Global Environmental Change 21 (2011) 701710

Contents lists available at ScienceDirect

Global Environmental Change

journal homepage: www.elsevier.com/locate/gloenvcha

Ethiopias growth prospects in a changing climate: A stochastic general

equilibrium approach
Channing Arndt a, Sherman Robinson b, Dirk Willenbockel b,*
a
Development Economics Research Group, Department of Economics, University of Copenhagen, DK-1455 Copenhagen K, Denmark
b
Institute of Development Studies, University of Sussex, Brighton BN1 9RE, UK

A R T I C L E I N F O A B S T R A C T

Article history: This study develops a stochastic economy-wide framework for analyzing economic impacts from
Received 15 June 2009 climate change and potential adaptation policies. For the stochastic analysis, particular attention is paid
Received in revised form 16 November 2010 to the development of a prior subjective distribution of future climate outcomes. The approach is applied
Accepted 18 November 2010
to Ethiopia. The results highlight the importance of cumulative processes and rates of growth. In
Available online 14 December 2010
particular, if climate change affects the rate of technical change and the rate of accumulation of capital,
the implications become signicant over time. Furthermore, analysis of the variability of the
Keywords:
components of GDP indicates that aggregate consumption always has a higher coefcient of variation
Climate change
than the other macro aggregates. The burden of adjustment appears to fall more heavily on consumers.
Sub-Saharan Africa
Computable general equilibrium analysis Poor consumers are likely to experience increased vulnerability.
Growth dynamics 2010 Elsevier Ltd. All rights reserved.
Development

1. Introduction facilitate the natural decline in the importance of agriculture as a

share of GDP thus reducing overall exposure to climate change by
It is now widely recognized that developing countries and in shrinking, at least in relative terms, a climate dependent sector?
particular low-income countries in tropical and sub-tropical Given the importance of climate sensitive sectors in developing
regions will be disproportionally affected by the adverse impacts countries, it is clear that effective adaptation programs need to be
of climate change. The combination of exposure to an already comprehensive and integrated with overarching national strate-
fragile environment, dominance of climate-sensitive sectors in gies. Furthermore, because the effects of climate change and the
economic activity and low autonomous adaptive capacity in these nature of adaptation options will vary across countries, adaptation
regions entail a high vulnerability to the harmful effects of global programs must take into account country-specic circumstances.
warming on agricultural production and food security, water This paper seeks to contribute an approach to evaluating
resources, human health, physical infrastructure and ecosystems. growth prospects in the context of climate change. We modify and
Recent authoritative scientic assessments emphasize that even extend a dynamic single country computable general equilibrium
under the most optimistic assumptions about the success of future (CGE) model to include stochastic elements that are characteristic
global mitigation action, an acceleration of adaptation efforts in of climate change. We apply the model to Ethiopia as a case
developing countries over the next decades is essential to build country. Ethiopia is heavily dependent on rain-fed agriculture, and
resilience and reduce damage costs (IPCC, 2007; Adger et al., 2007; its geographical location and topography in combination with low
Stern, 2007). adaptive capacity entail a high vulnerability to adverse impacts of
Two critical questions facing developing country policy makers climate change. Regional projections of climate models not only
are what exactly one should do and when should one do it? The predict a substantial rise in mean temperatures over the 21st
answers to these questions are not immediately clear. For example, century but also suggest an increase in rainfall variability with a
supposing that the likely impacts on agriculture are negative, rising frequency of both extreme ooding and droughts due to
should one invest in agriculture in order to combat the negative global warming. In contrast to existing comparative-static CGE
impacts of climate change or should one direct investment so as to assessments of climate impacts on agricultural production,1 the
dynamic approach presented here is able to capture the

* Corresponding author. Tel.: +44 1273 915700; fax: +44 1273 621202.
E-mail addresses: [email protected] (C. Arndt), [email protected]
1
(S. Robinson), [email protected] (D. Willenbockel). See, e.g., Mideska (2010), Hertel et al. (2010) and Bosello and Zhang (2005).

0959-3780/$ see front matter 2010 Elsevier Ltd. All rights reserved.
doi:10.1016/j.gloenvcha.2010.11.004
702 C. Arndt et al. / Global Environmental Change 21 (2011) 701710

intertemporal links between climate change and economic growth 3. Projections of long-run climate change for East Africa and
emphasized by Fankhauser and Tol (2005). In contrast to Ethiopia
respective global dynamic CGE-based assessments e.g., Reilly
et al. (2007) and the studies reviewed in Reilly and Willenbockel Table 1 summarizes long-run temperature and rainfall projec-
(2010) the single-country approach of the present study allows a tions for East Africa from IPCC (2007). An increase in annual mean
more detailed representation of the agricultural sector. A key area rainfall in East Africa is considered likely and is a robust result, with
explored is the role of uncertainty. 18 of 21 global circulation models (GCMs) models projecting an
The article is organized as follows. Section 2 provides a brief increase in the core of this region, east of the Great Lakes (IPCC, p.
overview of the Ethiopian economy. Section 3 summarizes 869). However, Ethiopia is not part of this core region and a closer
medium- and long-run projections of climate change for East look at country-specic projections reveals a less clear-cut picture,
Africa and Ethiopia including existing region-specic estimates of particularly with respect to precipitation. Table 2 reports projec-
climate change impacts on agricultural productivity. Section 4 tions for Ethiopia from a subset of ve higher-resolution GCMs for
provides an overview of the dynamic CGE model and outlines the two alternative SRES scenarios. While three models predict notable
simulation methodology including the treatment of uncertainty. long-run increases in average monthly precipitation, one model
Section 5 presents the results of the dynamic stochastic climate generates virtually no change and one model simulates a
change impact analysis. A nal section concludes. substantial decline.
For the prediction of agricultural productivity impacts of
2. Overview of the Ethiopian economy climate change, a seasonal breakdown of these long-period
averages is required, given that the reported gures may mask
Ethiopia is the country with the second largest population in signicant seasonal variability in predicted climate change
Sub-Sahara Africa (SSA), and, despite rapid economic growth over impacts. As Table 2 shows, this is indeed the case for the predicted
the past ve years, one of the poorest. The 2006/07 share of impacts of global warming on precipitation in Ethiopia from high-
agriculture in GDP is 46 percent while industry accounts for 13 resolution GCMs. The predicted decline in JuneSeptember rainfall
percent and services for 41 percent (CSA, 2008). Agriculture is even in Table 2 appears to be consistent with already observable trends
more dominant in terms of employment with 80 percent of the for parts of Ethiopia (Cheung et al., 2008; Seleshi and Zanke, 2004).
working population engaged in the sector. The main export is Funk et al. (2008) report a declining trend for growing-season
coffee with a share of 35.7 percent in total merchandise exports precipitation in crop-growing areas of Ethiopia and other Eastern
2006/07 (IMF, 2008). Ethiopia is a net importer of wheat. African countries over the period 19792005 and link this
Petroleum, coal and gas are only imported, capital goods have observation to warming of the Indian Ocean through mechanisms
very high import shares, and a number of consumer goods also not adequately captured by current GCMs. The study concludes
have high import shares. that recent climate change impact assessments, based on
The current government has explicitly instituted an agricul- optimistic precipitation simulations over Eastern Africa, may
ture development led industrialization (ADLI) policy. Under the underestimate yield reductions.
ADLI framework, a number of policy actions were taken to Warren et al. (2006) notes that the HadCM3 GCM, among
enhance agricultural productivity and ensure food security. others, predicts a dramatic increase in natural climate variability,
Despite these efforts, fast rural population growth in combination to the extent that global and regional temperatures will uctuate
with other factors detailed below has led to severe environmental on an annual basis over a range equal to more than half the
degradation. Deforestation and, soil erosion have increasingly increase in temperature predicted over the next 80 years. Challinor
become more pronounced over the past decade, in part due to the et al. (2007) consider extremes of temperature, and point out how
poor performance of other sectors to absorb the growing only a few days of high temperatures near owering in wheat,
population. groundnut and soybean can drastically reduce yield.
The agricultural sector is dominated by mixed rain fed small- Given the strong dependence of Ethiopias economy on rain-fed
scale farming based on traditional technologies. Small-scale agriculture, the dominant channel through which medium- and
subsistence farming accounts for 95 percent of the total area long-run changes in the means and variances of climatic variables
under crops and for more than 90 percent of the total agricultural affect economic performance will be the link from climate to
output. Production technologies are predominantly characterized agricultural yields. One method for estimating the agricultural
by the use of traditional farm implements, minimal application of yield impacts of climate change is the Ricardian approach
fertilizer and pesticides, weak extension services, and low use of pioneered by Mendelsohn et al. (1994), which provides cross-
improved seeds (Deressa, 2006). Only about 13 percent of section regressions of land values or net revenue against climate
potentially irrigable land is irrigated. and other exogenous characteristics.

