Content uploaded by Ayodele Odusola
Author content
All content in this area was uploaded by Ayodele Odusola on Aug 13, 2015
Content may be subject to copyright.
1
Effects of Temperature and Rainfall Shocks on
Economic Growth in Africa
Ayodele Odusola
, United Nations Development Programme ayodele.odusola@undp.org
Babatunde Abidoye
, University of Pretoria babatunde.abidoye@up.ac.za
Paper presented at the 29th Triennial Conference of the International Association of
Agricultural Economists (IAAE) in Milan, Italy from 8 to 14 August, 2015.
Abstract
This paper examines the impact of temperature and rainfall volatility on economic growth in 46
African countries. We employ the Bayesian hierarchical modeling approach which allows us to
estimate both country level and Africa-wide impact of climate change and extreme events on
economic growth in Africa. Our results show that a 1
0
Celsius increase in temperature leads to
1.58 percentage points decline in economic growth while temperature shock reduces economic
growth by 3.22 percentage points. A 1 percent change or shock in rainfall leads to a 6.7 percent
change in economic growth. The impact of temperature changes across the 46 countries ranges
from -1.24 percent to -1.82 percent in GDP. There are proximity effects on the impact. To
maximize the benefits of economies of scale, the paper suggests combined national, cross countries
and continental approaches to climate change adaptation in Africa.
Keyword: Climate Change; Economic Growth; Africa; Hierarchical Model; Bayesian framework;
Gibbs Sampling.
Classification: C1; C4; C5; O1; Q54, Q56
2
Introduction
The role climatic conditions play in the agricultural systems in Africa has been well documented.
Some studies, though not African specific, have examined the vulnerability of the overall economy
and key sectors (e.g. agriculture, forestry, energy, tourism, coastal and water resources) driving
economic growth to climate change.
1
The geographical location of most African countries on the
lower latitudes has already put the region at a disadvantaged position where about 80 percent of
damages from climate change are concentrated with any further warming posing serious threat to
productivity and livelihoods (Mendelsohn, 2009; Bansal and Ochoa, 2012).
African countries have experienced temperature and rainfall shocks that are large enough to change
agricultural, marine and other sectors productivity since the 1960’s.
2
For example, some countries
such as Algeria, Uganda and Malawi experienced less temperature anomalies between 1960 and
1977. However since 1977, they have been experiencing larger temperature anomalies (Figure 1).
It should be noted that the temperature anomaly of +0.6 degree C in Uganda is one of the highest
anomalies in the past 120 years from the global temperature data.
3
Similarly, looking at
temperature changes, Sudan, Chad, Uganda and Botswana have experienced substantial rise in
temperature – ranging from 1
o
to over 3
o
Celsius. Similarly, some other countries such as
Mauritania, Niger, Guinea and Sierra Leone have also experienced reduced level of precipitation
in the 2000s compared to the 1960s. For example, the average maximum rainfall in the 2000s in
Guinea was just 92.6 percent of the average minimum in the 1960s and 93.3 percent for Niger.
The Sahel and the Horn of Africa have also experienced substantial and frequent extreme events
in the form of droughts which often lead to famine in these regions. The latter decades of the
twentieth century in the Sahel were characterized by years in which annual rainfall totals were
consistently below the long term mean for the century, and punctuated by years of severe drought
(Brooks, 2004). Vizy and Cook (2012) show that the largest rise in heat wave days (ranging from
60 to 120 days) is in the Western Sahel.
A three degree warming for instance will have huge impact on any environment – biodiversity,
agriculture and the oceans. The UK Met Office have a map on the impact of global temperature
rise of 4 degree C in October 22, 2009.
4
The map shows the impact of forest fire, crops, water
availability, sea level rise, marine, drought, tropical cyclones and extreme temperature to name a
few. These impacts are based on global models that are based on scientific simulation.
1
See Dell et al (2012) and Koubi et al (2012) for the economy-wide impact and Boko, et al (2007), and
Schlenker
and Lobell (2010)
for sector specific effects.
2
Author’s computation using gridded data from CRU version 3.0 - Mitchell and Jones, 2005.
3
http://www.globalissues.org/article/233/climate-change-and-global-warming-
introduction#WhatarethemainindicatorsofClimateChange
4
See http://www.theguardian.com/environment/interactive/2009/oct/22/climate-change-carbon-emissions
3
Figure 1: 5 Year Mean of Temperature Anomaly – C degrees, 1960 - 2009
The science of the impact of climate change has been relatively conclusive as has been illustrated
in the previous paragraphs and various research. However, very minimal studies have been done
on the impact on each country in Africa and the continent as a whole. Analysis of countries and
regional impact is paramount for proper planning and adaptation strategies. The contribution of
this study is to provide estimates of the impact of temperature, precipitation and climate change
on 46 African countries’ GDP growth.
This paper is unique in several respects. First, the model captures observable and unobservable
factors affecting economic growth. Second, the framework of analysis (Bayesian hierarchical
model) allows us to pool all countries to obtain regional regression results while at the same time
generating specific impact for each country. Third, it is able to disaggregate climate change into
its various components (temperature and rainfall shocks), an issue that is rarely addressed in other
papers. Finally, the paper adopts current and medium term measures of temperature and rainfall
shocks. This paper uses data from 1961 to 2009.
This paper is divided into five parts. Following the introduction is Section 2 which touches on
review of empirical evidence on the effects of temperature and rainfall shocks on economic
growth. Section 3 presents the model and how our parameters of interest are estimated while
Section 4 describes the data and analysis of the findings. Section 5 concludes the paper.
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
Algeria Uganda Malawi
4
2. Literature Review
Weather conditions (high or low temperature, more or less precipitation and less intense or severe
storms) can affect economic activities (agriculture, industrial and services) in many ways. The
destruction of ecosystems from erosion, flood and drought, the extinction of endangered species
and deaths resulting from extreme weather can have a significant negative impact on economic
growth. The channels through which climate variability affects economic activities is varied and
diverse. Dell et al (2012) and Koubi et al (2012) show that the transmission channels between
weather conditions and economic activities can be clearly identified if the level of GDP is
considered, but are ambiguous for the growth rate. For the level of GDP, the short run effect of
increase in temperature (or fall in precipitation) could be offset by lower temperature (higher
precipitation) in the future thereby leaving the long-run GDP level unaffected. However, the story
is different when growth rate is affected because economic growth will be lower even if the level
of GDP returns to its normal level. Several factors account for this. The foregone consumption and
investment as a result of lower income during the period of higher temperature (lower
precipitation) distorts the growth process. Also, heavy investment on adaptation and mitigation
programmes will impose some opportunity costs, especially in terms of not investing such
resources on science, technology and innovations as well as human and physical capital investment
(Pindyck, 2011; Ali, 2012; and Abidoye and Odusola, 2012 and 2015). The resources spent on
climate change adaptation and mitigation have the tendency of crowding out investment on other
vital drivers of growth and development, especially spending on education, health and
infrastructure. The combined effects generate negative impact on economic growth (Frankhauser
and Tol, 2005).
The empirical literature has provided some evidence on the effects of temperature and rainfall
shocks on economic growth. But the evidence remains inconclusive in terms of results and
magnitude of effects. Using historical fluctuations in temperature, Dell et al (2012) find strong
linkages between temperature changes and aggregate economic growth. They establish that higher
temperatures substantially reduce the level and rate of economic growth in poor countries. Higher
temperatures have wide-ranging effects, reducing agricultural and industrial output, and political
stability. They conclude that the substantial negative impacts of higher temperatures on poor
countries are quite large to explain the cross-sectional temperature-income relationship between
rich and poor countries. In poor countries, for instance, a one degree Celsius rise in temperature
reduces per-capita income by about 8 percent and leads to a decline in growth rates by about 1.3
percentage points. The paper stresses that annual data on temperature could produce noisy results
than medium and long term data. The authors, however, conclude that precipitation has no effect.
Their finding on precipitation contrasts Miguel, Satyanath and Sergenti (2004) evidence of strong
positive relationship between rainfall and economic growth in Africa. In a similar vein, the finding
from Koubi et al (2012) does not produce any evidence to show that climate variability
(temperature) affects economic growth.
5
Some other studies have also examined that higher and rising temperature can significantly affect
agricultural productivity, farm income and food security. For instance, Schlenker and Lobell
(2010) provide evidence on the negative impact of climate change on African agriculture. The
mean estimates of aggregate production changes in Sub-Saharan Africa by 2050 to be 22 percent
for maize, 17 percent (sorghum), 17 percent (millet), 18 percent (groundnut) and 8 percent
(cassava). They find that in all cases, except cassava, the probability that the damages exceed 7
percent of total production is a 95 percent. Others such as Nordhaus and Boyer (2000), Tol (2002),
Mendelsohn et al (2006), and Barrios et al (2010) have also provided some evidence on the issue.
