Download The effect of development on the climate sensitivity of agriculture

Environment and Development Economics 6 (2001): 85–101
Copyright © 2001 Cambridge University Press
Policy Options
The effect of development on the climate
sensitivity of agriculture
ROBERT MENDELSOHN
Yale School of Forestry and Environmental Studies, 360 Prospect Street,
New Haven, CT 06511, USA
ARIEL DINAR
World Bank, 1818 H Street, Washington, DC, 20433 USA
APURVA SANGHI
NERA, 1255 23rd Street, Washington, DC, 20037 USA
ABSTRACT This paper examines whether a country’s stage of development affects its
climate sensitivity. The paper begins with a model of agriculture that shows that the
effect of development on climate sensitivity is ambiguous, depending on the substitution
between capital and climate. To resolve this issue, the climate sensitivity of agriculture
in the United States, Brazil, and India is measured using a Ricardian approach. Relying
on both intertemporal as well as cross-country comparisons, the empirical analysis suggests that increasing development reduces climate sensitivity.
1. Introduction
In anticipation of global warming, there has been an extensive amount of
research done examining the sensitivity of the economy to climate over the
last 15 years. The Intergovernmental Panel on Climate Change (IPCC)
(Pearce et al., 1996) reports that agriculture, energy, coastal structures,
water, and timber are all likely to be sensitive to climate change. Further,
this report hypothesizes that the damages will be greater in developing
countries because their level of capital and technology is lower. For
example, the results cited in Pearce et al. (1996) predict that OECD countries will suffer damages of 1.4–1.6 per cent of GDP but less-developed
countries will have damages between 1.6 per cent and 2.7 per cent.
Economies with less capital and technology could be more vulnerable to
climate change because they have less control over their environments,
because more of their economy is in vulnerable sectors, and because they
have warmer climates to begin with.
Although it is commonly believed that vulnerable sectors in developing
86
Robert Mendelsohn, Ariel Dinar, and Apurva Sanghi
countries are more climate sensitive (Nordhaus, 1991; Schelling, 1992), it
has never been empirically tested. In fact, a formal theory explaining why
climate sensitivity would fall as the level of development rises has not yet
been developed. This paper develops a theoretical model to examine how
development might affect the climate sensitivity of agriculture. Agriculture is an appropriate sector for such an analysis because it could be the
single most important market impact from warming (Mendelsohn and
Neumann, 1998; Mendelsohn and Schlesinger, 1999; Mendelsohn et al.,
2000) and because it has a significant role in developing country
economies, averaging 30 per cent of GDP (World Resources Institute,
1996). The model of agriculture includes technology, climate, and other
inputs. The model explores how technology could affect climate sensitivity. McKinsey and Evenson, (1998) argue that technology has not had a
direct effect on climate sensitivity because new technology has not historically been used to systematically move crops into warmer or cooler climate
zones. Nonetheless, if technology encourages capital to substitute for
climate, the climate sensitivity function might flatten and move higher
with development (see figure 1), so that developing countries would be
relatively less vulnerable to climate change. In contrast, if the marginal
productivity of technology is higher for farms in ideal climates, technology
and climate could be complements. In this case, technology would be targeted at ideal climate conditions, making farms in these environments
High technology
Crop
response
Low technology
Temperature
Figure 1 Technology and climate substitutes
Environment and Development Economics 87
High technology
Crop
response
Low technology
Temperature
Figure 2 Technology and climate complements
even more productive relative to more marginal locations. The overall
climate response function would become steeper as development increases
(see figure 2). Development consequently could increase or decrease the
climate sensitivity of agriculture depending upon how technology affects
the interaction between capital and climate.
In order to determine how development affects climate sensitivity, one
must first establish a measure of development. The literature on development provides several measures of development (e.g., Hicks and Streeten,
1979; Hayami and Ruttan, 1985; Nafziger, 1990). Some measures cover an
entire country whereas others are specific to a sector. For example, one
common measure of overall development is income per capita. As shown
in table 1, income per capita clearly ranks India, Brazil, and the United
States in increasing order. An alternative measure in the agricultural
Table 1. GNP per capita ($)
Year
India
Brazil
USA
1986
1991
1995
1997
297
323
320
354
1,854
2,882
3,427
4,283
22,786
23,941
25,404
26,080
Note: Values are in 1990 USD adjusted by GNP deflator.
Sources: World Bank (1988, 1993, 1997, 1998).
