2: The total factor productivity is determined by a constant A and an idiosyncratic term i across municipios which is independent from C i : A i = A exp ( i ). Assumption A3: The number of unskilled workers (those with type = 0) is large enough to assure that w i (0) = w for all i = 1; :::; M. Assumptions A1 and A2 determines that all municipios face the same capital return and do not present systematic technological dierences. Assumption A3 captures the fact 2 See, for instance, Banerjee and Newman (1993) and Galor and Zeira (1993). 6
that most of the Brazilian municipios have a signicant fraction of the population under poverty or even indigence conditions. Assumption A3 states that the subsistence activity yields the same income in all municipios. 3 This assumption is the same adopted in Jeong and Townsend (2003, 2007). As a result, it means that migration (and its eect to the labor market) does not aect our analysis of agricultural productivity in our case. As shown by equation (5), the whole wage schedule in this case is completely determined by the productivity parameters i () and the subsistence wage w. Under assumptions A1-A3, the direct eect of climate change on the agricultural output per hectare can be obtained by the following regression: ln Yi T i = 0 + (C i ) + u i ; (9) 1 where 0 ln A, (C 1 r w i ) + ln (E(exp(" 1 i)jc i )) and u i = " i + i. Specication (9) captures all the direct eects of climate on agricultural productivity. Our dependent variable in this rst exercise is the aggregate agricultural output per hectare for each municipio. Thus, we can interpret crop choice as embedded in the input choice. On the other hand, this analysis with assumptions A1-A3 ignores the eects of climate change that work through prices and technology adaptation. 3.2 Empirical Characterization We now estimate (9) using cross-section variation from the Brazilian municipios. The vector C i contains the available and observed variables regarding climate conditions for municipio i - temperature, rainfall and soil types. For the case of temperature and precipitation, we consider a quadratic specication following Mendelsohn et al. (1994). The soil information is introduced as a set of dummy variables indicating each one of the predominant soil types in each municipio. In the estimation of the eect of climate change on agricultural productivity, through equation (9), it is important to notice that we only have simulated data for temperature and precipitation. We do not have forecasts for soil types, athough it might be aected, in principle, by climate change. On the one hand, if we introduce it into the regression and keep it constant for the simulation, we may underestimate the eect of climate change on agricultural productivity. On the other hand, part of the soil conditions are not aected by climate and, therefore, it can improve our estimates. Thus, we report our main results with and without conditioning on soil conditions. Results are reported in table 3. A comparison of the marginal eects of temperature and rainfall are presented in gures 9 and 10, respectively. [Table 3 - Eect of Climate on Agricultural Output per Hectare] 3 Actually, the variation of the average income of poors across municipios is substantially lower than the variation of the per capita income. The percentile 10 and 90 for the average income of poors are R$53 and R$61, respectively, while the same indicators for per capita income are R$199 and R$640, respectively. 7
[Figure 9 - Marginal eect of temperature] [Figure 10 - Marginal eect of rainfall] As expected, the importance of temperature and rainfall decreases when we include soil dummies (in column 2) and geographical locations (in column 3). The largest change is observed with the latitude and longitude coordinates. The dierences between the marginal eects with and without soil types are comparatively less important, especially if we restrict our attention to the relevant values - more than 20 degrees Celsius (gure 1) and less than 150 mm per month (gure 3). Hereafter, we take the specication with soil types (column 2) as our preferred equation for agricultural productivity. 3.3 Simulations The impact of climate change on agricultural productivity is estimated through equation (9). We now consider the vector of climate information for each municipio, C i, as containing information on temperature, rainfall and soil types. Then, we use the projection of IPCC for temperature and rainfall for the period of 2030 to 2049 to build a dierent climate vector for each municipio, ^Ci. The expected agricultural output per hectare in each municipio, under the new climate conditions, is given by: E ln Yi T i = ^ 0 + ^ ^Ci ; where ^ 0 and ^ () corresponds to the estimates reported in the column 2 from table 3. From the predicted agricultural output per hectare, we can obtain the percent change in agricultural productivity given by the climate change in each municipio. A map with the impact for all Brazilian municipios is depicted in gure 11, while the results aggregated for Brazilian states, macro regions and the whole country are