Transcript On the Distributional Implications of Climate Change

On the Distributional Implications of
Climate Change:
A Methodological Framework and Application to Rural
India
Hanan Jacoby (DECRG)
Emmanuel Skoufias (PRMPR)
Mariano Rabassa (PRMPR)
World Bank
March 19, 2009
Motivation & Scope-1
 General consensus is that the main effect of climate
change will be to reduce agricultural productivity.
Given that the poor are concentrated in developing
country agriculture, they are likely to suffer the most.
 Within rural areas of developing countries there is
likely to be a great deal of heterogeneity in the
vulnerability to climate change.
 Studies to date useful for identifying vulnerable
countries or regions.
Motivation & Scope-2
Studies on the Impacts of CC:
Neo-Ricardian approach - Focus on impacts of
CC on agricultural productivity (land value, net
revenues etc) taking into account adaptation
Crop models - little or no adaptation
India: CC impacts range from + to – and
depend on crop and region studied
Impacts more modest when
adaptation is taken into account
Motivation & Scope-3
Yet, policy (e.g. targeting interventions)
must also be guided by information on
which types of households are more
vulnerable (e.g. according to physical
and/or human capital)
Household level data are essential
This study is the first attempt towards
estimating the distributional impacts of
climate change.
Poverty Rate (headcount)
Proportion of hh income from land - lamda

Framework
 Welfare measured by consumption per capita
 Consumption determined by resource
endowments (land and labor) and returns from
activities (farm and off-farm).
 We use a comparative statics approach to
estimate the impacts of climate change
on returns to land (agricultural productivity) taking
into account adaptation
on the returns to off-farm activities
 We trace the impacts of these productivity
changes on consumption in rural areas
Caveats -1
Cross-sectional variability in climate and
land values defines set of adaptation
possibilities in the Long Run
 Technological envelope of present is the same as in the future.
Climate scenario: uniform +1°C increase
in temperature (holding rainfall constant)
 Include higher-resolution climate change scenarios for India
(IITM, Pune). PRECIS model predicts higher temperature
monotonously spread over the country but substantial spatial
differences in projected rainfall changes
 Ignore potential change in rainfall variability
Caveats -2
Evolution of the distribution of
endowments (land, physical and human
capital) over time is not taken into
consideration,
 Impacts of CC derived based on current tock and distribution of
endowments . But different scenarios of such changes could
potentially be accommodated into the framework (e.g. more
educated hh members)
 Lanjouw & Murgai (2009): table 3: distribution of occupations in
rural India between 1983-2004 (period of trade reforms and
economic expansion) practically constant
Caveats -3
River basin flows and irrigation supply not
modeled in detail
Impact of climate change on prices not
considered
Typical in all applications of the neo-ricardian
approach
Methodology: basic model
budget constraint : c  a  pw
where
c  per capita consumption
a  per capita landholdin gs
 ( )  annualized net return on land
p  proportion of economically active members
w( )  annual wage
  climate
Note: HH labor optimally allocated between own and off-farm
activities
Comparative Statics of Climate
dc / c
d / 
dw / w

 (1   )
d
d
d
a
where  
a  pw
  V
Note: Shifts in hh labor allocation due to CC have no welfare
consequences as a result of Envelope Theorem.
Extension: Land heterogeneity
 Irrigated (i) and nonirrigated (n) land with
different returns and responses to climate.
(Need to assume increasing, convex cost of installing irrigation)
d i /  i
d n /  n
d / 

 (1   )
d
d
d
ai i
where  
ai i  an n
Note: Change in calculation of λ, i.e.,
a  ai i  an n
Proportion Irrigated
Extension: Labor heterogeneity
 If skilled labor (requiring greater human capital
investment) and unskilled labor earn different
wages and have different climate responses.
dws / ws
dwu / wu
dw / w

