Learning Center
Plans & pricing Sign in
Sign Out

Per Capita Income in India Statics


Per Capita Income in India Statics document sample

More Info
									On The Distributional Implications of Climate Change:
  A Methodological Framework and Application to
                    Rural India
                                         Hanan Jacoby∗
                                       Mariano Rabassa
                                      Emmanuel Skoufias

                                     The World Bank

                                  PRELIMINARY DRAFT

                                 This version: August 2009


          Climate change will severely depress productivity in developing country agricul-
       ture, a sector upon which much of the worlds poor depend. Though a large literature
       attempts to quantify the economic costs to agriculture as a whole, there has been lit-
       tle, if any, effort to understand how these potential losses will be distributed across
       households and how the poor will fare in particular. In this paper, we present a method-
       ological framework for tracing out the consequences of climate change for poverty and
       income distribution in the rural economy. We propose a two step approach, first esti-
       mating the impact of climate on the returns to land and labor, and then constructing,
       from these returns and household baseline characteristics, a counter-factual measure
       of household income (consumption) under alternative climate change scenarios.

     Contact Information: Jacoby; Skoufias
1     Introduction
There is wide consensus that significant climate change (CC) will occur over the next several
decades and that, in the developing world at least, much of its economic impact will be felt in
rural areas where agriculture is the dominant source of livelihood (e.g., Stern report). While
the deleterious effects of higher temperatures on crop yields are undisputed, efforts to map
out the consequences of this productivity shock for rural incomes have, to date, been crude
at best. Much of the literature in economics has focused on predicting the overall GDP
costs of reduced agricultural productivity; that is to say the average impact (Cline, 2007,
summarizes the global evidence; for the US, see Mendelsohn, et al, 1994 –henceforth MNS–
and the numerous commentaries thereupon). Comparatively little attention, however, has
been paid to heterogeneity in the costs of CC across the rural population of less developed
countries and, hence, to the distributional question. Such heterogeneity could arise from
both variation in the climate sensitivity of agricultural production across space as well as from
variation in the holdings of climate sensitive assets, especially farmland. Understanding how
these welfare costs will be distributed–i.e., which households are most vulnerable–is essential
to assessing how rural poverty and inequality will evolve as climate changes.
    This paper makes a first attempt at a distributional analysis of climate change in a
rural economy. Our launching point is the static farm household model, which provides a
natural framework for tracing through the implications of the fall in agricultural productivity.
Though highly stylized, the model delivers a simple and tractable decomposition of the
welfare effects of CC into components due to the changing returns on land and labor. With
estimates of these return functions in hand, we can take a household at any given point along
the initial rural welfare (per-capita expenditure) distribution and assign it a welfare change
based on its current characteristics, thus allowing a full accounting of the distributional shifts
that will ensue from a changing climate.
    India provides an ideal case study for the application of the empirical program just
outlined. A vast and diverse country with a wide range of agro-ecological zones, rural
India also exhibits substantial income inequality both across and within regions. Moreover,
Indian agriculture is expected to be especially hard-hit by warming temperatures in the
coming decades, although not necessarily uniformly (World Bank, 2009). The main data
sets we use are drawn from two rounds of India’s nationally representative National Sample
Surveys (NSS). The 2002-03 round collects detailed farm-level information and the 2004-05
round includes both a household expenditure survey and a labor force survey. These data
are supplemented with district-level panel data on agricultural productivity, rural wages,
weather and other variables.

     The main empirical challenge is to credibly estimate the returns to land and labor as
functions of climate. The literature provides two ways of doing so, both of which we adopt
in this paper. The first strategy follows the so-called neo-Ricardian approach of MNS by
assuming that farmer adaptation to future CC will be analogous to past adaptation to cross-
sectional spatial variation in climate. In other words, as the Punjab of tomorrow becomes
as warm as, say, the Rajasthan of today, Punjabi farmers will start behaving like Rajasthani
farmers do now. The second strategy, is to use panel data to obtain an upper bound on the
marginal impact of CC, focusing on responses to annual weather shocks, to which there is
little scope for adaptation (see Deschˆnes and Greenstone, 2007; henceforth DG).
     The plan of this paper is as follows. In the next section, we sketch the conceptual
framework, first in the starkest terms, where the household faces a single decision, followed
by extensions that incorporate more sectoral choices. Section 3 discusses issues in estimating
the relevant endowment prices as functions of climate, the key inputs into the distributional
analysis. Sections 4 and 5 describes the data and reports the results for the two approaches to
estimating the endowment price regressions. Given the climate change scenarios, discussed
in section 6, section 7 analyzes the resulting dynamics of poverty and inequality. Conclusions
and suggestions for future work are discussed in section 8.

2     Distributional Impacts in the Rural Economy
2.1    Insights from a Farm Household Model
A simple static farm household model highlights the key distributional issues. The house-
hold’s per-capita endowment of homogeneous land is a. Agricultural production is constant
returns to scale with farm labor as the sole input. An annual real wage of w per person can
be earned in the off-farm labor market with family and hired labor being perfect substitutes
on the farm. The household allocates its labor between own-farm and off-farm activities to
maximize its total income, which consists of farm production plus the real value of off-farm
earnings (the latter may be negative, if the household is a net hirer of labor). Alternatively,
maximized income can be expressed as the sum of profit (production net of the real cost of
labor, both family and hired) and the value of the household labor endowment. Using this
second formulation, let π = π(w; θ) be real farm profit per acre, which depends on the real
wage and on climate parameter θ (in practice, a vector including, e.g., mean temperature
and rainfall). And, let p be the proportion of economically active household members. The
household budget constraint in per capita terms is then simply

                                                c = aπ + pw,                                                 (1)

where c is consumption per household member, our preferred indicator of welfare.
    While climate affects income through farm profits, the real wage also may depend on θ;
it does so through supply and demand in the labor market. Consider the following spatial
equilibrium: Owners of land, however little they have, never move. Landless households
(a = 0), by contrast, are mobile, making their lifetime location decisions by comparing their
migration cost µ to the discounted present value of the infinite stream of farm labor earnings
at each location. Imagine a pool of migrant labor available to move anywhere from some
fixed central location. With a unitary discount rate, the net gains to moving to any location,
w−µ, must be equal to a fixed reservation utility in equilibrium. Thus, if migration costs have
the cumulative distribution function G(µ), the labor supply function to a location is simply
G(w). The equilibrium wage is determined by the intersection of this labor supply curve
with the labor demand curve, −Aπw (w; θ), where A is the average (per-capita) landholdings
among landowners at a given location. It follows that wθ = −Aπwθ /(Aπww + g(w)), where
g is the pdf of µ. In the very long run, it may be argued that the supply of labor to a
location is perfectly elastic (G is essentially flat), in which case wθ = 0. Whether this is
indeed so is, ultimately, an empirical question. The point here is that wage differentials
across climatic zones, to the extent that they exist at all, reflect differences in marginal labor
productivity (πwθ ) as well as the distribution of migration costs. Thus, in order to use the
cross-sectional relationship between w and θ to forecast the effects of future climate change,
one must assume that the distribution of migration costs to each location remains fixed into
the future.
    Given the dependence of profit and wages on θ, the comparative statics of changing
climate is straightforward. First, write the total effect of climate on profit as dπ = πw wθ +πθ .
Differentiating (1) with respect to θ gives cθ = a dθ + pwθ . Changes in the allocation of
labor between own and off-farm production have no welfare consequences as a result of the
envelope theorem.1 Dividing both sides through by equation 1 then delivers