presented in table 4. [Table 4 - Simulated Eects of Climate Change on Agricultural Output per Hectare] [Figure 11 - Eects of Climate Change on Agricultural Output] Considering the whole country, the increase of 6.57% in the average temperature along with the decrease of 0.71% in rainfall lead to a reduction of 18% in the agricultural productivity. We can also see that the impact is very heterogenous in the country. The state-level impacts range from -37% in Rond^onia State to 1% in Santa Catarina. The impact on the North region is substantially higher than the impact on the South region. An underlying assumption of the exercise above is absence of technological responses to the new climate conditions. The results are predicted using the imputation of climate forecasts on current agricultural technological status. However, the cross-section comparisons across municipios or states reveal how intense should be technological gains in order to mitigate the impact of climate change in each municipio or state. 8
4 Eects on Poverty While the previous section investigates the eect of climate change on agricultural productivity, we analyze its consequence in terms of poverty in this section. Changes in agricultural productivity aect income from rural households and, consequently, rural poverty. We rst investigate this eect assuming there is no labor mobility. Agents are only allowed to adapt to dierent environmental conditions through input choices. This is a polar case which is considered as a benchmark and is also the worse scenario for the estimated impact. However, rural households that work on agricultural sector can mitigate the consequences of the climate change adapting at least in two dimensions: (i) moving out agriculture towards more protable sectors and occupations (probably in the urban area); or (ii) migrating to more amenable municipios. Our analysis then takes into account those types of adaptation mechanisms, which are taken broadly as related to labor mobility - across sectors or municipios. 4.1 Economy without Labor Mobility 4.1.1 Theoretical Framework In the case without labor mobility we consider a purely agricultural economy (only one sector) without migration. Each municipio i has an endowment of T i hectares of land and produces Y i which is distributed to the N i individuals in the form of agricultural prots, wages and rents. The individual appropriation of the total income depends upon the individual productive skill and the (broadly speaking) individual political/economical power. Each individual is characterized by random types (; ), where the parameter represents the productive skill-level, which is independently and uniformly distributed in the [0; 1] interval, while the parameter captures the political/economical power, which is distributed among N i individuals according to the distribution function F i. Each individual n in municipio i is associated with a pair of types ( n;i ; n;i ). We assume the parameter is drawn before parameter. Thus, given the vector of skills in municipio i, [ n;i ] N i n=1, the political power types are drawn from F i which assures the income shares! ( n;i ; n;i ) are such that N i X n=1 Given the individual type ( n;i ; n;i ), the income is given by! ( n;i ; n;i ) = 1: (10) y n;i =! ( n;i ; n;i ) Y i : (11) An individual is considered poor in this economy if the income is below a threshold representing the (national) poverty line y. Thus, given the income vector [y n;i ] N i n=1 and the 9
agricultural output Y i, the poverty rate P i in municipio i is given by: P i j [y n;i ] N i n=1 = 1 N i N i X n=1 1 [y n;i y] ; where 8 >< 1; if y n;i ( n;i ) y; 1 [y n;i ( n;i ) y] = >: 0; otherwise. The expected poverty rate of municipio i, conditional on the aggregate agricultural output can be obtained as following: P i jy i = 1 N i N i X n=1 = 1 XN i N i n=1 Z 1 = 1 XN i N i n=1 0 Z = 1 XN i N i Z = [0;1] n=1 Z Pr [y n;i y] Pr! ( n;i ; n;i ) y Yi [0;1] Pr! (; n;i ) y j d Y i Z!(;) y Y i j!(;) y Y i df i () d df i () d P (y; Y i ; F i ) : (12) Notice that @ P (y; Y @ y i; F i ) > 0 and @ @Y i P (y; Y i ; F i ) < 0. Equation (12) suggests that, given a poverty line, the poverty rate is determined by the aggregate agricultural output and the distribution function F i which, ultimately, indicates how the aggregate production is allocated within each municipio. In order to assure that F i is properly dened, it depends on the population size of each municipio N i. Assumption A4: The political/economical distribution is characterized by only one (non-observed) parameter i (besides the population N i ). Under A4, we approximate the eect of agricultural output on poverty as following: P i = 0 + 1 ln (Y i ) + 2 ln (N i ) + v i : (13) 4.1.2 Empirical Characterization Table 5 presents the estimation of the relationship between agricultural output and poverty determined by equation (13). In panel A, the OLS estimate of 1 suggests that if a doubled agricultural output would be associated with a 7.4 percentage points of reduction in the poverty rate. [Table 5 - Agricultural Output and Poverty] 10
