WPS4279




                                THE IMPACT OF CLIMATE CHANGE

                           ON LIVESTOCK MANAGEMENT IN AFRICA:

                              A STRUCTURAL RICARDIAN ANALYSIS1




                              Sungno Niggol Seo and Robert Mendelsohn2




World Bank Policy Research Working Paper 4279, July 2007

The Policy Research Working Paper Series disseminates the findings of work in progress to encourage the exchange
of ideas about development issues. An objective of the series is to get the findings out quickly, even if the
presentations are less than fully polished. The papers carry the names of the authors and should be cited
accordingly. The findings, interpretations, and conclusions expressed in this paper are entirely those of the authors.
They do not necessarily represent the view of the World Bank, its Executive Directors, or the countries they
represent. Policy Research Working Papers are available online at http://econ.worldbank.org.




1An earlier version of this Working Paper was published as CEEPA Discussion Paper number 23.
2University of Aberdeen Business School, United Kingdom and School of Forestry and Environmental Studies,
Yale University, 230 Prospect Street, New Haven, CT 06511, USA, Seo e-mail: niggol.seo@abdn.ac.uk;
Mendelsohn tel: 203-432-5128, e-mail: robert.mendelsohn@yale.edu.
The authors want especially to thank Pradeep Kurukulasuriya, Rashid Hassan, James Benhin, Ariel Dinar,
Temesgen Deressa, Mbaye Diop, Helmy Mohamed Eid, K Yerfi Fosu, Glwadys Gbetibouo, Suman Jain, Ali
Mahamadou, Reneth Mano, Jane Mariara, Samiha El-Marsafawy, Ernest Molua, Mathieu Ouedraogo, and Isidor
Sène.
This paper was funded by the GEF and the World Bank. It is part of a larger study on the effect of climate change on
agriculture in Africa, managed by the World Bank and coordinated by the Centre for Environmental Economics and
Policy in Africa (CEEPA), University of Pretoria, South Africa.

SUMMARY

This paper develops the structural Ricardian method, a new approach to modeling agricultural
performance using cross-sectional evidence, and uses the method to study animal husbandry in
Africa. The traditional Ricardian approach measures the interaction between climate and
agriculture (Mendelsohn et al. 1994; Seo et al. 2005) but it does not reveal how farmers actually
adapt. It is consequently difficult to compare traditional Ricardian results with microeconomic
models built from the details of agronomic research (e.g. Adams et al. 1990, 1999; Reilly et al.
1996). The Model is intended to estimate the structure beneath Ricardian results in order to
understand how farmers change their behavior in response to climate. In this African livestock
example, the Structural Ricardian Model estimates which species are selected, the number of
animals per farm, and the net revenue per animal. All three of these elements are climate
sensitive.

A three-equation model is developed to estimate each of the choices facing a farmer. For each
farm, a primary animal is defined as the species that is observed to earn the greatest net revenue
on that farm. A multinomial logit is then estimated to predict which primary animal each farmer
selects. Given the primary animal chosen, the second equation estimates the number of animals
of that type per farm. The final equation estimates the net revenue per animal by species.

The model is used to study the sensitivity of African animal husbandry decisions to climate. A
survey of over 5000 livestock farmers in ten countries reveals that the selection of species, the
net income per animal, and the number of animals are all highly dependent on climate. As
climate warms, net income across all animals will fall but especially across beef cattle. The fall
in net income causes African farmers to reduce the number of animals on their farms. The fall in
relative revenues also causes them to shift away from beef cattle and towards sheep and goats.
All farmers will lose income but the most vulnerable farms are large African farms that currently
specialize in beef cattle.

Small livestock and large livestock farms respond to climates differently. Small farms are
diversified, relying on dairy cattle, goats, sheep and chickens. Large farms specialize in dairy and
especially beef cattle. Estimating a separate multinomial logit selection model for small and large
farms reveals that the two types of farm choose species differently and specifically have different
climate response functions. The regressions of the number of animals also reveal that large farms
are more responsive to climate.

Several climate scenarios are tested using the estimated three-equation model. Some simple
uniform climate change scenarios are tested that assume a warming of 2.5°C or 5°C and a change
in precipitation of +15% or -15%. The purpose of these scenarios is to see how different districts
across Africa respond to identical changes in climate. Uniform warming causes the probability of
choosing beef cattle to fall where these are currently being chosen. In contrast, warming causes
the probability of choosing sheep to rise, especially across the Sahel. Warming causes the
number of animals to fall but especially beef cattle. Finally warming causes the net revenue from
all animals to fall, but especially from beef cattle. Increasing precipitation causes the probability
of choosing beef cattle, dairy cattle and sheep to fall and that of goats and chickens to increase.
Wetter climatic conditions reduce the desired number and net revenue of beef cattle, dairy cattle,



                                                   2

sheep and chickens, but not goats. This effect is most likely due to the change in landscape,
associated with more precipitation, from savanna to forest. Combining all these changes, a 2.5°C
warming results in a 32% loss in expected net income and a 5°C warming leads to a 70% loss in
expected net income. Increasing precipitation by 15% results in a 1% loss in expected net
income.

We also examine climate change impacts using the separate regressions for small and large
livestock farms. With warming, small farms are expected to shift away from dairy cattle and
chickens to goats and sheep. Net incomes will fall for all animals except for sheep. The number
of animals will also fall. Expected income will fall by 13% with a warming of 2.5°C, but recover
with more warming to current levels of income. A 15% decrease in precipitation is expected to
increase small livestock farm incomes by 6%. For large farms, warming will cause a shift to
dairy cattle and sheep and away from goats, chickens and especially beef cattle. The income per
animal falls for all species as temperatures rise. With higher temperatures, large farms choose to
have fewer beef, chickens and sheep and choose more goats and dairy cattle. Large farmers'
incomes are expected to fall by an average of 26% with a 2.5°C warming and by 67% with a 5°C
warming, but a 15% decrease in precipitation is expected to increase these farmers' incomes by
2%.

The study also examines the consequences of a range of climate predictions from three
Atmospheric Oceanic General Circulation Models (AOGCMs). These models predict that
climate change will cause beef cattle to decrease in Africa and sheep and goats to increase. In
general, the climate models predict that the overall number of animals will fall although the
number of goats may increase. They also predict that the net revenue per animal will fall.
Combining all of these effects, the climate models predict average losses of 22% ($8 to $23
billion) in expected net income from livestock by 2020. These damages increase to 31% ($9 to
$24 billion) by 2060, and to 54% ($25 to $40 billion) by 2100.

Examining the effect on small and large farms reveals that small farms will choose dairy cattle
and sheep more often and goats and chickens less often as the primary animal. The income per
animal will tend to fall over time except for sheep. The number of animals will tend to fall with
warming with a few exceptions. The changes in the number of goats and sheep are relatively
negligible. The expected income for small farms will tend to increase over time with the
Canadian Climate Center (CCC) scenarios (34%), but fluctuate with the Parallel Climate Model
(PCM) and Center for Climate System Research (CCSR) scenarios depending on precipitation.
Large farmers, in contrast, will shift away from beef cattle and chickens in favor of dairy cattle,
sheep and goats. Net revenues will fall across animals, but especially for beef cattle. The
numbers of beef cattle and chickens will fall by large amounts, but the numbers of goats and
sheep will increase depending upon the scenarios. Putting all these results together, CCC will
lead to a $6000 reduction in expected net revenue per large farm (77%), CCSR to a $2,700
reduction (34%), and PCM to a $3,400 reduction (43%) by 2100.

The results indicate that warming will be harmful to commercial livestock owners, especially
cattle owners. Owners of commercial livestock farms have few alternatives either in crops or
other animal species. In contrast, small livestock farms are better able to adapt to warming or
precipitation increases by switching to heat tolerant animals or crops. Livestock operations will
be a safety valve for small farmers if warming or drought causes their crops to fail.


                                                  3

TABLE OF CONTENTS

Section                                         Page

  1     Introduction                               5

  2     Theory                                     5

  3     Data and empirical specification          10

  4     Empirical results                         10

  5     Climate simulations                       14

  6     Conclusion and policy implications        18

        References                                20




                                           4

1. Introduction

This paper develops a new empirical approach to studying agriculture, the Structural Ricardian
Model, and applies it to studying animal husbandry in Africa. This model, a variation of the
Ricardian approach (Mendelsohn et al. 1994), estimates the underlying profit functions of
specific animals or crops. The original Ricardian model examined the locus of profit maximizing
choices of farmers across all output choices. The Structural Ricardian Model estimates the
farmer's selection of the most profitable species, the number of animals chosen, and the
conditional net revenue per animal. Besides revealing how net revenue changes with climate, this
model also reveals details of how farmers adjust to climate. It explains farmer's choices across
animals (or crops) and measures how sensitive each animal (or crop) is to exogenous variables.
These animal specific results can be more directly compared to natural science based studies
(such as Reilly at al. 1996) and economic production studies of individual crops and animals
(such as Adams et al. 1999).

We use this new methodology to study the impact of climate change on animal husbandry in
Africa. Early analyses of the effects of climate change predicted extensive damage to the
agricultural sector across the globe (Pearce 1996). The bulk of agriculture studies on the effect of
climate change have focused on crops. However, a large fraction of agricultural output is from
livestock. Almost 80% of African agricultural land is used for grazing. African farmers depend
on livestock for income, food, animal products and insurance. Yet there are very few economic
analyses of climatic effects on livestock. An important exception to this gap is the study of the
effects of climate change on American livestock (Adams et al. 1999). American livestock appear
not to be vulnerable to climate change because they live in protected environments (sheds, barns
etc.) and have supplemental feed (e.g. hay and corn). In Africa, by contrast, the bulk of livestock
have no protective structures and they graze off the land. There is every reason to expect that
African livestock will be sensitive to climate change. This study analyzes the behavior of over
9000 African farmers in ten countries in order to measure the climate sensitivity of African
animal husbandry. Of the 9000 farmers interviewed, over 5000 were farming livestock.

The underlying theory of the Structural Ricardian Model is developed in the next section. Section
3 discusses how the data were collected and what variables are available. Section 4 discusses the
estimation procedure and the empirical results. Several climate change scenarios are then
examined in Section 5. The paper looks at both uniform changes in climate across Africa and
climate model predictions. It concludes with a summary of results and policy implications.