88
Robert Mendelsohn, Ariel Dinar, and Apurva Sanghi
Table 2. Tractors per hectare
Year
India
Brazil
USA
1960
1970
1980
1990
1994
0.004
0.013
0.075
0.087
0.096
0.013
0.031
0.065
0.092
0.135
0.354
0.472
0.616
1.120
1.142
Sources: From 1960–1980, Hayami and Ruttan (1985); for 1990, 1994 FAO
Website (http://apps.fao.org).
sector is technology such as tractors per hectare, table 2. Note that the
order of countries remains the same with both the income and the technological measure, and that, except for income in the early 1990s in India,
income per capita and tractors increased steadily over time in all three
countries suggesting that development has increased over time in each
country.
This study empirically examines agriculture in Brazil, India, and the
United States in order to test how development has affected climate sensitivity. Samples of farms from all three countries have been gathered to
conduct Ricardian analyses in each country (Mendelsohn, Nordhaus, and
Shaw, 1994; Dinar et al., 1998; Sanghi, 1998). First, we examine Indian and
Brazilian climate sensitivity over time as development has increased. The
climate sensitivity of farms in both countries has fallen over time. Second,
we compare the climate sensitivity of India and the United States. The
Ricardian function for India is more climate sensitive than the American
climate response function. These results suggest that development does
lead to lower climate sensitivity.
2. Theory
This section develops a model of agriculture to explore the relationship
between development and climate sensitivity. The model builds upon the
Ricardian approach (Mendelsohn, Nordhaus, and Shaw, 1994). Consumers
are assumed to have well-behaved utility functions and linear budget constraints. Assuming that consumers maximize utility subject to their
incomes, one can derive a system of inverse demand functions for all
goods and services
P1 F1(Q1, Q2, . . ., Qn, Y)
Pn F1(Q1, Q2, . . ., Qn, Y)
(1)
where Pi and Qi are respectively the price and quantity of good i, i 1,
. . ., n, and Y is aggregate income. The Slutsky equation is assumed to
apply, so that (1) is integrable.
Assuming a set of well-behaved production functions
Qi Qi(K, W, D), i 1, . . ., n
(2)
one can link technology or development (D), climate (W), and other purchased inputs (K) into the production of outputs (Qi) by a firm on a certain
Environment and Development Economics 89
site. Cost minimization of the production function leads to a cost function,
Ci
Ci Ci(Qi, R, W, D)
(3)
given the prices of other inputs (R). In this analysis, it is helpful to separate
land from the other variable inputs. We assume that land, Li, is heterogeneous with characteristic W and has an annual cost or rent of pLW. In this
analysis, we are specifically interested in the climate that is tied to each
piece of property. Firms are assumed to maximize profits given market
prices
Max PiQi Ci(Qi, R, W, D) pLW Li(W)
Qi
(4)
where Pi is the price of good i. Profit maximization leads firms to equate
prices and marginal cost as well as determine cost-minimizing levels of
production. We assume that there is perfect competition for land, which
implies that entry and exit will drive pure profits to zero
PiQi Ci(Qi, R, W, D) pLW Li(W) 0
(5)
If producing good i is the best use for the land given the climate, technology, and factor prices, the observed market rent on the land will be
equal to the annual net profits from production of good i. Solving for the
value of land rent per hectare yields
pLE [PiQi Ci(Qi, R, W, D)]/Li(W)
(6)
The land rent should be equal to the net revenue from the land. Taking the
present value of this stream of revenue over time suggests that land value,
VLW is equal to the present value of the stream of future net revenue
VLE p
0
ert dt LW
[P Q C (Q , R, W, D)] e
0
i
i
i
i
rt/L (W)
i
dt
(7)
By examining the relationship between land value (or net revenue) and
climate, one can measure its impact on the present value of net revenue.
The essence of the Ricardian model is (6) and (7).
We now wish to explore the relationship between technology and
climate sensitivity. McKinsey and Evenson (1998) built a technology–
climate model to measure how the green revolution affected crops in India.
They find that the green revolution in India increased farm net revenue
substantially but that technology had a neutral impact on climate sensitivity. In this case, technology and climate were largely independent of
each other. McKinsey and Evenson argue that the green revolution in India
did not try to move crops to new climate zones, it merely attempted to
increase productivity on a site.