 (1   )
d
d
d
ws ns
where  
ws ns  wu nu
Note: Change in calculation of λ.
Empirical Implementation
 Estimate marginal effects of θ on logendowment prices: log(πk) and log(wm).
 Calculate λ, φ, σ for each household (these
weights depend on household endowments
and on the associated endowment prices).
 Predict change in log(c) for given Δθ for each
household using formula. [Why not directly
estimate log(c) as function of θ?]
Neo-Ricardian Approach
 Reduced-form relation between land
productivity (net revenue or value) and climate
normals.
 Assumes cross-sectional relationship will
continue to hold into future  farmers will
adapt to CC along today’s technological
envelope.
 Only way to quantify the economic costs of CC
in agriculture while taking adaptation fully into
account. (Crop modeling takes only limited
account of adaptation).
Estimation Issues
 What to control for in Ricardian regressions?
Infrastructure (e.g., irrigation, roads) versus ‘immutable’
characteristics (e.g., soil, topography, irrigation
potential). Will infrastructure remain fixed as climate
changes? Will infrastructure adjust as it has in the past,
as reflected in the current long-run equilibrium?
 How to estimate log πk (θ) k =i,n ? Irrigation investment
is largely irreversible 
 Estimate log πi(θ) using data on irrigated plots only.
 Estimate log πn(θ) using data on both irrigated and nonirrigated
plots, thus allowing for the option of new irrigation investment
as climate changes. Using only nonirrigated plots artificially
holds irrigation infrastructure fixed at zero.
More Estimation Issues
 Panel versus cross-section: Dechenes and Greenstone
(2007) purge all locational characteristics using fixed
effects  estimate short-run response of farm revenue to
weather shocks. Since little adaptation occurs from year
to year, the SR impact is upper bound on LR impact of
CC. How informative is upper bound for, e.g., India?
 Tradeoff between more heterogeneous marginal effects
of climate and danger of overparameterization. (E.g.,
quadratic terms and interactions in quarterly
temp/precip.)
 Land values versus net revenues. Each subject to
measurement error of a different kind.
Existing Estimates for India
 Sanghi et al. (1998) using data from 271 districts find
that 1.0 °C warming would reduce net farm revenue by
9%. But Kumar and Parikh (1998) estimate only a 3%
decline using similar data and methodology.
 Guiteras (2008) uses district-panel data to estimate the
impact of weather shocks on gross productivity. Medium
term CC scenario (?)  crop yield will decline by 4.59%, but again this is an upper bound. A 1.0 °C temp.
increase would reduce rural wages by 2%.
 For both approaches, negative effect of temperature rise
far outweighs positive impact of precipitation increase.
Description of Key Variables

District-level analysis of endowment prices (~500 districts) based on
household and plot level data from 59th (2002-03) & 61st (2004-05)
rounds of nationally representative National Sample Survey.

Cropland values: 59th round gives data on area, value, and irrigation of
100+ thousand plots. We use log of district means. (“For assessing the value
of land acquired by the household through inheritance or otherwise…the informant, if
necessary, may be asked to take the help of the knowledgeable persons of the village to
ascertain the current market price of the type of land. This may be determined on the basis
of the transactions made within the village or in its vicinity during the recent past” NSS Field
Manual).

Net crop revenue: 59th round has info for ~40 thousand farm
households. Caveat: 2002-03 was a very poor harvest.

Rural wages: 61st round has daily wages earned in last week for ~50
thousand individuals. (~10k in skilled occupations: education, health,
public administration). Residuals of log wage regression on age-gender
dummies are averaged at district level.
Land value/ha vs. Net Revenue/ha
Poverty & Unskilled Rural Wages
Covariates
 Climate:
- Temperature: From 391 Indian weather stations (1951-1980; average
26 yrs/station). We take average of 3 nearest stations weighted by
inverse squared distance to district centroid.
- Precipitation: Gridded data from +1800 stations for 1960-2000
interpolated by IMD on 1° cells (CRU data is on 0.5 ° cell but based on
much fewer stations, including those outside India).
 Immutable characteristics:
-
Soil (FAO, soil map of world, 34 categories)
Topography (% of district with slope in 3 categories)
Elevation (% of district with elevation in 3 categories)
Rivers/km2 in district
Groundwater in thousands m3/km2 (state level)
Straight-line distance to nearest city of +1 million & +5 million. (Not
really immutable, but formation of lots of new big cities as a result of
CC seems unlikely in foreseeable future).
Temperature (annual average)
Rainfall (annual average)
Marginal effect of Temp on returns to
endowments
Marginal effect of Temp on returns to irrigated
land
Changes in PCE-Linear vs. Quadratic
Baseline Poverty &
Impacts on Poverty-Linear
Climate Change Incidence Curves
Rural India
Climate Change Incidence Curve - Linear Model
-2.10
-2.20
Change rate
-2.30
-2.40
-2.50
-2.60
-2.70
-2.80
-2.90
0
10
20
30
40
50
Percentile
60
70
80
90
100
CChange Incidence Curves-Linear
Rural Andhra Pradesh vs Punjab
Climate Change Incidence Curve - Linear Model
Climate Change Incidence Curve - Linear Model
Andhra Pradesh
Punjab
-2.15
-1.80
-2.20
-2.00
-2.25
-2.20
-2.40
-2.35
Change rate
Change rate
-2.30
-2.40
-2.45
-2.60
-2.80
-3.00
-2.50
-3.20
-2.55
-3.40
-2.60
-3.60
-2.65
0
10
20
30
40
50
Percentile
60
70
80
90
100
0
10
20
30
40
50
Percentile
60
70
80
90
100
Take-away messages
 We have proposed a flexible framework for quantifying
distributional implications of climate change in the rural
economies  worth applying in, e.g., Mexico, Brazil.
 Distributional impacts in India depend primarily on
proportion of household income derived from land.
Wealthier households will suffer proportionally greater
consumption declines because they hold more land (and
they are also concentrated in more affected areas).
 Changes in poverty rates are not highly localized (e.g.,
Punjab proportionally harder hit but richer to start with).
 Overall, the impacts on rural household income in the
medium term seem modest. It remains to be seen
whether impacts are robust to extensions such as
modeling increased rainfall variability.
Thank you