2. Theory

A farmer's optimization decision can be seen as a simultaneous multiple-stage procedure. The
farmer chooses the levels of inputs, the desired number of animals and the species that will yield
the highest net profit. Given the profit maximizing inputs from each farmer, one can estimate the
loci of profit maximizing choices for each animal across exogenous environmental factors such
as temperature or precipitation. These are the individual loci that lie beneath the overall profit
function for the farm (Mendelsohn et al. 1994). We call the approach `structural' because it
estimates the underlying profit response functions (the structure) that form the overall Ricardian


                                                 5

response. For example, in Figure 1 we display a traditional Ricardian response function with
respect to temperature. Underneath the loci of all choices is a set of animal specific response
functions. Given the climate, the farmer must choose the most profitable animal and also the
inputs that will maximize the value of that animal. We examine the individual net revenue
functions for each animal (Structural Ricardian Model) as well as the overall net revenue
function across all animals (Ricardian Model).

We assume that each farmer makes his animal husbandry decisions to maximize profit. Hence,
the probability that an animal is chosen depends on the profitability of that animal or crop. We
assume that farmer i's profit in choosing livestock j (j=1,2,...,J) is



ij =V(K ,Sj) +(K ,Sj)                                                                      (1)
           j            j




where K is a vector of exogenous characteristics of the farm and S is a vector of characteristics of
farmer i. For example, K could include climate, soils and access variables and S could include
the age of the farmer and family size. The profit function is composed of two components: the
observable component V and an error term, . The error term is unknown to the researcher, but
may be known to the farmer. The farmer will choose the livestock that gives him the highest
profit. Defining Z = (K,S), the farmer will choose animal j over all other animals k if:



*(Zji) >*(Zki)fork  j.[orif (Zki)-(Zji) <V(Zji)-V(Zki)for k  j]                            (2)




More succinctly, farmer i's problem is:



  argmax[      *(Z1i),*(Z2i),...,*(ZJi)]                                                   (3)
        j




The probability Pji for the jth livestock to be chosen is then



Pji = Pr[(Zki ) - (Z ) < Vj -Vk ]  k  j where Vj = V(Zji )                                 (4)
                      ji




                                                 6

Assuming  is independently Gumbel distributed3 and Vk = Zki k +k ,




Pji =        eZji j
             J   eZkik                                                                                     (5)
             k=1




which gives the probability that farmer i will choose livestock j among J animals (Chow 1983;
McFadden 1981).

The parameters can be estimated by the Maximum Likelihood Method, using an iterative
nonlinear optimization technique such as the Newton-Raphson Method. These estimates are
CAN (Consistent and Asymptotically Normal) under standard regularity conditions (McFadden
1999).

Note that farmers can choose more than one species of livestock among the five animals in our
study. That is, there are many combinations of animals that the farmer could choose. In this
analysis, we assume that farmers choose one primary animal from the five animals. A primary
animal is defined as the animal that generates the highest total net revenue in the farm (Train
2003). In Africa, 88% of total livestock net revenue is earned from a primary animal.4

Conditional on the livestock species chosen, we then estimate the optimal number of animals per
farm and the net revenue per animal. We rely on a two-stage model. In the first stage, we
estimate the probability of selecting an animal (equation 5). In the second stage, conditional on
the choice of a specific species, we estimate the optimal number of animals and the net revenue
per animal. Because the farmer may observe the error term that the researcher cannot observe,
one must correct for possible selection bias. That is, a farmer is more likely to choose a crop or
animal that is actually more profitable. Since the farmer maximizes net revenue conditional on
the choice of that species, the error in the second stage equation may be correlated with the error
in the first stage. According to Dubin and McFadden (1984), with the assumption of the
following linearity condition:5



3Two common assumptions about the error term are either the Normal or the Gumbel distribution. Normal random
variables have the property that any linear combination of normal varieties is normal. The difference between two
Gumbel random variables has a logistic distribution, which is similar to the normal, but with larger tails. Thus the
choice is somewhat arbitrary with large samples (Greene 1998).
4 Alternatively, one could model all the possible combinations of animals (McFadden 1999; Train 2003). Another
approach is to model the probability that each animal is chosen in a system of equations (Chib & Greenberg 1998).
Exploring these alternative approaches leads to similar results (Seo & Mendelsohn 2006b).
5 See Bourguignon et al. (2004) for the details of the selection bias corrections from the multinomial choice. They
find that Dubin and McFadden's method is preferable to the most commonly used Lee method, as well as to the
Dhal's semi-parametric method in most cases. Monte Carlo experiments also showed that selection bias correction
based on the multinomial logit model can provide fairly good correction for the outcome equation even when the IIA
hypothesis is violated.


                                                          7

                        J                      J
E(uj |1,...,J ) = j  rj( j -E( j))with rj =0
                                                                                             (6)
                        j=1                   j=1


where u j =error from the second stage,  =error from the first stage,  =standard error from
                                             j                                j

the unconditioned second stage regression, rj =correlation between the first stage error and
second stage error, then the conditional profit functions can be consistently estimated as:



 = xjj + x2jj +  ri       J     Pi ln Pi+lnPj + wj
                                               
                                                                                          (7a)
  j                   j  
                         i j    1- Pi



where the second term is the correction term and wj is the error term.



The optimal number of animals can be estimated in the same manner:



N j = xj j + x2j +  ri     J    Pi ln Pi+lnPj +vj
                                               
                                                                                         (7b)
                 j     j  
                          i j   1- Pi

where the second term is the correction term and vj is the error term.



The two stage model is composed of equation (5) and equations (7a) and (7b). Expected net
revenue is therefore:



               J
Wi (Ki,Si ) = Pj (Ki,Si ) (Ki ,Si )  N
                                             for all j.                             (8)
                               j          j
              j=1




Because climate is an independent variable in all three terms in (8), the marginal effect on
welfare of a change in a climate variable has three components: the effect on the probability of
the livestock to be chosen, the direct effect on the conditional profit per animal, and the effect on
the conditional number of animals:



                                                  8

Wj    = Pj                            N
             N +         j               j                                          (9)
zj      zj     j    j  zj   Pj  N +j  zj   Pj    j




where the marginal effect on the probability can be obtained by differentiating (5):



Pj              J
                                                                                    (10)
zj   = Pj[ - Pkk]
           j   
               k =1




and the marginal effect on the net revenue per animal can be obtained by differentiating (7a):



    j
zj   =  + 2xj                                                                       (11)
         j         j




and the marginal effect on the number of animals per farm can be found by differentiating (7b):



    j
zj   =  + 2xj                                                                       (12)
         j         j




The change in welfare resulting from a non-marginal change in climate can be computed as the
difference in the expected net revenues in two states. Suppose that climate changes from CA to
CB . Then the change in welfare can be approximated as:



Wi =Wi(CB)-Wi(CA)                                                                    (13)



The uncertainty surrounding our measure of the welfare change can be described by the 95%



                                                9

confidence interval of the expected climate change impact. In principle, there are two ways to
calculate confidence intervals: parametric and non-parametric. It is difficult to calculate the
variance of the climate change impact parametrically in this model, because welfare is the
product of three predictions.6 Uncertainty estimates are provided by bootstrap methods by
resampling and calculating the mean and 95% intervals of the climate change impacts.



3. Data and empirical specification

The data this study relies on were part of a larger GEF project to study climate change impacts
on agriculture (Dinar et al. 2006). The countries included were Burkina Faso, Cameroon, Egypt,
Ethiopia, Ghana, Kenya, Niger, Senegal, South Africa and Zambia. (Zimbabwe had to be
dropped from the livestock analysis because of turbulent conditions in that country during the
survey.) They were selected as representative of the wide range of climate throughout Africa, and
districts within each country were selected to provide a wide range of climatic variation. The
original survey interviewed over 9000 farmers from the 11 countries. Within that sample, over
5000 were livestock farmers.

The data include information on livestock production and transactions, livestock product
production and transactions, and relevant costs. The data indicate that the five major types of
livestock in Africa are beef cattle, dairy cattle, goats, sheep and chickens. Other less frequently
recorded animals include breeding bulls, pigs, oxen, camels, ducks, guinea fowl, horses, bees,
and doves. The major livestock products sold were milk, meat, eggs, wool and leather. Others
included butter, cheese, honey, skins and manure.

Climate data came from two sources: US Defense Department satellites and weather station
observations. We relied on satellite temperature observations and interpolated precipitation
observations from ground stations (see Kurukulasuriya and Mendelsohn 2006 for a detailed
explanation). Soil data were obtained from the FAO digital soil map of the world CD ROM. The
soil data were extrapolated to the district level using GIS (Geographical Information System).
The data set reports 116 dominant soil types.



4. Empirical results

Although there are many varieties of farm animal in Africa, we focused on the five primary ones
that generated most of the income from livestock: beef cattle, dairy cattle, goats, sheep and
chickens. Altogether these five animals generated 92% of the total revenue from livestock. We
begin with an analysis of all farms. We then examine small and large farms separately.




6 See Seo et al. (2006a) for the parametric calculation of the confidence intervals.




                                                           10

All farms

We estimated three sets of equations in our first model. The first set of equations determined
whether a farmer chooses a particular animal, the second estimated the number of animals a
farmer chooses given that he has picked a particular type of animal, and the third estimated the
net revenue per animal given that a farmer has chosen that animal. Because the choice of an
animal and the net value of each type of animal are linked, we used a selection model (Heckman
1979). We followed McFadden's model of multinomial choice to estimate the probability each
animal is chosen. The choice of each type of animal was assumed to be independent of the
choice of any other animal. The probability of choosing each animal was assumed to be a
function of summer and winter temperature and summer and winter precipitation. Other
explanatory variables included a dummy variable for West Africa, a dummy for large farms, and
a dummy variable for electricity.

Table 1 shows the results of the multinomial logit regression of the probability of choosing each
of the five animals. The base case is a household that chooses chickens. Most of the estimates are
very significant. The test of global significance of the model verifies that the model is highly
significant. The positive coefficients imply that the probability of choosing the animal increases
as the corresponding variable increases. The amount of increase of the probability can be read
from the odds ratio. For example, the odds ratio of the electricity dummy for beef cattle is 2.2,
which implies that farms with electricity are 2.2 times (odds ratio) more likely to own beef cattle.