Even if new technology has not historically tried to move crops into new
climate zones, technology could affect climate sensitivity by changing the
production function. For example, suppose the production function for
crops is
Qi G(R) * (aW bW2*Dc) * H(D)
(8)
90
Robert Mendelsohn, Ariel Dinar, and Apurva Sanghi
where a, b, and c are parameters. G(R) measures the productivity of purchased inputs, H(D) is the direct productivity associated with
development, D, and (aW bW2*Dc) measures the interaction between
climate, W, and development. In general, dG/dK 0, d2G/dK2 0, dH/dT
0, and d2H/dT2 0. With respect to temperature, we expect that a 0
and b 0, so agriculture would exhibit a hill-shaped relationship with
respect to temperature. The issue in this paper is whether the parameter c
is greater or less than zero. If c 0, technology allows farmers to substitute
capital (or other inputs) for climate. Technology will increase productivity
in more marginal areas and will flatten the climate sensitivity function
(figure 1). More technology will result in climate change having a smaller
effect on farm net revenues. If c 0, new technologies will increase productivity in optimal climate locations more than in marginal locations. The
result will be a more steeply shaped climate response function (figure 2).
In this case, agriculture will get more sensitive to climate change with
increased development.
The relationship between development and climate sensitivity depends
upon whether new technology encourages capital to be a complement or a
substitute for climate. By examining Ricardian functions over time as the
level of technology increases and by examining Ricardian functions across
countries with different levels of technology, one can determine which
hypothesis is empirically correct.
There are many assumptions in the Ricardian approach. Perhaps the
strongest assumption is that output prices would remain constant as
climate changes. If this assumption of constant output prices is relaxed,
crops that face a supply increase would have falling prices and crops that
face a supply reduction would obtain higher prices. By failing to take these
price changes into account, the Ricardian climate response functions are
biased, underestimating damages and overestimating benefits (Cline,
1996). The bias, however, is likely to be small. For example, assuming that
an agricultural crop has a demand price inelasticity of 0.5, a supply price
elasticity of 0.5, and a quantity change of 25 per cent, the average bias in the
welfare estimate will be only 7 per cent (Mendelsohn and Nordhaus, 1996).
In a world where some crops may expand whereas others may contract,
the bias from holding prices constant consistently exaggerates the net benefits of change. For example, if warming increased corn production and
decreased wheat production, the constant price assumption would overstate the corn benefits and understate the wheat losses. As these two
welfare effects offset each other, the true welfare effect may be near zero
but the Ricardian model might predict a small net benefit. This could lead
to a large percentage error, but the absolute size of the welfare bias
remains small.
Another intriguing question plaguing all impact studies is whether they
should be modeling the world or just individual countries. Although it is
theoretically preferable to model the world, it is empirically demanding.
As a starting position, most agricultural studies have consequently limited
themselves to studying individual countries and have simply examined
alternative assumptions about international trade (Adams et al., 1995,
1998). A few studies have explored global agricultural models but they
Environment and Development Economics 91
have been forced to explore hypothetical impacts across countries (Reilly,
Hohmann, and Kane, 1994) or they have relied on limited economic processes (Rosenzweig and Parry, 1994). This study examines results on a
purely national scale and does not take into account global responses.
Cross-sectional techniques are also vulnerable to omitted data. If there
are important differences between one area and another which are not
observed by the analyst and these omitted variables are correlated with
climate, the analyst can reach biased conclusions. The cross-sectional
studies in this report attempt to account for important site factors, such as
soils and market access. However, inadequate data and oversight can lead
to some important variables being left out of the analysis. If the omitted
variables can be shown to be correlated with climate, the magnitude and
direction of the bias can be identified.
One variable that other analysts felt should be included in the Ricardian
analysis is irrigation (Cline, 1996; Darwin, 1999). The original Ricardian
study (Mendelsohn, Nordhaus, and Shaw, 1994) omitted irrigation intentionally. The Ricardian authors felt that irrigation is an endogenous
response to climate, not an exogenous factor that should be held constant.
However, in response to Darwin, Mendelsohn, and Nordhaus (1999)
include irrigation in a two-stage regression. As Darwin predicted, irrigation is largely a response to inadequate precipitation during the growing
season. However, Darwin was not correct that including irrigation would
result in agriculture being more temperature sensitive. The results showed
that the temperature sensitivity of agriculture in the United States did not
change when irrigation was included in a land-weighted Ricardian model.
This is an important result since irrigation is difficult to control completely.
Interestingly, the importance of precipitation increased with irrigation
included in the model. Accounting for the high cost of installing irrigation
reveals that low precipitation farms are less valuable.
The Ricardian method informs about climate sensitivity by comparing
one farm with another. The method is not dynamic. It studies climate
effects by comparing one equilibrium with another. The Ricardian method
itself does not reveal much about the dynamics of agriculture, and it
cannot answer questions such as how quickly farmers would adjust to
climate change. Evidence about the speed of agriculture adjustments is
clearer when examining how rapidly farmers adjust to new market conditions and new agricultural policies. Market evidence suggests that
agricultural systems tend to adjust rapidly, adapting to changes in prices
within a year or two.