In addition, Bernauer, et al., (2010) find mixed results on the impact of temperature variability on
economic growth: the moving average-based measure of temperature for Africa is associated with
negative effects but no impact when they used the CRU Miguel dataset. Evidence from Ayinde et
al. (2011), reveals that a rise in temperature generates negative effect while an increase in rainfall
exerts positive effects on agricultural productivity. Ali (2012) also finds that a fall in rainfall
magnitude and changes in variability have a long term drag-effect on growth in Ethiopia. Evidence
from Ouraich and Tyner (2014), for instance, shows climate change shocks have altered regional
agricultural production pattern in Morocco. Their projections further reveal the impact of climate
change on GDP (in the absence of any adaptation) to range from -3.1 per cent (worst-case scenario)
to +0.4 per cent (best case scenario).
The effect differs across temperate and tropical areas. In mid and high latitudes, the suitability and
productivity of crops are projected to increase and extend northwards while the opposite holds for
most countries in tropical regions (Gornall et al 2010). They find that a 2
o
Celsius rise in
temperature in mid and high latitudes could increase wheat production by about 10 percent while
in low latitude regions, it could reduce by the same amount. Their projection, taking the effect of
technology into account, reveals that rising temperature in Russia Federation could increase wheat
yield by between 37 and 101 percent by 2050s. Similarly, Waldinger (2013) provides an analysis
of the effect of low temperature on economic growth in Europe. Although the effect of temperature
varies across climate zones, on average however, further temperature decreases in particularly cold
period generate negative effects. The result is strongly negative in cities already experiencing cold
climate while cities in relatively warm climate zones benefit from colder temperatures. Cities and
small towns depending heavily on agriculture without much access to long distance trade networks
are mostly affected.
Bansal and Ochoa (2011) reveal that temperature is an aggregate risk factor that adversely affects
equity returns and overall economic growth both at country and global levels. The study shows
that the covariance between country equity returns and temperature contains useful information
about the cross-country risk premium. For instance, countries closer to the Equator carry a high
temperature risk premium which decreases as a country is further away from the Equator. The
differences in temperature or temperature shocks mirror exposures to aggregate growth and equity
risks. Simply put, portfolios with larger exposure to aggregate growth risks are also exposed to
larger temperature shocks. In this study, countries closer to the Equator have larger risk premium
6
while it is negligible in countries with high latitudes. The paper also shows that economies of
countries closer to the Equator depend more on climate sensitive sectors, thereby exposing them
to higher risk premiums.
Several studies (e.g. Hirvonen, 2014) have also examined the effect of temperature shocks on
households’ welfare. It examines how fluctuations in temperatures affect household consumption
pattern and rural-urban migration in Tanzania. The paper establishes a co-movement between
household consumption and temperature. His evidence shows, controlling for rainfall, household
fixed effects and various time-varying factors, a one standard deviation increase in the mean
monthly growing season temperature decreases household per capita consumption by 4.9 per cent.
This is an indication that temperature shocks make rural households more vulnerable in Tanzania.
The temperature-induced income shocks are then found to inhibit long-term migration among men.
This therefore prevents them from tapping into and benefiting from the opportunities associated
with geographical mobility in the country, including consumption and income premiums.
Similarly, liquidity constraint associated with rainfall shocks shape aggregate temporary
international migration flows from rural Indonesia (Bazzi, 2013), influences men migration in
Ethiopia (Gray and Mueller, 2012).
In conclusion, the impact of climate change variability on economic growth in Africa remains
inconclusive. The differences in measurement of climate change or climate variability,
methodological approach, models employed and scope could account for this inconclusiveness in
findings. Addressing the conceptual, methodological, scope and coverage gaps associated with
some of the papers on this subjects, our paper brings a different perspective to the effects of
temperature and rainfall shocks on economic growth in Africa.
3. Analytical framework
This section examines the standard cross-country growth models that can be used to estimate the
relationship between economic growth and its key determinants. In addition to an analytical model
to assess how temperature and rainfall shocks affect economic growth, it also proposes a
methodology that controls for a specific type of omitted variable bias on parameters of interest.
3.1 The Basic Cross-Country Growth Regression Model
Following the framework in Barro (1991), Levine and Renelt (1992) and Sala-i-Martin (1997b),
we model
, economic growth of country i, as follows:
(1)
Where
7
In the above,
denotes the average growth rate of GDP of country i over a certain year range. In
line with Levine and Renelt (1992),
denotes a vector of explanatory variables of country i over
the same year range that are believed to influence growth. This typically involves sets of variables
that are always included in economic growth regression and a subset of variables chosen from a
pool of variables identified by past studies as potentially important in explaining growth, which
we denote as
.
The single cross-section growth regression specification appropriately models differences in
growth patterns of countries when there is no correlation between the variable of interest and other
explanatory variables. However, when the variable of interest is potentially correlated with
unobserved variables, the single cross-section growth regression specification will lead to
inconsistent estimate of the former. In the following section, we describe a Bayesian estimation
algorithm which properly accounts for the impact of correlation between unobserved variables and
temperature and rainfall shocks. This specification is important to study the impact of temperature
and rainfall shocks on economic growth.
3.2 Linear Hierarchical Model
Using Bayesian framework, this paper first assumes that the parameter on temperature and rainfall
will have a different impact on GDP across countries and should be permitted to vary across
countries. However, based on geography and similarity in practices in many African countries
especially with regards to contribution of agriculture to GDP, we expect some level of
commonality across the continent on its impact. On the other hand, climate variables such as
temperature and rainfall may also have some impact on many of the explanatory variables that
may be included (observed) or excluded (unobserved) in the model. Consistent estimate of the
parameters of temperature or precipitation and observed explanatory variables such as initial GDP
per capita or economic growth will require that these variables be uncorrelated with the unobserved
variables. This condition is unlikely to hold especially given that we cannot control for all the
variables due to unavailability of data on such variables that can potentially influence economic
growth and related to temperature and rainfall. This is the classic omitted variables bias and
inconsistency problem
5
, which are often associated with most of the studies reviewed in section
two above.
This paper proposes a linear hierarchical model that is similar to the non-Bayesian fixed effects
model but exploits the hierarchical prior framework to estimate the parameters of the observed
variables that influence economic growth.
6
We proceed with a model where all the regression
coefficients for temperature can vary across countries (random coefficients model), country effects
model in which the regression intercepts are allowed to vary across countries combined with a
5
Abidoye, Herriges, and Tobias (2012) illustrate this problem in a Random Utility Maximization setting.
6
A hierarchical prior on the parameters in this case makes the parameter vectors with high dimension
8
pooled model on the impact of temperature and rainfall on Africa. The effect of temperature lags,
rainfall and their shocks are also estimated as a pooled model. This model introduces a country-
specific constant term that captures both the observed and unobserved explanatory variables that
influence economic growth as described in Lindley and Smith (1972) and Abidoye et al (2012).
7
Rewriting equation (1) to reflect all variables of interest, we have:
!""
Where
has a hierarchical prior that makes it similar to a cross-country growth regression
model. This is specified as:
#
$
….. … ..(3)
Equation (2) is also called a mixed model with the random effects
and
varying across
countries but also imposes a restriction that’s in equation (3) are constant across countries.
This model resolves the omitted variable bias since
is no longer correlated with the variable of
interest (
and
) and also allows for separately identifying the impact of the observed
explanatory variables on economic growth using a hierarchical prior framework.
8
Equation (3) is estimated using a Bayesian framework and it adopts the blocking strategy used in
Abidoye et al (2012) proceeding in a manner that is similar to the classic fixed effects model by
isolating the impact of the unobservable (capturing them entirely in the country- specific constants)
and to insulate the climate parameters from their effects.
9
The blocking strategy will draw both
and
and the set of parameters that do not vary across countries - in a single block of draws. It
also has the added advantage of facilitating the mixing of the chain.
3.3 Hierarchical Priors
High dimensional parameter spaces are usually problematic for nonlinear models because of the
high number of parameters to estimate. The model above will require the estimation of 2*N + k
(i.e., N intercepts, N country effects, k slope parameters on rainfall and other variables in L and Z
plus the pooled effects and error precisions) parameters using N*T data points. Even with large T
relative to N, the number of parameters is still large relative to the sample size. The sparseness of
data in high-dimensional spaces can result in lower convergence of regression function estimators.
7
Detailed description of this model and similar hierarchical models in the Bayesian framework can be found in
Koop, Poirier, and Tobias (2007).
8
This is one of the benefits of using the Bayesian framework over the classical fixed effects specification.
9
As is pointed out in Abidoye, Herriges, and Tobias (2012), this simply echoes standard result that the fixed effects
estimator is unbiased even when correlation exists between the fixed effects and other explanatory variables included
in the model.
9
Hierarchical priors become highly valuable in cases like this with high dimensional parameter
spaces and it is one of the main attractions of the Bayesian framework. One added advantage of
the hierarchical prior for estimation purposes is that it places more structure on the distribution by
assuming that the random parameters are drawn from the same distribution. This additional
structure allows for more accurate estimation, especially if the assumption is consistent with
patterns of the data
10
.
Starting with the country-specific constants, the expected value of these variables, as is typical of
cross-section regressions are the observed variables while the unobserved variables are embedded
in the error term. The interactions of all country level variables (excluding temperature and
rainfall) typically included in cross-country growth models are solely captured in the country-
specific constants. This imposes the extra structure needed for the estimation of the country-
specific constants. We are also interested in estimating the relationship between the climate
variables (temperature and rainfall) and the unobserved variables that may not be captured in the
regression.