The climate variables are mostly significant. The linear term on summer temperature for sheep is
negative but the quadratic term is positive. The same is true for winter temperature. The sheep
temperature response function is U-shaped. In contrast, for beef cattle, the linear term is positive
but the quadratic term is negative in summer temperature. The cattle temperature response
function is hill-shaped. The negative coefficients for West Africa for beef cattle and dairy cattle
indicate that West African farms are less likely to have cattle. The coefficients are positive for
goats and sheep, indicating that West African farms are more likely to choose goats and sheep.
One reason why this region is different from the rest of Africa may be livestock diseases such as
nagana (trypanosomiasis), also known as sleeping sickness. Thirty percent of Africa's 160
million cattle population are said to be at risk from these diseases.7 Large farmers are more likely
to choose large animals than chickens. Farms with electricity are more likely to choose beef
cattle, dairy cattle, and sheep, but not goats.

Although this is not included in Table 1, we also tested a set of soil variables, a set of household
characteristics, a set of labor use variables, and other social statistics such as religions. These
variables were dropped since they were not significant.

Figure 2a graphs the relationship between the probability of choosing a species and annual
temperature. Note that the mean temperature in sub-Saharan Africa is 22°C. The probability of
choosing beef cattle and dairy cattle decreases rapidly as temperature rises. In contrast, the
probability of choosing goats and sheep climbs as temperature rises. With chickens, the
estimated probability is hill-shaped, with a maximum at the current mean temperature of Africa.
The graph clearly reveals that the choice of animals in Africa today is very temperature sensitive.


7International Livestock Research Institute website at http://www.ilri.org.



                                                         11

Figure 2b displays the estimated relationship between the probability of choosing an animal and
annual precipitation. The probability of choosing beef cattle, dairy cattle and sheep all decrease
as precipitation increases. More rain increases the probability of disease (Ford and Katondo
1977) and, perhaps more importantly, shifts the ecosystem from savanna to forest (Sankaran et
al. 2005). All three of these animals are clearly more productive in grasslands. In contrast to the
above results, goats and especially chickens are more likely as rain increases. Goats may be able
to forage more successfully in wetter climates.

In the second stage of the analysis, we estimated the conditional net revenue functions. The net
revenue per animal for each chosen species was regressed on climate, the West Africa dummy
variable, and a livestock product dummy variable. We accounted for selection bias by using the
Dubin-McFadden selection bias correction. These conditional net revenue regressions used only
annual and not seasonal climate variables. The seasonal climate variables were not statistically
significant. The functional form is quadratic in both temperature and precipitation.

Table 2 summarizes the results of the regression of the conditional net revenue per animal. These
regressions confirm that the conditional net incomes from the five animals are sensitive to
climate. For all the animals except dairy cattle, the linear terms are negative and the quadratic
terms are positive, implying a U-shaped response function with respect to both temperature and
precipitation. The coefficients on the West African dummy reveal that West African farmers earn
relatively less from dairy cattle and relatively more from goats and sheep compared to other
farmers. The coefficient of the livestock product dummy suggests that farms that sell products
earn more revenue, except for farms with beef cattle.

The selection bias coefficients reveal interactions among the species. If the coefficients are
negative, they suggest the two animals would normally compete with each other. If the
coefficients are positive, they suggest the two animals are complementary. That is, the more
profitable the farm is for one animal, the more likely the farmer will also choose the other
animal. For example, in the beef cattle regressions, the coefficient on the selection term for dairy
cattle is positive, suggesting the two types of cattle are complementary. In contrast, the
coefficient on the selection term for goats and chickens is negative. Farms which find it
profitable to farm beef cattle are less likely to select goats and chickens.

Because the quadratic terms make the climate variables difficult to interpret, we present the
estimated climate response functions in Figure 3. Figure 3a shows how conditional net income
per animal responds to temperature for each of the five species. The conditional net income per
animal is generally higher for beef cattle but it decreases rapidly as temperatures rise. Income per
animal has a relatively stable hill-shape for dairy cattle with respect to temperatures. The
conditional net income per animal for goats, sheep and chickens decreases with temperature, but
the slope of the decrease is relatively small compared to cattle. Although the profitability of
goats, sheep and chickens all decrease with warming, they become relatively more attractive to
African farmers as temperatures rise.

Figure 3b shows how conditional net revenue responds to precipitation. The conditional net
revenue of beef cattle decreases precipitously the wetter it gets. Sheep conditional net revenue
also decreases with precipitation. However, the conditional net revenue of dairy cattle, goats, and
chickens do not seem to be affected by precipitation. Again, although increased precipitation in



                                                   12

general appears to reduce the profitability of African livestock, the relative profitability of dairy
cattle, goats, and chickens increases.

We also estimated a third set of regressions that predict the number of animals of the chosen
species a farmer has. As reported in Table 3, farms in districts with more pasture chose more beef
cattle and sheep per household, but fewer goats and chickens. Farms with electricity owned more
animals. The climate variables were often significant, with the exception of dairy cattle. Other
than for sheep, temperature responses are hill-shaped. Precipitation response functions are U-
shaped except in the case of goats.

In Figures 4a and 4b we present how the estimated numbers of animals change in response to
temperature and precipitation. Figure 4a shows that the number of beef cattle decreases sharply
as temperature increases while sheep decrease slightly and then increase in number. There are
slight increases in the numbers of goats and dairy cattle. Chickens have a hill-shaped response
function with respect to temperature and a U-shaped response to precipitation, but we omit them
from both figures because they are at an incompatible scale. Figure 4b shows that the numbers of
dairy cattle and goats are quite stable over a large range of precipitation, but the number of beef
cattle and sheep decrease rapidly with more rainfall.



Small and large farms

In this second analysis, we explored whether small and large livestock farms make different
livestock choices. We defined a farm with less than US$630 worth of animals as a small farm
and a farm with more than US$630 worth as a large farm. On small farms in Africa the livestock
is worth US$230 on average and on large ones US$7800 on average. Large farms earn over 95%
of the gross revenue from livestock in Africa. Note that pastoralists would generally be included
as large farms. Because large farms own a considerable amount of livestock, they tend to be
more commercially oriented. In contrast, farms with few livestock tend to be household farms
that rely more heavily on household labor and are less engaged in market activities.

We tested whether small and large farms are alike by creating a dummy variable for large farms.
In the species choice equation, we introduced a set of interaction terms which are the product of
this large farm dummy and each climate variable. These interaction terms measure whether large
farms have a different climate response from that of small farms. The total climate response of a
large farm is consequently the sum of the original climate coefficient and this interaction term.

Table 4 presents the multinomial logit regression results of species choice with the climate
interaction terms. Looking at the large farm­climate interaction terms, we can see that the choice
of some species depends on farm size. Large farms react differently from small farms to summer
temperature and precipitation for beef cattle, summer precipitation and winter temperature for
dairy cattle and sheep, and summer temperature for goats. Looking at the climate coefficients
alone, which reflect the sensitivity of small farms, we can see that the choice of beef cattle is not
sensitive to climate, whereas the choice of the remaining species is sensitive to summer
temperature and summer and winter precipitation. Dairy cattle are also sensitive to winter
temperature.



                                                 13

The number of each species a farm owns also depends on the farm size, as shown in Table 5.
Large farms, in general, are more sensitive to hotter temperature. Some large farm interaction
terms are significant such as the annual mean precipitation for beef cattle, sheep, and chickens.

Conditional net income is not dependent on farm size. We consequently rely on Table 2 to predict
conditional net income per animal.



5. Climate simulations

Uniform scenarios

This analysis examined the consequences of some uniform climate change scenarios. In the
warming scenarios, we increased existing temperatures by a constant amount across Africa. In
the precipitation scenarios, we changed rainfall proportionally by the same amount across Africa.
Although these climate scenarios are simplistic, they provide a sense of how climate change
affects model predictions. The scenarios include a uniform increase in temperature of +2.5°C and
+5.0°C and a uniform change in precipitation of -15% and +15% across all of Africa.

We calculate the probability of choosing a particular species at the current climate and for each
uniform climate change scenario using the climate parameters in Table 1. The changes in
probabilities for each species are shown in Table 6a. With a 2.5°C warming, the probability of
choosing beef cattle and chickens is predicted to decrease, while the probabilities of choosing
goats and sheep are expected to increase. The change in dairy cattle is positive but insignificant.
With a 5.0°C warming, these effects increase even further. A 15% increase/decrease in rainfall
causes a decrease/increase in the likelihood of choosing beef cattle, dairy cattle and sheep.
Chickens and goats, by contrast, are more likely/less likely to be chosen as precipitation
increases/decreases. The effects of the tested temperature changes are larger than the effects of
the tested precipitation changes.

We calculated the change in conditional net revenue per animal using the climate coefficients in
Table 2. The results, shown in Table 6b, reveal that the conditional net revenues for all the
animals are sensitive to warming. The net revenue per animal falls with a warming of 2.5°C. The
net revenues for all five animals fall substantially more with an increase of 5°C, except for goats
which are more heat tolerant. The rainfall effect on net revenues is much smaller. A 15%
increase/decrease in rainfall decreases/increases the net revenue for beef cattle, sheep, and
chickens, but increases/decreases the net revenue for dairy cattle and goats. These net revenue
changes are statistically insignificant except for beef cattle.

We then calculate the number of animals that farms would have under current climate conditions
and uniform change future scenarios using the coefficients from Table 3. Table 6c shows the
change in the number of each animal chosen given that a farmer has chosen that species.
Warming reduces the number of beef cattle and chickens, but increases the number of goats and
sheep. Additional warming accelerates this trend. More rainfall also reduces the number of
animals per farm, and less rainfall increases the number, with the exception of goats. The rainfall
effects are smaller than the temperature effects. The response of many of the range animals (beef



                                                  14

cattle, dairy cattle and sheep) to increases in precipitation is worth noting. Although it is true that
the productivity of pasture may increase with more rainfall, increased rain causes natural
ecosystems to move from grassland to forests. In a natural foraging system such as that of Africa,
moving from moderate to large amounts of rainfall is therefore not helpful for most livestock
(Sankaran et al. 2005).