Another criticism leveled at the Ricardian model concerns whether or
not adaptation costs are included (Quiggin and Horowitz, 1999). The
Ricardian model does include adaptation costs accounted for by the
market. For example, if a farmer shifts to a new crop in order to adapt to
climate change, the model does take account of any changes in net revenues. If the new crop requires more inputs, for example, this would be
accounted for. The model, however, would not take into account capital
equipment abandoned to make an adjustment to a new crop. For example,
if a farmer shifted from wheat to corn, farm machinery that could only be
used on wheat would have to be abandoned. If this machinery lasted many
92
Robert Mendelsohn, Ariel Dinar, and Apurva Sanghi
decades, there could be substantial adjustment costs associated with rapid
responses. However, in practice, most farm machinery has an expected
lifetime of only five to ten years. Farmers can consequently adjust their
capital stock frequently over a century at virtually no cost. We consequently believe that adjustment costs are not likely to be an important
factor explaining the effect of a slow-moving change, such as climate
change, on a versatile sector such as agriculture.
3. Empirical analysis
We engage in several analyses to test the relationship between development and climate sensitivity. First, we use a panel of districts in India to
explore what happens to the annual climate response function as technology increases, between 1966 and 1986. Second, we examine a panel of
municipios in Brazil to test whether the climate response function in Brazil
has shifted over time. Third, we compare the climate response functions
estimated in India and the United States. We do not include Brazil in these
comparisons because the climate measurements for Brazil are not consistent with the Indian and American data. To illustrate the differences found
in these models, we use the Indian and American models to predict the
outcome of warming in both the United States and India.
In order to appreciate the data upon which the analysis rests, we present
the range of climates observed for each country in table 3. The temperature
range for each country is large compared to temperature changes predicted by warming. The difference between the United States and the two
tropical countries is also large, especially in winter. These three countries
represent two distinct bands in the temperature range of the earth, the temperate and the tropical. Fitting the climate response function to this wide
range of observed temperatures suggests the results would be applicable
even in a severe climate change scenario. However, one must be careful
not to extrapolate the results beyond the observed range of the data. For
example, the positive squared terms in some of the Indian regressions
imply that the temperature response function is convex for India. It is not
correct to extrapolate that therefore extreme hot temperatures would be
good for India because these temperatures would put India on the rising
part of that function.
Table 3. Country climates
Winter
Country
Low
Summer
Mean
High
Low
Mean
High
Temperature (C)
Brazil
7.2
India
9.2
US
218.1
20.0
18.3
20.2
32.8
25.3
20.3
15.9
21.9
13.8
24.4
27.9
24.3
33.4
32.5
33.9
Precipitation (mm/mo)
Brazil
0.1
India
0.3
US
6.1
73.3
18.4
66.5
426.9
76.0
355.6
1.1
0.0
0.0
173.4
307.0
92.2
495.8
916.9
233.7
Environment and Development Economics 93
Several cross-section panels are used in this analysis. There are 21
annual cross sections of net revenue in India from 271 districts (Dinar et al.,
1998). The first analysis breaks this Indian data set into two parts
(1966–1976) and (1977–1986) and compares results of early versus later
years. In Brazil, there are four cross sections of property values from 3,941
municipios collected in Censuses taken in 1970, 1975, 1980, and 1985
(Sanghi, 1998). The second analysis compares the Brazilian results for 1970
and 1975 with the results for 1980 and 1985. The third analysis compares
property values in India and the United States. Property values are
approximated for the Indian data set using the present value of annual net
revenues from 1966 to 1986. In the United States, we rely on a single cross
section of property values in 1,900 countries taken in 1983 (see
Mendelsohn, Nordhaus, and Shaw, 1994). The final analysis compares predictions using the coefficients from the Indian and American regressions.
Using both models, we predict warming results for both India and the
United States.
Panel data analysis
We use a semi-log quadratic functional form to explain Indian net revenue,
NR
Log(NR) ai Ci bi Ci2 dj Xj e
(9)
where Ci are climate variables and Xj are control variables such as soils and
economic variables (see Sanghi, Mendelsohn, and Dinar, 1998). The quadratic form tests a hill-shaped response function to the climate variables.
The regression is estimated using WLS, weighted by the number of
hectares of cultivated land in each district. The coefficients from the
regression are presented in table 4.
The means have been subtracted from the climate variables in (9). The
linear coefficients consequently represent the marginal effect of each
climate variable evaluated at the mean of Indian climate. The coefficients
of both temperature and precipitation are different in the 1966–1975 versus
1977–1986 periods for India. The sum of the marginal temperature coefficients across seasons is 0.272 in the earlier period and 0.081 in the later
period. The difference in the temperature coefficients is statistically significant (F 7.38). The sum of the marginal precipitation coefficients
increases over this period from 6.77e-3 to 11.38e-3. The precipitation
coefficients in the two periods are also significantly different (F 30.10).