                                d log(c)    d log(π)           d log(w)
                                         =λ          + (1 − λ)                                               (2)
                                   dθ          dθ                  dθ
     One may legitimately question whether climate change is sufficiently ”marginal” that the envelope the-
orem applies. Deaton (1997) argues in the context of price or tax policy reforms that the second order
effects of large price changes (i.e., those that do not ”envelope away”) will alter the conclusions of the distri-
butional analysis only to the extent that behavioral elasticities (in the present case, the elasticity of supply
of off-farm labor) vary substantially across income classes. Within the climate change literature, it has also
been pointed out that comparative statics analyses do not account for adjustment costs, as adaptation is not
likely to be instantaneous (see Quiggin and Horowitz, 1999; Kelly et al., 2005). Modelling these dynamics
is beyond the scope of the present exercise.

where λ = aπ/(aπ + pw). In sum, although climate change affects household welfare only
through changes in endowment prices, these changes can have heterogeneous consequences.
In the special case where the marginal effect of climate on log endowment prices is constant
across households, all the distributional impacts of climate change work through λ, the
proportion of income derived from own-farm production.

2.2     Extensions
The basic story can be enriched along two dimensions. First, suppose that land is now of
two types, irrigated (I) and nonirrigated (N ), such that aI + aN = a, with irrigated being
the more profitable; i.e., πI > πN . With constant returns, the only way to get households
to hold both types of land at the same time is to posit an increasing private annualized (per
capita) cost of irrigated land, κ(aI ).2 The optimal choice of aI will equate the per acre return
differential πI − πN to the marginal cost κ . The comparative statics of climate change is
as before with a dπ replaced by aI dπI + aN dπN . One does not need to know by how much
                   dθ                  dθ      dθ
households will switch into (or out of) irrigated land in response to climate change as the
cost of any such compositional shift will just balance the benefit at the margin (again the
envelope theorem). The formula for d log(c) is the same as (2) with the replacement of d log(π)
                                          dθ                                                  dθ

                          d log(π)       d log(πI )           d log(πN )
                                    =ϕ              + (1 − ϕ)                               (3)
                             dθ              dθ                   dθ
where ϕ = aI πI /(aI πI + aN πN ) is the share of irrigated land in total income from land.
    A second extension is to distinguish labor by sector. Suppose that there is an unskilled
sector that includes both agricultural and non-agricultural labor. Within this sector, work-
ers are strongly substitutable across occupations and, because of the derived demand from
agriculture, wages depend on climate. By contrast, wages in the skilled sector may be rel-
atively insulated from climate. At any rate, entry into skilled occupations requires costly
human capital investment, hence labor is imperfectly substitutable across sectors. Now, let
wU and wS be the unskilled and skilled wage, repectively, and e the proportion of econom-
ically active household members with the requisite human capital to earn the skilled wage.
Suppose that the household’s private annualized per capita cost of education is ψ(e).3 Total
wage income is p((1 − e)wU + ewS ) and the income-maximizing allocation of active household
     Think of κ as a ”reduced form” representing the increasing opportunity cost of investment or the de-
creasing suitability of land for irrigation.
     A more satisfactory overlapping generations treatment would recognize that the human capital of current
adults is predetermined and that ψ is a cost to educating current children that only yields a payoff in the
future. Once again, our static framework abstracts from the empirically difficult issues of adjustment costs.

members across sectors must satisfy ψ (e) = p(wS −wU ). Following the same logic as before,
we would now replace d log(w) in equation (2) by

                          d log(w)     d log(wS )           d log(wU )
                                   =σ             + (1 − σ)                                (4)
                              dθ           dθ                   dθ
where σ = ewS /((1 − e)wU + ewS ) is the share of skilled labor income in total labor income.
   Inspection of the last two equations reveals that when the marginal effect of climate on log
endowment prices are identical across sectors (irrigated/nonirrigated or skilled/unskilled),
the overall expression for the percentage change in consumption reverts to formula (2).
However, the calculation of λ is, in general, not the same when there is either land or labor
heterogeneity; in particular

                                           aI π I + aN π N
                          λ=                                       .                       (5)
                               aI πI + aN πN + p((1 − e)wU + ewS )

    The upshot of these derivations is that once we have estimates of the marginal effects
from the regression of the respective log endowment price on θ, we can predict the percentage
consumption response due to climate change for different types of households, indexed by
λ, ϕ, and σ. The advantages of this two-step approach can be appreciated by considering
the alternative of estimating a regression of log(c) on θ in just one step. Of course, a linear
regression will deliver only limited distributional effects because a given change in θ has the
same impact on (log) consumption for all households subject to the same climate. To go
further, one would have to expand the consumption model to include interactions between
climate variables and, in the base case, λ. In addition to all the climate variables, the other
determinants of endowment prices would also have to be interacted with λ. The extensions
to the model imply additional three-way interactions between θ, λ, and ϕ and/or between
θ, λ, and σ. Thus, a proliferation of interaction terms may be required to capture the
same distributional impacts as a two-step approach; the one-step regression model would,
consequently, be both cumbersome and likely subject to severe multicolinearity.

3     Estimating Changes in Endowment Prices
3.1    The Neo-Ricardian Approach
MNS develop the so-called neo-Ricardian framework to forecast the impact of climate change
on agricultural productivity. The idea is to first estimate the reduced form relationship
between π, as measured by land values or net farm revenues, and climate θ. Using this
regression, one then predicts the effect of the climate change scenario (∆θ) on productivity.

In other words, spatial differences in agricultural productivity between warmer and colder (or
wetter and dryer) regions are assumed to be analogous to the corresponding intertemporal
differences. Thus, the neo-Ricardian approach presupposes that exactly the same menu of
agricultural technologies (crop mixes, irrigation types, etc.) available today will continue to
be relevant for farmers as they adapt to future climate change; technological innovation is
    There is a link to the older hedonic literature that considers what can be inferred about
the value of location-specific characteristics from price data (Rosen, 1974, and Roback, 1982;
see Bartik, 1987, for a review). For instance, in the model of the previous section, the rela-
tionship between w and θ is an hedonic equation in the sense that it emerges from a market
equilibrium, albeit a very simple one (for one thing, we abstract from any worker preferences
toward climate). As we have shown, wθ captures a combination of demand (production)
side and supply side parameters. Nothing can be inferred about the distribution of migra-
tion costs, for instance, from w(θ) alone. If, however, we are willing to assume that this
distribution, as well as the set of production technologies, is fixed over time, then w(θ) is
valid for prediction purposes.
    An alternative approach to forecasting the productivity impacts of climate change in
agriculture is based on individual crop yield modelling using temperature and precipitation
responses taken directly from experimental field trials. More sophisticated studies (e.g.,
Parry et al. 1999) also account for certain adaptations to climate change, such as shifts
in planting dates, increased irrigation, and changes in crop varieties. However, even this
limited set of adaptations is assumed a priori, rather than necessarily reflecting how farmers
would actually behave on the ground. By contrast, the neo-Ricardian methodology allows
for as much (but no more) adaptive response as is revealed in empirical data.

