Vous êtes sur la page 1sur 150

African Development Review / Revue africaine de développement

The African Development Bank

The African Development Bank is a regional multilateral development finance institution


the members of which are all of the 54 countries in Africa and 25 countries from Asia,
Europe, North and South America. The purpose of the Bank is to further the economic
development and social progress of African countries, individually and collectively. To this
end, the Bank promotes the investment of public and private capital for development, Editor Vol . 29 No. 4 December/Décembre 2017
primarily by providing loans and grants for projects that contribute to poverty reduction John C. Anyanwu
and broadly based development in Africa. Managing Editor Contents
Abebe Shimeles
Elections and Subjective Living Conditions in Sub-Saharan Africa 545
Associate Editors Tugba Zeydanli
La Banque africaine de développement Augustin Fosu Oil and the Naira: A Markov Switching Perspective 562
Akin Iwayemi Idowu Oluwasayo Ayodeji
Abdul B. Kamara Foreign Direct Investment in Ghana: The Role of Infrastructural Development
La Banque africaine de développement est une institution régionale multilatérale de Stephen Karingi and Natural Resources 575
financement du développement. Elle compte 54 pays membres africains et 25 pays Laeticia Mukarasi Ebenezer Bugri Anarfo, Abel Mawuko Agoba and Robert Abebreseh
Germano Mwabu
membres non africains situés en Asie, en Europe et en Amérique du nord et du sud. FDI Determinants in Least Recipient Regions: The Case of Sub-Saharan
Tchétché Nguessan Africa and MENA 589
L’objectif de la Banque est de promouvoir le développement économique et le pro- Catherine Pattillo Godwin Okafor, Jenifer Piesse and Allan Webster
grès social des pays africains, pris individuellement et collectivement. A cet effet, elle Advisory Board Quantifying the Road Influence Zone on Socio-economic Developments
encourage l’investissement de capitaux publics et privés en faveur du développement, Olu Ajakaiye in Rural Tigray, Ethiopia 601
Ernest Aryeetey Fredu Nega Tegebu and Edris Hussein Seid
surtout à travers l’octroi de prêts et de dons à des projets et programmes qui contribuent
Jean-Paul Azam Croissance économique et dégradation de l’environnement au Cameroun 615
à la réduction de la pauvreté et au développement économique du continent. Arne Bigsten Edmond Noubissi Domguia and Henri Njangang Ndieupa
Paul Collier Inequity in Maternal and Child Health Care Utilization in Nigeria 630
Hakim Ben Hammouda Rifkatu Nghargbu and Olanrewaju Olaniyan
The African Development Bank/
Lahouel Mohamed Hedi Financement Bancaire des Petites et Moyennes Entreprises au Cameroun 648
Banque africaine de développement Ravi Kanbur André Dumas Tsambou, Christian Zamo Akono, Ludwick Ndokang Esone et Roger
Rue Joseph Anoma Louis Kasekende Tsafack Nanfosso
Steve Kayizzi-Mugerwa Semi-parametric Regression-based Decomposition Methods: Evidence
01 BP 1387 Kupukile Mlambo from Regional Inequality in Tunisia 660
Abidjan 01, Côte d’Ivoire

Vol . 2 9 No . 4 December/Décembre 2017


Victor Murinde Amal Jmaii, Damien Rousselière and Christophe Daniel
Léonce Ndikumana Analyse des effets de la qualité des institutions sur la volatilité des Investissements
Temitope Oshikoya Directs Etrangers en Afrique 674
Telephone (+225) 20 26 10 20 John Page Bruno Emmanuel Ongo Nkoa and Jacques Simon Song
Email economics-research@afdb.org Jeffrey Sachs
Désiré Vencatachellum Appreciation to Article Reviewers 689
Internet www.afdb.org
Corrigendum 691