Table 1
Projected increase in temperature and precipitation in East Africa for the period 20802099 in relation to the period 19801999.

Season Temperature response (8C) Precipitation response (percent)

Min 25 percent Median 75 percent Max Min 25 percent Median 75 percent Max

DJF 2.0 2.6 3.1 3.4 4.2 3 6 13 16 33

MAM 1.7 2.7 3.2 3.5 4.5 9 2 6 9 20
JJA 1.6 2.7 3.4 3.6 4.7 18 2 4 7 16
SON 1.9 2.6 3.1 3.6 4.3 10 3 7 13 38

Annual 1.8 2.5 3.2 3.4 4.3 3 2 7 11 25

Source: Christensen et al. (2007), Tab 11.1.

Notes: The gures show averages of temperature and precipitation projections from a set of 21 global models for the A1B scenario. The mean temperature and precipitation
responses are rst averaged for each model over all available realizations of the 19801999 period from the 20th Century Climate in Coupled Models (20C3M) simulations and
the 20802099 period of A1B. Computing the difference between these two periods, the table shows the minimum, maximum, median (50 percent), and 25 and 75 percent
quartile values among the 21 models for temperature (8C) and precipitation (percent) change. Seasons: DJF: December, January, February, etc. East Africa is dened as a grid
cell extending from 12S 22E to 18N 52E.
 C. Arndt et al. / Global Environmental Change 21 (2011) 701710 703

Table 2
GCM predictions of climate change in Ethiopia by month and season.

Average monthly and seasonal temperature (8C)

19611990 20702099 Change 19611990 20702099 Change Percent

Feb 22.62 26.29 3.67 Spring 23.66 27.41 3.75 15.9

Mar 23.93 27.65 3.72
Apr 24.42 28.29 3.87
May 24.45 28.31 3.86 Summer 23.98 28.02 4.04 16.8
Jun 24.21 28.40 4.19
Jul 23.28 27.34 4.06
Aug 23.04 26.93 3.89 Fall 23.11 27.04 3.93 17.0
Sep 23.38 27.31 3.93
Oct 22.92 26.88 3.96
Nov 21.95 25.58 3.63 Winter 21.58 25.20 3.62 16.8
Dec 21.28 24.85 3.57
Jan 21.52 25.18 3.66

Average monthly precipitation by month and season (mm per month)

19611990 20702099 Change 19611990 20702099 Change Percent

Feb 17.08 24.64 7.56 Spring 44.99 47.84 2.85 6.3

Mar 37.20 39.68 2.48
Apr 80.70 79.20 1.50
May 89.28 72.85 16.43 Summer 93.58 74.46 19.13 20.4
Jun 68.40 41.40 27.00
Jul 123.07 109.12 13.95
Aug 127.10 117.18 9.92 Fall 92.98 87.77 5.21 5.6
Sep 85.50 70.80 14.70
Oct 66.34 75.33 8.99
Nov 31.50 51.60 20.10 Winter 17.63 30.22 12.59 71.4
Dec 10.54 18.91 8.37
Jan 10.85 20.15 9.30

Note: Own calculations based on Cline (2007: Tables H1H4. The predictions for 20702099 are averages of the results from the six GCMs (ECHAM4/OPYC3, HadCM3, CSIRO-
Mk2, CGCM2, GFDL-R30 and CCSR/NIES) available from the IPCC Data Distribution Center for the SRES A2 scenario. The delineation of seasons follows Kurukulasuriya et al.
(2006).

For present purposes, the most relevant study of this type is clearly visible. The issue of CO2 fertilization is subject to notorious
Kurukulasuriya et al. (2006) which uses survey data for around debate. Recent free-air carbon enrichment (FACE) studies have
9000 farms across 11 African countries including Ethiopia. The called into question the results of earlier closed-laboratory
study takes account of non-linearities in the relationship between experiments that suggest the presence of strong positive
agricultural production and climate by including quadratic forms productivity effects for major crops due to higher CO2 concentra-
for precipitation and temperature by season. When the estimated tion levels (Long et al., 2006). Yet, the validity of FACE results has
Ricardian model from this study is linked with interpolations of the likewise been questioned (Tubiello et al., 2007), and the debate
GCM-based predicted seasonal changes in rainfall and temperature appears to be unresolved at present.
from Table 2, the resulting agricultural productivity effect for As Semenov and Porter (1995) point out, non-linearity of crop
dryland crops is on the order of 7.4 percent over the period 2001 responses entails the necessity to preserve the variability of
2030. weather sequences to estimate the effect of climate on agricultural
An alternative approach to the prediction of long run climate production and to assess agricultural risk. In their study, the
change impacts on agricultural productivity links the temperature authors couple a crop simulation model for wheat with a stochastic
and precipitation estimates from GCMs to process-based crop weather generator. This approach allows changes not only in mean
models that capture the main physiological processes responsible values but also in the variance or type of distribution for climate
for plant growth in a stylized form and enable the prediction of variables. Their results indicate that changes in climatic variability
yield as a function of climate variables, soil type, moisture and crop can have a more profound effect on yield and its associated risk
management practices. The yield effects for four major crop types than changes in mean climate. Porter and Semenov (2005) cite
in an African sub-region including Ethiopia in Table 3 are drawn simulation results for wheat, in which a doubling of the standard
from Rosenzweig and Iglesias (2006). The sensitivity to the deviation of temperature is estimated to generate the same
assumed presence or absence of positive CO2 fertilization effects is decrease in yield as a 4 8C increase in mean temperature.