3.2    Treatment of Infrastructure
 As climate varies, many man-made factors of production could also vary endogenously:
irrigation, electrification, roads, population density (and, therefore, market size), and so
on. We refer to such productive factors, generically, as infrastructure. By contrast, there
are exogenous agro-ecological factors that are not directly caused by climate, but which
may be correlated with it, even if incidentally. Slope, for example, may be related to
temperature (i.e., via mountains) and could directly influence land productivity. In sum,
both infrastructure and agro-ecological factors are potentially correlated with climate in
a cross-section, but only infrastructure responds to changes in climate over the relevant

    Within the neo-Ricardian literature, the treatment of infrastructure, and irrigation in
particular, has generated considerable debate (see Cline, 2007, for an overview). Should one
control for irrigation in a regression of farm profits or land values on climate? Schlenkler et
al. (2005) answer in the affirmative; irrigated farms are not only likely to be more productive
than dryland farms, but also differ in their responsiveness to climate variation (DG follow
this lead in their reworking of the MNS analysis). However, a regression that conditions
on irrigation asks: What is the impact of climate change on farm productivity holding
the stock of irrigation fixed? This question is relevant only in a setting where the stock
of irrigation is expected to remain fixed, which may not be the case in countries in which
the agricultural sector is still developing. In the latter context, it is preferable to control
for observable indicators of irrigation potential (and perhaps their interaction with climate
variables) rather than for irrigation infrastructure.4,5 Similarly, it would be better to control
for access to markets using geographical variables rather than using possibly endogenous
transport infrastructure.
    There is an argument for distinguishing irrigated and unirrigated land in the context of
a distributional analysis. As shown in the previous section, distributional impacts may be
sensitive to the disaggregation of the marginal effects into dπI and dπN . If irrigation is, as a
                                                               dθ       dθ
first approximation, an irreversible investment, then farmers of land that is already irrigated
cannot adapt to climate variation by adopting or removing irrigation. From their point of
view, the stock of irrigation is indeed fixed and, hence, the relevant marginal effect is that
for irrigated land only. Owners of unirrigated land have the option of installing irrigation
as climate varies, so it is the overall marginal effect, in which land is not distinguised by
type, that is relevant. In other words, dπI must be estimated from data on irrigated land,
whereas dθ must be estimated on data from irrigated and unirrigated land. Estimating
     only using data on unirrigated land would, again, presume that adaptation occurs on
all margins except irrigation investment.
      This approach still may be ‘optimistic’ in assuming a nonincreasing cost of exploiting the existing
irrigation potential. Kurukulasuriya and Mendelsohn (2007) use a somewhat related strategy by controlling
for irrigation infrastructure but treating it as endogenous, using irrigation potential (water flow) as an
instrument. However, since the primary interest is in the overall effect of climate on farm productivity, it is
not clear what advantage this structural modelling has over a reduced form that just controls for irrigation
      Dinar et al. (2008), in a series of neo-Ricardian studies in Africa, control for surface water runoff as
calculated from a hydrological model. Since the inputs to this model (temperature, rainfall, soil type, rivers,
groundwater, etc.) can each be controlled for separately in the regression, the value-added of including a
complicated nonlinear function of these variables (i.e., runoff) is unclear. In the Indian context, more rapid
Himalyian glacial melt may increase surface water runoff in the medium term and decrease it in the longer
term, in which case modelling runoff separately would be of value. Unfortunately, predicting the extent and
implications of glacial melt is subject to even more uncertainties than predicting climate change (see World
Bank, 2009).

3.3    Panel versus Cross-Sectional Estimators
Deschˆnes and Greenstone (2007) argue that it is essential to control for all locational char-
acteristics that may be correlated with climate, whether man-made or immutable, and that
the only way to do this is with panel data. To fix ideas, suppose that, as with most neo-
Ricardian studies, data are available on net farm revenues per acre rather than on land
values. The long-run average of farm revenues should be proportional to the price of land
insofar as crop production is the only significant use of farmland. Climate, θi , in turn,
determines average farm revenues. Actual farm revenues Rit in cross-sectional unit i in
year t also depends on weather realized in that particular year, ξit . Consider the following

                                 Rit = αθi + βξit + ωi + υit ,                            (6)

where ωi and υit are permanent and transitory error components. All of the infrastructure
and agro-ecological factors, observed or otherwise, are impounded in ωi . Since ωi may be
correlated with θi , cross-sectional neo-Ricardian estimates may be biased.
     Embodied in this specification is the notion that permanent variation in climate has a
different impact on productivity than transitory variation in weather. This is the essence
of adaptation: Long run adjustments by farmers mitigate the effects of climate variation
whereas in the very short-run there is practically no scope for such mitigation, hence α < β.
Recognizing this, DG point out that a first-differenced over time version of equation (6)
provides an unbiased estimate of β, purged of the effect of ωi , which can be taken as an
upper bound for the parameter of true policy-interest, α. Their approach thus only uses
information on short-run (i.e., year-to-year) net revenue responses to weather shocks.
     Of course, an upper bound is only as useful as it is tight. In applying their methodology
to U.S. farmland, DG find that β is a relatively small number, implying that α must, in
turn, be close to zero. While the upper bound is informative in the U.S. case, what about
in a setting where α is large? In this case, a correspondingly large estimate of β would not
tell us whether α is actually large or small. The panel data estimator is, therefore, only
informative in settings where the impact of climate change on agricultural productivity is
small! For much of the developing world, we cannot presume that this will be so.
     Finally, in reflecting back on DG’s critique of cross-sectional neo-Ricardian estimators,
it is crucial to bear in mind the economic model underlying the empirical specification. As
discussed above, conditioning on infrastructure is tantamount to assuming a future in which
climate changes but infrastructure remains fixed at current levels. If this assumption is
correct, then unobserved infrastructure is an “omitted variable” (part of ωi ) that biases

cross-sectional estimates of α. If this assumption is invalid, then infrastructure variables do
not belong in neo-Ricardian regressions at all. Under these circumstances, the panel data
estimator is justified only if there are important immutable charactistics, such as soil quality,
that cannot be observed and that are correlated with climate.