However, it is important to emphasize that the consistent estimation of (13) requires that Y i and v i are not correlated. Considering the way Y i is determined, it means that the error terms in (9) and (13) should not be correlated. This assumption is not true, for instance, if municipios with higher total productivity factor have systematically better (positive correlation) or worse (negative correlation) distribution of political/economical power. In order to account for this possibility, we use C i as an instrument for Y i in the estimation of 1. In this case, we only consider the variation of the agricultural output that comes from dierences in climate conditions in the estimation procedure. Panel B presents the instrumental variable estimation of (13), in which we use temperature, rainfall and soil dummies as instruments for agricultural output. The estimate of 1 has increased from -7.4 percentage points, in the OLS, to -12.8 percentage points, in the instrumental variables estimation. The comparison between the two estimates suggests that the v i and u i are positively correlated. Non-observed factors determine simultaneously higher agricultural output and less poverty. This eect was producing a downward bias in the eect of agricultural output on poverty. 4.1.3 Simulations Table 6 presents the simulated impacts on poverty. For each municipio, we have considered the new vector of predicted climate conditions for the 2030-2049 along with the instrumental variables estimates reported in table 5 to simulate the poverty in each municipio. We then averaged the predicted eects for states, macro regions and the whole country. [Table 6 - Simulated Eects of Climate Change on Rural Poverty] According to the estimates reported in table 6, climate change is expected to increase in 3.2 percentage point the poverty of rural areas in Brazil. Considering that the baseline expected poverty rate of 40, it represents an increase of 8% of poor in Brazilian rural areas. More interesting, however, is the geographical variation of the impact. While the North region is the more aected area in absolute terms, with an increase of 6.2 percentage points in the rural poverty rate, the South region, on the other hand, is beneted with a reduction of 0.9 percentage point in the poverty. The state-level eects range from -1.3 percentage point in Santa Catarina state (in the South region) to 7.2 percentage point in Amazonas state (in the North region). In relative terms, however, the impact is stronger in the Central-West region. The change of 4.5 percentage points for the whole region represents an increase of 16% in rural poverty. Goias, which is also in the Central-West region, is the state with the highest relative impact - 18% increase in rural poverty. 4.2 Economy with Labor Mobility The analysis in the previous section considers an economy in which individuals cannot adapt to climate change. However, there are at least two ways of mitigating the adverse eects of the new economic environment: migration and sector/occupational change. In this section, we compute the impacts of climate change on poverty, allowing adaptation 11
to the new climate conguration. 4.2.1 Theoretical Framework We extend our analysis in order to incorporate labor mobility, either across municipios or across sectors. We consider a very simple structure where agents choose among two alternative income sources. The agricultural income source has the same structure of the economy described above, providing income y A ( n;i ; n;i ) =! ( n;i ; n;i ) Y i. The alternative activity available for the individual of type n;i yields yi B ( n;i ). This alternative activity can be interpreted as the possibility of living in another municipio, or a dierent occupation. The nal income of a type ( n;i ; n;i ) individual is: 8 >< y A ( n;i ; n;i ) ; y A ( n;i ; n;i ) yi B ( n;i ) ; y ( n;i ; n;i ) = >: yi B ( n;i ) ; otherwise. (14) Equation (14) shows that labor mobility can occur dierently among individuals of different types. The expected poverty rate of municipio i conditional on the aggregate agricultural output, in this case, is: P i jy i = 1 N i N i X n=1 Z 1 = 1 XN i N i n=1 0 Z = 1 XN i N i Z = Pr [y ( n;i ; n;i ) y] n=1 [0;1] " Z [0;1] Pr [y (; n;i ) yj] d Z P y; Y i ; F i ; y B i y(;)y y B i ()!(;)Y iy df i () d Z # df i () + df i () d!(;)y i yi B()y : (15) The alternative activity oers a pathway out of poverty for the agent of type ( n;i ; n;i ) whenever y A i ( n;i ; n;i ) y y B i ( n;i ). Therefore, the incorporation of this adaptation mechanism in our analysis can reduce the impact of climate change on poverty rate in: where ^C i is the simulated climate conditions. Pr n y A i ( n;i ; n;i ) y y B i ( n;i ) j ^C i o ; (16) 4.2.2 Empirical Characterization The empirical analysis of section 4.1 still holds under labor mobility in theory. We can approximate (15) as in equation (13). However, for the empirical application, we need to adjust the sample used in the estimation to account for those who escaped from poverty y A i ( n;i ; n;i ) y y B i ( n;i ). Since we observe individuals after their choices (ex post), it is possible to rebuild the poverty measure including the individuals who have changed their income sources. This sample adjustment is made in two ways. 12