The intertemporal Indian data set thus provides support for the proposition that development has reduced India’s climate sensitivity over time.
The remaining studies analyze property values. In all cases, we rely on
a log-linear functional form. Time dummy variables control for socioeconomic factors that might vary over time as well as for inflation. Given the
hyperinflation in Brazil in the eighties, inflation is a difficult factor to
control for explicitly. Just subtle distinctions such as the month of the
survey could alter values by several percentage points. The regressions
were weighted by the number of hectares in cultivation.
With the Brazilian analysis, the climate variables for 1970 and 1975 were
allowed to differ from the climate coefficients for 1980 and 1985. These two
94 Robert Mendelsohn, Ariel Dinar, and Apurva Sanghi
Table 4. Indian annual regression
Temperature
January
Jan Sq.
April
Apr Sq.
July
July Sq.
October
Oct. Sq.
Precipitation
1966–1975
1977–1986
1966–1975
1977–1986
9.41
(0.23)
2.85
(0.92)
144.
(5.11)
16.4
(3.30)
121.
(2.65)
23.3
(4.57)
15.8
(0.28)
21.6
(1.78)
32.0
(0.79)
11.4
(3.62)
102.
(3.58)
8.81
(1.75)
299.
(6.43)
6.04
(1.16)
353.
(5.97)
12.9
(1.05)
10.0
(4.01)
0.076
(1.06)
2.30
(1.28)
0.055
(2.67)
1.40
(7.94)
0.003
(9.49)
6.94
(8.45)
0.030
(5.53)
18.5
(7.30)
0.227
(3.13)
9.81
(5.40)
0.079
(3.61)
1.16
(6.33)
0.002
(5.64)
3.85
(4.70)
0.005
(0.92)
Control variables
1966–1976
5691.
(15.78)
Literacy
119.
(11.50)
Bulls/ha
106.
(3.84)
Soils1
303.
(15.05)
Soils4
26.6
(0.95)
R2
0.93
1977–1986
Latitude
Tractors/ha
Soils2
Soils5
5189.
(14.42)
49.0
(3.19)
2095.
(0.79)
193.
(8.06)
89.5
(2.24)
N
Popden
High Yield
#hectares
Soils3
Soils6
28.1
(8.10)
4.86
(0.08)
146.
(9.27)
123.
(3.58)
103.
(2.22)
5624
a t-statistic
in parenthesis. Dependent variable is the log of real annual
Note:
net revenue. The mean has been subtracted from each climate variable.
Coefficients multiplied by 1,000.
sets of climate coefficients are compared in table 5 along with the
remaining variables used in the regression to control for soils and economic variables. The means of the climate variables have been subtracted
from the climate variables. The sum of the marginal temperature coefficients is equal to 50.5 in the earlier period and 43.2 in the later period.
The sum of the marginal precipitation coefficients is equal to 1.5 in the
early period and 1.9 in the later period. The temperature coefficients and
the precipitation coefficients were not significantly different. The results
suggest that development in Brazil has reduced climate sensitivity over
time slightly.
In both the Indian and Brazilian regressions, there were also some
changes in the coefficients on the squared climate terms. In India, the temperature and precipitation squared terms tended to become less negative,
Environment and Development Economics 95
Table 5. Brazilian panel regression
Temperature
March
Mar Sq.
June
Jun Sq.
September
Sep Sq.
December
Dec Sq.
1970–1975
1980–1985
1970–1975
1980–1985
99.2
(3.95)
9.47
(3.43)
131.
(9.38)
17.4
(9.15)
146.
(8.89)
36.8
(21.97)
164.
(6.34)
2.35
(0.68)
90.0
(4.31)
7.13
(2.99)
123
(11.88)
12.5
(8.15)
172.
(12.77)
36.1
(25.85)
182.
(8.43)
14.2
(5.05)
2.07
(6.43)
0.24
(0.19)
2.95
(6.51)
3.32
(1.52)
0.51
(0.70)
47.5
(9.31)
3.02
(10.92)
4.20
(2.91)
2.44
(9.13)
4.37
(4.42)
1.72
(4.91)
0.22
(0.12)
0.92
(1.64)
52.3
(12.35)
2.75
(11.53)
0.02
(0.02)
Control variables
Soils
28.1
(1.54)
Soils2
140.
(7.26)
Soils3
422.
(14.61)
Soils4
71.7
(1.91)
Soils5
479.
(7.45)
Soils6
408.