4       Neo-Ricardian Estimation Results
4.1     Data and Variables
The main data sets we use are drawn from India’s National Sample Surveys (NSS). The NSS,
initiated in 1950, is a nationally representative survey conducted in the form of successive
rounds, covering the entire rural and urban areas of the country. The subject covered in each
rounds is decided on the basis of a 10 year time frame. In this cycle, 2 years are devoted to
quinquennial surveys on household consumer expenditure, employment and unemployment
situation. These full-scale household expenditure surveys, also called ‘thick’ rounds, are the
primary source of information on per capita expenditure and poverty in India. Besides the
quinquennial surveys, smaller socio-economic surveys –called ‘thin’ rounds– are conducted in
the remaining years. Subjects covered periodically include: land holding and livestock; debt
and investment; social consumption; health, morbidity and vital statistics; manufacturing,
trade and services in unorganized sectors; and family living surveys.
    In this paper we use data from two NSS rounds. The 59th round, a ‘thin’ round, admin-
istered in 2003 focused on rural households and their livelihoods. It provides detailed data
on landholdings and self-assessed valuation for almost 105,000 cultivated plots, as well as
extensive information on agricultural production –by agricultural season– for about 50,000
farm households.6 In India, there are two main agricultural seasons. Kharif, the most im-
portant, extends from early June, with the beginning of monsoon rains, to November. Rabi,
the winter or ‘dry’ season, extends from November to April. As opposed to Kharif harvest,
which is totally dependent on the quantity and timing of monsoon rains, Rabi crops rely
heavily on irrigation waters.
    The 61st ‘thick’ round administered in 2004-05 provides data on rural wages, the price of
the other key household endowment, labor. This survey constitutes the most comprehensive
    Different samples of households were surveyed in different schedules. Plot-level data on household
landholdings and land values were collected in the “Debt and Investment” Schedule, and household-level
data on agricultural production and net farm revenues were collected in the “Situation Assessment Survey
for Farmers” Schedule. Further, each household in each sample was visited twice. While land values were
recorded only during the first visit, information on agricultural production (net revenues) was gathered in
both visits for the previous agricultural season.

labor force survey in India. Detailed time allocation data, based on a one-week recall, was
collected for each household member.7

4.1.1    Agricultural Data

Although data were collected at the household level using clustered –i.e., village– sampling,
confidentiality prevents identifying the precise location of each village within a district, thus
making the district the lowest level of geographic dissagregation for which household endow-
ments can be linked to climate variables and other determinants of agricultural productivity.
This forces us to estimate the effect of climate on log endowment prices (land and labor) at
the district-level. Table 1 describes the variables used in this study. In 2002, India had 576
districts –excluding districts in Lakshadweep, Andaman and Nicobar Islands, and Jammu
and Kashmir– with an average size of 5,265 squared kilometers.
    For reasons already mentioned, land values from cultivated plots can be divided further
into two categories according to their irrigation status. Table 2 shows that, as expected, the
average price –in real terms– of an irrigated hectare is higher than a non-irrigated hectare.8
In general, irrigation in India is very common with 54 percent of the plots, accounting for
41 percent of the total cultivated area, under any kind of irrigation system. Additionally,
geographic areas show a high variability in their land prices. The average value of an hectare
of cropland is higher in the Indo-Gangetic Plain, extending from Punjab to Assam, as seen
in Figure 1. This part of India is the most intensively farmed zone of the country and one of
the most intensively farmed in the world. Much of the land has access, or potential access, to
irrigation waters from wells and rivers, ensuring crops even in years of drought and making
possible a Rabi crop as well as a Kharif harvest.
    Are land markets capable of reflecting the productivity impacts of different climates bet-
ter than farm revenues? In principle, the value of land is equal to the present discounted
stream of rental rates. In empirical work, however, land values and net farm revenues are
each subject to measurement error of a different kind. Net farm revenues, though usually
more available from household surveys than land prices, are largely determined by weather
realizations, which may depart greatly from climate normals. In the case that climate shocks
have a large impact on local prices, as is typical of underdeveloped agricultural markets, the
cross-section variability in net farm revenues will not uniquely reflect productivity differen-
tials. In the event, the 2002-03 harvest was very poor. The monsoon rains failed during July,
      Lanjouw and Murgai (2009) argue that calculation of rural wage rates directly from the NSS surveys
yields more reliable figures than are available from alternative sources.
      Otherwise stated, all prices in this study have been deflated using official state-specific rural poverty

resulting in a season rainfall deficit of 19 percent –the most deficient monsoon since 1987–
causing profound loss of agricultural production, with a drop of over 3 percent in India’s
    Land values better reflect long-term climate characteristics than net revenues. The main
drawback is that, since the value of farmland is self-assessed by farmers rather than derived
from actual market transactions, such estimates might be unreliable.9,10 The NSS, however,
took special care to assure reporting accuracy. If needed, farmers where allowed to consult
knowledgeable persons of the village to ascertain the market price of the type of land (NSS
Field Manual).
    The correlation between district-level land values and net revenues is XX. District-wise
mean net farm revenues –in real terms– are shown in Figure 2. The different spatial dis-
tribution, relative to land prices, is attributable, in part, to the abovementioned deficit in
monsoon rains. Though most parts of India suffered from severe reduced rainfall, especially
the south and the northwest, rainfalls where close to normal in the northeast.

4.1.2    Labor data

In rural India most adults are self-employed, either in farm work or non-agriculture activities.
The survey recorded a wage income, whether in cash or in kind, for about 45 thousand
adults. The vast majority of these workers, approximately 44 percent, are casual agricultural
workers. Regular employment is more common in the non-farm sector, especially in public
sector services such as education and public administration.
    For wage workers, we construct a daily wage based on the earnings from their main
activity, that is, the activity in which the individual worked at least four days in the week.
To ascertain time disposition the questionnaire asks the activities pursued by each household
member along with the time intensity in quantitative terms for each day of the reference
week. Since a person may be engaged in more than one type of activity on a single day,
the time intensity was measured in half-day units. About 70 percent of wage or salaried
laborers declare working in only one activity during the reference week, and 91 percent of
those working in more than one activity worked at least four days in one of them. Both of
these proportions are higher in the non-farm sector than in the agricultural sector. As was
the case with land prices, there exists a great deal of heterogeneity in rural daily real wages
across Indian regions. Rural wages are higher in the Northern Indian Plain and in districts
     Although the NSS questionnaire recorded plot-level acquisition and disposal values for 494 plots since
the beginning of the sample period (July 2002), the number of transactions per district is not large enough
to credibly estimate a cross-section regression.
     Jacoby (2000) provides evidence that self-assessed plot values accurately capture the rental income from
land in the South Asian context.

closer to the seas, especially in the southern state of Kerala (see Figure 3).
    We classify skilled workers as those with at least secondary school completed. In our
sample, the fraction of skilled workers among all rural wage or salaried laborers is 27 percent
(though this fraction is smaller in the whole rural population since self-employed farmers are
mostly uneducated). The daily average real wage for skilled workers is 2.6 times higher than
for unskilled workers, as seen in Table 2.
    In order to purge the effect of individuals’ characteristics on labor prices, we first regress
daily log-wages (in separate regressions for skilled and unskilled workers) on dummies for
age categories, gender, and social status (whether or not the household belongs to scheduled
tribe, scheduled caste or other backward class). The district-level average residual for skilled
and unskilled workers are used as dependent variables in the neo-Ricardian analysis.