ISSN 1017-6772
African Development Review, Vol. 29, No. 4, 2017, 545–561

Elections and Subjective Living Conditions in Sub-Saharan Africa

Tugba Zeydanli

Abstract: In sub-Saharan Africa (SSA), competitive elections can cause considerable violence and widespread
destruction of property, most of which is ethnically motivated. Recent literature shows that ethnic identification is more
prominent during competitive election periods in comparison to other identifying categories such as gender, religion, and class/
occupation. This paper utilizes data from 12 SSA countries and over 40,000 respondents taken from the Afrobarometer. It asks
if individual perceived living conditions changes in the run-up to competitive elections. Strong evidence shows that a perceived
living condition does change. It is positively related to the proximity to an election and this proximity effect depends on the
competitiveness of the election. The paper further investigates the background mechanisms behind this positive relationship,
that is, to what extent does living conditions of the individual change if the party that the individual supports wins the election
and is there a change in living conditions of the individual before and after the election? In addition, this paper documents that
ethnic identification also has a positive impact on individual perceived living conditions after controlling for electoral cycle
variables.

1. Introduction
Elections in sub-Saharan Africa (SSA) are characterized by uncertainties because of the possibility of election-related violence.
Election-related violence could take place at every stage of the electoral process: before, during or after elections. The intense
environment of elections in SSA depends on several factors, including the economic performance of the country, the provision of
public goods, institutional, social, historical, geographical and identity factors. Identity factors such as ethnicity, religion, and
race can be manipulated by the political elite to gain votes. The most pronounced identification factor among countries is ethnic
identification during competitive elections (see Eifert et al., 2010).
Ethnic identity is an important determinant of people’s lives in SSA. It affects who they trust, conduct business with, and
vote for. Moreover, it can influence individuals’ overall well-being. It is often debated in the literature whether the
importance of ethnic identity is driven by social or political affairs. Some scholars argue that ethnic identification comes
from culture; that is, how people have lived throughout the centuries (see Geertz, 1993; Fishman, 1996; and Ajayi, 2000).
Others argue that it is a political construct; political parties in many African countries use ethnic identities as a tool to gain
access to political power (see Young, 1976; Bates, 1983; and Horowitz, 1985). Using the Afrobarometer,1 Eifert et al. (2010)
show that ethnic identification is more prominent during election periods in comparison to other identifying categories such
as gender, religion, and class/occupation. Ethnic attachments become even stronger if elections are in a competitive
environment. More specifically, they show that respondents are 1.8 percentage points more likely to identify ethnically for
every month closer the country is to a competitive presidential election. This suggests that ethnic identities in Africa are
strengthened by political competition.
Competitive elections in ethnically diverse societies usually involve manipulation and violence. These elections result in
loss of life, physical injuries, psychological issues, violation of freedom of speech and of human rights. For instance, Gabon


Department of Economics, Martin Luther University Halle-Wittenberg, IWH, Große Steinstraße 73, D-06099 Halle (Saale), Germany. E-mail: tugba.
zeydanli@wiwi.uni-halle.de. I would like to thank Pedro Vicente for his supervision and guidance as well as Olivier Bargain, Andrew Clark, Louis
L^evy-Garboua, Chou Nuon, Roxana Gutierrez-Romero, Ana Balcao Reis, Semih Sakalli, Claudia Senik, Semih Tumen, Christoph Wunder, two
anonymous referees and seminar participants at the Scottish Economic Society 2016, Centre for the Study of African Economies: Economic
Development in Africa 2016, 1st World Comparative Economics, 4th LCSR International Workshop, 4th LCSR International Summer School, and
NOVA Research Group for very helpful comments and suggestions. Any errors are mine.
© 2017 The Authors. African Development Review © 2017 African Development Bank. Published by Blackwell Publishing Ltd,
9600 Garsington Road, Oxford OX4 2DQ, UK and 350 Main Street, Malden, MA 02148, USA. 545
546 T. Zeydanli