(7.45)
R2
Precipitation
0.76
Soils7
Soils8
Soils9
Soils10
Soils11
Soils12
N
143.
(4.09)
275.
(11.70)
125.
(5.76)
714.
(38.01)
553.
(26.81)
587.
(13.05)
Latitude
Population
Pop sq
Yr70
Yr75
Yr80
113.
(28.69)
0.39
(18.87)
1.16 e-5
(17.93)
1506.
(43.50)
285.
(8.30)
14.5
(1.06)
14827.
Note: t-statistic in parenthesis. Dependent variable is the log of property
values per hectare. The mean has been subtracted from each climate variable.
Coefficients are multiplied by 1,000.
implying a flattening of the climate sensitivity function. In Brazil, there
were only two significant changes. The December temperature squared
term became more positive and the March precipitation squared term
became more negative.
Cross-sectional analysis
We now compare the evidence between the United States and India. The
regression for the United States involves just one year of data, 1983. The
annual net revenue data from India for 1966 to 1986 is used to compute a
net present value for each district (assuming a 5 per cent interest rate).
The results of the United States regression are shown in table 6 and the
results for the Indian regression are shown in table 7. The climate variables
96
Robert Mendelsohn, Ariel Dinar, and Apurva Sanghi
Table 6. United States cross-sectional regression
January
Jan Sq.
April
Apr Sq.
Temperature
Precipitation
85.9
(5.31)
1.04
(1.52)
54.0
(1.16)
3.71
(2.03)
2.33
(2.03)
0.69
(0.19)
1.07
(0.40)
16.2
(3.51)
Control variables
Population
Pop sq.
Latitude
Salinity
Flood prone
Constant
R2
2.18
(24.98)
5.35 e-4
(13.60)
10.2
(3.69)
2463.
(12.97)
213.
(3.52)
5804
(6.63)
.98
Temperature
Precipitation
6.31
(3.99)
28.2
(3.51)
5.88
(2.06)
74.0
(4.08)
July
127.
(1.55)
July Sq. 4.77
(2.60)
October 25.7
(0.35)
Oct Sq.
7.19
(2.74)
Water capacity
5.21 e-2
(1.83)
792.
(5.30)
426.
(1.98)
24.6
(3.80)
13.7
(0.21)
55.6
(2.14)
Wetland
Erosion
Slope length
Sand
Clay
N
2938
Note: Dependent variable is log of property value. The coefficients are
multiplied by 1000. t-statistics are in parenthesis.
Table 7. Indian cross-sectional regression
Temperature Precipitation
January
Jan Sq.
April
Apr Sq.
151.
(0.43)
3.52
(0.45)
408.
(0.61)
9.11
(0.82)
Control variables
Literacy
1150.
(3.11)
High Yield
202.
(0.61)
Tractors
3070.
(0.24)
Constant
9120.
(0.79)
R2
.38
8.57
(0.82)
0.21
(1.15)
8.04
(1.12)
0.04
(0.65)
Temperature Precipitation
July
1980.
(3.08)
July Sq.
37.0
(3.25)
October 1050.
(0.75)
Oct Sq.
21.6
(0.82)
Hectares
Soils1
Soils2
N
0.27
(0.32)
0.00
(0.33)
6.31
(1.63)
0.01
(0.54)
203.
(2.16)
204.
(3.18)
150.
(1.84)
269
Note: Dependent variable is log of present value of net revenues. The
coefficients are multiplied by 1000. t-statistics are in parenthesis.
Environment and Development Economics 97
have not been transformed in these regressions so that results from one
country can be compared with the other. In order to understand what the
complex set of coefficients implies for temperature, we compare a range of
warming scenarios. For each temperature change scenario, we examine
what each regression implies for both countries. In order to make the
results comparable across countries, we express the impacts as a percentage of the original agricultural value. Applying both climate response
functions to India reveals that the Indian climate function is more sensitive
than the American climate function (figure 3). A 2C warming implies
damage of 6.6 per cent according to the American function but damage of
36.0 per cent according to the Indian function. A similar result is found for
the United States (figure 4). The American climate response function
Figure 3 Warming impact to India—effect of temperature on farm value
Figure 4 Warming impact to United States—effect of temperature on farm value
98
Robert Mendelsohn, Ariel Dinar, and Apurva Sanghi
implies a 2C warming would result in damages of 1.2 per cent whereas
the Indian climate response function predicts damages of 40.4 per cent.
The estimates in figures 3 and 4 take only the temperature change into
account and do not factor in the effects of precipitation or carbon dioxide.