4.1.3    District-Level Data

Temperature data, available by meteorological stations, were obtained from the India Me-
teorological Department (IMD). Specifically, the dataset includes average daily maximum
and minimum temperature readings from 391 Indian weather stations for each month from
1951 through 1980, with an average of 26 years of data per station. For reasons that will
be clarified below, we use data from 370 stations; those with temperature records spanning
more than 20 years. The network is evenly distributed across Indian regions with the ex-
ceptions of the Thar Desert, mostly in the state of Rajasthan, and the mountainous state of
Arunachal Pradesh.
    A high resolution rainfall dataset was also obtained from the IMD. It contains 53 years,
from 1951 through 2003, of daily gridded (1◦ by 1◦ latitude-longitude cells) data, interpolated
using rainfall records from more than 1800 weather stations across India. There are XX cells
over India, each covering an approximate area of 12,100 squared kilometers.
    Since our focus is on the long-run impacts of temperature and precipitation, we examine
climate normals. To create district-level temperature records we use a weighted average from
the three nearest stations to the district’s geographical center. The weights are the inverse of
the squared distance to the district’s centroid.11 To create district-level precipitation records
we first compute the average monthly rainfall for each cell from 1960 through 2000, and then
     District-level records do not noticeable change if we use five or ten stations instead of three, or if we
use those stations that are located within a 200 or 300 kilometer radius from each district’s centroid. A
somehow different approach is taken by MNS to interpolate districts’ temperature and precipitation from
weather station. Normal climate at a district’s centroid was estimated by running a weighted regression
across all the weather stations within a 500 miles radius. Each regression included 14 exogenous variables:
latitude, longitude, altitude, distance from nearest shoreline, and the corresponding square and interaction

interpolate rainfall records from those cells over a district, using the proportion of district’s
area on each cell as weights. Other climate datasets: CRU Guiteras.
    That India is a diverse country with a wide range of agro-ecological zones is reflected in
the extreme variability of both temperature and rainfall across its vast geography. Annual
average temperature ranges from 13◦ C to 29◦ C, with the mildest weather being across the
Indo-Gangetic Plain (see Figure 4). The southwestern summer monsoons, extending from
June through September, play a critical role in determining whether the Kharif harvest will
be bountiful, average, or poor in any given year. Monsoon rains also determine the annual
rainfall spatial distribution, which varies from 100 millimeters in western Rajasthan to over
8,000 millimeters at Cherrapunji in Meghalaya, considered the wettest spot on earth (see
Figure 5).
    Data on soil quality and other topographical factors, such as terrain slope, are drawn from
the Food and Agriculture Organization’s Soil Map of the World (SMW). The SMW provides
detailed data for each soil mapping unit –defined as a geographical area with similar soil
classes. The number of soil type classes which compose the FAO Soil Map legend is 106, and
these are often grouped into 26 major categories. The dataset contains additional information
about soil mapping units: three textural classes (coarse, medium, and fine), which reflect
the relative proportions of clay, silt, and sand in the soil, and three slopes classes (level to
gently undulating, rolling to hilly, and steeply dissected to mountainous). We use the digital
SMW –converted into digital format by ESRI– and GIS software to assign to each district a
dominant soil type, as well as a dominant soil texture and slope class.
    Irrigation is central to agricultural productivity. It is estimated that over 70 percent
of India’s food grain production comes from irrigated agriculture. As previously discussed,
potential access to irrigation waters, in lieu of irrigation use, is what should be accounted for
in a neo-Ricardian study. For this reason, we use ground water and surface water availability
as indicators of irrigation potential.
    District-wise ground water data was obtained from the Indian Ministry of Water Re-
sources. Specifically, the dataset contains the annual total replenishable ground water in the
district. These shallow aquifers in the active recharge zone play an important part in tube
well irrigation, and hence on agricultural production. In the case of surface water runoff,
the other main source of irrigation water, we compute the length of perennial rivers within
the district. Surface water irrigation has natural limitations imposed by a huge variabil-
ity of yearly precipitation, making length rather than flow a preferable proxy for irrigation
potential. The spatial distribution of both ground water and surface water availability de-
pict, not surprisingly, a similar picture (see Figure 6 and Figure 7). Irrigation potential
per squared kilometer is higher in the Northern Plain where meltwater from the Himalayan

glaciers provides up to 85 percent of the water in the great rivers during the dry summer
    Finally, the distance from the district’s geographical center to the closest city with popu-
lation over one million, is included to proxy the access to agricultural markets. As mentioned
before, in the neo-Ricardian context, it is better to control for access to markets using ge-
ographical variables rather than using possibly endogenous transport infrastructure. There
might still remain some endogeneity, however, since current population distribution across
space is affected by current –or recent past– climate. To explore this issue, we also use the
distance from the geographical center of a district (as in 2002) to the cities with population
over 150 thousands in 1901.

4.2    Results for Land Values (Net Revenues)

4.3    Results for Wages by Skill Level

5      Panel Data Estimation Results
5.1    Data and Variables

5.2    Results for Net Revenues

5.3    Results for Wages

6      Climate Change Scenarios
In the previous two sections we have estimated the reduced form relationship between π, as
measured by land values or net farm revenues, and climate θ. With the marginal effects in
hand, the next step –before undertaking the distributional analysis– is to predict the effect
of the climate change scenarios (∆θ) on overall agricultural productivity.
    For this paper, we have chosen to use the climate change predictions produced for the
Third Assessment Report of the United Nation’s Intergovernmental Panel on Climate Change
(IPPC). In particular, we use the output of two of the most reputable Coupled Atmosphere-
Ocean General Circulation Models (AOGCMs): the HadCM3 model, developed by the
Hadley Centre for Climate Prediction and Research, and the CCRS/NIES model, produced
by the Center for Climate System Research at the University of Tokyo.12 Our measure
    AOGCMs experiments are available through the IPCC’s Data Distribution Center http://www. Climate prediction data from the HadCM3 comes in a gridded format with a spatial
resolution of 2.5◦ latitude by 3.75◦ longitude (approximately 417km by 278km at the equator). Also in

of “climate change” is the mean monthly change in surface temperature (abstracting from
changes in precipitation for the time being) computed for three future periods relative to
the baseline period 1961-1990. These time periods are defined as short-run (2010-2039),
medium-run (2040-2069), and long-run (2070-2099).
    Recognizing the important role of anthropogenic emissions on future climate, the IPCC
introduced in 2000 several emissions scenarios. These scenarios explore alternative develop-
ment pathways, covering a wide range of demographic, economic and technological driving
forces, and resulting GHG emissions (IPCC SRES, 2000). For this empirical application, we
use the A1F1 (high emissions path), A2a (medium emissions path), and B1a (low emissions
path) scenarios.13
    Finally, to construct district-level temperature changes from gridded data we linked each
district’s geographical center to the closest cell center. In Table 6 we report the district-
level projections. Two points are worth mentioning. First, all three scenarios predict a
future warming for India with substantial variation across regions and seasons –with a higher
increase during Rabi. Second, all three scenarios predict a mean temperature increase of
approximately 1◦ C in the short-run. As one moves further into the future, however, the
predicted warming from each scenarios varies substantially, with temperature changes that
range from 2.4◦ C (HadCM3 B1a scenario) to 5.7◦ C (CCSR A1F1 scenario), by the end of
the century.