Table 3
Crop model predictions (yield change in percent relative to 1980).

Yield changes in percent (base 1980) All Soy beans Coarse grains Ricea Wheat

No CO2 fertilization
2020 (475 ppm) 9 10 10 8 8
2050 (574 ppm) 17 18 18 16 16
2080 (712 ppm) 28 28 28 28 28
With CO2 fertilization
2020 (475 ppm) 4 2 6 4 2
2050 (574 ppm) 6 1 14 5 4
2080 (712 ppm) 13 5 22 12 11

Notes: Estimates are for Africa low-income calorie exporters including Ethiopia, Sudan, Mozambique, Uganda, Benin, Gambia and Togo. For methodological details see
Rosenzweig and Iglesias (2006).
a
Not produced in Ethiopia.
704 C. Arndt et al. / Global Environmental Change 21 (2011) 701710

4. Simulation methodology Table 4

Comparison of weather variation impacts across modelling approaches.

4.1. The CGE model framework Crop model Regression model

Std dev. Min Max Std dev. Min Max

The simulation approach starts from a recursive-dynamic
Barley 0.054 0.121 0.104 0.163 0.253 0.592
version of the standard single-country CGE model of Lofgren Maize 0.105 0.278 0.349 0.236 0.449 0.626
et al. (2002) and extends it for the stochastic analysis of climate Oilseeds 0.073 0.152 0.249 0.114 0.239 0.360
change impacts. For each time period, the model gives a Wheat 0.039 0.114 0.077 0.130 0.272 0.312
comprehensive, internally consistent account of decisions and
related payments involving production, consumption by house-
holds and the government, private and public investment, trade, commodity basis. However, the index of aggregate agricultural
taxation, and transfers between households, government and the production, as well as the components food, cereals, and livestock,
rest of the world. Producers are price takers in intermediate input, show a signicant positive relationship between variability in the
factor and output markets and maximize intra-temporal prots growth of production and time. There is no systematic evidence of
subject to sectoral CES-Leontief production technologies. Consum- auto-correlation at either the commodity level or the more
er behavior is derived from intra-temporal utility maximizing aggregate level regressions.
behavior subject to within-period budget constraints. Utility For our purposes, the specication assumes that the variability
functions take the StoneGeary form, yielding an LES demand captured in the error term is due primarily to climate outcomes
system specication. Domestic goods and imports are imperfect occurring in each crop year. This overstates variability to some
substitutes in demand. Intra-temporal equilibria are linked (unknown) degree as factors unrelated to weather outcomes can
intertemporally through growth of the labour force, capital clearly contribute to production variability. For example, the
accumulation, and productivity effects arising from climate shocks. expectation of low prices could reduce production in a given year
The model is calibrated to a benchmark dataset based on the despite an average or even above average weather realization.
2001/02 Social Accounting Matrix (SAM) for Ethiopia (Ahmed et al., Unfortunately, these events are practically impossible to control
2008). The aggregated SAM for the present study identies 22 for in the Ethiopian context.
production activities (11 of which are agricultural), 24 commodity In order to investigate this issue, we compare the weather
groups, 2 household groups and 5 primary factors, including a shocks predicted by the econometric model described above with
separate agricultural labour factor. The dynamic simulation runs predictions from a process based crop modeling exercise for
cover a time period of 25 years. Ethiopia for selected crops (World Bank, 2010). The results are
shown in Table 4. The variation in productivity in the crop models,
4.2. The treatment of uncertainty based on historic weather variation, is indeed much less than the
variation remaining in the statistical model (which we attribute to
The approach for treating climate uncertainty and its implica- weather variation). It is useful to reect on what is happening.
tions for agricultural production in the Ethiopian context involves Perhaps the most obvious interpretation is that a great deal of
two steps. First, we examine historical data on agricultural unobserved variation unrelated to weather remains in the error
production by product category represented in the social terms. A second possibility is that the crop models are
accounting matrix in order to determine the historical variation insufciently sensitive to weather variation (and by implication
in production trends of each commodity and the tendency for climate variation). Along this second line, disenchantment with
production of various commodities to co-vary. Second, we consider crop models has led some analysts to pursue purely statistical
appropriate treatment of future climate outcomes in the context of approaches to precipitation, temperature, and crop yield.2 For
climate change. example, Lobell et al. (2008) nd that monthly data on precipita-
To determine historical climate variability, data on indices of tion and temperature provide, on their own, substantial explana-
agricultural production and land use by commodity for Ethiopia tory power. This is true even at fairly broad scales (even though the
were obtained from FAOStat for the period 19612004 (the most average for a broad region could differ substantially from actual
recent year for which complete data is available). The following realizations at specic points). Schlenker and Lobell (2010)
regression was run for each agricultural commodity present in the explicitly compare results from statistical and process based
Ethiopia Social Accounting Matrix: models with a focus on Africa. In their analysis, the statistical
models indicate substantially more sensitivity to precipitation and
P it ai bi t g i t 2 dP it1 hi d eit (1) particularly temperature than a series of prominent crop models.
In short, our investigations lead us to an exceedingly important
where subscript i represents crops, t represents time both as a gap in the literature. Specically, estimates of the impact of
subscript and within the regression (1961 = 0), t2 represents time weather variation or climatic change on crop yields vary widely
squared, Pit represents growth in production in percentage change according to the approach pursued. A group of economists is
from period t  1, Pit1 represents lagged growth in production, d unlikely to resolve the issue. Instead, we choose to remain with the
represents a dummy variable accounting for political regime statistical modeling approach as described above. The analysis has
change in 1991 and eit represents an error term. The specication the property of being at the upper end of the range of the
permits production growth to follow a quadratic time trend. The implications of weather variation for cropping outcomes, which
lagged term for percentage growth captures shorter run trends and proves useful at later stages in the analysis.
the tendency for growth trends to revert towards average after a In addition, for the modeling purposes considered here the
good year or a bad year due to the effect on the base for the specication is convenient as the error term relates directly to the
percentage change calculation in the following year.
Results from the regressions typically estimate a negative value
2
for the d parameter though the parameter is statistically The concerns highlighted by Mendelsohn et al. (1994) when they introduced
the Ricardian approach to crop modeling are different. They were concerned with
signicantly different from zero in only a few cases. Analysis of capturing the possibility of substitution across crops. Current disenchantment
the residuals squared over time illustrates little tendency for focuses on the ability of process based crop models to adequately predict response
production variance to expand with time on a commodity by to weather across economically signicant spatial scales.
 C. Arndt et al. / Global Environmental Change 21 (2011) 701710 705