First, we consider a measure of total poverty considering all residents in each municipios, incorporating all urban households. This alternative measure of poverty captures the fact that some individuals might adapt to the new climate conditions changing sectors or occupations. Second, we build a migration-adjusted poverty measure. For each municipio, we consider a sample comprised of the non-migrant households and those who out-migrate to other municipios - migrants from other municipios are also excluded. We then compute a poverty measure using this adjusted sample for each municipio, for both urban and rural areas. Results are presented in table 7. Panel A presents the OLS estimates while panel B depicts the instrumental variables for the two alternative poverty measures. The comparison between the OLS and the instrumental variable regressions reveals the same pattern we get in table 5, suggesting again that there is a positive correlation between the error terms in (9) and (13). [Table 7 - Agricultural Output and Poverty] Comparing the results obtained using the three measures, we obtain a pattern which is compatible with the theory. The coecient of agricultural output is -0.128 for the case of rural poverty, -0.126 for the total (urban and rural) poverty and -0.123 for the migrationadjusted measure of poverty. As expected, when we go from the rst to the latter, we allow households to adjust in more dimensions. So, it is reasonable to get less intense eects of agricultural output on poverty. 4.2.3 Simulations The simulated eects of climate change on poverty for the case with labor mobility are presented in table 8. As in the other cases, simulations were computed considering the new vector of climate variables for each municipio that comes from IPCC for the period 2020-2039. The poverty impacts for each municipio are obtained from the models estimated in the previous section and reported in table 7. Based on the predicted eects for each municipio, we aggregate the data for states, macro regions and the whole country. [Table 8 - Simulated Eects of Climate Change on Poverty] In absolute terms, the poverty variation induced by climate change is much less intense when we allow households to adjust both in terms of occupations and sectors as well as in terms of migration. The absolute impact reduces from 3.2 percentage points, for the case without labor mobility, to 2 percentage points, when we allow for labor mobility across sectors or municipios. We also observe the same pattern of mitigation for macro regions and states. It seems that the ability of changing sectors and occupations is what really matters. Although there is a substantial change between the estimates obtained from rural poverty and total (urban and rural) poverty indices, there is no important variation when we allow for the more general case based on the migration-adjusted poverty measure. In addition, results show substantial geographical variation. The impact on the North region is the highest while a reduction on poverty is expected in the South region. A better account of the geographical heterogeneity of the impacts is shown by gure 12, where we 13
plot the eects based on the migration-adjusted poverty measure for each municipio. [Figure 12 - Eects of Climate Change on Poverty] 5 Conclusion The paper investigates the impact of climate change on agricultural productivity and poverty in Brazil. We present a structural model to guide the empirical analysis. The model is estimated and simulated based on IPCC predictions about temperature and rainfall. Results suggest that not only the global warming is expected to generate signicant eects in the country, but also the impact is heterogeneous in the territory. Although the average eect is adverse, there are gainers and loosers in the process. We also show that adaptation is an important issue. The impact on poverty for the whole country reduces from 3.2 percentage points to 2.0 percentage points when we allow for labor mobility. References [1] Adams, Richard; Cynthia Rosenzweig; Robert Pearl. (1990). \Global climate change and U.S. agriculture", Nature, 345 (6272): 219-224. [2] Adams, Richard. \Global climate change and Agriculture: an economic perspective", American Journal of Agricultural Economics, 71 (5): 1272-1279. [3] Assunc~ao, J. J. (2008). \Rural Organization and Land Reform in Brazil: The Role of Nonagricultural Benets of Landholding", Economic Development and Cultural Change, 56 (4): 851-870. [4] Banerjee, A. V. and A. Newman. (1993). \Occupational Choice and the Process of Development", Journal of Political Economy, 101: 274-298. [5] Berry, R. A. and W. R. Cline (1979). Agrarian structure and productivity in developing countries, Baltimore: The Johns Hopkins University Press. [6] Callaway, John; F. Cronin; J. Currie; J. Tawil (1982). \An analysis of methods and models of assessing the direct and indirect economic impacts of CO 2 -induced environmental changes in