If one assumes that precipitation will increase by 8 per cent and that
carbon fertilization will increase productivity by 30 per cent, the net effect
of a 2C warming in India would be a benefit of 23 per cent according to
the US function and no net effect according to the Indian function. The
effects in the United States would be a benefit of 29 per cent according to
the American function and a loss of 20 per cent according to the Indian
function. The Indian climate response function is far more temperature
sensitive than the American climate response function.
4. Conclusion and policy implications
This paper analyzes the link between development and climate sensitivity.
The paper begins by developing a theoretical model of agriculture to
understand how development could affect climate sensitivity. Technology
could encourage capital to either substitute or complement climate in agricultural production. If capital becomes more of a substitute (complement)
for climate with new technology, development will reduce (increase)
climate sensitivity. The effect of development on climate sensitivity is
therefore an empirical question.
Several empirical analyses are conducted to measure the interaction
between development and climate using the Ricardian approach. Two of
the studies rely on intertemporal comparisons following agricultural
climate sensitivity over a long time period in India and Brazil. Development has increased and climate sensitivity has fallen in both countries
over time. The analysis consequently suggests that development may
reduce climate sensitivity. However, this intertemporal test is an indirect
measurement. The analysis does not directly link development to a change
in climate sensitivity. It remains possible that other factors that also
changed monotonically over time could have caused climate sensitivity to
change.
The paper also examines cross-country effects. This analysis compares
the climate sensitivity between India and United States. The study reveals
that the Ricardian function for Indian agriculture is far more climate sensitive than the American Ricardian function. These differences are
significant and substantial. The predicted effect from warming is much
more severe using the Indian estimated coefficients compared to the
American coefficients.
These results suggest that development does have an important effect on
climate sensitivity. Farmers in less-developed countries are currently more
climate sensitive than the farmers in more-developed countries. Combining this climate sensitivity with the higher temperatures at low
latitudes suggests that low-latitude agriculture would be hurt if warming
occurred today. The results however provide a slightly more optimistic
picture if warming occurs far into the future. If developing countries can
continue to improve their agricultural systems, they are likely to be less
climate sensitive when warming actually occurs. With continued develop-
Environment and Development Economics 99
ment, global warming has an ambiguous effect on farmers in low-latitude
countries.
These results have important implications for policy. As countries continue to develop, future damages from warming are likely to be smaller
than they would have been if the climate changed today. Since most
warming scenarios imply climate changes primarily in the latter half of the
next century, damages in developing countries may not be as large as first
thought. The quadratic nature of the climate response functions also
implies that warming will cause more severe damages if temperatures rise
more than currently expected.
As the likelihood of global warming increases, adaptation is becoming
an increasingly important strategy. Farmers in developing countries are
more vulnerable to climate impacts precisely because they have less
capital-intensive technologies and management practices. The international community could respond to climate change by helping
developing countries adopt more modern farming techniques. Not only
will this have direct benefits to current farmers, but also it will likely
reduce their climate sensitivity. The international community can also help
by conducting empirical studies of impacts in developing countries. There
have been an inadequate number of studies done on developing countries
to date. Additional impact studies in developing countries are badly
needed to understand the consequences of climate changes around the
world. Most conspicuously, there have been almost no empirical studies in
Africa despite it being one of the most vulnerable regions in the world.
National governments can explore two different roles. First, they can
support public research and development into modern technologies. Even if
this effort were climate neutral, it would help private farmers develop more
options for handling climate change. Further, some research could directly
explore augmenting new management practices and varieties of crops to
cope with warming. Since climate change is a relatively slow process, countries can gradually invest in modern approaches to meet this need.
Governments could also address the efficiency of their agricultural
sectors. By reducing subsidies and removing barriers for adaptation, governments could encourage stronger healthier agricultural sectors. As
certain crops become less competitive with warming and new crops
become feasible, governments should encourage farmers to adjust to new
market and ecological conditions. For example, governments could
provide local farmers with good predictions of future climate change as
well as information about appropriate adaptations.
References
Adams, R.M., R. Fleming, B.A. McCarl, and C. Rosenzweig (1995), ‘A reassessment
of the economic effects of climate change on US agriculture’, Climatic Change, 30:
147–167.
Adams, R., B. McCarl, K. Segerson, C. Rosenzweig, K. Bryant, B. Dixon, R. Conner,
R. Evenson, and D. Ojima (1998), ‘The economic effect of climate change on US
agriculture’, in R. Mendelsohn and J. Neumann (eds.), The Economic Impact of
Climate Change on the United States Economy, Cambridge: Cambridge University
Press.
100
Robert Mendelsohn, Ariel Dinar, and Apurva Sanghi
Cline, W. (1996), ‘The impact of global warming on agriculture: a Ricardian
Analysis: Comment.’ American Economic Review, 86: 1308–1311.