deviations from trend production due to climate variability. The agricultural output though the number of consecutive dry days is
challenge is to use these error terms to construct a distribution of expected to decline. Positive outcomes are also possible. If ooding
likely climate outcomes from which one can draw for the is avoided, increased frequency of strong precipitation events, such
simulation of future outcomes in a Monte Carlo analysis. This as days with more than 10 mm of precipitation, may result in an
distribution should preserve the information captured by the increased frequency of strong positive outcomes given the
moments, including covariances, of the empirical distribution of important role of water constraints on total production in Ethiopia.
errors. This is potentially important as climate outcomes In summary, we should project a distribution of climate
unfavorable to commodity A are not necessarily unfavorable to outcomes that is more uniform than the historical data would
commodity B. indicate. This comes about for two reasons. First, our high
One approach is to assume that the error terms are normally condence in the existence of climate change combined with
distributed and drawn from a normal distribution with appropri- our lack of knowledge about the implications of global climate
ate covariance structure. However, the estimated error terms fail change for climate outcomes in Ethiopia requires us to consider a
tests for normality. As a result, a bootstrap approach was more uniform and potentially broader distribution of outcomes.
employed. In particular, in the baseline forward-looking stochastic Second, the information that we do have with respect to possible
simulations, vectors of error terms were drawn with replacement implications of global climate change for Ethiopia points to greater
from the empirical distribution with equal probability for each weight at the extremes of the distribution.
observed outcome from 1963 to 2004 (42 possible vectors). A 25- The critical operational question is exactly how and by how
year forward-looking simulation might repeat the realizations much the historical distribution should be modied in order to
observed in particular years. This approach allows one to consider account for climate change. Unfortunately, as the range of
the volatility of agricultural production outcomes under the outcomes presented in Section 3 suggests, this is currently
assumption that underlying climate variability remains the same. unknown (and may be unknowable). We have no choice but to
In other words, the history of climate realizations is, at rst, develop a subjective prior distribution of future climate outcomes
assumed to be a reasonable guide to future climate realizations. with existing data and technology under the knowledge that
The next step is to consider the implications of climate change climate change renders history an imperfect guide to the future. To
for the future distribution of climate outcomes. As discussed address this conundrum, we turn to nancial economics where a
earlier, the accumulation of GHGs in the atmosphere implies that roughly analogous situation prevails.
underlying climate variability in Ethiopia is likely to change. On a continuous basis, US investors must determine appropri-
Globally, climate is, with high condence, likely to become ate allocations between low risk allocations, such as government
warmer, wetter, and more volatile (IPCC, 2007). At more rened bonds, and high risk allocations such as equities. Long historical
levels of disaggregation, such as Ethiopia, the magnitude, timing experience indicates that investors demand return premiums for
and even direction of climate change is more uncertain, as was investments in equities over government bonds that exceed levels
illustrated in Section 3. Further, as emphasized in Section 3, consistent with generally accepted parameters for time prefer-
beyond the uncertainty associated with the distribution of climate ences and risk aversion combined with historically observed
outcomes, there remains uncertainty associated with how differ- volatility in the growth of output (consumption) by an order of
ent climate outcomes will translate into economic outcomes on the magnitude. Similarly, investors are willing to accept a much lower
ground. return on safe assets than theory, historical volatility in output/
For our purposes, it is most useful to consider how the prospect consumption and reasonable parameters would suggest. Finally,
of climate change should inuence the way we consider the prior equity prices are far more volatile than standard models would
distribution of future climate outcomes for Ethiopia. Two points predict. These inconsistencies are, respectively, labeled the equity
are particularly relevant. First, with high condence, the scientic premium, risk free rate, and equity volatility puzzles.
community believes that the global climate is changing due to Recently, Weitzman (2007) offered a unied explanation for all
anthropogenic factors. As Ethiopia is, most certainly, a part of this three puzzles. He shows that Bayesian updating of nonergodic
global system, the parameters underlying climate outcomes in systems causes rational agents to signicantly thicken the tails of
Ethiopia are changing as well. Two implications arise based only on their subjective prior distributions of future outcomes. This tail
the information that the underlying parameters of the distribution thickening maintains even if the evolution of the underlying
are shifting. system is arbitrarily slow and the volume of data on the underlying
systems tends toward innity. In other words, despite substantial
1. The range of possible outcomes (the support of the distribution) data on economic outcomes in, for example, the United States, the
may expand. knowledge that the underlying parameters of the economic system
2. The probability associated with any given set of outcomes may are evolving causes investors to signicantly atten the center of
change. the distribution of expected future economic growth rates and
distribute this weight to the tails. Or in Weitzmans words, people
In short, because the distribution itself is changing, we are less are acting in the aggregate as if there is much more marginal-
well informed about the nature of the distribution. The uniform utility-weighted subjective variability about future growth rates
distribution across a given support range is the least informative than past observations seem to support (Weitzman, 2007, p. 1).
distribution. As climate change leaves us less well informed about Though imperfect, this observation would appear to be a
the nature of the distribution of climate outcomes for Ethiopia, our potentially useful guide to the construction of future expectations
prior distribution of future climate outcomes should become more with respect to climate change. Helpfully, Weitzman provides rough
like the uniform distribution, reecting our lack of knowledge. The empirical orders of magnitude. Specically, US investors behave as if
support of that distribution may also expand, for the same reason. the standard deviation of fundamental economic volatility, dened
Second, as has been discussed, there is some information on the as the volatility of aggregate consumption, were more than eight
likely implications of global climate change for climate realizations times larger than historical data series would suggest. In other
in Ethiopia. The available information points to higher tempera- words, investors, in their projection of the future distribution of
tures and greater probability of extreme precipitation events, with outcomes, subjectively expand the historical variance of consump-
greater probability of ooding (ceteris paribus). Higher tempera- tion growth by a factor of roughly 70. This massive expansion in
tures are likely to accentuate the impact of consecutive dry days on variance neatly explains the three aforementioned puzzles.
706 C. Arndt et al. / Global Environmental Change 21 (2011) 701710