(2009), Guinea-Bissau (2009), Kenya (2007), Lesotho (2008), Malawi (2009), Nigeria (2003–2011), Somalia (2009), and
Zimbabwe (2008) were characterized by violent national (or presidential) elections and more violence after elections. During
Nigeria’s 2003 federal and state election, at least 100 people were killed and many were injured. In Kenya’s 2007
presidential elections, 1,200 people were killed and over a quarter of a million people were displaced. The dispute caused
serious violence among supporters. As a result, leaders from the opposition and ruling parties were charged by the
International Criminal Court for inciting violence. In Zimbabwe’s election in March 2008 there was a violent dispute on the
election results. A civil conflict erupted and had to be resolved partly through an external military intervention. Election-
related violence in Africa is very high. Some elections considered to be free and fair may still experience violence during and
after the election is over.
The main focus of the paper is to investigate how individuals’ subjective living conditions change when competitive elections
are approaching. Due to the intense environment of competitive elections, the general tendency is to expect to observe a fall in the
individual living conditions the closer in time the survey is to an election and further that this decrease is greater in a competitive
election than a landslide election, which leads us to our main hypothesis. Moreover, the paper elaborates background
mechanisms of the relationship, which will be explained further in more detail.
The econometric framework captures the effects of proximity, the competitiveness of national elections, and the interaction of
both variables, while controlling for large sets of individual variables such as gender, age, age-squared, rural/urban areas,
education levels, employment status, and economic conditions. The Afrobarometer enables employing country-fixed effects that
control for country-level features, including unobservable characteristics that cannot be measured since they have been collected
not only across multiple countries but also at multiple points in time for the same countries. This creates a major advantage in
testing the election variables that vary within countries across survey rounds.2
To my knowledge, this is the first paper to shed light on the effects of elections on living conditions. This hypothesis is tested
with several living conditions questions provided by the Afrobarometer across 12 African countries. The main hypothesis is
rejected. This paper provides strong and robust evidence that political competition increases individual-level subjective living
conditions. The change in living conditions is related to how close in time the survey is to an election and this proximity effect
depends on the competitiveness of the election. Individual subjective living conditions increases more in a competitive election
period, in which the margin of victory is near zero compared to a landslide election. For every month closer a country is to a
competitive election, on an average the individual-level subjective living conditions demonstrates a 0.015 standard deviation
increase.
There are several possible mechanisms that account for these relationships. The first possibility is to test the effects of
winning elections on individual subjective living conditions. Kahneman et al. (1999) suggest that partisan identity has
considerable implications for the growing literature on well-being in economics, psychology, and other fields. The results
show that winning the national election increases subjective living conditions. The second possibility may be that individual
living conditions increases as election day approaches due to expectations and aspirations, but then starts to fall gradually.
The result shows that the proximity—before and after the election—is positively related to subjective living conditions, but
the impact before the election is greater than that after the election. The third possibility may be focused on whether there is
any link between public expenditure and individual living conditions. The result shows that the public expenditure on
defence increases the individual-level living conditions. The fourth possibility may be that having participated in politics can
improve living conditions. Stutzer and Frey (2006) show that in Switzerland, engaging directly in the democratic process
through referenda increases life satisfaction. Discussing politics and being interested in public affairs have a positive impact
on perceived living conditions in SSA. The fifth mechanism is that voting in free and fair elections improves the living
conditions of the individual.
In addition, this paper investigates in which direction ethnicity has an impact on one’s living conditions. One factor that
some political scientists and economists identify as a cause of instability and poor economic growth is ethnicity. There is
considerable literature documenting an inverse relationship between social heterogeneity and economic growth (see
Section 3.2 for details). Since the salience of ethnic identification grows stronger during political competition, this leads us
to expect that the higher the identification is, the lower is the individual living conditions. Finally, the sensitivity analyses
check whether the salience of ethnicity is also pronounced in the data and we run an analysis to test the econometric
model.
The plan of the paper is as follows. Section 2 provides an overview of the dataset, justifies the construction of dependent and
independent variables, and explains the details of the econometric model. Section 3 presents the estimates, discusses in detail the
results summarized above, and Section 4 runs some sensitivity analyses. Section 5 concludes the research.
© 2017 The Authors. African Development Review © 2017 African Development Bank
Elections and Subjective Living Conditions 547