7     Distributional Analysis
7.1     Changes in Poverty

7.2     Climate Change Incidence Curves

8     Conclusions

gridded format, the CCSR/NIES model has a resolution of 5.6◦ latitude-longitude.
     The A1 scenario, usually referred as ‘business-as-usual’, assumes a world of very rapid economic growth, a
global population that peaks in mid-century and rapid introduction of new and more efficient fossil-intensive
technologies. The A2 scenario describes a very heterogeneous world with high population growth, slow
economic development and slow technological change. Finally, the B1 scenario describes a convergent world,
with the same population as the A1 scenario, but with more rapid changes in economic structures towards
a service and information economy. No likelihood has been attached to any scenario (IPCC, 2001).

                                     Table 1 —Description of Variables

Land                   Log of mean land value (Rs. per ha) for irrigated and unirrigated plots owned by rural
                       households. Values deflated using state-specific rural poverty lines.
                       Source: NSS 59th round.

Human Capital          Mean residual from daily rural log-wage regressions for skilled and unskilled workers on age,
                       gender, and social group. Rural wages deflated using states-specific rural poverty lines.
                       Source: NSS 61th round.

Temperature            Normal monthly mean temperature (◦ C) from 1951 through 1980. Interpolated from the
                       three closest weather stations weighted by proximity to districts’ geographical centroid.
                       Source: India Meteorological Department.

Rainfall               Normal monthly precipitation (millimeters) from 1960 through 2000. Interpolated from 1◦
                       gridded data weighted by districts’ area in each cell.
                       Source: India Meteorological Department.

Rainfall deviation     2002-03 monthly rainfall minus normal monthly rainfall (millimeters). Used in net farm
                       revenues regressions.
                       Source: India Meteorological Department.

Soil type              Dominant soil type according to FAO classification (12 categories).
                       Source: Soil Map of the World, FAO 1992.

Soil texture           Dominant soil texture. Dummies: coarse, medium, and fine.
                       Source: Soil Map of the World, FAO 1992.

Slope                  Dominant slope (steepness). Dummies: flat (0-8%), undulating (8-30%), and hilly (above
                       Source: Soil Map of the World, FAO, 1992.

Elevation              Height from sea level. Percentage of land within three elevation categories: 0-250m, 250-
                       1500m, and above 1,500m.
                       Source: Digital Elevation Map of the World.

River density          Perennial rivers and channalized rivers (km per km2 ).
                       Source: Digital Chart of the World, ESRI, 1997.

Ground water           Total replenishable ground water (bcm per year per km2 ).
                       Source: Dynamic Ground Water Resources of India, Ministry of Water Resources, 2004.

Distance to city       Linear distance (km) from districts’ geographical centroid to closest city with population
                       above a million.
                       Source: Population Census 2001.

 Notes: All variables are district-level variables. FAO soil categories within continental India include acrisols, cam-
 bisols, gleysols, lithosols, fluvisols, luvisols, nitosols, arenosols, regosols, vertisols, xerosols, and yermosols. Skilled
 workers are those with at least secondary school completed.

                       Table 2 —District-Level Summary Statistics
                                                         Mean       Std Dev          Districts

                         Number of plots                 186.39      158.09             549
                         Fraction irrigated               0.54
                         Value of land (Rs. per ha)
                         Irrigated                      339,361     323,646
                         Unirrigated                    245,672     282,863

                         Number of farm households       116.74      83.54              561
                         Fraction cultivate Kharif        0.94                          561
                         Fraction cultivate Rabi          0.71                          555
                         Net revenue (Rs. per ha)
                         Annual                          20,626      17,307
                         Kharif                          13,137      11,393
                         Rabi                            15,491      12,051
Human Capital
                         Number of wage workers          75.61       55.61              566
                         Fraction skilled                 0.27
                         Daily rural wage (Rs.)
                         Skilled                         186.67      72.87              564
                         Unskilled                        72.05      40.28              565
                         Temperature (◦ C)                24.75       2.91              576
                         Rainfall (mm)                   108.82      59.79              576
                         Fraction flat slope               0.48        0.33              576
                         Fraction fine texture             0.19        0.39              576
                         Fraction low elevation           0.55        0.44              576
Irrigation Potential
                         River density (km per km2 )     0.089       0.052              576
                         Ground water (bcm per km2 )     19.66       15.40              562
Access to Markets
                         Distance to city (km)           224.6       189.9              576

 Notes: All entries are simple averages over the 576 continental districts (excluding districts in
 Jammu and Kashmir) when available.

                                    Table 3 —Land Price Analysis
                                          All Land                         Irrigated Land         Net Revenues
                           (1)         (2)        (3)           (4)              (5)                   (6)

Temperature               -0.0912    -0.0677     -0.1501       -0.0913         -0.0830                -0.0458
                         (0.0118)   (0.0124)    (0.0232)      (0.0191)        (0.0165)               (0.0151)
                         [0.0196]   [0.0179]    [0.0304]      [0.0239]        [0.0242]               [0.0204]
Rainfall                  -0.0038    -0.0045     -0.0056       -0.0023         -0.0024                 0.0024
                         (0.0012)   (0.0012)    (0.0010)      (0.0009)        (0.0008)               (0.0008)
                         [0.0018]   [0.0018]    [0.0014]      [0.0011]        [0.0011]               [0.0009]
Texture: fine                                      0.2544       -0.1536          0.0393                0.0276
                                                (0.2036)      (0.1590)        (0.1746)               (0.1642)
                                                [0.2423]      [0.2005]        [0.2078]               [0.2028]
Texture: medium                                   0.4762        0.1939          0.1577                 0.2864
                                                (0.1023)      (0.0833)        (0.0853)               (0.1035)
                                                [0.1180]      [0.1084]        [0.1202]               [0.1470]
Slope: 8-30%                                     -0.2878         0.048          0.1891                 0.3545
                                                (0.2051)      (0.1651)        (0.1653)               (0.1314)
                                                [0.2196]      [0.1562]        [0.1594]               [0.1534]
Slope: >30%                                      -0.2678       -0.2124         -0.2445                 0.0334
                                                (0.0855)      (0.0711)        (0.0663)               (0.0699)
                                                [0.0850]      [0.0668]         [0.0662               [0.0695]
Elevation: 250-1500m                             -0.6363       -0.1718          0.0009                -0.2405
                                                (0.1035)      (0.0932)        (0.0817)               (0.1012)
                                                [0.1101]      [0.1041]        [0.1099]               [0.1183]
Elevation: >1500m                                -0.7474        0.1362          0.1857                -0.0879
                                                (0.2909)      (0.2761)        (0.2655)               (0.2586)
                                                [0.3902]      [0.3310]        [0.3414]               [0.3077]
River density                                                   2.3763          1.3749                 3.6357
                                                              (0.6683)        (0.6438)               (0.6854)
                                                              [0.7651]        [0.6556]               [0.8528]
Groundwater                                                     0.0110          0.0116                0.0054
                                                              (0.0026)        (0.0025)               (0.0029)
                                                              [0.0029]        [0.0030]               [0.0038]
Distance city                                                  -0.0027         -0.0022                -0.0008
                                                              (0.0002)        (0.0003)               (0.0002)
                                                              [0.0003]        [0.0003]               [0.0004]
Soil dummies                No        Yes         Yes             Yes            Yes                    Yes
Adjusted R2                0.11       0.25        0.43           0.59            0.51                   0.43
Observations               542        542         542             542             536                    536