The translation of these insights into the climate adaptation constant implying that fundamental observed covariance relation-
problem at hand is far from straightforward. For example, while ships are preserved though the estimated covariance of the
truly catastrophic economic outcomes could be expected to render distribution changes due to the reweighting.
the distinction between safe and risky investments moot, the
distinction is likely to maintain far into the left tail of the 5. Climate shocks and autonomous adaptation: simulation
distribution of economic outcomes. On the other hand, in Ethiopia, results
the impacts of truly dismal climate outcomes are likely to be
similar regardless of policies pursued. As a result, there may be 5.1. Scenarios
little information in the far left tail that would have signicant
content for public decision-making from the perspective of the The focus in this section is on identifying the implications of
Ethiopian government. There is also the related issue of motivating climate shocks on economic performance. Some adaptation by
public decision-making on the basis of aggregate observed farmers within the individual agricultural activities is already built
behavior of investors. Nevertheless, the fundamental point into the estimates of productivity impacts derived from the crop
remainsrational decision-makers who must construct subjective simulation and Ricardian studies reviewed in Section 4. Moreover
prior distributions of outcomes of nonergodic systems signicantly adaptation in the form of migration of workers to other sectors in
atten and expand the distribution of historical outcomes. response to climate-related productivity shocks is endogenously
Our choices are presented Fig. 1. It shows a smoothed determined in the simulation analysis by allowing inter-sectoral
estimated distribution function for the impact of climate outcomes labour mobility, as are consumption responses to climate-induced
on total agricultural value added. The gures were derived by relative price changes on the demand side.
averaging the estimated percent productivity shocks across In the model, a climate shock involves changing various
commodities, obtained from the historical data via the regression productivity parameters and capital stocks, and then producers are
analysis described above, using 2001 shares in agricultural value assumed to be able to re-optimize given the changes. For example,
added as weights. The resulting average provides an estimate of given a drought, farmers are assumed to be able to adjust factor
the direct impact of climate variation on agricultural value added inputs (e.g., use less intermediate inputs) rather than have their
as a whole. Fig. 1 shows that, despite high volatility, the mass of yield simply be less than expected. The model thus incorporates
probability remains near the mean using historical information. anticipative adaptation on an annual basis, which is probably an
The projected distribution, on the other hand, more closely optimistic assumption, given the unexpected nature of many
resembles the uniform distribution. Weight has been shifted from climate shocks. At the same time, agents (including farmers) do not
the center of the distribution towards more extreme outcomes. have perfect foresight into the future. Investment allocations are
This is consistent with the discussion above. made on the basis of adaptive expectations.
The details on the procedure employed to derive the projected Initial parameters in the model are rst calibrated to the 2001/
distribution are described in Ahmed et al. (2009). In brief, an 02 SAM-based benchmark dataset, such that the model equilibri-
information theoretic procedure was employed to reweigh each um for the initial period exactly replicates the benchmark. A base
possible outcome such that the projected distribution reects dynamic simulation is then constructed which involves specica-
enhanced variance (33 percent greater standard deviation) and the tion of annual time paths for labour force and arable land growth,
observed small positive skewness in the historical distribution was productivity growth by production activities (Table 6), govern-
set to zero. Golan et al. (1996) provide a detailed treatment of how to ment expenditure, and trade balances over the 25-year simulation
estimate such probability weights using a cross-entropy criterion horizon. The world market prices of traded goods are held xed in
with moment constraints, based on information theory. The result is all simulation experiments.
greater weight in the tails of the distribution. While reweighted, the The CC-CGE model incorporates stochastic elements, as
points in the distribution remain historical observations with more described in the previous section. For each dynamic simulation,
extreme outcomes having a higher probability of being observed. the model is solved repeatedly (thirty times per annual time step)
The implications of these events across commodities remain for the 25-year period with different random draws of sectoral
[()TD$FIG] total factor productivity (TFP) parameters for each agricultural
0.06
activity in all years. For each dynamic run, there are thus 750
historical random draws for each of the 11 agricultural activities. We
projected experimented with different numbers of iterations for a sub-set of
0.05 scenarios with no signicant changes in the results. The random
draws are from the empirical joint distributions based on historical
data, and so reect actual historical distributions of shocks rather
0.04
than tted distributions. For these bootstrap simulations, we
weight

compute the means and standard deviations of the results for all
0.03 endogenous variables.
Five experiments are reported and discussed below. The
simulations are designed to explore the potential negative impact
0.02
of climate shocks. While the model agents will engage in
autonomous adaptation, the experiments are conducted in the
0.01 absence of specic public investments designed to ameliorate
them. The ve experiments are:

0  Simulation S1 (Historical): Historical pattern of agricultural

productivity variability, drawing sectoral agricultural productiv-
0

.6
8

.5

.4

.3

.2

.1
1.

3.

5.

7.

9.
5.

3.

1.

11
-9

-7

-5

-3

-1
-1

-1

-1

ity shocks due to climate shocks from the historical record.

TFP shock
 Simulation S2 (Thicker Tails): Increased variance of agricultural
Fig. 1. Smoothed distribution functions for variation in agricultural value added due productivity shocks by increasing the weight of extreme shocks
to climate outcomes. in sampling from the historical record. Standard deviation of
 C. Arndt et al. / Global Environmental Change 21 (2011) 701710 707