2. Data and Methodology

2.1 Data and Summary Statistics

The paper utilizes the Afrobarometer from Round 1 to Round 4. The Afrobarometer is a pan-African, non-partisan research
network. It measures the social, political, and economic atmosphere in Africa at an individual-level with a cross-sectional
approach in 35 African countries. The survey collects detailed information about the respondents’ individual characteristics,
their views about democracy, governance, livelihoods, economic concerns, social capital, conflict and crime, their participation
in the electoral process, and perceptions about national identities. Each survey employs the same sampling methodology and
includes a large, nationally representative sample of individuals.
Twelve SSA countries are used in this study3: Botswana, Ghana, Lesotho, Malawi, Mali, Namibia, Nigeria, South Africa,
Tanzania, Uganda, Zambia, and Zimbabwe.4 The data spans from 1999 to 2009, covering almost ten years of information from
each country. In order to account for country-fixed effects and have the variation of a different election in a given country, all
countries are included in each survey round. To achieve national representativeness, appropriate weights have been used.
Weights are calculated as 1/(number of observations for that country).
The Afrobarometer includes several candidates for living conditions:

1. Your Present Living Conditions—‘In general, how would you describe your own present living conditions?’
2. Your Living Conditions in 12 Months—‘Looking ahead, do you expect the following to be better or worse: Your living
conditions in twelve months’ time?’
3. Your Living Conditions Compared to 12 Months Ago—‘Looking back, how do you rate the following compared to twelve
months ago: Your living conditions?’
4. Your Living Conditions Compared to Others—‘In general, how do you rate your living conditions compared to those of
others in your country?’
5. Ethnic Group Economic Conditions—‘Think about the condition of your ethnic group. Are their economic conditions
worse, the same as, or better than other groups in this country?’
The first three questions’ responses are based on a five-point scale with 1 representing ‘very bad’; 2 ‘fairly bad’; 3 ‘neither
good nor bad’; 4 ‘fairly good’; and 5 ‘very good’. ‘Your living conditions compared to others’ and ‘ethnic group economic
conditions’ are ranked on a five-point scale on which 1 indicates ‘much worse’ and 5 indicates ‘much better’.
To streamline the interpretation and draw a general conclusion, the paper analyses hypotheses with two different
dependent variables to create indices. Summary indices are created by aggregating information across related outcomes of
similar subjective living conditions (Kling et al., 2007). The main motivation for this grouping is to improve the statistical
ability to detect effects that are consistent across specific outcomes when these specific outcomes also have idiosyncratic
shocks. Following the methodology of Kling et al. (2007), summary indices are created based on specific outcomes, in
which specific outcomes are normalized by subtracting the mean of the group and then dividing by the standard deviation
of the group. Formally, X i is the ith of I outcomes; let mi be the group mean and X let s i be the standard deviation of the
group. The normalized outcome is X i ¼ ðX i  mi Þ=si . The summary index is X i ¼ X =I . Overall, the summary index
i i i
is defined as the weighted average of z-score of its components. The z-scores are normalized scores based on the group
mean and standard deviation. As stated in Table 1, each component of the index has a mean of zero and a standard
deviation of one.
The first outcome is for ‘your living conditions’, which is a combination of the first three questions. These questions
respectively evaluate the individual’s current situation, future condition, and the comparison of current and past living
conditions, in an attempt to measure living conditions in a time perspective. In this respect, they are from the same domain, which
enable aggregate information across related outcomes. The mean of ‘your living conditions in 12 months’ is higher than the other
two living condition variables, most probably due to the individual’s high expectations and aspirations for the future. The mean
of comparison of current and past living conditions using ‘your living conditions compared to 12 months ago’ is higher than the
current living conditions, which might roughly be interpreted as an indication that the people are becoming happier. The second
outcome is for ‘your conditions compared to others’, which is a combination of the last two questions. This assumes a reference
group of language/tribe/ethnic group. This assumption becomes stronger with the similar mean approximately 2.8, and high
correlation 68 percent.5
© 2017 The Authors. African Development Review © 2017 African Development Bank
548 T. Zeydanli