Combining the results on the probability of species choice, the number of animals, and the
conditional net revenue (Tables 6a, 6b, and 6c), we calculate the expected net revenue per farm
in Table 6d for various climate change scenarios. The expected net revenue is the product of the
conditional net revenues per animal times the number of animals times their probability of being
chosen, summed across all species. The expected net revenue from animals on African farms is
currently $3000 per farm. Our model predicts a 32% loss in expected net revenue with a 2.5°C
warming, and a 69% loss with a 5°C warming. These predictions take into account the change in
probability of choosing each species, the drop in net revenue per animal, and the reduction in the
number of animals of that species. The expected revenues fall with temperature because of
reductions in the conditional net income of all species and because of shifts away from highly
profitable beef cattle. Rainfall effects are comparably smaller. A 15% increase in rainfall leads to
a loss of 2% in expected net revenue per household from livestock and a 15% decrease in rainfall
leads to a gain of 2%.

To see how these effects are distributed, we present in Figure 5 the probability of choosing beef
cattle in South Africa under the current climate, with + 2.5°C, and with +5°C warming. With the
current climate, beef cattle are evident in large numbers in the southeastern part of the country.
However, with 2.5°C warming the area suitable for beef cattle shrinks dramatically. With 5°C
warming, beef cattle disappear almost entirely from Africa. In Figure 6 we show the impact of
warming on sheep for all of Africa. There is a widespread reduction of sheep in South Africa
with warming. However, near the Sahel sheep flourish with warming. As temperatures warm,
sheep are distributed across a much wider area. A similar result applies to goats.

In the remaining uniform simulations, we examine the results for small and large farms. The
probabilities of choosing different species is calculated from the climate coefficients in Table 4.
Table 7a reveals that warming causes small livestock farms to shift to goats and sheep and away
from dairy cattle and chickens. The conditional net revenue effects per animal are calculated
from Table 2. Table 7b indicates that the net revenue per animal falls as temperatures rise for
beef cattle, goats, and chickens, but increases for dairy cattle, especially sheep. The number of
animals per farm are calculated from the climate coefficients in Table 5. Table 7c reveals that
farmers choose to have fewer of all animals. Finally, Table 7d calculates the expected income of
small farmers from the results in Tables 7a, 7b and 7c. The expected income of small farmers
falls by an average of 13% with 2.5°C warming, but by a negligible amount with 5°C warming.
A 15% decrease in precipitation is expected to increase small livestock farm incomes by 6%.

The calculations for Table 8 for large farms are similar to the calculation in Table 7 for small
farms except that they also rely on the climate coefficients that apply to large farms. Table 8a
reveals that warming causes large livestock farms to shift to dairy cattle and sheep and away
from goats, chickens and especially beef cattle. Table 8b indicates that the income per animal
falls for all species as temperatures rise. The percentage reduction is largest for goats, sheep and
chickens but the absolute amount is larger for beef cattle. Table 8c reveals that with higher


                                                  15

temperatures farmers choose to have fewer beef cattle, chickens and sheep but more goats and
dairy cattle. Finally, in Table 8d, large farmers' expected income falls by an average of 26% with
2.5°C warming and 67% with a 5°C warming. A 15% decrease in precipitation is expected to
increase large livestock farmers' incomes by 2%.



AOGCM scenarios

We also examined a set of climate change scenarios predicted by AOGCMs. The climate
scenarios reflect the A1 SRES scenarios from the following models: CCC (Boer et al. 2000),
CCSR (Emori et al. 1999), and PCM (Washington et al. 2000). For each model, we examined
country level climate change scenarios in 2020, 2060, and 2100. For each climate scenario, we
added the change in temperature predicted by each climate model to the baseline temperature in
each district. We also multiplied the percentage change in precipitation predicted by each climate
model by the baseline precipitation in each district or province. This gave us a new climate for
every district in Africa for each model and each time period.

Table 9 summarizes the climate scenarios of the three models for the years 2020, 2060, and
2100. The models predict a broad set of scenarios consistent with the range of outcomes in the
most recent IPCC (Intergovernmental Panel on Climate Change) report (Houghton et al. 2001).
In 2100, PCM predicts a 2°C temperature increase in Africa, CCSR a 4°C increase, and CCC a
6°C increase. Rainfall predictions are noisier: PCM predicts a 10% increase in rainfall in Africa,
CCC a 10% decrease, and CCSR a 30% decrease. In addition to the mean rainfall in Africa
varying substantially across the scenarios, there is also substantial variation in rainfall across
countries within each scenario.

Examining the path of climate change over time reveals that temperatures are predicted to
increase steadily until 2100 for all three models. Precipitation predictions, however, vary across
time: CCC predicts a declining trend, CCSR an initial decrease, and then increase, and decrease
again, and PCM an initial increase, and then decrease, and increase again.

We used the parameters from our estimated models to simulate the impacts of climate change on
the expected revenue of livestock management under various scenarios. Table 10 examines the
results using the models estimated for all farms. Table 10a shows the changes in the probabilities
of choosing a particular animal for each climate scenario. All the scenarios predict that farmers
would choose fewer beef cattle and chickens, but more goats and sheep. The choice of dairy
cattle, however, depends on the scenarios. On average, the probabilities of choosing beef cattle
decrease by 1% in 2020, 2% in 2060 and 3% in 2100. The probability of choosing sheep
increases over the same period.

We show the changes in the conditional net incomes per animal in Table 10b. The CCC model
predicts net income from beef cattle will steadily decrease over the next century. PCM predicts a
substantial drop in income from beef cattle until 2060, but the drop stabilizes afterwards. CCSR
predicts a substantial drop in income from beef cattle through 2060, but an increase afterward.
Net incomes for sheep and chickens also decrease over time, but the decreases are smaller. The
responses for dairy cattle and goats are dependent on the scenarios, but generally change little.



                                                  16

Table 10c calculates the changes in the number of animals of each species for each climate
scenario. Across all models, the number of beef cattle is predicted to decrease by an average of
five per farm in 2020, 10 in 2060, and 15 in 2100. Chickens also decrease in number. In contrast,
numbers of goats and sheep are predicted to increase over time. The number of sheep increases
rapidly after 2060 according to CCC and CCSR scenarios.

Combining the results from Tables 10a, 10b and 10c, the expected change in net income is
shown in Table 10d for each AOGCM scenario. In all cases, there are losses by the year 2020.
The CCC scenario predicts a 13% loss, the PCM scenario a 37% loss, and the CCSR scenario a
15% loss in expected net income per household. African farmers are expected to lose income
because they must switch to animals that provide lower returns (thus lowering the net income per
animal) and reduce the number of animals for most species. These net results are consistent with
a traditional Ricardian analysis of the same data (Seo & Mendelsohn 2006a). The damage
increases over time: by 2060 between 15 and 40% of expected income is lost, and between 40
and 70% by 2100.

The 95% confidence interval was calculated for all of these estimates using 200 bootstrap runs.
The impact estimates are significant for both uniform temperature change scenarios and uniform
precipitation change scenarios. They are also significant for the AOGCM scenarios in 2020,
2060, and 2100.

Table 10d also shows the aggregate impact across Africa. The results suggest that the livestock
damage will vary from a loss of $8 to $23 billion in livestock income in 2020, from to $9 to $24
billion in 2060, and from $25 to $40 billion in 2100. The PCM scenario predicts little change
from 2020 to 2060, and the CCSR scenario predicts the smallest loss in 2100 among the models
due to the large decrease in rainfall predicted in this period. The expected income from livestock
will fall because of warming but it will rise in some dry scenarios as farmers shift from crops to
livestock.

The remaining analysis looks at the consequences of the AOGCM scenarios for small and large
livestock farms. Table 11a examines how species choice changes with each AOGCM for small
farms. In 2100, the CCC model predicts that dairy cattle and sheep will be chosen more often as
the primary animal and goats and chickens less often. The income per animal will tend to fall
over time except for sheep. With the CCC scenario, dairy cattle and sheep will provide a higher
income in 2100. With the CCSR scenario in 2100, sheep and dairy cattle also provide higher
incomes but the effect is much smaller. The number of animals per farm, shown in Table 11c,
will tend to fall with warming, again with a few exceptions. With the CCSR and PCM scenarios,
the numbers of beef cattle will increase. The changes in the number of goats and sheep are
relatively negligible. Finally, in Table 11d, the expected income for small farms will tend to
increase over time with the CCC scenario up to $36 per farm or 34% in 2100. The income
fluctuates with the PCM and CCSR scenarios over time, resulting in an 8% increase with the
PCM scenario and a 5% decrease with the CCSR scenario in 2100.

Table 12a reveals that large farms in 2100 will shift away from beef cattle, goats and chickens in
favor of dairy cattle and sheep with the CCC scenario. However, with the CCSR scenario, large
farms will move away from beef cattle in favor of dairy cattle and goats. Finally, there will be
little shifting of species in the PCM scenario. Table 12b shows that there is a large absolute



                                                 17

reduction in the net revenue per animal for beef cattle and dairy cattle in the CCC scenario but
there is an even larger percentage loss for sheep and chickens in this scenario. In the CCSR
scenario, the effects are smaller but the net revenues for sheep and dairy cattle increase while the
net revenue for chickens falls. Finally, with the PCM scenario, the net revenue effects are even
smaller, with revenues for beef cattle, sheep and goats all falling but increasing for dairy cattle.
Table 12c reveals that in the CCC scenario the numbers of beef cattle and chickens will fall on
large farms but the number of goats will increase. With the CCSR scenario, the number of sheep
will increase. With the PCM scenario, the number of chickens will fall. Table 12d pulls all these
results together to predict that the CCC scenario will lead to a $6000 reduction in expected net
revenue per large farm (77%), the CCSR to a $2700 reduction (34%), and the PCM to a $3400
reduction (43%).

The effect of global warming on large farms is considerably more severe than on small farms in
Africa. In order to understand this result, it is helpful to know which species small and large
farms depend upon for income. Currently, small farmers get 60% of their livestock income from
dairy cattle, 13% from goats and 13% from sheep. Only 5% of their income comes from beef
cattle. In contrast, large livestock farms get 61% of their income from beef cattle and another
32% from dairy cattle. Small farms are well diversified, whereas large farms are specialized in
cattle, especially beef cattle. Commercial beef cattle prosper in temperate climates but respond
poorly to high temperatures. Large farms consequently suffer large losses in climate scenarios
that involve high temperatures, such as CCC. Small farmers, in contrast, have their portfolio in
more heat tolerant animals and will therefore gain from climate change according to the CCC
scenario.