 Notes: The dependent variable in columns 1 through 4 is the log of district mean real land value per hectare
 (using both unirrigated and irrigated plots), and in column 5 is the log of district mean land value per hectare
 for irrigated land. Specifications are weighted by the number of plots in the district. The dependent variable
 in column 6 is the log of district mean annual net real farm revenue per hectare, weighted by the number of
 farm households in the district. Robust standard errors are reported in parenthesis, and Conley’s standard
 errors are reported in brackets (two degrees lag window in both dimensions).

Table 4 —Land Price Analysis: Alternative Climate Variables
                      IMD Data               CRU data

Temperature            1.6158
Temperature squared    -0.0797
Temperature cubic      0.0012
Rainfall               0.0023
Rainfall squared       0.0000
Rainfall cubic         0.0000
Temperature Kharif     -0.0264
Temperature Rabi       -0.0535
Rainfall Kharif        -0.0008
Rainfall Rabi           0.0020


                       Table 5 —Labor Price Analysis
                                     Unskilled Labor                     Skilled Labor
                           (1)       (2)         (3)           (4)             (5)

Temperature              -0.0581    -0.0448        -0.0415    -0.0345        -0.0273
                        (0.0059)   (0.0067)       (0.0101)   (0.0103)       (0.0064)
                        [0.0126]   [0.0114]       [0.0144]   [0.0139]       [0.0077]
Rainfall                 0.0012      0.0009         0.0005     0.0010         0.0010
                        (0.0003)   (0.0003)       (0.0004)   (0.0004)       (0.0003)
                        [0.0008]   [0.0005]       [0.0005]   [0.0005]       [0.0003]
Texture: fine                                       0.1214      0.0650        -0.0232
                                                  (0.0660)   (0.0608)       (0.0694)
                                                  [0.0731]   [0.0596]       [0.0702]
Texture: medium                                    0.1850      0.1424         0.0814
                                                  (0.0459)   (0.0451)       (0.0429)
                                                  [0.0600]   [0.0535]       [0.0369]
Slope: 8-30%                                        0.3485     0.3870         0.1590
                                                  (0.0663)   (0.0672)       (0.0554)
                                                  [0.0697]   [0.0751]       [0.0530]
Slope: >30%                                         0.0759     0.0689         0.0900
                                                  (0.0341)   (0.0317)       (0.0553)
                                                  [0.0443]   [0.0377]       [0.0332]
Elevation: 250-1500m                               -0.2092    -0.1795        -0.0926
                                                  (0.0446)   (0.0505)       (0.0459)
                                                  [0.0669]   [0.0678]       [0.0522]
Elevation: >1500m                                  0.1432     0.2246          0.1012
                                                  (0.1508)   (0.1660)       (0.1125)
                                                  [0.2148]   [0.2051]       [0.1068]
River density                                                 0.7100         -0.0449
                                                             (0.3127)       (0.2625)
                                                             [0.4770]       [0.3159]
Groundwater                                                   -0.0012         0.0006
                                                             (0.0012)       (0.0010)
                                                             [0.0014]       [0.0014]
Distance city                                                 -0.0005        -0.0001
                                                             (0.0001)       (0.0001)
                                                             [0.0002]       [0.0001]
Soil dummies               No         Yes           Yes         Yes            Yes
Adjusted R2               0.31        0.41          0.50        0.53           0.39
Observations              554         554           554         554             553

 Notes: The dependent variable in columns 1 through 4 is the mean residual of log
 daily rural wage regressions on age, gender, and social group for unskilled salaried
 workers, and in column 5 for skilled workers. Skilled workers are those with at least
 secondary school completed. All specifications are weighted by the number of wage
 workers in the district. Robust standard errors are reported in parenthesis, and
 Conley’s standard errors are reported in brackets (two degrees lag window in both

  Table 6 —District-Level Temperature Change Projections
                     HadCM3                                     CCSR/NIES
          Mean     Std Dev Min        Max          Mean       Std Dev  Min        Max
                                        A1:2010 - 2039
Annual    1.26       0.17     0.79    1.59          0.87        0.22       0.48    1.80
Rabi      1.25       0.20     0.77    1.51          1.17        0.23       0.85    2.27
Kharif    1.28       0.24     0.76    1.84          0.57        0.29      -0.17    1.39

                                        A1:2040 - 2069
Annual    2.64       0.34     1.83    3.73          2.93        0.69      1.99     5.31
Rabi      3.28       0.46     1.87    3.90          3.87        0.78      2.19     6.46
Kharif    2.00       0.58     1.00    4.11          1.99        0.79      0.71     4.23

                                        A1:2070 - 2099
Annual    4.67       0.45     3.13    6.01          5.66        1.16      3.85     9.52
Rabi      5.23       0.56     3.19    6.64          7.15        1.47      3.96    11.40
Kharif    4.10       0.56     2.97    6.21          4.17        1.08      2.11    7.64
                                        A2:2010 - 2039
Annual    0.96       0.14     0.48    1.48          0.96        0.25       0.42    1.96
Rabi      1.14       0.20     0.50    1.61          1.44        0.33       0.81    2.78
Kharif    0.79       0.21     0.14    1.37          0.48        0.31      -0.33    1.27

                                        A2:2040 - 2069
Annual    2.40       0.24     1.60    2.83          2.38        0.63      1.57     4.38
Rabi      2.63       0.28     1.69    2.91          3.16        0.71      1.72     5.27
Kharif    2.17       0.31     1.31    3.05          1.60        0.72      0.20     3.57