the agricultural sector of the U.S. economy", Pacic Northwest Laboratory Working Paper, 4384. [7] Darwin, Roy (1999). \The impact of global warming on agriculture: a ricardian analysis: comment", The American Economic Review, 89 (4): 1049-1052. [8] De Janvry, A.; N. Key and E. Sadoulet. (1997). \Agricultural and Rural Development Policy in Latin America: new directions and new challenges", FAO Agricultural Policy and Economic Development Series - 2. [9] Desch^enes, Olivier; Michael Greenstone. (2007). \The economic impact of climate change: evidence from agricultural prots and random uctuations in weather", The American Economic Review, 97 (1): 354-385. [10] Galor, O. and J. Zeira (1993). \Income Distribution and Macroeconomics", Review of Economic Studies, 60: 35-52. 14
[11] Kurukulasuriya, Pradeep; Robert Mendelsohn; Rashid Hassan; James Benhin. (2006). \Will African agriculture survive climate change?", The World Bank Economic Review, 20 (3): 367-388. [12] Jeong, H. and Robert Townsend. (2007). \Sources of TFP growth: occupational choice and nancial deepening," Economic Theory, 32(1): 179-221. [13] Jeong, H. e R. M. Townsend (2003). \Growth and Inequality: Model Evaluation Based on an Estimation-Calibration Strategy", IEPR Working Papers, Institute of Economic Policy Research (IEPR). [14] Mendelsohn, Robert; Alan Basist; Ariel Dinar; Pradeep Kurukulasuriya; Claude Williams (2004). \What explains agricultural performance: climate normals or climate variance". mimeo. [15] Mendelsohn, Robert; Ariel Dinar; Alan Basist; Pradeep Kurukulasuriya. (2004b). \Crosssectional analysis of climate change impacts", World Bank Policy Research Working Paper 3350. [16] Mendelsohn, Robert; Ariel Dinar; Apurva Sanghi (2001). \The eect of development on the climate sensitivity of agriculture", Environment and Development Economics, 6: 85-101. [17] Mendelsohn, Robert; Alan Basist; Pradeep Kurukulasuriya; Ariel Dinar (2004c). \Climate and Rural Income". mimeo. [18] Mendelsohn, Robert; William Nordhaus; Daigee Shaw (1994) \The impact of global warming on agriculture: a ricardian approach", The American Economic Review, 84 (4): 753-771. [19] Mendelsohn, Robert; William Nordhaus (1999). \The impact of global warming on agriculture: a ricardian analysis: replay", The American Economic Review, 89 (4): 1046-1048. [20] Mendelsohn, Robert; William Nordhaus (1999b). \The impact of global warming on agriculture: a ricardian analysis: replay", The American Economic Review, 89 (4): 1053-1055. [21] Quiggin, John; John K. Horowitz (1999). \The impact of global warming on agriculture: a ricardian analysis: comment", The American Economic Review, 89 (4): 1044-1045. [22] Seo, Niggol; Robert Mendelsohn (2007). \An analysis of crop choice: adapting to climate change in Latin American farms", World Bank Policy Research Working Paper 4162. [23] Seo, Niggol; Robert Mendelsohn (2007b). \A Ricardian analysis of the impact of climate change on Latin American farms", World Bank Policy Research Working Paper 4163. 15
Appendix: GCMs used in the analysis Model ID, country, year Atmosphere, resolution Ocean, resolution Temperature (annual and seasonal) Precipitation (annual and seasonal) BCCR-BCM2.0, Norway, 2005 T63 (~1.9 1.9 ) L31 0.5-1.5 1.5 L35 X CCSM3, USA, 2005 T85 (~1.4 1.4 ) L26 0.3-1 1 L40 X CGCM3.1(T47), Canada, 2005 T47 (~2.8 2.8 ) L31 1.9 1.9 L29 X X CNRM-CM3, France, 2004 T63 (~1.9 1.9 ) L45 0.5-2 2 L31 X CSIRO-Mk3.0, Australia, 2001 T63 (~1.9 1.9 ) L18 0.8 1.9 L31 X X ECHAM5/MPI-OM, Germany, 2005 T63 (~1.9 1.9 ) L31 1.5 1.5 L40 X X ECHO-G, Germany/Korea, 1999 T30 (~3.9 3.9 ) L19 0.5-2.8 2.8 L20 X GFDL-CM2.0, USA, 2005 2.0 2.5 L24 0.3-1.0 1.0 L20 X X GFDL-CM2.1, USA, 2005 2.0 2.5 L24 0.3-1.0 1.0 L20 X GISS-AOM, USA, 2004 3.0 4.0 L12 3.0 4.0 L16 X X GISS-EH, USA, 2004 4.0 5.0 L20 2.0 2.0 L16 X GISS-ER, USA, 2004 4.0 5.0 L20 4.0 5.0 L13 X INM-CM3.0, Russia, 2004 4 5 L21 2 2.5 L33 X IPSL-CM4, France, 2004 2.5 3.75 L19 1-2 2 L31 X MIROC3.2 (medres), Japan, 2004 T42 (~2.8 2.8 ) L20 0.5-1.4 1.4 L44 X X MIROC3.2 (hirres), Japan, 2004 T106 (~1.1 1.1 ) L56 0.2-1.4 0.3 L47 X X MRI-CGCM2.3.2, Japan, 2003 T42 (~2.8 2.8 ) L30 0.5-2.0 2.5 L23 X PCM, USA, 1998 T42 (~2.8 2.8 ) L26 0.5-0.7 1.1 L40 X X UKMO-HadCM3, UK, 1997 2.5 3.8 L19 1.5 1.5 L20 X Emission Scenario: A1B Spatial Extent: Global Present Time Period: 1980-1999 Future Time Period: 2030-2049 Methodology: All of the GCM outputs as netcdf files were brought into ArcGIS 9.2 as point coverages, due to the spatially-variable grain cell sizes of many of the GCM outputs. The scripting language, Python 2.4.1, was used to select time ranges, calculate range means for each model and automate the subsequent process. The point coverages were converted to grids, with the resolutions specified above for each model, except that square grid cell resolutions were used. (The dimension was the minimum if the original resolution was rectangular.). The grids were interpolated using kriging to fill in any missing values globally. All grids were resampled to a 2º resolution. The ensemble means (temperature) and medians (precipitation) were then calculated.