6. Conclusion

This paper develops a new technique, the Structural Ricardian Model, to model a farmer's choice
as a simultaneous decision process. Farmers choose the profit maximizing level of inputs for
each animal (or crop if applied to crop farming), the species that provides the highest net revenue
and the number of animals of that species. The Structural Ricardian Model shows how each of
these decisions is influenced by climate, not just the way expected net revenue changes. The
model reveals the farmers' underlying decision making and gives insights into how they might
adapt to climate change.

We applied this analysis to livestock management in Africa. The multinomial choice model
revealed that the probability of selecting beef cattle and chickens will diminish sharply with
warming. This is completely consistent with the observation that current beef cattle operations
are located only in temperate locations across Africa. Further, the model predicts that numbers of
goats and sheep will increase with warming. This again is consistent with observations of where
goats and sheep are currently located, in relatively hot locations such as Burkina Faso, Niger and
Senegal.

The conditional net revenue analysis supports the multinomial choice results. Although the net
revenues for all the five major animals will decrease with warming, beef cattle will be the most
severely affected. Consequently, farmers will switch from beef cattle as temperatures rise.



                                                 18

Finally, the results of the model predicting the number of animals of each species is also
consistent with the results from the multinomial logit choice and conditional net revenue
analysis. All species will be hurt by warming and so there will be fewer animals per farm. Beef
cattle is especially vulnerable. The net profitability of livestock will be reduced and so farmers
will reduce their investments in livestock and shift away from beef cattle.

There has been very little quantitative research on animal husbandry in Africa so there are few
empirical studies to compare these results with. A standard Ricardian analysis was done by Seo
& Mendelsohn (2006a), using the same data. The results of the standard Ricardian model are
quite similar to the results in this paper. That is, the model predicts that net revenues will fall
with either rising temperature or rising rainfall levels.

All the AOGCM predictions suggest that African livestock will be damaged as early as 2020.
Even small changes in temperature will be sufficient to have a relatively large effect on beef
cattle operations. Additional warming is expected to exacerbate these damages. Farmers
dependent on beef cattle will be especially hard hit. In contrast, small farms who can switch to
sheep or goats may not be as vulnerable to higher temperatures compared to large farms that
cannot make this switch. Precipitation also plays an important role in the AOGCM results.
Scenarios with less precipitation are less harmful. Because pastures and ecosystems in general
are more productive with more rain, this result may seem counterintuitive. However, in Africa,
lower precipitation may reduce animal diseases that are quite significant for livestock. Perhaps
more importantly, less rain shifts forest ecosystems to savanna or grasslands. These grasslands
are more productive for sheep, dairy cattle and beef cattle. Reductions in precipitation from large
to moderate levels appear to be beneficial to livestock. As long as there is sufficient precipitation
to support grasslands, the livestock will gain.

It is important to note that the economic viability of large livestock operations is more vulnerable
to warming, largely because they depend on beef cattle. Although commercial scale livestock
operations do well in temperate parts of Africa today, they have few alternatives with warming.
Warming will force reductions in beef and dairy cattle, critical to many commercial enterprises.
In contrast, small farmers have many substitutes. If it gets warmer, they can shift to heat tolerant
animals such as goats and sheep. In these circumstances, small farmers in Africa are actually
better able to adapt to climate change than their larger more modern counterparts.

It is interesting to compare the results of this analysis with the Ricardian study of livestock (Seo
& Mendelsohn 2006a). This study predicts that large livestock farms will be hurt by warming but
that small farms will in fact gain. These qualitative results are also supported in this paper. The
Ricardian study predicts a range of expected income effects from climate change depending on
how the model is estimated. The structural results are in the middle of this range. The structural
results are different but consistent with the more traditional Ricardian model predictions.

This analysis reveals that farmers will most likely adapt to climate change. It suggests that
farmers will switch species and move away from cattle and towards goats and sheep. Small
farmers will be able to make these changes without much change in expected income. However,
these changes are predicted to reduce the net incomes of large farms considerably. African policy
makers must be careful to encourage private adaptation during this period of change. There may
be nothing that can be done to sustain the large cattle operations that depend on current climate.



                                                  19

Providing subsidies or other enticements for such operations to continue once the climate
changes would only compound the problem. Instead, governments should encourage farmers to
change the composition of animals on their farms as needed. That is, they should inform farmers
about how other livestock owners have coped with higher temperatures and share indigenous
knowledge. Governments should anticipate that farmers will make changes on their lands and do
whatever is needed to facilitate these changes.




                                               20

REFERENCES

Adams R et al., 1990. Global climate change and US agriculture. Nature 345: 219­224.

Adams R et al., 1999. The economic effects of climate change on US agriculture. In Mendelsohn
      R & Neumann, J (eds), The Impact of Climate Change on the United States Economy,
      Cambridge, UK: Cambridge University Press.

Boer G, Flato G & Ramsden D, 2000. A transient climate change simulation with greenhouse gas
      and aerosol forcing: Projected climate for the 21st century. Climate Dynamics 16: 427­
      450.

Bourguignon F, Fournier M & Gurgand M, 2004. Selection bias corrections based on the
      multinomial logit model: Monte-Carlo comparisons. DELTA Working Paper No. 20,
      École Normale Supérieure.

Chib S & Greenberg E, 1998. An analysis of multivariate probit models. Biometrika 85: 347­
      361.

Chow G, 1983. Econometrics. New York: McGraw-Hill.

Dinar A, Hassan R, Kurukulasuriya P, Benhin J & Mendelsohn R, 2006. The policy nexus
      between agriculture and climate change in Africa. A synthesis of the investigation under
      the GEF / WB Project, Regional Climate, Water and Agriculture:           Impacts on and
      Adaptation of Agro-ecological Systems in Africa. CEEPA Discussion Paper No. 39,
      Centre for Environmental Economics and Policy in Africa, University of Pretoria.

Dubin JA & McFadden DL, 1984. An econometric analysis of residential electric appliance
      holdings and consumption. Econometrica 52(2): 345­362.

Emori, S, Nozawa T, Abe-Ouchi A, Namaguti A & Kimoto M, 1999. Coupled ocean-atmospheric
      model experiments of future climate change with an explicit representation of sulfate
      aerosol scattering. Journal of the Meteorological Society of Japan 77: 1299­1307.

Ford J & Katondo K, 1977. The distribution of tsetse flies in Africa. Organization of African
      Unity, Nairobi, Kenya.

Greene WH, 1998. Econometric Analysis (3rd edition). New Jersey: Prentice Hall.

Heckman JJ, 1979. Sample selection bias as a specification error. Econometrica 47(1): 153­162.

Houghton JT et al. (eds), 2001. Climate Change 2001: The Scientific Basis. Contribution of
      Working Group I to the third assessment report of the Intergovernmental Panel on
      Climate Change. Cambridge: Cambridge University Press.

Kurukulasuriya P & Mendelsohn R, 2006. A Ricardian analysis of the impact of climate change
      on African cropland. CEEPA Discussion Paper No. 8, Centre for Environmental



                                              21

       Economics and Policy in Africa, University of Pretoria.

McFadden DL, 1981. Econometric models of probabilistic choice. In McFadden, DL, Structural
       Analysis of Discrete Data and Econometric Applications. Cambridge: MIT press.

McFadden DL, 1999. Chapter 1. Discrete Response Models. University of California at
       Berkeley, Lecture Notes.

Mendelsohn R (ed), 2001. Global Warming and the American Economy: A Regional Analysis.
       Edward Elgar, Northhampton, MA, USA.

Mendelsohn R, Nordhaus WD & Shaw D, 1994. The impact of global warming on agriculture: A
       Ricardian analysis. American Economic Review 84: 753­771.

Pearce D et al., 1996. The social costs of climate change: Greenhouse damage and benefits of
       control. In Bruce J, Lee H & Haites E (eds), Climate Change 1995: Economic and Social
       Dimensions of Climate Change. Cambridge: Cambridge University Press, pp.179­224.

Reilly J et al., 1996. Agriculture in a changing climate: Impacts and adaptations. In Watson R,
       Zinyowera M, Moss R & Dokken D (eds), 1996. Climate Change 1995: Impacts,
       Adaptations, and Mitigation of Climate Change: Scientific-Technical Analyses,
       Cambridge: Cambridge University Press for the Intergovernmental Panel on Climate
       Change (IPCC).

Sankaran M ET AL., 2005. Determinants of woody cover in African savannas, Nature 438: 846­
       849.

Seo SN & Mendelsohn R, 2006a. Climate change impacts on animal husbandry in Africa: A
       Ricardian analysis.     CEEPA Discussion Paper No. 9, Centre for Environmental
       Economics and Policy in Africa, University of Pretoria.

Seo S N & Mendelsohn R, 2006b. Climate change adaptation in Africa: A microeconomic
       analysis of livestock choice. CEEPA Discussion Paper No. 19, Centre for Environmental
       Economics and Policy in Africa, University of Pretoria.

Seo SN, Mendelsohn R & Munasinghe M, 2005. Climate change impacts on agriculture in Sri
       Lanka. Environment and Development Economics 10: 581­596.

Train K, 2003. Discrete Choice Methods with Simulation. Cambridge, UK: Cambridge
       University Press.

Washington W et al., 2000. Parallel Climate Model (PCM): Control and transient scenarios.
       Climate Dynamics 16: 755­774.