For the estimation of
, we assume that each country share some degree of “commonality” in the
impact of temperature and or rainfall and economic growth by assuming that the country-specific
effect of climate change shocks across Africa are drawn from the same distribution. In addition to
this structure, we also allow for correlation between the impact of temperature, rainfall and other
factors that may influence economic growth.
Rewriting equation (1) in matrix form gives:
%
&
'
(
)*
+
…… (4)
Equation (4) seeks to draw
,
and in a single block. The mean and variance matrix of +
will
incorporate the hierarchical priors explained earlier.
Specifically:
+
,
-./
0
1
23
4
5
4
6
5
4
6
6
7
1
89":
The variable
includes a constant term and the observed/included explanatory variables that
influence growth in country i. The correlation between climate change shocks and the intercept is
10
See Koop (2003) for more information on this.
10
captured with5 and the pooled impact of temperature on Africa is captured with the
parameter
and prior for defined as0
1
7
1
. There are some silent features of this model that is worth
mentioning – our specification helps controls for the problem of potential correlation between
variable of interest and the unobserved variables which may potentially bias
and
that we are
interested. However, as is the case with most cross-country growth model, will not solve the
problem of potential correlation between the included explanatory variables and the excluded
variables. It is typically assumed that this assumption holds. However, if this assumption does not
hold, our specification can be extended to make use of instrumental variables approach to
consistently estimate. Even when such correlation between the observed variables and
unobserved variables exists, the inclusion of country-specific constants and our posterior simulator
will yield consistent estimates of the parameters of interest.
To complete our model, we specify priors for the remaining parameters. These are enumerated
below: 0
;
7
;
0
6
<
7
6
<
∑
=>
),
4
5
4
6
5
4
6
6
-?%5
@&
=>
5
ABC
D
E
F
G
F
H
"I
The hyper-parameters of the priors above, such as0
;
7
;
5
C
D
e.t.c., are supplied by the
researchers and are in general chosen to be relatively vague to allow for dominance of the
information from the data. The notation N refers to the normal distribution, whereas W (.,.)
represents a Wishart distribution and IG(.,.) represents the inverse gamma distribution
parameterized as in Koop, Poirier, and Tobias (pp. 336-339).
11
These particular families of priors
are chosen primarily because when combined with the likelihood function yield conditional
posterior distributions that are easily recognized and sampled. These proper priors also make
model comparison and calculation of Bayes Factor relatively easy.
Our prior means 0
;
and0
6
are set to zero matrices with the appropriate size and respective
variance 7
;
and7
6
set relatively large to allow vague and proper prior. The priors
(hyperparameters) on the variance term are also selected by choosing C
J andD
KL.
12
5
is set to be equal to 5 and the prior is chosen to reflect some degree of variability in the
temperature and economic growth across countries. These priors are chosen to be reasonably
11
Let be an N X N positive definite (symmetric) random matric, A be a fixed (nonrandom) N X N positive definite
matric, and v >0 be a scalar degrees-of-freedom parameter. Then H has a Wishart distribution, denoted H ~ W (A,v)
with a defined pdf and reduces to the gamma distribution if N=1. The inverted gamma distribution on the other hand
has the property that, if Y has an inverted gamma distribution ~ IG (a,b), then 1/Y has a gamma distribution with a
mean – E(Y) = [b(a-1)]
-1
and Var(Y) = [b
2
(a-1)
2
(a-2)]
-1
for a>2.
12
This chooses the prior mean for sigma^2 equal to 20 with standard deviation also equal to 20
11
diffuse and non-informative. Appropriate prior sensitivity analysis carried out shows the results
are robust as presented below.
3.4 The Posterior Simulator
13
Bayesian inference and posterior simulator is a process of updating researchers’ prior beliefs of
the parameters to be estimated into posterior beliefs based on observed data. The updating -
typically termed posterior simulation – involves working in terms of probability densities. The
framework involves a joint distribution of all quantities of interest –parameters and data using the
principles of probability – Bayes theorem to back out the posterior density of interest. These
posterior densities are approximated by a combination of a likelihood function and a prior
14
.
The model is fitted using the Gibbs sampler
15
and employing a number of blocking steps to
mitigate autocorrelations and consistently estimate our parameters of interest. Specifically, we fit
the model via Markov Chain Monte Carlo (MCMC) methods that utilizes the Gibbs sampler. The
idea is to draw from the posterior conditional distributions rather than the joint posterior
distributions themselves that are usually difficult to draw from.
4. Data, estimation techniques, descriptive statistics and analysis of results
4.1 The Data
This section describes the data used to run the models specified above. Temperature and rainfall
data for each African country is deduced from the database of Climate Research Unit (CRU) using
observed gridded monthly mean temperature and rainfall data (CRU, version 3.0 as outlined in
Mitchell and Jones, 2005).
16
The CRU dataset is based on station data and composed of monthly
0.50 latitude/longitude gridded series of climatic parameters over the period 1901-2009. However
the data used for this paper runs from 1961-2009.
17
13
For readers interested in detailed model specification see Appendix 1.
14
See
Koop et al (2007) and other Bayesian econometrics texts for further reading on this.
15
The Gibbs sampler is an iterative algorithm that has become an indispensable tool to Bayesians
and researchers undertaking simulation based inference. For more information see Koop, et al
(2007).
16
According to the Climatic Research Unit (CRU) project team, the reference for CRU version 3.0 is Mitchell and
Jones, 2005.
17
The Global Gridded Climatology data is presented at a new high resolution and made available by the Climate
Impacts LINK project, Climate Research Unit, University of East Anglia, Norwich, UK (Mitchell and Jones, 2005).
12
Data for other explanatory variables are obtained from the Africa Development Indicators (ADI)
(2011). Economic growth is measured as the annual percentage growth rate of GDP at market
prices based on constant local currency. The population values are midyear estimates.
The primary and secondary school enrolment rates, and life expectancy are used as proxies for
human capital investment. Although previous research (e.g. Mankiw et al (1992) and
Gemmell (1996)) has argued that using school enrolment as a proxy for the level of human capital
can be problematic. Because it has been used in many other studies, we therefore allow the model
likelihood to dictate if it should be included or not.
The model also controls for availability of port, language spoken, and initial private savings as a
ratio of GDP. Availability of a port is used to proxy for geography and savings and language are
typically controlled for in the growth literature. Savings is an increasing function of economic
growth but is also endogenous because higher economic growth can lead to higher savings. As
with other variables, we avoid the endogeneity problem by using initial savings. Language can
capture trade opportunities and heterogeneity in growth patterns with francophone African
countries typically with high similarity which can be observed in the growth patterns.
The data is available for 46 countries
18
. The choice of the countries is based on data availability
on the economic growth variables. However, the panel was unbalanced because of gaps in data
for some countries.
4.3 Estimation and Testing
The algorithm described in Section 3 has been used to run our posterior simulator for 500 000
iterations discarding the first 50 000 of these as the burn-in.
19
Results from these runs suggest that
the Markov Chan - Monte Carlo (MCMC) simulation chain from the posterior mixed reasonably
well and appears to converge within a few hundred iterations.
Although our point estimates are suggestive of good performance, any MCMC-based inference
can be affected by the degree of correlation among the parameter draws over sequential iterations.
The mixing of the posterior simulations has been used to determine how many draws are needed
to achieve the same level of numerical precision that would be obtained under an independent and
identically distributed (iid) sampling. When the degree of correlation is high it leads to a slow
mixing that may limit the simulator from exploring all areas of the posterior as may be needed.
18
The countries are: Algeria, Angola, Benin, Botswana, Burkina Faso, Burundi, Cameroon, Central African Republic,
Chad, Congo, Democratic Republic of Congo, Cote d'Ivoire, Egypt, Equatorial Guinea, Eritrea, Ethiopia, Gabon,
The Gambia, Ghana, Guinea, Guinea-Bissau, Kenya, Lesotho, Liberia, Libya, Madagascar, Malawi, Mali, Mauritania,
Morocco, Mozambique, Namibia, Niger, Nigeria, Rwanda, Senegal, Sierra Leone, South Africa, Sudan, Swaziland,
Tanzania, Togo, Tunisia, Uganda, Zambia, Zimbabwe.
19
``Burn-in'' is a colloquial term that describes the practice of throwing away some iterations at the beginning of an
MCMC run to discard the iterations before convergence is reached.
13
These inefficiency factors, can be calculated by using the definition of the numerical standard
errors (NSE) of a Monte Carlo estimate with correlated draws. The mean estimates can be obtained
as:
MNOP
Q
R
S
T
Q
RUVW
X
Q
Y5
X
Q=>
XZ>
….. …. ….. (6)
Where O represents an arbitrary scalar parameter of interest, [ denotes the number of post-
convergence simulations, OP
Q
represents our estimate of NO\ as the sample average of our post-
convergence draws, 5
X
represents the correlation between simulations ] periods (iterations) apart
and ^
)7C_O\.
The NSEs for our models are small relative to the mean estimates which indicate our simulation
estimates accurately approximate the posterior means of the selection parameters. This, again,
suggests that our algorithm mixes quite well. The values for the NSEs for the country effect
parameters are presented in Appendix 2.