                                        A2:2070 - 2099
Annual    4.05       0.39     2.58    4.82          5.00        1.07      3.18     8.24
Rabi      4.43       0.48     2.72    5.48          6.39        1.26      3.43     9.77
Kharif    3.67       0.45     2.44    4.87          3.61        1.15      1.97     6.72
                                        B1:2010 - 2039
Annual    0.91       0.13     0.54    1.60          0.82        0.22       0.43    1.75
Rabi      1.04       0.17     0.70    1.34          1.14        0.27       0.66    2.41
Kharif    0.78       0.25     0.20    2.02          0.49        0.26      -0.18    1.22

                                        B1:2040 - 2069
Annual    1.55       0.19     1.15    2.17          1.76        0.51       1.05    3.58
Rabi      1.87       0.24     1.19    2.43          2.37        0.57       1.30    4.50
Kharif    1.24       0.32     0.60    2.37          1.16        0.59      -0.20    2.96

                                        B1:2070 - 2099
Annual    2.43       0.28     1.54    3.14          3.40        0.79      2.08     5.89
Rabi      2.66       0.27     1.60    3.54          4.51        0.88      2.36     7.02
Kharif    2.20       0.41     1.31    3.32          2.30        0.90      1.05     4.80

 Notes: District-level temperature projections based on HadCM3 and CCSR/NIES
 models for A1F1, A2a, and B1a scenarios (IPCC Third Assessment Report). Tem-
 perature changes with respect to the baseline period 1961-1990 are expressed in degree
 Celsius. Rabi season is from December to May, and Kharif season is from June to
 November. Number of districts is 576 (districts from Andaman & Nicobars Islands,
 Lakshadweep, and Jammu & Kashmir excluded).

   Table 7 —Estimated Percentage Change in Per Capita Consumption
                              HadCM3                                   CCSR/NIES
                 Mean      Std Dev   Min       Max           Mean     Std Dev  Min         Max
Scenario A1F1
2010-2039        -0.0447   0.0093    -0.1364     0          -0.0293    0.0078    -0.1096    0
2040-2069        -0.0920   0.0184    -0.3365     0          -0.0988    0.0258    -0.3643    0
2070-2099        -0.1623   0.0296    -0.5425     0          -0.1931    0.0474    -0.6624    0

Scenario A2a
2010-2039        -0.0334   0.0075    -0.1348     0          -0.0325    0.0086    -0.1242    0
2040-2069        -0.0844   0.0159    -0.2587     0          -0.0800    0.0230    -0.3072    0
2070-2099        -0.1417   0.0260    -0.4403     0          -0.1704    0.0434    -0.5934    0

Scenario B1a
2010-2039        -0.0321   0.0067    -0.1368     0          -0.0272    0.0075    -0.1110    0
2040-2069        -0.0538   0.0103    -0.1852     0          -0.0589    0.0176    -0.2234    0
2070-2099        -0.0853   0.0162    -0.2738     0          -0.1151    0.0301    -0.4092    0

 Notes: Population weights are used to calculate average percentage change in consumption. Es-
 timates based on 73,778 households. The change in per capita consumption is based on climate
 projections from HadCM3 and CCSR/NIES models for A1F1, A2a, and B1a scenarios (see notes
 to Table 7).

               Figure 1 —Mean Cropland Value Per Hectare

Notes: This map shows quantiles of cropland values in thousand rupees per hectare
(in real terms). District-level land values are computed from almost 105 thousand
cultivated plots in rural areas (data source: NSS 59th round). Outliers –those plots with
value per hectare beyond five standard deviations– have been removed. On average,
there are 186 cultivated plots per district. The value of an hectare is higher across
the Indo-Gangetic Plain in northern India. This part of India is the most intensively
farmed zone of the country. Much of the land has access, or potential access, to
irrigation waters from wells and rivers.

             Figure 2 —Mean Net Farm Revenue Per Hectare

Notes: This map shows quantiles of net farm revenues in rupees per hectare (real
terms) for the 2002-03 agricultural season. District-level farm revenues are computed
from about 50 thousand farm households in rural areas (data source: NSS 61st round).
On average there are 117 farm households in a district. Net revenues are higher in
the Northern Plains (especially in Punjab, Haryana, and western Uttar Pradesh), and
in regions where monsoon rains are heavier, such as the northeastern states and the
southwest costs. In the 2002-03 season, monsoon rains where below historic averages
in most parts of India, with the exception of the northeastern states.

                      Figure 3 —Mean Daily Rural Wage

Notes: This map shows quantiles of daily wages for rural workers –in agriculture and
off-farm activities– for 2004-05 (in real terms). Daily rural wages are computed from
earnings in the main activity –at least 4 days in the reference week. District-level wage
rates are averaged from about 41 thousand workers (data source: NSS 61st round).
The average number of salaried workers in a district is 76. Rural wages are highly
(negatively) correlated with rural poverty rates (see map below).

                 Figure 4 —Average Monthly Temperature

Notes: This map shows the average monthly temperature from 1951 through 1980.
District-level temperature records are interpolated from the three closest weather sta-
tion from a network of 370 stations across India. Input data include the average daily
maximum and minimum temperature readings for each station (data source: India
Meteorological Department). The network is fairly distributed across Indian regions
with the exceptions of the Thar Desert, mostly in the state of Rajasthan, and the
mountainous state of Arunachal Pradesh.

                  Figure 5 —Average Monthly Precipitation

Notes: This map shows the average monthly rainfall from 1960 through 2000. District-
level rainfall records are constructed from a high resolution (1◦ by 1◦ latitude-longitude)
gridded rainfall dataset (data source: India Meteorological Department). Each grid cell
covers an area of approximately 12,100 km2 . Input data include daily precipitation (in
millimeters) interpolated from more than 1800 Indian weather stations.


     (a) River Density                                    (b) Ground Water

                         Figure 6 —Irrigation Potential
Figure 7 —Average Monthly Household Per Capita Expenditure

                Figure 8 —Poverty Rate (Head Count Ratio)

Notes: This map shows the district-wise rural poverty rates. Official 2004-05 state-
specific rural poverty lines are used to calculate district-level head count ratios from
monthly household per capita expenditure (data source: NSS 61st round). Twenty
eight percent of the rural population was below the poverty line in 2004-05. Poverty
vary substantially across regions, showing a higher incidence in the east-central states
of Madhya Pradesh, Chattisgarh, Jharkhand, Orissa, and Bihar.


     (a) Short-Run: 2010-2039                                   (b) Long-Run: 2070-2099

             Figure 9 —Predicted Change in Per Capita Expenditure (HadCM3 A1)
                 (a) Short-Run: 2010-2039

                 (b) Long-Run: 2070-2099

Figure 10 —Climate Change Incidence Curves (HadCM3 A1)

                               (a) Short-Run: 2010-2039

                               (b) Long-Run: 2070-2099

Figure 11 —Climate Change Incidence Curves Under Perfect Labor Mobility (HadCM3 A1)


To top