                                               22

Table 1: Multinomial logit selection model


                                         Beef cattle                             Dairy cattle


Variable                     Estimate        2       Odds ratio      Estimate           2     Odds ratio


Intercept                     -2.916       1.450                      13.336         74.287

Temperature summer             0.496       6.711         1.642        -1.145        100.922     0.318

Temperature summer sq         -0.014       11.261        0.986         0.019         58.532     1.019

Precipitation summer           0.015       14.592        1.015        -0.022         49.731     0.978

Precipitation summer sq        0.000       28.270        1.000         0.000         15.070     1.000

Temperature winter            -0.556       19.712        0.573         0.175          2.290     1.192

Temperature winter sq          0.018       23.700        1.018         0.004          1.571     1.004

Precipitation winter          -0.004       0.501         0.996        -0.032         43.425     0.969

Precipitation winter sq        0.000       0.439         1.000         0.000          9.098     1.000

West Africa                   -1.088       22.092        0.337        -3.092        226.729     0.045

Large farms                    4.097      540.972       60.178         2.654        416.886    14.209
Electricity                    0.785       17.854        2.192         0.599         13.550     1.821



Table 1: (continued)


                                           Goats                                      Sheep


Variable                     Estimate        2       Odds ratio      Estimate           2     Odds ratio


Intercept                      6.564       14.871                     12.307         51.906

Temperature summer            -0.804       50.026        0.448        -0.803         43.024     0.448

Temperature summer sq          0.016       48.575        1.016         0.014         33.097     1.014

Precipitation summer          -0.007       6.668         0.993        -0.007          4.020     0.993

Precipitation summer sq        0.000       9.204         1.000         0.000          0.185     1.000

Temperature winter             0.174       1.511         1.191        -0.404         12.038     0.668

Temperature winter sq         -0.001       0.019         0.999         0.015         22.158     1.015

Precipitation winter          -0.024       24.591        0.977        -0.027         23.462     0.974

Precipitation winter sq        0.000       16.397        1.000         0.000          6.173     1.000

West Africa                    0.446       6.735         1.562         0.935         20.778     2.547

Large farms                    0.888       51.060        2.429         1.694        182.293     5.440
Electricity                   -0.048       0.112         0.953         0.320          4.606     1.378


Note: Likelihood ratio test: P<0.0001, Lagrange multiplier test: P<0.0001, Wald test: P<0.0001




                                                        23

Table 2: Conditional net revenue per animal regression

                              Beef cattle           Dairy cattle           Chickens

Variable                  Estimate  T-statistic Estimate  T-statistic Estimate   T-statistic


Intercept                 846.23        2.55    -479.90      -1.57     6.92        2.73
Temperature mean          -12.12       -0.37     66.37        2.24     -0.41       -1.73
Temperature mean sq        -0.08       -0.10     -1.67        -2.3     0.01        1.03
Precipitation mean         -6.64       -6.30     -3.40       -5.03     -0.02       -4.16
Precipitation mean sq      0.03         3.93     0.02         4.48     0.00        3.74
West Africa                11.01        0.22     -7.75       -0.54     0.00        0.01
Livestock products        -36.86       -2.74     43.45        3.84     1.20        8.83
Cattle beef ­ selection                         -113.57      -3.62     -1.40       -2.97
Cattle dairy ­ selection   77.43        1.39                           1.45         2.4
Goats ­ selection         -134.78      -0.89     72.22        0.66     -1.70       -2.37
Sheep ­ selection         222.06        1.05    135.43        1.36     1.25        1.29
Chickens ­ selection      -129.12      -1.84    -116.15      -2.69
ADJ RSQ                    0.54                  0.15                  0.20
N                           548                  1084                  1047



Table 2: (continued)


                                Goats                 Sheep

Variable                  Estimate  T-statistic Estimate  T-statistic


Intercept                  70.29        4.45    119.95        6.51
Temperature mean           -4.69       -3.69     -6.90       -4.13
Temperature mean sq        0.09         3.24     0.12         3.27
Precipitation mean         -0.17       -4.01     -0.49       -8.48
Precipitation mean sq      0.00         6.05     0.00         7.88
West Africa                7.49         2.46     5.16         0.92
Livestock products         10.54       12.88     8.71         7.55
Cattle beef ­ selection    -7.38       -2.32     -0.88       -0.32
Cattle dairy ­ selection   2.34         0.67     9.17         2.23
Goats ­ selection                               -17.61       -2.93
Sheep ­ selection          7.43         1.35
Chickens ­ selection       -0.14       -0.06     10.73        2.31
ADJ RSQ                    0.28                  0.22
N                          1096                   908




                                                24

Table 3: Conditional number of animals regression

                              Beef cattle           Dairy cattle            Chickens


Variable                  Estimate  T-statistic Estimate  T-statistic Estimate    T-statistic


Intercept                 -21.27       -0.09    -49.65       -1.30    -3283.10      -1.57
Annual temperature         24.49        1.12     4.11         1.10     604.30       3.13
Temperature sq             -0.73       -1.31     -0.08       -0.86     -18.65       -4.09
Annual precipitation       -1.50       -1.89     -0.03       -0.31     -34.89       -5.36
Precipitation sq           0.00         0.69     0.00         0.11      0.10        2.88
Electricity dummy          66.98        5.79     7.61         3.57      86.82       0.49
% pasture                  9.30         0.18                          -1061.09      -2.02
Livestock products                               36.71        4.40
Cattle beef ­ selection                          -6.97       -1.55    -4075.22      -9.02
Cattle dairy ­ selection   54.40        2.47                           1742.12      4.74
Goats ­ selection         -155.51      -1.53     29.85        2.16    -3724.64      -4.48
Sheep ­ selection          74.73        0.91    -21.25       -1.62     4735.25      6.00
Chickens ­ selection       44.75        1.05     -1.46       -0.27
ADJ RSQ                    0.43                  0.21                   0.09
N                           508                  1084                   1047



Table 3: (continued)

                                Goats                 Sheep


Variable                 Estimate   T-statistic Estimate  T-statistic


Intercept                 -68.63       -4.25    131.95        1.94
Annual temperature         3.64         3.02    -10.24       -1.81
Temperature sq            -0.06        -2.49     0.21         1.75
Annual precipitation       0.03         0.75     -1.32       -6.23
Precipitation sq           0.00         2.05     0.01         5.15
% pasture                 -3.45        -2.95    203.23       11.64
Electricity dummy          2.28         2.57     2.02         0.38
Livestock products                               -2.34       -0.59
Cattle beef ­ selection    5.25         1.52    -126.91     -11.06
Cattle dairy ­ selection  -2.95        -1.27     70.85        6.60
Goats ­ selection                                21.63        1.03
Sheep ­ selection         -24.52       -6.19
Chickens ­ selection       8.99         4.04     17.42        1.04
ADJ RSQ                    0.56                  0.38
N                          1096                   908




                                                25

Table 4: Multinomial logit species choice with large farm interactions


                                   Beef cattle                   Dairy cattle



Variable                  Estimate     2      Odds ratio Estimate    2       Odds ratio


Intercept                  3.996     0.939               26.743   153.151

Temperature summer         -0.262    0.671     0.770     -1.673   109.630     0.188

Temperature summer sq      0.004     0.299     1.004      0.026    58.193     1.027

Precipitation summer       -0.002    0.043     0.998     -0.044    93.672     0.957

Precipitation summer sq    0.000     0.250     1.000      0.000    51.533     1.000

Temperature winter         -0.375    1.856     0.688     -0.428    7.394      0.652

Temperature winter sq      0.014     2.880     1.014      0.022    22.795     1.022

Precipitation winter       -0.004    0.129     0.996     -0.032    26.882     0.968

Precipitation winter sq    0.000     0.069     1.000      0.000    4.051      1.000

Temperature summer* B      0.947     4.032     2.578      0.585    2.786      1.795
Temp summer sq* B          -0.019    3.565     0.981     -0.004    0.286      0.996

Precipitation summer *B    0.038    13.869     1.038      0.047    37.306     1.048

Prec summer sq* B          0.000     9.337     1.000      0.000    32.427     1.000

Temperature winter *B      -0.269    0.642     0.764      0.750    8.575      2.117

Temp winter sq *B          0.007     0.432     1.007     -0.022    8.545      0.978

Precipitation winter* B    -0.007    0.156     0.993     -0.004    0.078      0.996

Prec winter sq *B          0.000     0.852     1.000      0.000    1.379      1.000

West Africa                -1.647   41.268     0.193     -3.767   248.382     0.023

Large farms                -6.469    1.115     0.002     -16.155   11.626     0.000
Electricity                0.912    21.854     2.490      0.741    17.890     2.097



 Note: B is large farm dummy.




                                               26

Table 4: (continued)


                                    Goats                       Sheep



Variable                  Estimate    2    Odds ratio Estimate    2    Odds ratio


Intercept                  2.284    1.269   9.813      6.876    8.708

Temperature summer         -0.622   22.297  0.537     -0.788    28.011  0.455

Temperature summer sq      0.013    25.081  1.014      0.014    21.281  1.014

Precipitation summer       -0.004   1.710   0.996     -0.013    12.773  0.987

Precipitation summer sq    0.000    5.730   1.000      0.000    3.931   1.000

Temperature winter         0.338    3.236   1.403      0.122    0.358   1.129

Temperature winter sq      -0.006   1.289   0.994      0.004    0.434   1.004

Precipitation winter       -0.019   13.052  0.982     -0.019    9.311   0.981

Precipitation winter sq    0.000    8.604   1.000      0.000    0.576   1.000

Temperature summer *B      -0.838   5.553   0.433     -0.387    1.136   0.679

Temp summer sq *B          0.016    4.569   1.016      0.011    2.109   1.011

Precipitation summer *B    -0.004   0.238   0.996      0.029    13.436  1.029

Prec summer sq *B          0.000    0.171   1.000      0.000    15.884  1.000

Temperature winter *B      -0.211   0.442   0.810     -0.760    6.891   0.468

Temp winter sq *B          0.008    0.931   1.009      0.017    4.363   1.017

Precipitation winter *B    -0.023   2.634   0.977     -0.020    1.990   0.980

Prec winter sq *B          0.000    2.601   1.000      0.000    4.199   1.000

West Africa                0.421    5.727   1.523      0.601    8.144   1.824

Large farms                13.186   7.040   532881    11.440    5.313   92997
Electricity                0.034    0.054   1.034      0.457    9.076   1.579


 Note: B is large farm dummy.




                                            27

Table 5: Conditional number of animals regression with large farm dummy


                                Beef cattle           Dairy cattle          Chickens



Variable                  Estimate    T-statistic Estimate  T-statistic Estimate T-statistic


Intercept                    25.46        0.10    -59.01       -1.53    -109.48    -0.07
Annual temperature           0.64         1.03     -0.25       -1.19     13.12     0.10
Temperature sq               -0.02       -1.02     0.01         1.18     -0.51     -0.17
Annual precipitation         -0.04       -1.39     -0.01       -2.39     2.79      0.78
Precipitation sq             0.00         1.45     0.00         1.48     -0.02     -0.84
Annual temp *B               18.66        1.02     6.91         1.31    395.30     2.70
Temperature sq *B            -0.39       -0.85     -0.15       -1.14    -12.92     -4.40
Annual prec *B               -1.71       -1.91     -0.13       -0.94    -79.69    -11.80
Precipitation sq *B          0.01         1.74     0.00         0.06     0.35      10.42
Electricity dummy            68.03        5.81     8.04         3.78     63.25     42.89
% pasture                    78.96        1.88                          -279.58    -1.08
Livestock products                                 40.75        5.26
Cattle beef ­ selection                            11.38        0.61    -437.39    -0.75
Cattle dairy ­ selection     17.95        0.19                          146.64     0.61
Goats ­ selection           -147.97      -0.99     -2.51       -0.15     50.29     0.15
Sheep ­ selection            28.11        0.36    -22.20       -1.59    141.02     0.41
Chickens ­ selection        107.65        1.05     13.36        0.94
ADJ RSQ                      0.22                  0.10                  0.85
N                            508                   1084                  1047


Note: B is large farm dummy.