The posterior mean is commonly used to interpret a moment of the posterior distribution. The
posterior probabilities, while similar to the classical p-value, provide information on the degree of
posterior certainty that the impact of the parameter is negative. The algorithm described above is
used to run the growth models.
4.3 Descriptive Analysis
This section presents the main feature of temperature and dynamics in the 46 African countries
used in this paper. The focus is on yearly temperature and five-year average rainfall including their
deviations from that average. This is because gross domestic product (GDP) growth data is mainly
available on yearly basis for the African countries.
20
Table 1 shows the minimum and maximum,
the difference between the minimum and maximum, the mean (1961 and 2009) and the absolute
change between 1961 and 2009 of the yearly average temperature.
21
Based on the mean value for
the sample period, Mali, Burkina Faso, Senegal, and Mauritania are among the hottest countries
in Africa on average while Lesotho, Morocco, South Africa, Rwanda and Tunisia appear to be the
coldest. Sudan, Botswana and Zimbabwe experienced the highest change over the period of 49
years when we take the difference between the maximum and minimum yearly average
temperature as in column 3 of Table 1. Countries that changed by more than 2
o
Celsius between
1961 and 2009 are Sudan (3.04), Chad (2.61), Niger (2.47) and Egypt (2.15).
20
Apart from data availability, average temperature nets out the effect of seasonality and climate change by
definition focuses on average temperature differential and deviation. Thus, while we recognize that yearly
temperature and rainfall may not accurately capture daily or growing season temperature fluctuation, we argue that
they adequately reflect the influence of temperature and rainfall averages on economic growth.
21
This does not imply that the temperature for that country within a year does not go below or above the minimum
or maximum but the mean of the yearly average in that period is reported.
14
Figure 1 shows the series of temperature for countries with the top 5 countries with the highest
change between the maximum and minimum yearly average temperature as in column 3 of Table
1. Sudan and Chad have the highest levels and the yearly average have been rising consistently
during the period. They are followed by Niger, Egypt, Uganda and Libya. Countries that
experienced some relative stability in temperature during the period of analysis include
Madagascar, Congo Democratic Republic, Gabon, Liberia and Sierra Leone (see Figure ).
As shown in Table 3, the unconditional effect of temperature change lag appears to have an inverse
relationship with the change in current output. We include temperature lags in the regression to
aid our understanding of the impact of temperature dynamics and economic growth.
Figure 3 presents a simple summary statistics for rainfall. Liberia has the highest average yearly
rainfall of all the African countries but also experienced the highest fluctuation (as captured by the
standard deviation) for the period. Guinea Bissau and Equatorial Guinea, The Gambia are among
the countries with the highest rainfall and variation in rainfall during the period of analysis. In the
next section we used a five-year moving average and 5-year average deviation of rainfall to capture
the long run impact of rainfall on economic growth.
4.4 Analysis of the results
This paper answers the following questions: (1) what is the impact of temperature and rainfall on
economic activities in Africa? (2) What is the impact of climate shock as measured by long run
deviation from the mean on economic growth in Africa? (3) What is the residual/lag impact of
temperature on economic growth in Africa? (4) Given recent interventions and adaptation
strategies, is there any difference in the impact of temperature shocks between 1960’s to 2000
compared to the whole sample period?
The analysis below is based on parameter posterior means and posterior probabilities of the
parameter being negative [denoted P (. <0|y)]. It provides the link between temperature, rainfall,
and their long run deviations on the one hand and economic growth on the other based on the
pooled regression parameters, the slope and intercept results for 46 African countries.
Table 2 presents the result of common parameter estimates. The presentation of different variants
of the model provides some robustness checks to test our different hypothesis, as is typical of
cross-section economic growth models. In the first three columns of Table 2, we control for initial
GDP per capita, population, primary school enrolment and life expectancy. Although evidence is
not strong, the initial conditions of human capital (proxied by initial primary school enrolment and
life expectancy) contribute positively to economic growth. The evidence is strongest for life
expectancy with the probability of it being positive at about 70 percent. This evidence may be
suggesting that life expectancy may not only serve as a proxy for human capital but also an
indicator of quality of life. There is little or no evidence in support of the initial condition of net
primary school enrolment and population growth influencing economic growth in Africa. The
results show the importance of initial condition (the log of initial GDP per capita) in the continent
15
growth process. These results are generally consistent with previous studies on determinants of
growth in Africa.
In addition to these variables, we also control for geography as measured by port, language, initial
private savings as a ratio of GDP and technology transfer as measured by foreign direct investment,
secondary school enrolment and their interaction. While there is little evidence that port, language
and savings have a significant impact on economic growth, technology transfer measures provide
interesting results. The initial conditions of foreign direct investment show a negative impact on
growth with the probability of this being negative ranging between 70 per cent and 90 percent
across the various models. This is in line with the literature on foreign direct investments (FDIs).
FDI without adequate human capital for the transfer to take place will potentially stunt economic
growth. As reported in the results, there is strong evidence that FDI reduces economic growth
when its interaction with secondary school enrolment is not controlled for. When the interaction
is controlled for, the negative impact on growth fell. The inclusion of this interactive variable
reduces the negative impact of temperature on economic growth (see models 5 and 7). It shows
the variable has both direct and indirect effects on economic growth. This clearly suggests that
countries with high quality of secondary school education are likely to reap the benefits of
enhanced economic growth. Even if the right human capital is in place, strong national institutions
are needed to avoid expropriation through clandestine capital outflows. There is also evidence
that high secondary school enrolment increases growth.
The role temperature, rainfall, and their respective shocks play in explaining economic growth in
Africa is pivotal. In column 1, when only temperature and rainfall are the included variables, a 1
0
Celsius in temperature tends to reduce economic growth by 1.28 percentage points and the
relationship is always established at a probability level of 92.3 percent. The relationship is even
more pronounced across most of the seven models, the impact of a 1
0
Celsius ranges between -
1.25 percentage points and -1.59 percentage points on economic growth in Africa. The results from
models 3, 5 and 7 reveal that the serious negative impact of a rise in temperature is certain, with
probability levels of 100.00 percent.
To capture the residual impact of temperature, we introduce five-year temperature lags. However,
the relationship of the lag temperature on economic growth is non-linear – becoming positive in
the first year lag, turning negative in the second year lag and changing to positive trend in the third
to the fifth year lags. The probability level becomes relatively weaker after the second year lag
(Table 3). Based on the foregoing and using the current temperature and the first two-year lags,
the cumulative net impact of temperature on economic growth can be crudely calculated as -0.6141
(-1.586 + 1.2616 – 0.2897), which still remain quite high for the continent.
The impact of rainfall is positive across all the seven models. A one percentage change in the
rainfall medium term (5-year moving average) mean volume appears to increase economic growth
by 2.8 percent in Model 1. This positive relationship is established at a probability level of 93.60
percent. For all the models the impact ranges between 2.74 percent and 6.73 percent with the
16
relationship being established at 92.7 per cent and 95.8 percent probabilities. This result tends to
underscore what African economy is losing from absence of irrigated farming and the frequent
extreme droughts in the Sahel and the Horn of Africa.
In order to estimate the impact of climate change shocks on economic growth, we control for the
temperature and rainfall deviations from their respective 5-year moving averages. Unexpected
change in temperature and rainfall (rise or fall) produce significant impact on economic growth in
Africa. An unexpected rise of one standard deviation from the average temperature reduces
economic growth by 3.22 percentage points with 99 percent of the mass in the positive region.
This implies that an unexpected reduction in temperature by at least one standard deviation from
the mean will raise GDP by 3.22 percentage points. Any shock (rise or fall) in rainfall (deviation)
of at least a magnitude of one percent from its 5-year mean value may lead to a rise or fall in
economic growth by 6.76 percent (Table 3). The effect of any unfavorable deviations from
temperature or rainfall is quite damaging to the African economy. In addition, to be fully engaged
in efforts that will lead to enhanced climate change adaptation, heavy investment in meteorological
services and weather indexed insurance to farmers will help to ameliorate the excruciating effect
of weather shocks to the economy.
The impact of climate change is not only on economic growth. It also affects other determinants
of economic growth. The correlation between temperature and other factors that influence
economic growth is mostly negative but with weak probability. The probability that the
relationship is negative is established at about 60 percent across all the models (Table 2). This
implies that African countries with lower temperature increases tend to have higher growth rates
compared to those with a high rise in temperature. This suggests the combined direct and indirect
effects of climate change could be more serious than envisaged especially if the impact of
temperature increase on growth fundamentals – particularly those with irreversible consequences
– is negative (especially life expectancy). Finally, there is evidence of individual heterogeneity
across countries as shown by the estimates of δ
2α
and δ
2β
with δ
2α
= 0.17 and δ
2β
= 0.13 on average.
The country level impact of temperature on economic growth and their probabilities of being
negative (Pr (: < 0|y)) is overwhelmingly negative (Table 4). It shows the continental average
blurs the individual countries performance which most other studies have not been able to unravel.