Table 1: Summary statistics — Afrobarometer, Round 1-2-3-4


Variable Mean Std. dev. Available round

Dependent variable
1. Your living conditions 0 1 2-3-4
Your present living conditions 2.61 1.22 2-3-4
Nor. your present living conditions 0 1 2-3-4
Your living conditions in 12 months 3.39 1.25 1-2-3-4
Nor. your living conditions in 12 months 0 1 2-3-4
Your living conditions compared to 12 months ago 3.01 1.131 1-2-3-4
Nor. your living conditions compared to 12 months ago 0 1 2-3-4
2. Your conditions compared to others 0 1 3-4
Your living conditions compared to others 2.81 1.08 1-2-3-4
Nor. your living conditions compared to others 0 1 3-4
Ethnic group economic conditions to others 2.8 1.017 3-4
Nor. ethnic group economic conditions to others 0 1 3-4
Individual characteristics
Male 0.50 0.5 1-2-3-4
Age 36.79 14.89 1-2-3-4
Urban 0.37 0.48 1-2-3-4
Postgraduate 0.004 0.06 1-2-3-4
University 0.03 0.16 1-2-3-4
High school 0.07 0.25 1-2-3-4
Secondary school 0.37 0.48 1-2-3-4
Primary school 0.36 0.48 1-2-3-4
Informal schooling 0.03 0.18 1-2-3-4
No schooling 0.13 0.34 1-2-3-4
Employed 0.37 0.48 1-2-3-4
Interested in public affairs 1.78 1.10 2-3-4
Trust national electoral commission 1.56 1.1 2-3-4
Economic conditions 0 1 2-3-4
How often gone without food 3.02 1.07 1-2-3-4
Nor. how often gone without food 0 1 2-3-4
How often gone without water 3.08 1.11 1-2-3-4
Nor. how often gone without water 0 1 2-3-4
How often gone without medical care 2.95 1.1 1-2-3-4
Nor. how often gone without medical care 0 1 2-3-4
How often gone without cash income 2.58 1.17 1-2-3-4
Nor. how often gone without cash income 0 1 2-3-4
Personal identification
Occupation/Class 0.35 0.47 1-2
Language/Ethnic/Tribe Group 0.26 0.44 1-2
Religion 0.16 0.36 1-2
Gender 0.04 0.20 1-2
Other 0.17 0.38 1-2
No. of observations 60,050
Notes: Weights are calculated as 1/(number of observations of that country). Stated number of observations is for independent variables in all rounds. Number of
observations for each dependent variable is noted in estimation results.

Gender, age, age-squared, rural/urban areas, education levels, employment status, and economic conditions are controlled for
the individual characteristics. The variables of economic conditions are the indices of the following questions, ‘Over the past
year, how often, if ever, have you or anyone in your family gone without enough food to eat, enough clean water for home use,
medicines or medical treatment, and a cash income?’ To test the relationship between living conditions and ethnic identification,
© 2017 The Authors. African Development Review © 2017 African Development Bank
Elections and Subjective Living Conditions 549