                                                  28

Table 5: (continued)


                                 Goats                Sheep



Variable                  Estimate  T-statistic Estimate T-statistic


Intercept                   70.34     4.29       1.96      0.03
Annual temperature          -5.36     -4.13      0.10      0.12
Temperature sq              0.10      3.53       -0.01     -0.41
Annual precipitation        -0.01     -0.09      0.03      1.29
Precipitation sq            0.00      0.04       0.00      -0.25
Annual temp *B              2.38      1.55       -8.43     -0.71
Temperature sq *B           -0.05     -1.54      0.10      0.42
Annual prec *B              -0.23     -0.69      -2.31     -5.13
Precipitation sq *B         0.00      0.61       0.01      3.93
% pasture                   0.62      0.49      245.12     14.34
Electricity dummy           7.06      1.80       33.58     5.11
Livestock products                               -4.55     -1.16
Cattle beef ­ selection     10.13     0.58      254.11     2.22
Cattle dairy ­ selection    3.02      0.52      -68.75     -1.69
Goats ­ selection                               -148.55    -4.27
Sheep ­ selection           -4.70     -1.00
Chickens ­ selection        -6.18     -0.62     -59.90     -1.01
ADJ RSQ                     0.29                 0.39
N                           1096                  908


Note: B is large farm dummy.




                                                29

Table 6a: Predicted change in the probability of selecting each animal from uniform
climate scenarios

                         Beef cattle Dairy cattle   Goats     Sheep      Chickens

 Baseline probability      11.8%        23.1%       23.4%     19.4%       22.3%
 Increase temp 2.5°C       -1.7%         0.4%       0.8%      3.3%        -2.8%
 Increase temp 5°C         -3.8%         2.1%       0.0%      8.7%        -7.0%
 Decrease rain 15%         -0.3%         1.8%       -1.2%     1.1%        -1.4%
 Increase rain 15%         -0.1%        -1.5%       1.5%      -1.1%        1.0%


Table 6b: Predicted change in net income per animal from uniform climate scenarios
(US$/animal)

                         Beef cattle Dairy cattle   Goats     Sheep      Chickens


Baseline income            145.54       132.09       6.49     11.77        1.14
Increase temp 2.5°C        -27.80        -3.40      -0.81     -2.55        -0.34
Increase temp 5°C          -47.09       -21.36      -0.54     -3.49        -0.55
Decrease rain 15%           13.91        -0.42      -0.06      1.03        0.03
Increase rain 15%           -7.08         3.78       0.29     -0.58        -0.02


Table 6c: Predicted change in number of each animal per farm from uniform climate
scenarios (animals/household)

                         Beef cattle Dairy cattle   Goats     Sheep      Chickens


Baseline number             63.47        23.84      15.36     34.05       790.09
Increase temp 2.5°C         -9.00         1.84       1.45      0.35       -112.61
Increase temp 5°C          -18.96         2.88       2.14      3.20       -183.84
Decrease rain 15%           7.52          0.13      -0.90      2.99        50.30
Increase rain 15%           -6.33        -0.11       0.99     -1.80       -45.20


Table 6d: Predicted change in expected income from uniform climate scenarios (US$)

                           Mean          Total     % change  Bootstrap  Bootstrap
                         (US$/farm) (billions US$)          lower 95 %  upper 95%


Expected income             3023          60
Increase temp 2.5°C         -964          -19      -31.90%    -1077        -722
Increase temp 5°C           -2083         -41      -68.89%    -2452        -1631

Decrease rain 15%            184           2        2.49%      139          207
Increase rain 15%           -108          -2       -1.88%      -111         -60




                                              30

Table 7a: Predicted change in the probability of selecting each animal from uniform
climate scenarios for small farms

                          Beef cattle Dairy cattle   Goats     Sheep     Chickens

 Baseline probability        2.20%      15.40%      29.60%     18.27%    34.53%
 Increase temp 2.5°C         0.17%      -3.27%       2.13%     4.03%     -3.07%
 Increase temp 5°C           0.36%       0.09%       1.27%     7.57%     -9.29%
 Decrease rain 15%           0.02%       1.69%      -1.69%     1.48%     -1.50%
 Increase rain 15%          -0.07%      -0.95%       1.79%     -1.39%    0.62%


Table 7b: Predicted change in net income per animal from uniform climate scenarios for
small farms (US$)

                          Beef cattle Dairy cattle   Goats     Sheep     Chickens


Baseline income              151.38      140.98       6.55      12.89      0.86
Increase temp 2.5°C          -51.12        1.34      -1.63      0.18      -0.18
Increase temp 5°C            -80.99        7.90      -2.24      3.78      -0.18
Decrease rain 15%            18.70         3.37      -0.43      0.48       0.03
Increase rain 15%            -13.60       -3.14       0.66      -0.30     -0.01


Table 7c: Predicted change in number of each animal per farm from uniform climate
scenarios for small farms (animals/household)

                          Beef cattle Dairy cattle   Goats     Sheep     Chickens


Baseline number               1.48         2.92       7.09      5.83      31.10
Increase temp 2.5°C          -0.16        -0.02      -0.66      -0.77     -8.15
Increase temp 5°C            -0.05        -0.00      -1.17      -1.64     -12.13
Decrease rain 15%            -0.03         0.03      -0.06      -0.21      1.35
Increase rain 15%             0.08        -0.02       0.06      0.20      -1.01


Table 7d. Predicted change in expected income from uniform climate scenarios for small
farms (US$)

                             Mean         Total     % change  Bootstrap Bootstrap
                          (US$/farm) (billions US$)          lower 95 % upper 95%


Expected income               102          6.0
Increase temp 2.5°C           -13          -0.8     -12.7%       -24        3
Increase temp 5°C              -1          -0.1      -0.9%       -8         6

Decrease rain 15%              6           0.4       5.9%         1        13
Increase rain 15%              -3          -0.2      -2.9%       -6         4




                                               31

Table 8a: Predicted change in the probability of selecting each animal from uniform
climate scenarios for large farms

                           Beef cattle Dairy cattle  Goats     Sheep     Chickens

 Baseline probability        24.5%       33.4%       15.1%     20.8%      6.1%
 Increase temp 2.5°C         -4.6%        4.5%       -1.5%     3.4%       -1.8%
 Increase temp 5°C           -10.0%       8.7%       -3.8%     8.7%       -3.5%
 Decrease rain 15%           -1.2%        1.2%        0.1%     0.5%       -0.6%
 Increase rain 15%            0.6%       -1.0%        0.3%     -0.6%      0.7%


Table 8b: Predicted change in net income per animal from uniform climate scenarios for
large farms (US$)

                           Beef cattle Dairy cattle  Goats     Sheep     Chickens


Baseline income              146.01      120.18       6.04     12.08       1.68
Increase temp 2.5°C          -15.30       -3.75       -1.54    -5.19      -0.43
Increase temp 5°C            -28.11      -21.10       -2.48    -8.29      -0.87
Decrease rain 15%             12.19       -1.31       0.39      1.41      -0.14
Increase rain 15%             -4.60        9.01       -0.25    -0.59       0.21


Table 8c: Predicted change in number of each animal per farm from uniform climate
scenarios for large farms (animals/household)

                           Beef cattle Dairy cattle  Goats     Sheep     Chickens


Baseline number               77.58       35.82       23.81    49.40     1907.25
Increase temp 2.5°C           -3.64        1.26       6.85     -1.27     -178.14
Increase temp 5°C            -13.91        1.31      15.96     -4.91     -369.30
Decrease rain 15%             4.86         0.36       0.76      5.75      47.37
Increase rain 15%             -2.20       -0.54       -1.20    -2.69      -10.11


Table 8d. Predicted change in expected income from uniform climate scenarios for large
farms (US$)

                             Mean         Total     % change  Bootstrap Bootstrap
                          (US$/farm) (billions US$)          lower 95 % upper 95%


Expected income               7826         60
Increase temp 2.5°C          -2033         -16       -26.0%    -2504      -1617
Increase temp 5°C            -5206         -40       -66.5%    -6843      -4664

Decrease rain 15%             160           1         2.0%      121        199
Increase rain 15%             -152         -1        -1.9%      -189       -77




                                               32

Table 9: African average AOGCM climate scenarios


                         Current      2020         2060     2100


Temperature (°C )

          CCC             23.29       24.94        26.85    29.96
          CCSR            23.29       25.27        26.17    27.39
          PCM             23.29       23.95        24.94    25.79

Rainfall (mm/month)

          CCC             79.75       76.84        71.86    65.08
          CCSR            79.75       73.99        76.67    62.44
          PCM             79.75       89.58        80.72    83.18




                                         33

Table 10a: Predicted change in the probability of selecting each animal from AOGCM
climate scenarios


                          Beef cattle Dairy cattle  Goats     Sheep    Chickens

   Baseline probability    11.8%        23.1%       23.4%     19.4%      22.3%


   2020
             CCC           -1.59%      1.31%        1.75%     1.90%      -3.37%
            CCSR           -0.54%      -0.10%       1.59%     1.86%      -2.81%
             PCM           -2.10%      -1.37%       5.51%     1.24%      -3.28%

   2060
             CCC           -1.70%      -1.46%       4.69%     2.87%      -4.40%
            CCSR           -1.24%      4.41%        -1.90%    2.10%      -3.38%
             PCM           -2.09%      3.99%        -2.65%    8.62%      -7.87%

   2100
             CCC           -3.93%       6.81%       -5.73%   16.69%     -13.84%
            CCSR           -3.29%      -0.10%       4.60%     1.88%      -3.10%
             PCM           -2.08%      -0.87%       4.34%     4.01%      -5.39%