The results shows that the largest impact of temperature on economic growth is in the Democratic
Republic of Congo followed by Sierra Leone, Madagascar and Central African Republic. Evidence
from the 46 countries is largely negative with β
i
ranging from -1.822 for Democratic Republic of
Congo and -1.244 for Equatorial Guinea. The worst hit five countries are Congo Democratic
Republic, Sierra Leone, Tanzania, Madagascar, and Central African Republic. A 1
0
Celsius rise in
temperature reduces economic growth by between 1.75 percent and 1.82 percent for these five
countries. The negative impact is more severe than the continental average of 1.58 percent in 19
countries (see Table 4). Five countries with the least impacts are Equatorial Guinea, Egypt, Eritrea,
Angola, and Algeria. Egypt and Algeria have one of the largest irrigation schemes in Africa while
17
Equatorial Guinea and Angola, apart being blessed with swampy forests also rely on oil money,
when to suggest a better capacity to cope with the effect of weather shocks.
The intensity of temperature change varies from country to country. Yet, it has no respect for
boundaries. Similarity in impact on economic growth, based on geographic proximity, provides
a strong basis for grouping countries into at least nine sub-groups (Figure 4): (i) Mali and
Mauritania; (ii) Niger and Libya; (iii) Algeria and Morocco; (iv) Cameroon and CAR; (v) Senegal,
Guinea, Cote d’Ivoire and Ghana; (vi) Nigeria, Benin and Togo; (vii) Ethiopia, Somalia, Kenya
and Uganda; (viii) Zimbabwe, Zambia, Rwanda and Burundi; and (ix) Namibia, Botswana, South
Africa, Lesotho, Swaziland, Mozambique and Malawi. Similar multi-countries impact calls for
economies of scale in climate change adaptation. Combined national, regional and continental
adaptation strategies are more appealing to reap the synergy associated with economies of scale.
The regional approach also helps to mitigate the risks of asymmetric capacity to adapt to climate
change in Africa.
To determine if the impact of climate change on economic growth has been improving or
worsening over the past five decades, we divided the period into two: one smaller sample (1961-
2000) and a full sample (1961 and 2009). We then compare the results from these two samples
(see Table 4 and Figure 5). The impact of temperature on economic growth in Africa was found
to be higher in the full sample than the small sample. Evidence from the small sample (1961-2000)
tends to show lower level of damages to economic growth than the larger sample. A 1
o
Celsius
rise in temperature slows down economic growth by 1.42 percentage point for the small sample
with a probability value of 0.99 compared with 1.59 for the full sample period for Africa. Despite
the substantial drag on growth emanating from change in temperature, agricultural productivity in
Africa has increased since 2000 (Block, 2010). It shows that without the damaging effects of
climate change on agriculture, agricultural productivity and production would have been quite
substantial. However, there is a relatively stronger evidence that the net effect of a change in
temperature incorporating the 5-year lags is higher in the small sample (-0.121) than in the full
sample (-0.041).
22
This tends to suggest adaptation to extreme weather changes is improving.
Finally, across the two samples, there is no significant difference in terms of the long run
temperature shock impact as measured by a 5-year deviation from the mean.
22
This ignores the fact that there is weak evidence that the probability that the parameters of the 3
rd
, 4
th
and 5
th
lags in the full sample are positive. We report the estimates for the lags and rainfall for the sub-sample in the
appendix.
18
5. Conclusions
Africa is at the centerpiece of climate change and it exhibits a good case for climate change
paradox – contributes marginally to greenhouse gas emission but bears excruciating impacts with
limited capacity to manage them. The vulnerability of the African economy and key sectors
driving economic performance (such as agriculture, forestry, energy, tourism, coastal and water
resources) to climate change is substantial. Yet, in the past five decades, many countries in Africa
such as Sudan, Chad, Uganda and Botswana have experienced high rise in temperature – ranging
from 1
o
Celsius to over 3
o
Celsius. During the same period, countries such as Mauritania, Niger,
Guinea and Sierra Leone also experienced substantial decline in rainfall - average annual
maximum rainfall in the 2000s in Guinea and Niger fell short of their average annual minimum in
the 1960s. The impact of changes in temperature and rainfall on Africa’s economy is considerably
large. A 1
0
Celsius increase in temperature leads to 1.58 percentage points decline in economic
growth while an unexpected one degree standard deviation from the average shock tends to
generate 3.22 percentage points decline in GDP. On the other hand a one percent change (rise/fall)
in rainfall leads to a 6.7 percent (increase/decline) in economic growth. Any rainfall shock also
generates a similar effect. The impact of temperature changes is even more excruciating at the
country level – ranging from -1.24 (Equatorial Guinea) and -1.82 (Democratic Republic of Congo).
These developments make proactive management of climate change adaptation and the impact of
climate change imperative in Africa.
Given that very few African countries have the capacity to deal with climate change adaptation,
the possibility of using economies of scale to deal with this challenge offers some bilateral, multi-
countries or regional oriented strategies. The regional approach helps to mitigate the risks of
asymmetric capacity to adapt to climate change in Africa.
19
References:
Abidoye, B. O., J. A. Herriges, and J. L. Tobias. (2012), “Controlling for Observed and
Unobserved Site Characteristics in RUM Models of Recreation Demand.” American Journal of
Agricultural Economics 94 (5): 1070–1093.
Abidoye, B.O and A.F. Odusola. (2012), “Climate Change and Economic Growth in Africa: An
Econometric Analysis.” Paper presented at the African Economic Conference 2012. November
2012, Kigali, Rwanda
Abidoye, B.O and A.F. Odusola. (2015), “Climate Change and Economic Growth in Africa: An
Econometric Analysis”, Journal of African Economies, 2015, p.:1-25.
Adger, W. N., N. W. Arnell, and E. L. Tompkins (2005), “Successful adaptation to climate change
across scales”, Global Environmental Change, Volume 15, Issue 2, July 2005, Pages 77–86.
Ali, Seid Nuru, (2012), “Climate Change and Economic Growth in a Rain-Fed Economy: How
Much Does Rainfall Variability Cost Ethiopia?” (February 8, 2012). Available at SSRN:
http://ssrn.com/abstract=2018233 or http://dx.doi.org/10.2139/ssrn.2018233
Ayinde, O. E., O. O. Ajewole, I. Ogunlade, and M. O. Adewumi. (2010), “Empirical Analysis of
Agricultural Production and Climate Change: A Case Study of Nigeria.” Journal of Sustainable
Development in Africa 12 (6).
Ayinde, O.E, M. Muchie and G. B. Olatunji. (2011), “Effect of Climate Change on Agricultural
Productivity in Nigeria: A Co-integration Model Approach”, Journal Human Ecology, 35(3): 189-
194 (2011).
Bansal, R. and M. Ochoa (2011). "Temperature, Aggregate Risk, and Expected Returns". Working
Paper 17575. National Bureau of Economic Research
Barrios, Salvador; Luisito Bertinelli & Eric Strobl (2010) “Trends in rainfall and economic growth
in Africa: A neglected cause of the African growth tragedy.” Review of Economics and Statistics
92(2): 350–366.
Barro, R. J. (1991), “Economic Growth in a Cross Section of Countries.” The Quarterly Journal
of Economics 106 (2): 407–443.
Barro, R. J., and X. Sala-i-Martin. (1992). “Convergence.” Journal of Political Economy: 223–
251.
Bazzi, Samuel. (2013). “Wealth heterogeneity, income shocks, and international migration: theory
and evidence from Indonesia.” Mimeo, University of California, San Diego.
20
Bernauer, T., Kalbhenn, A., Koubi, V., & Ruoff, G. (2010). “Climate change, economic growth,
and conflict.” International Studies Association, New Orleans, Feb, 16-20.
Block, Steven (2010). "The Decline and Rise of Agricultural Productivity in Sub-Saharan Africa
Since 1961," NBER Working Papers 16481, National Bureau of Economic Research, Inc.
Boko, M., I. Niang, A. Nyong, C. Vogel, A. Githeko, M. Medany, B. Osman-Elasha, R. Tabo and
P. Yanda, (2007), “Africa. Climate Change 2007: Impacts, Adaptation and Vulnerability.”
Contribution of Working Group II to the Fourth Assessment Report of the Intergovernmental Panel
on Climate Change, M.L. Parry, O.F. Canziani, J.P. Palutikof, P.J. van der Linden and C.E.
Hanson, Eds., Cambridge University Press, Cambridge UK, 433-467.
Brooks, Nick (2004), “Drought in the African Sahel: Long term perspectives and future
prospects”, Tyndall Centre for Climate Change Research Working Paper 61, October 2004.
Burke, Marshall, John Dykema, David Lobell, Edward Miguel, and Shanker Satyanath. (2011).
“Incorporating climate uncertainty into estimates of climate change impacts, with applications to
US and African agriculture.” National Bureau of Economic Research (NBER) Working Paper No.
17092.
Dell, Melissa, Benjamin F Jones, and Benjamin A Olken. (2012) “Temperature Shocks and
Economic Growth: Evidence from the Last Half Century.” American Economic Journal:
Macroeconomics 4.3 (2012): 66–95. Web. http://dspace.mit.edu/handle/1721.1/73620#files-area
[Accessed 22 May 2014]
Fankhauser, S., and R. SJ Tol. (2005), “On Climate Change and Economic Growth.” Resource and
Energy Economics 27 (1): 1–17.