the personal identification question is used; ‘Besides being [a citizen of X], which specific group do you feel you belong to first
and foremost?’ The responses are grouped into five categories: language/ethnic group/tribe, religion, occupation/class, gender,
and other. The other category stands for race, region, age, ‘I’m my own person’. This paper adopts a country-fixed effect
framework, which automatically controls for many other aspects of a country: level of economic development, history, civil war,
etc.6 All regressions include round dummies.
Nearly half of the individuals in the sample are male with an average age of 37. This paper restricts the sample
minimum age to 18, which means that all individuals have the right to vote. Approximately 37 per cent of individuals in
the sample live in urban areas. Education is grouped into seven categories. Postgraduate refers to graduate studies with 0.4
per cent of the sample meeting this criterion. Nearly 3 per cent graduated from a university. The highest share in education
falls into primary and secondary school. Less than half of the sample is unemployed. As a proxy of income variable,
economic conditions are controlled for, which is an index that averages together income-related variables such as how
often the respondent had gone without food, water, medical care, and income. Nearly 35 per cent of individuals in the
dataset identified themselves by occupation/class, 26 per cent by language/ethnic/tribe, 16 per cent by religion and 0.4 per
cent by gender.
Table 2 summarizes economic and political characteristics of sample countries. The average GDP per capita of sample
countries is higher than the average of SSA, which is mainly driven by Botswana, Namibia, and South Africa. The other
countries are poorer on an average in the SSA. Rates of urbanization are almost the same level as the SSA average.
Utilizing at least two7 round surveys brings variations in the months to election variable, which is called ‘proximity to the
election’ in the regression. In Botswana, for example, 1 means that the survey round occurred one month before the
election and 15 means the survey round occurred 15 months after the election. The competition of presidential election is
measured by vote margin, and is simply the vote share difference between the winner and the runner-up. The
competitiveness level in sample countries is similar to the African average. Utilizing three rounds of the Afrobarometer
brings more variation within country in the competitiveness variable as analyses can make use of more elections. Out of 48
countries, 16 have an election in less than 10 months; 6 out of these 16 countries have a vote margin of more than 0.34,
which is the average vote margin of SSA. The last column stands for the name of the ruling party during that election
period. This variable is utilized for examining whether winning a competitive election changes individual-level subjective
living conditions.

2.2 Empirical Methodology


The econometric model employs the same model as that of Eifert et al. (2010), which is designed to illuminate the influence of
proximity of election, competitiveness of election, and interaction of these two effects to subjective living conditions. In the
model, i represents the individual respondent, c is for country, and t denotes the survey round as attached to the individual
subjective living conditions SLCict . Within this setup, the paper systematically analyses the extent to which individual-level
suective living conditions is related to observable characteristics and a country’s political environment.

SLCict ¼ X 0ict a þ C 0ct b þ mict ð1Þ

The vector X ict represents individual-level variables,8 the vector C ct is for country-level factors, and mict is individual’s
idiosyncratic level. The focus of the paper is on election variables and are represented as C 0ct b ¼ b1 pct þ b2 cct þ b3 ðpct x cct Þ: pct
is a proximity variable that measures months until the election in the country compared to the survey round. In Table 2, negative
numbers indicate the most recent past election. Proximity is coded as 1absðmonths to=f rom the most recent electionÞ so that
larger numbers imply increasing proximity. cct is a competitiveness variable and defined as vote margin, which is the gap
between the vote share of the winner and the runner-up in the most recent election. The competitiveness variable is calculated
from the vote margin as 1ðvote marginÞ. Larger numbers indicate increasing competitiveness. pct  cct is the interaction
variable of proximity and competitiveness.
Since the dependent variable is ordinal rather than cardinal, the ideal way to carry out analyses is through ordered
probit. However, Ferrer-i-Carbonell and Frijters (2004) demonstrate that the results from cardinal analysis using ordinary
least squares (OLS) is very similar to those from ordinal. For ease of interpretation, the equation is estimated by using
OLS.
© 2017 The Authors. African Development Review © 2017 African Development Bank
550 T. Zeydanli