productivity shocks increases by 41 percent. Essentially no shift period, is symmetric. Negative shocks reduce savings/investment
of mean. while positive climate shocks increase savings/investment. The net
 Simulation S3 (S2 + DMean TFP): S2 plus negative 7.25 percent effect over time is an average level of investment in S2 that is
shift in the mean of productivity shocks. essentially identical to the level observed in S1 implying, alongside
 Simulation S4 (S3 + DTFP Trend): S3 plus a reduction in the constant growth in labour and underlying technology across S1
underlying rate of agricultural productivity growth by 25 and S2, a very similar production possibility frontier in the nal
percent. year.
 Simulation S5 (S4 + DCapital): S4 plus loss of capital due to S3 shifts the distribution of climate outcomes to the left by 7.25
extreme events. percent, which is in between the ranges of crop models with and
without CO2 fertilization in Table 3. Even though this shift in the
The rst stochastic simulation, S1, essentially reproduces the distribution of shocks does not change the underlying rate of
historical stochastic experience of Ethiopia. S2 assumes that the technical progress in agriculture, it has an effect on growth within
shocks are within the historical boundaries, but with more the 25-year time frame. Compared to S1 and S2, the overall GDP
frequent large shocks chosen from the historical record. S3 shifts growth rate as well as the growth rate of consumption is reduced
the distribution away from the historical record, assuming a 7.25 by about 0.2 percentage points.
percent negative shift in the mean of the agricultural productivity In S4, the rate of underlying technical progress in agriculture is
shocks in addition to increased sampling weight on extreme reduced by 25 percent. Over the 25-year time frame considered,
shocks as in S2. S4 4 builds on S3 and assumes that there is a further this effect is more powerful than the distribution shift. Compared
25 percent reduction in the underlying rate of growth in total with S3, annual GDP and consumption growth rates fall by about
factor productivity (TFP) in the agricultural sector. This simulation 0.3 percent.
assumes that the need to adapt to a changing climate claims The introduction of the possibility of capital stock destruction is
resources that would otherwise have been devoted to achieving presented in S5. As currently modeled, capital stock destruction
greater productivity under similar climatic conditions. S5 builds on reduces GDP and consumption by a similar magnitude compared
S4 and includes the possibility of elevated destruction of capital to the reduction in the underlying growth rate of TFP.
due to an increased frequency of extreme events. In particular, the Table 6 shows average annual growth rates of factors of
capital installed in the agricultural sector is assumed to undergo production as well as the economy-wide rate of technical change.
accelerated depreciation when climate shocks are negative. The By assumption, labour by skill category and land all grow at the
rate of accelerated depreciation is equal to one minus the TFP same rates across the simulations. Technical change rates decline
shock raised to the power of 0.5. So, if the multiplicative TFP shock consistent with the assumptions of the simulation. The capital
to sector k is 0.9, the shock to installed capital in sector k would be stock grows more slowly with each simulation due to, rst, lower
0.90.5 = 0.95. The sequence of the ve scenarios presented here savings as a result of slower technical progress and, second, in S5,
allows for an intuitive decomposition of impacts across results. the explicit modeling of accelerated depreciation of capital due to
The empirical basis underlying the shocks applied in simula- climate shocks.
tions 4 and 5 needs to be improved in future work. However, it is Table 6 shows a decomposition of the sources of growth over
plausible that the need to adapt to climate change would reduce the 25-year period, using aggregate GDP data from the simulations
underlying technical progress and that enhanced climate volatility and applying the simple Solow growth decomposition methodol-
may effectively lead to more rapid depreciation rates for capital ogy. Factor growth rates are weighted by factor shares in real value
(including land improvements) in the agricultural sector. As the added, and the weighted sum is subtracted from the aggregate
results will show, these impacts on rates of accumulation tend to growth rate of GDP at factor cost. The difference, or residual, is
result in fairly profound impacts. We turn now to the results. attributed to total factor productivity growth.
The results in Tables 57 highlight the importance of
5.2. Simulation results for climate shocks cumulative processes. Over time, even relatively small changes
in rates of growth cumulate to large overall effects. As shown in
Table 5 provides data on aggregate growth rates. S1 generates a simulations S1 and S2, a pure increase in variation will not affect
7.5 percent average annual real GDP growth rate over 25 years. the growth rate through time if the variation has no impact on the
Private and government consumption grow at 6.9 and 6.4 percent, accumulation of productive capacity. However, as indicated above,
respectively, while aggregate investment grows at 7.5 percent and there are reasons to believe that adaptation to climate change may
exports grow rapidly at 10.1 percent. The various climate shocks distract from process improvement resulting in a lower underlying
have differing impacts on growth rates depending on how they are rate of technical change. In addition, climate change may also
modeled. In S2, the variance of the distribution of climate shocks to result in effectively higher rates of depreciation of capital in
agriculture is increased, but the distribution remains (essentially) agriculture (particularly livestock capital as a result of drought and
centered on zero. These shocks are assumed to have no effect on infrastructure vulnerable to damage from ooding). However,
underlying total factor productivity growth and no direct effects on while plausible, these empirical relationships merit further study.
the capital stock (e.g., livestock herds are unaffected by drought). With respect to the functional distribution of income, average
The impact on savings, and hence installed investment in the next rates of growth of real factor income are presented in Table 8.

Table 5
Average annual real growth rate of macroeconomic variables (in percent).

S1 Historical S2 Thicker Tails S3 S4 S5

S2 + DMean TFP S3 + DTFP Trend S4 + DCapital

Consumption 6.9 6.9 6.7 6.4 6.1

Investment 7.5 7.5 7.4 7.2 6.9
Exports 10.1 10.1 10.0 9.9 9.5
Imports 6.9 6.9 6.9 6.8 6.4
GDP 7.5 7.5 7.3 7.0 6.8
Absorption 6.9 7.0 6.8 6.5 6.3
708 C. Arndt et al. / Global Environmental Change 21 (2011) 701710

Table 6
Annual growth rates of production factors and rate of technical change (in percent).

S1 Historical S2 Thicker Tails S3 S4 S5

S2 + DMean TFP S3 + DTFP Trend S4 + DCapital

Agricultural labour 2.5 2.5 2.5 2.5 2.5

Unskilled labour 2.5 2.5 2.5 2.5 2.5
Skilled labour 2.5 2.5 2.5 2.5 2.5
Capital 7.3 7.3 7.2 7.1 6.4
Land 1.0 1.0 1.0 1.0 1.0
Technical change 3.4 3.4 3.3 3.1 3.1

Table 7
Solow decomposition of growth (percent).

Sim1 Sim2 Sim3 Sim4 Sim5

Agricultural labour 4.1 4.1 4.2 4.3 4.5

Unskilled labour 2.6 2.6 2.7 2.8 2.9
Skilled labour 9.3 9.3 9.5 9.8 10.2
Capital 37.7 37.6 37.9 38.6 36.5
Land 1.5 1.5 1.6 1.6 1.7
Technical change 44.8 44.9 44.2 42.9 44.2

The progressively stronger climate change effects simulated mean) for the macro aggregates for the beginning and ending years
in scenarios S2 to S5 have relatively mild effects on the for each simulation. As expected, the variation increases signi-
functional distribution of income, though, as noted above, small cantly from S1 to S2 with increased variability of climate shocks
changes in rates can have large effects over extended periods of generating increased variability in economic performance. Vari-
time. While income of all factors declines as a result of climate ability is further increased in S5 due to the role of extreme climate
change, the declines are slightly less pronounced for agricultural events in the destruction of capital.
labour and unskilled labour across S3, S4, and S5 indicating that, Aggregate consumption always has a higher CV than the other
on average and over time, the climate shocks do not depress the macro aggregates while government and investment expenditure
returns to factors owned by the poor in particular. However, this have lower values. The burden of adjustment falls more heavily on
result does not contradict the common perception that as a consumers, which is to be expected since the productivity shocks
result of high exposure and low adaptive capacity rural poor all occur in agriculture, most of which is used for consumption.
households are particularly vulnerable to adverse climate change Because the poor spend a disproportionate share of their income
impacts. on food and because they are more likely to be employed in
Nevertheless, the implications of climate change for the poor agriculture, the enhanced volatility of consumption is likely to be
may be harsh and highly negative in specic years. Table 9 presents particularly pronounced amongst the poor. These observations
the coefcient of variation (CV, standard deviation divided by the point strongly to accentuated vulnerability.