Fernandez, C., E. Ley, and M. F. J. Steel. (2001), “Model Uncertainty in Cross-country Growth
Regressions.” Journal of Applied Econometrics 16 (5): 563–576.
Gallup, J. L., and J. D. Sachs. (2000), “Agriculture, Climate, and Technology: Why Are the
Tropics Falling Behind?” American Journal of Agricultural Economics 82 (3): 731–737.
Gallup, J. L., J. D. Sachs, and A. D. Mellinger. (1999), “Geography and Economic Development.”
International Regional Science Review 22 (2): 179–232.
Gemmell, N. (1996), “Evaluating the Impacts of Human Capital Stocks and Accumulation on
Economic Growth: Some New Evidence.” Oxford Bulletin of Economics and Statistics 58 (1): 9–
28.
Gornall Jemma, Richard Betts, Eleanor Burke, Robin Clark, Joanne Camp, Kate Willett and
Andrew Wiltshire. (2010), “Implications of climate change for agricultural productivity in the
early twenty-first century”, Philosophical Transactions of the Royal Society of biological Sciences,
Volume 365, No. 1554 2973-2989 , 27 September 2010
21
Gray, Clark L, and Valerie Mueller. (2012). "Drought and Population Mobility in Rural Ethiopia."
World Development no. 40 (1):134-145.
Hirvonen, Kalle. (2014), “Temperature Shocks, Household Consumption and Internal Migration:
Evidence from rural Tanzania.” CSAE Conferences & Workshops: 2014 Annual Conference.
Hoeffler, A. (2002), “The Augmented Solow Model and the African Growth Debate.” Oxford
Bulletin of Economics and Statistics 64 (2): 135–158.
Hoegh-Guldberg, Ove and John F. Bruno (2010): “The Impact of Climate Change on the World’s
Marine Ecosystems”, Science. 18 June 2010, vol. 328, no. 5985 pp. 1523-1528.
Intergovernmental Panel on Climate Change –IPCC. (2007), “Climate Change Impacts,
Adaptation and Vulnerability”, Report of Working Group II to the Fourth Assessment Report of
the IPCC. Cambridge University Press, Cambridge: UK and New York.
IPCC. (2007). Climate Change 2007: Synthesis Report. Contribution of Working Groups I, II and
III to the Fourth Assessment Report of the Intergovernmental Panel on Climate Change. Ed. R. K
Pachauri and Reisinger, A. Vol. 446. November. IPCC Geneva, Switzerland.
Koop, G. (2003). “Bayesian Econometrics”, Wiley.
Koop, Gary, Dale J. Poirier, and Justin L. Tobias. (2007), “Bayesian Econometric Methods.”
Cambridge University Press.
Koubi, Vally, Thomas Bernauer, Anna Kalbhenn and Gabriele Spilker (2012), “Climate
variability, economic growth, and civil conflict”, Journal of Peace Research 2012 49: 113, p.113-
127.
La Rovere, Roberto, Genti Kostandini, Abdoulaye Tahirou, John Dixon, Wilfred Mwangi, Zhe
Guo, and Marianne Bänziger. (2010), “Potential impact of investments in drought tolerant maize
in Africa”; International Institute of Tropical Agriculture (IITA) and International Maize and
Wheat Improvement Center (IMWIC)
Levine, R., and D. Renelt. (1992), “A Sensitivity Analysis of Cross-country Growth Regressions.”
The American Economic Review: 942–963.
Lindley, D. V., and A. F. M. Smith. (1972), “Bayes Estimates for the Linear Model.” Journal of
the Royal Statistical Society. Series B (Methodological): 1–41.
Mankiw, N. G., D. Romer, and D. N. Weil. (1992), “A Contribution to the Empirics of Economic
Growth.” The Quarterly Journal of Economics 107 (2): 407–437.
Mendelsohn, R. 2009. “The Impact of Climate Change on Agriculture in Developing Countries”
Journal of Natural Resources Policy Research 1: 5-19.
22
Mendelsohn, Robert; Ariel Dinar & Larry Williams (2006). “The distributional impact of climate
change on rich and poor countries.” Environment and Development Economics 11(1): 159–178.
Miguel, Edward & Shanker Satyanath (2010) “Understanding transitory rainfall shocks, economic
growth and civil conflict.” NBER working paper (16461) (http://www.nber.org/papers/w16461).
Miguel, Edward, Shanker Satyanath & Ernest Sergenti (2004) “Economic shocks and civil
conflict: An instrumental variables approach.” Journal of Political Economy 112(4): 725–753.
Mitchell, T. D., and P. D. Jones. (2005), “An Improved Method of Constructing a Database of
Monthly Climate Observations and Associated High-resolution Grids.” International Journal of
Climatology 25 (6): 693–712.
Nordhaus, William & Joseph Boyer (2000) “Warming the World: Economic Models of Global
Warming.” Cambridge, MA: MIT Press.
Odusola, A.F. and A.E. Akinlo, (2001), “Output, Inflation and Exchange Rate in Developing
Countries: An Application to Nigeria.” The Developing Economies, XXXIX, 2001.
Odusola, A.F. and Akinlo E.A. (1994), “Food Supply and Inflation in Nigeria.” International
Review of Economics and Business, Volume XLI No.8 August, 1994
Ouraich, Ismail and Wallace E. Tyner (2014), “Climate change impacts on Moroccan agriculture
and the whole economy: An analysis of the impacts of the Plan Maroc Vert in Morocco”, WIDER
Working Paper 2014/083.
Parry, M. L., O. F. Canziani, J. P. Palutikof, P. J. van der Linden, and C. E. Hanson. (2007), IPCC,
2007: Climate Change 2007: Impacts, Adaptation and Vulnerability. Contribution of Working
Group II to the Fourth Assessment Report of the Intergovernmental Panel on Climate Change.
Cambridge University Press, Cambridge.
Pindyck, R. S. (2011), “Fat Tails, Thin Tails, and Climate Change Policy.” Review of
Environmental Economics and Policy 5 (2): 258–274.
Rabassa, M., E. Skoufias, and H. G. Jacoby. (2012), “Weather and Child Health in Rural Nigeria.”
World Bank Policy Research Working Paper (6214). http://www-
wds.worldbank.org/servlet/WDSContentServer/WDSP/IB/2012/10/02/000158349_20121002133
547/Rendered/PDF/wps6214.pdf.
Rosenzweig, C. and M.L. Parry (1994), “Potential impact of Climate Change on world food
supply.”, Nature, Volume 367. 13. January 1994, p.133-138.
Sala-i-Martin, X. X. (1997a). “I Just Ran Two Million Regressions.” The American Economic
Review: 178–183.
23
Sala-i-Martin, X. X. (1997b), “I Just Ran Four Million Regressions.” National Bureau of
Economic Research. http://www.nber.org/papers/w6252.
Savvides, A. (1995). “Economic Growth in Africa.” World Development 23 (3): 449–458.
Schlenker, Wolfram, and David B Lobell. 2010. "Robust negative impacts of climate change on
African agriculture." Environmental Research Letters no. 5 (1):014010.
Schlenker, Wolfram, and Michael J Roberts. 2009. "Nonlinear temperature effects indicate severe
damages to US crop yields under climate change." Proceedings of the National Academy of
Sciences no. 106 (37):15594-15598.
Tol, Richard (2002) “Estimates of the damage costs of climate change.” Part II: Dynamic
estimates. Environmental and Resource Economics 21(1): 135–160.
Tompkins, E. L. and W. N. Adger (2004) “Does adaptive management of natural resources
enhance resilience to climate change?” Ecology and Society 9(2): 10,
http://www.ecologyandsociety.org/vol9/iss2/art10.
Vizy, Edward K. and Kerry H. Cook, (2012). “Mid-Twenty-First-Century Changes in Extreme
Events over Northern and Tropical Africa.” J. Climate, 25, 5748–5767.
Waldinger, Maria (2013). “The Long Term Effects of Climatic Change on Economic Growth:
Evidence from the Little Ice Age, 1500 – 1750.”
http://personal.lse.ac.uk/fleischh/climateandgrowthSept2013.pdf
Weisbach, D.A, E.J., Moyer, M.D. Woolley, and M.J. Glotter, (2013), “Climate impacts on
economic growth as drivers of uncertainty in the social cost of carbon”, Coase-Sandor Working
Paper Series in Law and Economics, http://chicagounbound.uchicago.edu/law_and_economics
[Accessed, 28 May 2014].