Darwin, R. (1999), ‘The impact of global warming on agriculture: a Ricardian
analysis: Comment’, American Economic Review, 89: 1049–1052.
Dinar, A., R. Mendelsohn, R. Evenson, J. Parikh, A. Sanghi, K. Kumar, J. McKinsey,
and S. Lonergan (1998), ‘Measuring the impact of climate change on Indian agriculture’, World Bank Technical Paper No. 402, Washington, DC.
Hayami, Y. and V. Ruttan (1985), Agricultural Development, An International
Perspective, Baltimore: Johns Hopkins University Press.
Hicks, N. and P. Streeten (1979), ‘Indicators of development: the search for a basic
needs yardstick’, World Development, 7: 567–580.
McKinsey, J. and R. Evenson (1998), ‘Technology–climate interactions: was the
green revolution in India climate friendly?’, in A. Dinar, R. Mendelsohn, R.
Evenson, J. Parikh, A. Sanghi, K. Kumar, J. McKinsey, and S. Lonergan,
‘Measuring the impact of climate change on Indian agriculture’, World Bank
Technical Paper No. 402, Washington, DC.
Mendelsohn, W., W. Nordhaus, and D. Shaw (1994), ‘The impact of global warming
on agriculture: a Ricardian analysis’, American Economy Review, 84: 753–771.
Mendelsohn, R. and W. Nordhaus (1996), ‘The impact of global warming on agriculture: reply’, American Economic Review, 86: 1312–1315.
Mendelsohn, R. and J. Neumann (eds.) (1998), The Impact of Climate Change on the
Economy of the United States, Cambridge: Cambridge University Press.
Mendelsohn, R. and W. Nordhaus (1999), ‘The impact of global warming on agriculture: reply’, American Economic Review, 89: 1053–1055.
Mendelsohn, R. and M. Schlesinger (1999), ‘Climate response functions’, Ambio, 28:
362–366.
Mendelsohn, R., W. Morrison, N. Andronova, and M. Schlesinger (2000), ‘Countryspecific market impacts from climate change’, Climatic Change 45: 553–569.
Nafziger, W.E. (1990), The Economics of Developing Countries, 2nd edition, Englewood Cliffs: Prentice Hall.
Nordhaus, W. (1991), ‘Too slow or not too slow: the economics of the greenhouse
effect’, Economic Journal, 101: 920–937.
Pearce, D., W. Cline, A. Achanta, S. Fankhauser, R. Pachauri, R. Tol, and P. Vellinga
(1996), ‘The social cost of climate change: greenhouse damage and the benefits of
control’, in J. Bruce, H. Lee, and E. Haites (eds.), Climate Change 1995: Economic
and Social Dimensions of Climate Change, Cambridge: Cambridge University Press.
Quiggin, J. and J. Horowitz (1999), ‘The impact of global warming on agriculture: a
Ricardian analysis: Comment’, American Economic Review (forthcoming).
Reilly, J., N. Hohmann, and S. Kane (1994), ‘Climate change and agricultural trade:
who benefits and who loses?’, Global Environmental Change, 4: 24–36.
Rosenzweig, C. and M. Parry (1994), ‘Potential impacts of climate change on world
agriculture’, Nature, 367: 133–138.
Sanghi, A. (1998), ‘Global warming and climate sensitivity: Brazilian and Indian
agriculture’, Ph.D. dissertation, Department of Economics, University of
Chicago.
Sanghi, A., R. Mendelsohn, and A. Dinar (1998), ‘The climate sensitivity of Indian
agriculture’, in A. Dinar, R. Mendelsohn, R. Evenson, J. Parikh, A. Sanghi, K.
Kumar, J. McKinsey, and S. Lonergan, ‘Measuring the impact of climate change
on Indian agriculture’, World Bank Technical Paper No. 402, Washington, DC.
Schelling, T. (1992), ‘Some economics of global warming’, American Economic
Review, 82: 1–14.
Watson, R., M. Zinyowera, R. Moss, and D. Dokken (1996), Climate Change 1995:
Impacts Adaptations, and Mitigation of Climate Change. Scientific-Technical Analysis.
Cambridge: Cambridge University Press.
Environment and Development Economics 101
World Bank (1988), World Development Report, Washington, DC: World Bank.
World Bank (1993), World Development Report, Washington, DC: World Bank.
World Bank (1997), World Development Report, Washington, DC: World Bank.
World Bank (1998), World Development Report, Washington, DC: World Bank.
World Resources Institute (1996), World Resources, 1996, New York: Oxford University Press.