Table 1: Description of the Data Variable Definition Source Geography Rainfall Estimates of the average quantity of water precipitation in Climate Research Unit-University each municipality for the period of 1961-1990, expressed of East Anglia (CRU-UEA) in millimeters per month Temperature Estimates of the average temperature in each municipality for the period of 1961-1990, expressed in Celsius grade Climate Research Unit-University of East Anglia (CRU-UEA) Soil A set of dummy variables indicating the kind of soil. EMBRAPA Agricultural Output per Ha Productivity Mean of total value of agricultural production per planted area from 1997 to 2006 Annual Survey of Municipal Agricultural Production (PAM) IBGE Agricultural Output per Worker Mean of total value of agricultural production from 1997 to 2006 per total number or worker in agricultural sector in year 2000. Annual Survey of Municipal Agricultural Production (PAM) IBGE; and Demographic Census, 2000 -IBGE] Average Agricultural Wages Mean of wages paid in agricultural sector in year 2000. Demographic Census, 2000 IBGE Poverty Rural Poverty (Less than US$ Total of poor (those who earn less than US$ 1 dollar) in Demographic Census, 2000 IBGE 1 dollar per day) rural areas/total of rural population Poverty (Less than US$ 1 Total of poor (those who earn less than US$ 1 Demographic Census, 2000 IBGE dollar per day) dollar)/total of population Poverty Adjusted for Total of poor (those who earn less than US$ 1 Demographic Census, 2000 IBGE Migration (Less than US$ 1 dollar per day) dollar)/total of population, considering only the nonmigrants and the out-migrants GCM Forecast INM-CM3.0, Russia, 2004 Forecast of temperature change for the period between 2030 2049, based on 20 General Circulation Models, expressed in Celsius Grade. IPCC (WMO/UNEP) Average Change in Rainfall Forecast of average percent changes for the period IPCC (WMO/UNEP) between 2030 2049, in Rainfall based on 20 General Circulation Models
Table 2: Basic Statistics Mean Sd Min Max Temperature (Celsius Grade) 22,73 2,99 14 28,04 Rainfall (mm/month) 116,11 36,63 28,87 282,43 Average Temperature Change Forecast (Celsius Grade) 1,43 0,23 0,92 1,95 Average Rainfall Change Forecast (%) -1,44 3,15-8,24 5,47 Rural Poverty (Less than US$ 1 dollar per day) 0,32 0,22 0 0,88 Poverty (Less than US$ 1 dollar per day) 0,26 0,19 0 0,8 Poverty Adjusted for Migration (Less than US$ 1 dollar per day) 0,25 0,18 0 0,8 Agricultural Output per Ha (R$ 2006) 3,07 2,62 0 37,16 Agricultural Output per Worker (R$ 2006) 1,88 4 0 55,31 Average Agricultural Wages (R$ 2000) 238,12 215,38 0 4083,21 INM-CM3.0, Russia, 2004