World Bank Group. (2011), “Africa Development Indicators 2011.” World Bank Publications
24
Table 1: Descriptive analysis using average yearly temperature between 1961 and 2009 in
Africa
Countries Min Max Max -
Min Mean Standard
Deviation
Absolute
change (1961 –
2009)
Algeria 21.7183 24.0408
2.3225 22.9593 0.5498 1.0092
Angola 21.1717 22.4442
1.2725 21.6619 0.2948 0.6150
Benin 26.6167 28.6092
1.9925 27.5625 0.4599 1.0200
Botswana 20.3900 23.2067
2.8167 21.8570 0.6175 1.4567
Burkina Faso 27.5367 29.1192
1.5825 28.3158 0.3950 1.3367
Burundi 19.8467 21.7325
1.8858 20.4821 0.4582 0.9625
Cameroon 23.9950 25.5050
1.5100 24.7096 0.3263 1.0117
Central African
Republic 24.2825 26.0192
1.7367 25.0949 0.4455 1.0608
Chad 25.7200 28.3292
2.6092 26.9862 0.5752 2.6092
Congo, Dem. Rep. 23.7883 25.3300
1.5417 24.6242 0.3007 0.6442
Congo, Rep. 23.7483 25.0975
1.3492 24.2292 0.3327 1.0075
Cote d'Ivoire 25.5775 27.1658
1.5883 26.4062 0.3228 0.2133
Egypt, Arab Rep. 21.5442 23.7383
2.1942 22.5724 0.5603 2.1450
Equatorial Guinea 23.7967 25.3725
1.5758 24.5874 0.2805 0.6792
Eritrea 25.3542 27.4750
2.1208 26.5285 0.5428 1.8333
Ethiopia 21.8142 23.5233
1.7092 22.6137 0.3842 1.4750
Gabon 24.1667 25.9117
1.7450 25.0922 0.3118 0.4558
Gambia, The 26.5900 28.4700
1.8800 27.4524 0.4575 0.4725
Ghana 26.4450 28.1433
1.6983 27.2854 0.3723 0.6758
Guinea 25.0483 26.5592
1.5108 25.7261 0.3405 0.6742
Guinea-Bissau 26.1575 27.8758
1.7183 26.9548 0.4151 0.3925
25
Kenya 23.4600 25.5508
2.0908 24.5894 0.4281 1.0558
Lesotho 11.4783 13.3975
1.9192 12.3937 0.4861 0.4900
Liberia 24.7108 26.1017
1.3908 25.3811 0.2940 0.4175
Libya 21.2167 23.0992
1.8825 22.2126 0.4904 1.8825
Madagascar 21.6717 22.8117
1.1400 22.2972 0.3214 0.0533
Malawi 21.1992 22.9067
1.7075 22.0151 0.4000 0.7092
Mali 27.4425 29.3650
1.9225 28.5028 0.4787 1.2508
Mauritania 26.7217 29.0292
2.3075 27.9403 0.5555 0.7558
Morocco 16.0350 18.4650
2.4300 17.3518 0.5302 0.2858
Mozambique 23.1583 24.8175
1.6592 23.8753 0.3713 0.2883
Namibia 19.1475 20.9667
1.8192 20.2395 0.3716 0.9458
Niger 26.2017 28.6750
2.4733 27.4515 0.4876 2.4733
Nigeria 26.1875 27.8358
1.6483 26.9258 0.3789 1.5208
Rwanda 18.3283 20.2417
1.9133 18.9906 0.4815 1.0875
Senegal 27.1425 29.0617
1.9192 28.0759 0.4617 0.4650
Sierra Leone 25.6000 26.9650
1.3650 26.2442 0.3212 0.5967
Somalia 26.2883 27.5167
1.2283 26.9508 0.2649 0.6600
South Africa 16.9583 18.5950
1.6367 17.8460 0.4205 0.8250
Sudan 25.8158 28.8592
3.0433 27.2606 0.7315 3.0433
Swaziland 19.3950 21.1558
1.7608 20.2124 0.4443 0.3408
Tanzania 21.8308 23.3808
1.5500 22.5235 0.4159 0.6550
Togo 26.2367 28.2742
2.0375 27.1916 0.4424 0.8367
Uganda 22.0092 24.5800
2.5708 23.0009 0.6691 1.9025
Zambia 20.9608 23.2917
2.3308 21.8409 0.5243 0.9167
Zimbabwe 20.2942 22.9133
2.6192 21.2825 0.5556 1.1375
26
Table 2: Dependent Variable is GDP growth rate using data from 1961-2009 (P (. <0|y) in parentheses)
Explanatory Variables
M1
M2
M3
M4
M5
M6
M7
Temperature (“Pooled” impact on Africa)
-
1.2845
-
1.3987
-
1.2781
-
1.3180
-
1.4206
-
1.2544
-
1.5861
(0.9228)
(0.9992)
(1.0000) (0.9768)
(1.0000)
(0.9846)
(1.0000)
Rainfall 5-year moving average (“Pooled” impact on
Africa)
0.0283
0.02818
0.02738 0.0331
0.0334
0.0673
0.0338
(0.0639)
(0.0659)
(0.0728)
(0.0438)
(0.0452)
(0.0440)
(0.0424)
Constant
0.1545
0.0966
0.05184
0.0158
0.0366
0.0377
0.0230
(0.4268)
(0.4606)
(0.4792) (0.4939)
(0.4849)
(0.4848)
(0.4909)
Log Initial GDP per capita
0.5245
0.33811 0.3196
0.4066
0.3575
0.3628
(0.2142)
(0.3186)
(0.3390)
(0.2940)
(0.3197)
(0.3214)
Population Growth
0.0615
0.02718
-
0.0548
-
0.0316
-
0.0291
-
0.0675
(0.4716)
(0.4900) (0.5240)
(0.5131)
(0.5156)
(0.5286)
Primary School Enrolment (log)
0.14404 0.0338
0.0788
0.0490
0.0677
(0.4349)
(0.0484)
(0.4625)
(0.4736)
(0.4697)
Life expectancy (log)
0.29136
0.1259
0.1971
0.1981
0.1455
(0.3704) (0.4457)
(0.4133)
(0.4137)
(0.4340)
Port
-0.0388
-0.0163
(0.5153)
(0.5069)
Foreign Direct Investment GDP ratio
-
0.4493
-
0.3835
-
0.4471
-
0.3899
27
(0.8923)
(0.7096)
(0.9044)
(0.7119)
Language 0.0622
0.0752
(0.4742)
(0.4651)
Savings -0.0002
-0.0023
(0.5042)
(0.5046)
Secondary School Enrolment (log) 0.2840
0.2915
0.2434
0.3072
(0.3711)
(0.3673)
(0.3908)
(0.3621)
FDI X Secondary School Enrolment -0.0923
-0.0807
(0.5534)
(0.5478)
Sigma square alpha 0.1680
0.1696
0.17271 0.1702
0.1684
0.1695
0.1699
(0.0000)
(0.0000)
(0.0000) (0.0000)
(0.0000)
(0.0000)
(0.0000)
Sigma beta 0.1240
0.1240
0.1243 0.1298
0.1272
0.1267
0.1308
(0.0000)
(0.0000)
(0.0000) (0.0000)
(0.0000)
(0.0000)
(0.0000)
correlation (rho) -0.0395
-0.0406
-0.0405 -0.0397
-0.0402
-0.0406
-0.0397
(0.6025)
(0.6047)
(0.6047) (0.5999)
(0.6026)
(0.6036)
(0.6002)
28
Table 4: Estimation results used for robustness – using Model 7.
Full Sample Sub-sample
(1961 -2009) (1961 - 2000)
Variables Posterior
Mean Pr(:<0|y)
Posterior
Mean Pr(:<0|y)
Temperature Lag 1
1.2616
0.0058
1.1376
0.0223
Temperature Lag2 -0.2897
0.7179
-0.4487
0.7868
Temperature Lag 3
0.0926
0.4264
-
0.2014
0.6381
Temperature Lag 4 0.2336
0.3209
0.3724
0.2592
Tempe
rature Lag 5
0.2473
0.3026
0.4394
0.2076
Rainfall 5- year
MA* 0.0338
0.0424
0.0406
0.0278
Rainfall shocks**
0.0676
0.0595
0.0829
0.0514
Temperature shocks -3.2186
0.9898
-3.2621
0.9795
Note: * MA is moving average
• Shocks for both temperature and rainfall are measured based on standard deviation from their
respective five-year moving averages.
Table 4: Country Level result - Dependent Variable is GDP Growth Rate
All sample
(1961 – 2009) 1961 – 2000
Countries / Variable Posterior
Mean Pr(:<0|y)
Posterior
Mean Pr(:<0|y)
Pooled Mean
Temperature Effect -1.586 1.000 -1.420 1.000
Algeria
-1.482 1.000 -1.277 0.998
Angola
-
1.464
1.000
-
1.452
1.000
Benin
-1.590 1.000 -1.414 1.000
Botswana
-
1.533
1.000
-
1.298
0.997
Burkina Faso
-1.539 1.000 -1.359 1.000
Burundi
-
1.684
1.000
-
1.528
1.000
Cameroon
-1.703 1.000 -1.538 1.000
Central African
Republic -1.749 1.000 -1.571 1.000
Chad
-1.484 1.000 -1.337 0.999
Congo, Dem. Rep.
-1.822 1.000 -1.697 1.000
Congo, Rep.
-1.670 1.000 -1.516 1.000
Cote d'Ivoire
-1.639 1.000 -1.442 1.000
Egypt, Arab Rep.