Table 10b: Predicted change in net income per animal from AOGCM climate scenarios
(US$/animal)


                          Beef cattle Dairy cattle  Goats     Sheep    Chickens

 Baseline income           145.00      132.09        6.49     11.77       1.14


 2020
             CCC           -12.45        0.48        -0.36    -1.49       -0.24
            CCSR           -22.68       -2.59        -0.22    -0.92       -0.28
            PCM            -14.07       12.49        0.01     -1.12       -0.11

 2060
             CCC           -21.46       12.66        -0.21    -1.72       -0.22
            CCSR           -58.13       -6.06        -0.33    -2.68       -0.39
            PCM            -27.58       -4.62        0.24     -2.29       -0.43

 2100
             CCC           -42.92       -22.32       1.90     -1.74       -0.67
            CCSR            -9.32       53.50        2.13     4.29        -0.32
            PCM            -27.61       17.21        0.25     -1.30       -0.31




                                             34

Table 10c: Predicted change in number of each animal per farm from AOGCM climate
scenarios (animals/farm)


                         Beef cattle Dairy cattle  Goats      Sheep     Chickens

 Baseline number             63           24         15         34        790


 2020
            CCC              -3            1          1         1         -57
            CCSR             -7            1          0         3         -64
            PCM              -8            0          1         -2        -84

 2060
            CCC              -9            1          1         -2       -103
            CCSR             -11           2          2         0        -128
            PCM              -9            2          1         3        -115

 2100
            CCC              -21           3          0         12       -171
            CCSR             -6            3          1         23        -6
            PCM              -12           2          2         2        -129



Table 10d. Predicted change in expected income from AOGCM climate scenarios (US$)


                           Mean          Total     % change  Bootstrap Bootstrap
                         (US$/farm) (billions US$)          lower 95 % upper 95%

Expected income             3023          60


2020
            CCC             -423          -8       -14.0%      -444      -278
           CCSR             -465          -9       -15.4%      -487      -296
            PCM            -1135          -23      -37.5%     -1710      -640

2060
            CCC            -1229          -24      -40.6%     -1768      -682
           CCSR             -445          -9       -14.7%      -528       -23
            PCM            -1091          -22      -36.1%     -1348      -806

2100
            CCC            -2200          -44      -72.8%     -2783      -1697
           CCSR            -1261          -25      -41.7%     -1604      -893
            PCM            -1437          -29      -47.5%     -1989      -847




                                              35

Table 11a: Predicted change in the probability of selecting each animal from AOGCM
climate scenarios for small farms


                          Beef cattle Dairy cattle  Goats     Sheep    Chickens

   Baseline probability      2.20%     15.40%       29.60%   18.27%     34.53%


   2020
             CCC             0.2%       -1.0%       0.7%      2.6%       -4.1%
            CCSR             0.3%       -1.7%       0.9%      2.8%       -3.0%
             PCM             -0.1%      2.0%        1.5%      0.6%       -6.0%

   2060
             CCC             0.0%       2.4%         1.7%     1.6%       -6.7%
            CCSR             -0.1%      4.9%        -3.9%     3.1%       -5.0%
             PCM             0.5%       5.7%        -4.5%     7.8%       -9.5%

   2100
             CCC             1.0%       20.0%       -8.5%     1.1%       -18.6%
            CCSR             0.0%       2.8%        -1.3%     0.6%       -5.7%
             PCM             0.1%       4.3%        -0.3%     1.6%       -8.3%



Table 11b: Predicted change in net income per animal from AOGCM climate scenarios for
small farms (US$)


                          Beef cattle Dairy cattle  Goats     Sheep    Chickens

 Baseline income             151.38    140.98        6.55     12.89       0.86


 2020
             CCC             -29.98      4.08       -0.89     0.52        -0.12
            CCSR             -37.88     -1.41       -1.09     1.67        -0.12
            PCM              -29.57     -14.45       0.14     -0.06       -0.07

 2060
             CCC             -41.28     -12.05       -0.88    -0.54       -0.12
            CCSR             97.12       0.50       -0.55     0.72        -0.19
            PCM              -68.07      5.34       -1.01     3.04        -0.14

 2100
             CCC             -94.27     21.20        -1.92    10.95       -0.09
            CCSR             -86.76      8.88       -0.90     5.91        -0.06
            PCM              -57.88     -12.26      -0.94     1.16        -0.15




                                             36

Table 11c: Predicted change in number of each animal per farm from AOGCM climate
scenarios for small farms (animals/farm)


                          Beef cattle Dairy cattle   Goats     Sheep     Chickens

 Baseline number             1.48          2.92       7.09      5.83      31.10


 2020
            CCC              0.04         -0.01      -0.57     -0.39      -3.98
            CCSR             0.05         -0.02      -0.70     -0.57      -4.28
            PCM              0.21         -0.01       0.01      0.34      -2.81

 2060
            CCC              0.25         -0.01      -0.06     -0.24      -4.67
            CCSR             0.07         -0.02      -0.12     -0.57      -6.82
            PCM              0.04          0.00      -1.10     -0.99      -5.63

 2100
            CCC             -0.09          0.19      -0.27     -1.96      -5.26
            CCSR             0.40          0.20      -0.23     -1.31      -4.55
            PCM              0.34         -0.03      -0.17     -0.49      -6.14



Table 11d: Predicted change in expected income from AOGCM climate scenarios for small
farms (US$)


                            Mean          Total     % change  Bootstrap Bootstrap
                          (US$/farm) (billions US$)          lower 95 % upper 95%

Expected income              106           6.00


2020
            CCC               -6          -0.34      -5.7%      -13         1
           CCSR              -13          -0.74     -12.3%      -26        -1
            PCM               13           0.74      12.3%       3         23

2060
            CCC               5            0.28      4.7%        -5        21
           CCSR               16           0.91      15.1%       -2        23
            PCM               -3          -0.17      -2.8%      -12        14

2100
            CCC               36           2.04      34.0%       -5       110
           CCSR               -6          -0.34      -5.7%      -24        54
            PCM               9            0.51      8.5%        -9        36




                                               37

Table 12a: Predicted change in the probability of selecting each animal from AOGCM
climate scenarios for large farms


                           Beef cattle Dairy cattle  Goats    Sheep    Chickens

   Baseline probability      24.5%       33.4%       15.1%    20.8%      6.1%


   2020
             CCC             -3.9%       3.2%        -0.7%    2.6%       -1.2%
            CCSR             -1.5%       1.5%        -0.1%    1.9%       -1.9%
             PCM             -2.8%       -2.0%       2.4%     2.1%       0.3%

   2060
             CCC             -2.0%       -1.7%       1.5%     3.6%       -1.3%
            CCSR             -2.1%       5.8%        -0.8%    -0.8%      -2.0%
             PCM             -6.2%       4.9%        -2.7%    7.3%       -3.2%

   2100
             CCC             -11.0%      8.3%        -7.2%    15.3%      -5.4%
            CCSR             -6.9%       3.6%        4.1%     1.4%       -2.2%
             PCM             -3.1%       0.2%        0.0%     4.3%       -1.4%



Table 12b: Predicted change in net income per animal from AOGCM climate scenarios for
large farms (US$/animal)


                           Beef cattle Dairy cattle  Goats    Sheep    Chickens

 Baseline income             146.01     120.18        6.04    12.08       1.68


 2020
             CCC              -4.50      -3.11       -1.01    -3.20      -0.35
            CCSR             -14.93      -4.89       -0.99    -3.30      -0.39
            PCM               -7.63      23.16       -0.59    -1.84       0.07

 2060
             CCC             -11.50      23.35       -1.01    -3.35      -0.03
            CCSR              7.02       -12.61      -1.76    -5.25      -0.48
            PCM              -13.73      -12.36      -1.69    -5.75      -0.48

 2100
             CCC             -20.75      -46.00      -2.26    -8.27      -1.02
            CCSR              1.40       63.36        0.23    3.43       -0.28
            PCM              -12.67      28.40       -1.23    -3.22      -0.08




                                              38

Table 12c: Predicted change in number of each animal per farm from AOGCM climate
scenarios for large farms


                         Beef cattle Dairy cattle  Goats      Sheep    Chickens

 Baseline number           77.58         35.82      23.81     49.40     1907.25


 2020
            CCC             1.49          1.04       5.00     0.98      -57.80
            CCSR           -2.16          0.91       6.01     1.82      -123.32
            PCM            -1.18         -0.92       0.00     -1.24     -96.66

 2060
            CCC            -0.51         -0.56       2.44     -1.43     -139.20
            CCSR           -3.41          1.53       6.64     -0.13     -184.73
            PCM            -1.56          1.42      11.59     -1.99     -141.62

 2100
            CCC            -14.18         1.17      25.53     -2.69     -289.22
            CCSR           -4.75         -1.98       5.78     41.22      -8.72
            PCM             0.21         -0.64       3.57     5.24      -180.19



Table 12d. Predicted change in expected income from AOGCM climate scenarios for large
farms (US$)


                           Mean          Total     % change  Bootstrap  Bootstrap
                         (US$/farm) (billions US$)          lower 95 %  upper 95%

Expected income


2020
            CCC             -779          -6       -10.0%      -932        -442
           CCSR            -1000          -8       -12.8%     -1068        -703
            PCM            -2817          -22      -36.0%     -4624       -2481

2060
            CCC            -3171          -24      -40.5%     -5358       -2926
           CCSR            -1232          -9       -15.7%     -2084        -345
            PCM            -2362          -18      -30.2%     -3431       -2040

2100
            CCC            -6023          -46      -77.0%     -8567       -5732
           CCSR            -2685          -21      -34.3%     -6003       -2184
            PCM            -3382          -26      -43.2%     -5156       -2896




                                              39

Figure1: Theoretical livestock response functions




                                            40

Figure 2a: Estimated probability of selecting species given annual temperature




                                             41

Figure 2b: Estimated probability of selecting species given annual precipitation




                                             42

Figure 3a: Net revenue response function to temperature




                                            43

Figure 3b: Net revenue response function to precipitation




                                            44

Figure 4a: Number of livestock response function to temperature




                                           45

Figure 4b: Number of livestock response function to precipitation




                                           46

Figure 5: The change in the probability of choosing beef cattle in South Africa with
uniform climate change scenarios




                                        47

Figure 6: The change in the probability of choosing goats in Africa with uniform climate
change scenarios




                                           48