Table 8
Average annual rate of growth of real factor income (percent).

S1 Historical S2 Thicker Tails S3 S4 S5

S2 + DMean TFP S3 + DTFP Trend S4 + DCapital

Agricultural labour 6.2 6.2 6.0 5.7 5.6

Unskilled labour 6.7 6.7 6.5 6.1 5.9
Skilled labour 7.5 7.5 7.2 6.8 6.5
Capital 7.2 7.3 7.0 6.5 6.1
Land 6.0 6.0 5.8 5.3 5.2

Table 9
Coefcient of variation of macroeconomic variables.

S1 Historical S2 Thicker Tails S3 S4 S5

S2 + DMean TFP S3 + DTFP Trend S4 + DCapital

t=1
Consumption 2.6 3.7 3.9 3.9 3.9
Investment 1.1 1.2 1.0 1.0 1.0
Exports 5.9 6.3 5.3 5.3 5.3
Imports 2.6 2.9 2.3 2.3 2.3
GDP 2.3 3.2 3.3 3.3 3.3
Absorption 2.0 2.8 2.9 2.9 2.9
t = 25
Consumption 2.1 2.4 2.6 2.5 2.7
Investment 1.1 0.9 1.0 1.1 1.5
Exports 0.8 0.9 0.9 0.8 1.4
Imports 0.7 0.8 0.8 0.7 1.2
GDP 1.7 1.8 1.9 1.9 2.2
Absorption 1.6 1.8 1.9 1.8 2.1
 C. Arndt et al. / Global Environmental Change 21 (2011) 701710 709

Table 10
Evolution of the share of agriculture in value added (in percent).

S1 Historical S2 Thicker Tails S3 S4 S5

S2 + DMean TFP S3 + DTFP Trend S4 + DCapital

t=1 46.3 46.8 45.0 45.0 45.0

t = 12 37.7 37.5 35.9 34.6 35.1
t = 25 30.7 30.7 29.3 26.9 27.2

Two other points merit mention. First, the CV for aggregate On top of this analysis of averages, analysis of the variability of
absorption is always smaller than for real GDP, indicating that the components of GDP indicates that aggregate consumption
international trade serves to dampen the impact of climate shocks always has a higher coefcient of variation (CV) than the other
on aggregate demand. Second, CV is systematically smaller in the macro aggregates. The burden of adjustment appears to fall more
nal time period (t = 25) than in the initial time period (t = 1). This heavily on consumers. Poor consumers are likely to experience
is partly explained by the ow of real resources out of agriculture. increased vulnerability.
As shown in Table 10, the share of agriculture in real value added Effective adaptation measures are likely to be highly cost
declines from more than 45 percent in the rst simulation year to effective. Adaptation policies designed to reduce vulnerability to
between 27 and 31 percent in the nal year. The decline in the climatic change are more likely to succeed if they are integrated
share of agriculture in total value added reduces the exposure of with efforts aimed at poverty reduction and general economic
the full economy to climate shocks. growth strategies. Ethiopias high level of unexploited water
With respect to robustness, a number of simulations were done resource management potential and low fraction of irrigated land
to test the robustness of the results concerning the impact of suggest that investments in water resource management infra-
climate shocks to different assumptions about the underlying structure should be an integral component of a climate change
parameters and trends. Three different issues were explored adaptation strategy aimed at the reduction of vulnerability to the
concerning sensitivity to: (1) different underlying growth rates, (2) effects of global warming.
choice of time horizon, and (3) different aggregations of the
agricultural sectors from the SAM. Acknowledgment
Lower rates of productivity growth (both disembodied and
embodied in land and capital) for both agricultural and non- The helpful comments of three anonymous referees are
agricultural sectors were chosen to generate a calibrated growth gratefully acknowledged.
rate of GDP of about ve percent. Climate change shock
simulations were run against this growth path. The results indicate References
that the impact of climate shocks is similar.
To test for sensitivity to time horizon, the various experiments Adger, W.N., Agrawala, S., Mirza, M.M.Q., Conde, C., OBrien, K.L., Pulhin, J., Pulwarty,
were repeated for 30 years instead of 25. There was essentially no R., Smit, B., Takahashi, K., 2007. Assessment of Adaptation Practices, Options,
Constraints and Capacity. In: Parry, M.L., Canziani, O.F., Palutikof, J.P., Hanson,
change in the impact on growth rates. Results for the terminal year, C.E., van der Linden, P.J. (Eds.), Climate Change 2007: Impacts, Adaptation and
of course, differ because of the longer time horizon, but Vulnerability. Contribution of Working Group II to the Fourth Assessment
comparisons across the simulations were essentially similar to Report of the Intergovernmental Panel on Climate Change, Cambridge Univer-
sity Press, Cambridge, pp. 719743.
the 25-year period. Ahmed, H., Amogne, A., Bekele, G., Deriba, G., Teferra, B., Robinson, S., Willenbockel,
To test for sensitivity to aggregation, a version of the model was D., 2008. Social Accounting Matrix for Ethiopia 2001/2002 (EFY 1994): Data
implemented with the eleven agricultural sectors aggregated to Sources and Compilation Process. Ethiopian Development Research Institute,
Addis Ababa.
ve. The stochastic draws were also aggregated. The results were Ahmed, H., Arndt, C., Robinson, S., Willenbockel, D., 2009. Ethiopia: a country study
strikingly different, with far less effect from climate change shocks. on the economic impacts of climate change. World Bank Report No.46946-ET.
The problem is that the aggregation eliminated the covariation The World Bank, Washington, DC.
Bosello, F., Zhang, J., 2005. Assessing Climate Change Impacts: Agriculture. Fonda-
across shocks affecting the various agricultural sectors, so that the
zione Eni Enrico Mattei Nota di Lavaro 94.2005.
impacts were essentially dampened by the averaging. The results Challinor, A.J., Wheeler, T.R., Garforth, C., Craufurd, P., Kassam, A., 2007. Assessing
indicate the importance of using as much agricultural detail as is the vulnerability to food crop systems in Africa to climate change. Climatic
needed to capture the links between climate change shocks and Change 83, 381399.
Cheung, W.H., Senay, G.B., Singh, A., 2008. Trends and spatial distribution of annual
particular sectors. and seasonal rainfall in Ethiopia. International Journal of Climatology 28, 1723
1734.
6. Conclusion Christensen, J.H., Hewitson, B., Busuioc, A., Chen, A., Gao, X., Held, I., Jones, R., Kolli,
R.K., Kwon, W.-T., Laprise, R., Magana Rueda, V., Mearns, L., Menendez, C.G.,
Raisanen, J., Rinke, A., Sarr, A., Whetton, P., 2007. Regional Climate Projections.
This study develops a stochastic economy-wide framework for In: Solomon, S., Qin, D., Manning, M., Chen, Z., Marquis, M., Averyt, K.B., Tignor,
analyzing economic impacts from climate change and potential M., Miller, H.L. (Eds.), Climate Change 2007: The Physical Science Basis. Contri-
bution of Working Group I to the Fourth Assessment Report of the Intergov-
adaptation policies that developing countries might undertake in ernmental Panel on Climate Change. Cambridge University Press, Cambridge,
the near and long term. Results point to three channels where pp. 847940.
climate change has substantial impact on economic growth. The CSA, 2008. National Accounts Statistics of Ethiopia: The 1999 EFY Update Estimates.
Central Statistical Agency, Federal Democratic Republic of Ethiopia, Addis
rst involves reductions in average agricultural yields due to less
Ababa.
favourable growing conditions. The second involves reductions in Deressa, T.T., 2006. Measuring the economic impact of climate change on Ethiopian
the rate of process improvement due to the demands of adaptation. agriculture: Ricardian approach. CEEPA Discussion Paper No.25. Centre for
Environmental Economics and Policy in Africa, Pretoria.
The third involves the possibility of destruction of capital stocks
Fankhauser, S., Tol, R.S.J., 2005. On climate change and economic growth. Resource
due to an increased probability of extreme events. The results and Energy Economics 27, 117.
highlight the importance of cumulative processes and rates of Funk, C., Dettinger, M.D., Michaelsen, J.C., Verdin, J.P., Brown, M.E., Barlow, M., Hoell,
growth. If climate change effects the rate of technical change and A., 2008. Warming of the Indian Ocean threatens Eastern and Southern African
Food Security but could be mitigated by agricultural development. Proceedings
the rate of accumulation of capital, the implications become of the National Academy of Sciences of the United States of America 105 (32),
signicant over time. 1108111086.
710 C. Arndt et al. / Global Environmental Change 21 (2011) 701710