-1.409 1.000 -1.197 0.996
Equatorial Guinea
-1.244 0.999 -1.145 0.992
Eritrea
-1.450 1.000 -1.170 0.990
29
Ethiopia
-1.537 1.000 -1.430 1.000
Gabon
-1.691 1.000 -1.497 1.000
Gambia, The
-1.574 1.000 -1.405 1.000
Ghana
-1.637 1.000 -1.475 1.000
Guinea
-1.681 1.000 -1.483 1.000
Guinea-Bissau
-1.727 1.000 -1.535 1.000
Kenya
-1.544 1.000 -1.348 1.000
Lesotho
-1.542 1.000 -1.415 0.996
Liberia
-1.582 1.000 -1.442 0.997
Libya
-1.514 1.000 -1.390 0.996
Madagascar
-1.771 1.000 -1.633 1.000
Malawi
-1.549 1.000 -1.388 1.000
Mali
-1.516 1.000 -1.325 0.999
Mauritania
-1.507 1.000 -1.295 0.997
Morocco
-1.492 1.000 -1.327 0.999
Mozambique
-1.572 1.000 -1.465 1.000
Namibia
-1.595 1.000 -1.419 0.999
Niger
-1.534 1.000 -1.351 0.999
Nigeria
-1.581 1.000 -1.410 1.000
Rwanda
-1.628 1.000 -1.517 1.000
Senegal
-1.612 1.000 -1.432 1.000
Sierra Leone
-1.800 1.000 -1.712 1.000
Somalia
-1.570 1.000 -1.377 1.000
South Africa
-1.517 1.000 -1.372 0.999
Sudan
-1.489 1.000 -1.253 0.998
Swaziland
-1.543 1.000 -1.421 0.999
Tanzania
-1.790 1.000 -1.605 1.000
Togo
-1.563 1.000 -1.376 1.000
Uganda
-1.508 1.000 -1.369 0.999
Zambia
-1.685 1.000 -1.533 1.000
Zimbabwe
-1.644 1.000 -1.380 0.999
30
Figure 1: Temperature Series for five of the Most Volatile (High variance) Countries in Africa
15
17
19
21
23
25
27
29
31
1961 1963 1965 1967 1969 19 71 1973 1975 1977 1979 1981 1983 1985 19 87 1989 1991 1993 1995 1997 1999 2001 20 03 2005 2007 2009
Sudan Tunisia Uganda Botswana Chad Linear (Sudan)
31
Figure 2: Temperature Series for five of the Least Volatile (Lowest variance) Countries in Africa
15
17
19
21
23
25
27
29
1961 1963 1965 1967 1969 1971 1973 1975 1977 1979 1981 1983 1985 1987 1989 1991 1993 1995 1997 1999 2001 2003 2005 2007 2009
Sierra Leone Madagascar Gabon Congo, Dem. Rep. Liberia
32
Figure 3: Minimum, Maximum and Standard Deviation of yearly rainfall average (1961-2009)
0.0000
5.0000
10.0000
15.0000
20.0000
25.0000
0.0000
50.0000
100.0000
150.0000
200.0000
250.0000
300.0000
Algeria
Angola
Benin
Botswana
Burkina Faso
Burundi
Cameroon
Central African Republic
Chad
Congo, Dem. Rep.
Congo, Rep.
Cote d'Ivoire
Egypt, Arab Rep.
Equatorial Guinea
Eritrea
Ethiopia
Gabon
Gambia, The
Ghana
Guinea
Guinea-Bissau
Kenya
Lesotho
Liberia
Libya
Madagascar
Malawi
Mali
Mauritania
Morocco
Mozambique
Namibia
Niger
Nigeria
Rwanda
Senegal
Sierra Leone
Somalia
South Africa
Sudan
Swaziland
Tanzania
Togo
Uganda
Zambia
Zimbabwe
Min Max Standard Deviation
33
Figure 4: The intensity of Temperature Impact on Economic Growth in Africa
34
Figure 5: Distribution of the "Pooled" Mean Effect of Temperature on GDP Growth in Africa
35
Appendix
Appendix 1: Additional information on the model specification
For estimation purposes, we rewrite equation (1) in matrix form as:
%
&
'
(
)*
+
…… (A1)
The above equation (A1) will seek to draw
,
and in a single block. The mean and variance
matrix of +
will incorporate the hierarchical priors explained earlier.
Specifically:
+
,
-.+)/
0
1
2U
`)3
4
5
4
6
5
4
6
6
7
1
89"a
Posterior Simulator
Before describing the posterior simulator, first let b%c+
d
Z>
e
U
`
=>
^
& and define
b
=f
as all the elements of b other thang. The joint posterior distribution for all the parameters
of this model can be written as:
hb\
i/jh
\*
+
^
hk+
l
0
1
7
1
U
`
=>
m
n
Z>
2hkl0
;
7
;
4
mhk
l0
6
<
7
6
<
6
mhkl0
;
7
;
4
m
h^
\C
D
h∑
=>
\5
@ "aJ
Step 1: Draw c+
d
Z>
e
\b
=co
p
d
This complete conditional is proportional to the joint posterior distributionhb\. Absorbing all
the terms that do not involve+
into the normalizing constant of this condition gives us the
complete posterior conditional for+
. We have stacked the observations over time for each
country so that:
36
3
>
q
8, *
r
>
>
q q q
s
s
t.
Thus, using the result of Lindley and Smith (1972) we obtain:
hk+
lb
=o
p
mku
o
p
o
p
u
o
p
m"aL
Where
u
o
p
'*
v
*
^
U
`
=>
(
=>
o
p
*
v
^
U
`
=>
+
We sample each of the +
by drawing from the corresponding complete conditional.
Step 2: Complete Posterior Conditional for and
will follow by conditioning on
and
respectively.
The complete conditionals for: \b
=;
@_@
Where
@'
v
^
4
7
;=>
(
=>
_
v
w
^
4
7
;=>
0
;
And w is all the country specific constants stacked.
Step 3: Complete Posterior Conditional for ^
^
\b
=S
xT
AByz!
C
{":|
W*
+
v
W*
+
D
}
=>
~
Step 4: Complete Posterior Conditional for ∑
=>
∑
=>
\b
=∑
•€
?y{|k+•
W+•mk+•
W+•m‚@5
}
=>
5
~
Where +•
refers to only elements of
and
in the vector+
.
37
Appendix 2: Country Estimates with Model Diagnostics - Posterior mean, Probability that the parameter is less than zero,
Posterior standard deviation and Numerical standard errors.
Values Mean Pr(:<0|y)
Std NSE
Algeria
-1.4817
1.0000
0.4779
0.0630
Angola
-1.4637
1.0000
0.4844
0.0644
Benin
-1.5899
1.0000
0.4643
0.0627
Botswana
-1.5332
1.0000
0.4869
0.0600
Burkina Faso
-1.5390
1.0000
0.4664
0.0632
Burundi
-1.6838
1.0000
0.4702
0.0628
Cameroon
-1.7031
1.0000
0.4664
0.0623
Central African Republic
-1.7486
1.0000
0.4640
0.0624
Chad
-1.4842
1.0000
0.4683
0.0633
Congo, Dem. Rep.
-1.8219
1.0000
0.4667
0.0620
Congo, Rep.
-1.6702
1.0000
0.4697
0.0618
Cote d'Ivoire
-1.6393
1.0000
0.4706
0.0629
Egypt, Arab Rep.
-1.4086
1.0000
0.4739
0.0617
Equatorial Guinea
-1.2443
0.9992
0.4806
0.0632
Eritrea
-1.4499
1.0000
0.4882
0.0625
Ethiopia
-1.5365
1.0000
0.4665
0.0625
Gabon
-1.6912
1.0000
0.4837
0.0637
Gambia, The
-1.5744
1.0000
0.4662
0.0623
Ghana
-1.6373
1.0000
0.4690
0.0622
Guinea
-1.6812
1.0000
0.4670
0.0623
Guinea-Bissau
-1.7275
1.0000
0.4624
0.0623
Kenya
-1.5435
1.0000
0.4683
0.0621
Lesotho
-1.5420
0.9996
0.5403
0.0630
Liberia
-1.5825
0.9998
0.5391
0.0674
Libya
-1.5140
1.0000
0.4931
0.0634
Madagascar
-1.7713
1.0000
0.4710
0.0621
Malawi
-1.5489
1.0000
0.4771
0.0623
Mali
-1.5157
1.0000
0.4629
0.0626
Mauritania
-1.5068
1.0000
0.4743
0.0633
Morocco
-1.4915
1.0000
0.4858
0.0633
Mozambique
-1.5721
1.0000
0.4687
0.0623
Namibia
-1.5945
1.0000
0.4983
0.0637
Niger
-1.5340
1.0000
0.4707
0.0633
Nigeria
-1.5810
1.0000
0.4653
0.0629
Rwanda
-1.6280
1.0000
0.4749
0.0628
Senegal
-1.6124
1.0000
0.4634
0.0622
Sierra Leone
-1.8005
1.0000
0.4717
0.0627
Somalia
-1.5703
1.0000
0.4766
0.0621
South Africa
-1.5171
1.0000
0.4830
0.0632
Sudan
-1.4889
1.0000
0.4636
0.0618
38
Swaziland
-1.5432
1.0000
0.4876
0.0640
Tanzania
-1.7900
1.0000
0.4750
0.0604
Togo
-1.5634
1.0000
0.4670
0.0623
Uganda
-1.5084
1.0000
0.4679
0.0625
Zambia
-1.6850
1.0000
0.4843
0.0633
Zimbabwe
-1.6438
1.0000
0.4727
0.0627