Table 3 - Climate and Agricultural Output per Hectare Dependent Variable: Agricultural Output per Hectare (in log) (1) (2) Average Temperature 0.640*** 0.506*** (0,056) (0,059) Average Temperature Squared -0.016*** -0.014*** (0,001) (0,001) Average Precipitation 0.019*** 0.015*** (0,001) (0,002) Average Precipitation Squared -0.000*** -0.000*** (0,000) (0,000) Soil Dummies No Yes Constant -7.091*** -4.978*** (0,640) (0,699) Observations 4948 4948 INM-CM3.0, Russia, 2004 0,32 0,42 Note: Standard errors in parentheses. * significant at 10%; ** significant at 5%; *** significant at 1%
Table 4 - Simulated Effects of Climate Change on Agricultural Output per Hectare Rainfall Change Temperature Change Agricultural Output per Hectare Forecast Agricultural Output per Hectare Percent Change Brazilian States Acre -0,22% 7,09% 1.713 1.131-34,3% Alagoas -6,48% 4,98% 1.475 1.116-24,9% Amapá -0,78% 5,52% 1.611 1.146-28,9% Amazonas 2,22% 6,73% 1.643 1.056-36,0% Bahia -6,30% 5,94% 1.444 1.076-25,6% Ceará -1,82% 5,65% 724 507-30,6% Distrito Federal -3,59% 8,16% 2.137 1.723-19,4% Espírito Santo -4,12% 5,83% 2.143 1.730-18,6% Goiás -1,91% 7,63% 1.666 1.222-26,9% Maranhão -2,03% 5,90% 1.054 703-33,4% INM-CM3.0, Russia, 2-2,12% 7,59% 1.639 1.123-31,8% Mato Grosso do Sul 0,35% 6,95% 1.652 1.277-23,0% Minas Gerais -1,42% 7,27% 2.027 1.718-16,6% Paraiba -4,27% 5,43% 1.140 856-25,4% Paraná 1,87% 6,87% 1.886 1.781-6,5% Pará -1,88% 6,39% 1.552 1.031-33,7% Pernambuco -6,15% 5,54% 1.197 914-24,8% Piauí -2,39% 6,04% 777 515-33,9% Rio Grande do Norte -2,83% 5,01% 735 532-28,3% Rio Grande do Sul 3,23% 5,70% 2.247 2.187-2,7% Rio de Janeiro -0,85% 5,13% 3.022 2.592-14,1% Rondônia -1,46% 7,44% 1.491 938-37,1% Roraima 0,47% 7,02% 1.642 1.075-34,9% Santa Catarina 2,94% 6,43% 2.210 2.230 1,0% Sergipe -6,26% 4,81% 1.008 752-25,3% São Paulo 0,05% 6,76% 1.938 1.625-16,9% Tocantins -2,82% 6,69% 1.122 718-36,2% Macro Regions North -1,49% 6,71% 1.486 969-35,0% Northeast -4,65% 5,70% 1.162 851-27,7% Central-West -1,54% 7,47% 1.655 1.192-28,3% Southeast -0,77% 6,86% 2.006 1.686-16,8% South 2,54% 6,33% 2.069 1.995-4,2% Brazil -0,71% 6,57% 1.752 1.475-18,2%
Table 5: Agricultural Output and Poverty Rural Poverty Panel A: Ordinary Least Square Agricultural Output (in log) -0.074*** (0,002) Population (in log) 0.117*** (0,003) Constant 0,001 (0,024) R-squared 0,41 Observations 5441 Panel B: Instrumental Variables Agricultural Output (in log) -0.128*** (0,003) Population (in log) 0.141*** (0,003) Constant 0.268*** (0,027) R-squared 0,26 Observations 4913 First Stage (Dep. Variable: agricultural output (in log)) Average Temperature 0.873*** (0,113) Average Temperature Squared -0.021*** (0,003) Average Precipitation 0.029*** (0,003) Average Precipitation Squared -0.000*** 0,000 Soil Dummies Yes Constant -2.529* (1,299) R-squared 0,29 Observations 4913 Note: Standard errors in parentheses. * significant at 10%; ** significant at 5%; *** significant at 1%.