Golan, A., Judge, G., Miller, D.R., 1996. Maximum Entropy Econometrics: Robust Reilly, J., Paltsev, S., Felzer, S., Wang, B., Kicklighter, X., Melillo, D., Prinn, J., Sarom,
Estimation with Limited Data. John Wiley and Sons, New York. R., Sokolov, M., Wang, A.C., 2007. Global economic effects of changes in crops,
Hertel, T.W., Burke, M.B., Lobell, D.B., 2010. The poverty implications of climate- pasture, and forests due to changing climate, carbon dioxide, and ozone. Energy
induced crop yield changes by 2030. Global Environmental Change 20, 577 Policy 35, 53705383.
585. Reilly, M., Willenbockel, D., 2010. Managing uncertainty: a review of food system
IMF, 2008. The Federal Democratic Republic of Ethiopia: statistical appendix. IMF scenario analysis and modelling. Philosophical Transactions of the Royal Society
Country Report 08/260. International Monetary Fund, Washington, DC. B 365, 30493063.
IPCC, 2007. Climate Change 2007: Synthesis Report. Intergovernmental Panel on Rosenzweig, C., Iglesias, A., 2006. Potential Impact of Climate Change on World Food
Climate Change, Geneva. Supply: Data Sets from a Major Crop Modeling Study. Columbia University,
Kurukulasuriya, P., Mendelsohn, R., Hassan, R., Benhin, J., Deressa, T., Diop, M., http://beta.sedac.ciesin.columbia.edu.
Mohamed Eid, H., Yer Fosu, K., Gbetibouo, G., Jain, S., Mahamadou, A., Mano, R., Schlenker, W., Lobell, D.B., 2010. Robust negative impacts of climate change on
Kabubo-Mariara, J., El-Marsafawy, S., Molua, E., Ouda, S., Ouedraogo, M., Sene, I., African agriculture. Environmental Research Letters 5 (1), 014010.
Maddison, D., Seo, S.N., Dinar, A., 2006. Will African agriculture survive climate Seleshi, Y., Zanke, U., 2004. Recent changes in rainfall and rainy days in Ethiopia.
change? World Bank Economic Review 20 (3), 367388. International Journal of Climatology 24, 973983.
Lobell, D.B., Burke, M.B., Tebaldi, C., Mastrandrea, M.D., Falcon, W.P., Naylor, R.L., Semenov, M.A., Porter, J.R., 1995. Climatic variability and the modeling of crop
2008. Prioritizing climate change adaptation needs for food security in 2030. yields. Agricultural and Forest Meteorology 73, 265283.
Science 319, 607610. Stern, N., 2007. The Economics of Climate Change: The Stern Review. Cambridge
Lofgren, H., Harris, R.L., Robinson, S., 2002. A Standard Computable General Equi- University Press, Cambridge.
librium (CGE) Model in GAMS. Microcomputers in Policy Research 5, IFPRI, Tubiello, F., Amthor, J.S., Boote, K.J., Donatelli, M., Easterling, W., Fischer, G., Gifford,
Washington, DC (With the assistance of M. El-Said and M. Thomas). R.M., Howden, M., Reilly, J., Rosenzweig, C., 2007. Crop response to elevated CO2
Long, S.P., Ainsworth, E.A., Leakey, A.D.B., Nosberger, J., Ort, D.R., 2006. Food for and world food supply: a comment. European Journal of Agronomy 26 (3), 215
thought: lower-than-expected crop yield stimulation with rising CO2 concen- 223.
tration. Science 312 (5782), 19181921. Warren, R., Arnell, N., Nicholls, R.J., Levy, P., Price, J., 2006. Understanding the
Mendelsohn, R., Nordhaus, W.D., Shaw, D., 1994. The impact of global warming on regional impacts of climate change. Tyndall Centre for Climate Change Research
agriculture: a Ricardian analysis. American Economic Review 84 (4), 753771. Working Paper No.90. UEA, Norwich.
Mideska, T.K., 2010. Economic and distributional impacts of climate change: the Weitzman, M., 2007. Subjective expectations and asset return puzzles. American
case of Ethiopia. Global Environmental Change 20, 278286. Economic Review 97 (4), 11021130.
Porter, J.R., Semenov, M.A., 2005. Crop responses to climatic variation. Philosophical World Bank, 2010. Economics of Adaptation to Climate Change: Ethiopia (Consul-
Transactions of the Royal Society B 360, 20212035. tation Draft). World Bank, Washington, DC.