Table 6 - Simulated Effects of Climate Change on Rural Poverty Rural Poverty Brazilian States Predicted Current Poverty (%) Simulated Poverty (%) Change (p. p.) Acre 40 46 5,9 Alagoas 44 49 4,2 Amapá 40 45 5,0 Amazonas 44 51 7,2 Bahia 44 48 3,9 Ceará 54 59 5,4 Distrito Federal 34 36 1,9 Espírito Santo 38 40 2,3 Goiás 24 29 4,4 Maranhão 48 54 5,8 Mato Grosso 34 40 5,8 Mato Grosso do Sul 23 26 3,4 Minas Gerais 31 33 1,4 Paraiba 48 52 4,3 Paraná 25 24-0,7 Pará 46 51 5,7 Pernambuco 49 53 3,9 Piauí 48 55 6,3 Rio Grande do Norte 48 53 5,0 Rio Grande do Sul 26 25-0,8 Rio de Janeiro 36 36 0,8 Rondônia 41 48 6,7 Roraima 40 46 6,4 Santa Catarina 28 27-1,3 Sergipe 47 51 4,3 São Paulo 36 37 0,4 Tocantins 37 44 6,5 Macro Regions North 44 50 6,2 Northeast 47 52 4,6 Central-West 28 32 4,5 Southeast 34 35 1,0 South 26 25-0,9 Brazil 40 43 3,2
Table 7: Agricultural Output and Poverty Municipal (Rural + Urban) Poverty Migration Adjusted Municipal Poverty Panel A: Ordinary Least Square Agricultural Output (in log) -0.045*** -0.045*** (0,002) (0,002) Population (in log) 0.006** 0.013*** (0,003) (0,003) Constant 0.593*** 0.526*** (0,021) (0,020) R-squared 0,15 0,15 Observations 5479 5479 Panel B: Instrumental Variables Agricultural Output (in log) -0.126*** -0.123*** (0,004) (0,003) Population (in log) 0.036*** 0.042*** (0,003) (0,003) Constant 1.012*** 0.927*** (0,033) (0,032) R-squared Observations 4948 4948 First Stage (Dep. Variable: agricultural output (in log)) Average Temperature 0.835*** 0.835*** (0,116) (0,116) Average Temperature Squared -0.020*** -0.020*** (0,003) (0,003) Average Precipitation 0.028*** 0.028*** -0,003-0,003 Average Precipitation Squared -0.000*** -0.000*** 0 0 Soil Dummies Yes Yes Constant -2.476* -2.476* (1,336) (1,336) R-squared 0,27 0,27 Observations 4948 4948 Standard errors in parentheses * significant at 10%; ** significant at 5%; *** significant at 1%
Brazilian States Table 8 - Simulated Effects of Climate Change on Poverty with Mobility Municipal (Rural + Urban) Poverty Migration Adjusted Municipal Poverty Predicted Current Poverty (%) Simulated Poverty (%) Change (p. p.) Predicted Current Poverty (%) Simulated Poverty (%) Change p.) Acre 25 30 5,5 25 31 5,3 Alagoas 24 28 3,4 25 28 3,3 Amapá 29 34 5,1 29 34 5,0 Amazonas 25 31 6,7 26 32 6,6 Bahia 26 30 3,5 27 30 3,4 Ceará 33 38 5,0 34 39 4,9 Distrito Federal 5 6 1,7 8 9 1,7 Espírito Santo 18 21 2,8 19 21 2,7 Goiás 14 17 3,7 15 18 3,7 Maranhão 33 38 5,3 33 38 5,2 Mato Grosso 25 30 5,4 25 31 5,3 Mato Grosso do Sul 13 17 3,2 14 17 3,1 Minas Gerais 21 22 1,2 21 23 1,2 Paraiba 33 37 3,7 33 37 3,7 Paraná 16 15-0,6 17 16-0,6 Pará 27 32 4,9 28 32 4,8 Pernambuco 29 32 3,3 29 33 3,2 Piauí 33 39 5,8 33 39 5,7 Rio Grande do Norte 30 35 4,5 31 35 4,4 Rio Grande do Sul 18 17-0,8 18 18-0,8 Rio de Janeiro 19 20 1,3 21 22 1,3 Rondônia 25 31 6,1 25 31 6,0 Roraima 25 31 5,4 26 31 5,3 Santa Catarina 20 18-1,1 20 19-1,1 Sergipe 30 34 4,0 30 34 3,9 São Paulo 20 20 0,2 21 21 0,2 Tocantins 32 38 6,0 31 37 5,8 Macro Regions North 27 32 5,6 27 33 5,5 Northeast 30 34 4,1 30 34 4,0 Central-West 14 18 3,6 15 19 3,5 Southeast 20 21 0,8 21 22 0,8 South 18 17-0,8 18 17-0,8 Brazil 22 24 2,0 23 25 2,0 (p.
Figure 1 Density.05.1.15 Distribution of Temperature Brazilian Municipalities 0 10 15 20 25 30 Average Temperature in Celsius Degree Normal Temperature Density
Figure 2
Figure 3 Density.005.0 01.015 Distribution tion of Rainfall Brazilian Municipalities 0 0 100 200 300 Rainfall in mm per month Normal Rainfall Density
Figure 4
Figure 5 Distribution of Change in Temperature Brazilian Municipalities 0 Density.5 1 1.5 2 1 1.5 2 Temperature Change in Celsius Degree Normal Temperature Density
Figure 6
Figure 7 Density.05.1.15 Distribution of Rainfall Change Brazilian Municipalities 0-20 -10 0 10 20 Rainfall Change in mm per month Normal Rainfall Change Density