Table 2: Economic and political characteristics of countries


Economic characteristics Political characteristics
Country and survey year GDP per capita ($) %Urban Month to election Vote margin Ruling party

Botswana, 1999 7,727 52 1 0.31 BDP


Botswana, 2003 9,366 56 15 0.25 BDP
Botswana, 2005 11,177 57 8 0.25 BDP
Botswana, 2008 14,104 60 12 0.31 BDP
Ghana, 1999 1,390 43 12 0.04 NPP
Ghana, 2002 1,560 45 22 0.08 NPP
Ghana, 2005 2,030 57 8 0.08 NPP
Ghana, 2008 2,486 50 4 0.00 NDC
Lesotho, 2000 1,019 19 23 0.36 LCD
Lesotho, 2003 1,172 22 10 0.32 LCD
Lesotho, 2005 1,330 23 19 0.28 NIP
Lesotho, 2008 1,648 25 20 0.28 NIP
Malawi, 1999 556 14 6 0.14 UDF
Malawi, 2003 561 15 12 0.08 Coalition
Malawi, 2005 605 15 13 0.08 Coalition
Malawi, 2008 727 15 6 0.36 DPP
Mali, 2001 727 29 16 0.92 ADEMA
Mali, 2002 747 29 6 0.30 Coalition
Mali, 2005 914 31 22 0.52 ADP
Mali, 2008 998 33 20 0.52 ADP
Namibia, 1999 3,872 32 3 0.66 SWAPO
Namibia, 2003 4,405 34 14 0.69 SWAPO
Namibia, 2006 5,998 36 15 0.69 SWAPO
Namibia, 2008 6,596 37 13 0.64 SWAPO
Nigeria, 2000 1,131 42 10 0.26 Coalition
Nigeria, 2003 1,597 44 6 0.30 PDP
Nigeria, 2005 1,795 46 18 0.51 PDP
Nigeria, 2008 2,149 48 13 0.51 PDP
S. Africa, 2000 6,653 57 13 0.57 ANC
S. Africa, 2002 7,195 58 19 0.57 ANC
S. Africa, 2006 9,319 60 22 0.57 ANC
S. Africa, 2008 10,250 61 6 0.49 ANC
Tanzania, 2001 823 23 7 0.53 CCM
Tanzania, 2003 938 23 29 0.69 CCM
Tanzania, 2005 1,073 24 4 0.69 CCM
Tanzania, 2008 1,313 25 28 0.36 CCM
Uganda, 2000 763 12 10 0.42 YKM
Uganda, 2002 866 13 17 0.42 NRM
Uganda, 2005 1,014 13 10 0.22 NRM
Uganda, 2008 1,268 14 30 0.42 NRM
Zambia, 1999 899 35 26 0.04 MMD
Zambia, 2003 1,055 36 17 0.04 MMD
Zambia, 2005 1,073 37 13 0.14 Coalition
Zambia, 2009 1,367 38 8 0.02 MMD
Zimbabwe, 1999 885 33 7 0.02 ZANU-PF
Zimbabwe, 2004 314 35 26 0.14 ZANU-PF
Zimbabwe, 2005 477 36 1 0.53 ZANU-PF
Zimbabwe, 2009 425 38 14 0.05 MDC

Ave. sample 2,840 35 14 0.35
 
Ave. SSA 1,606 37 0.34
Notes: Macroeconomic variables are taken from World Development Indicators. Political variables come from the African Election Database.

© 2017 The Authors. African Development Review © 2017 African Development Bank
Elections and Subjective Living Conditions 551

3. Results and Discussion


This section documents the empirical results and provides an extensive discussion of the impact of election variables on
subjective living conditions and the possible underlying mechanism of this relationship. Equation (1) is estimated by using the
Afrobarometer controlled for individual characteristics, country, and round dummies on different living condition measures.

3.1 Effects of Proximity to Competitive Elections


This section provides a discussion of electoral cycle variables. All individual subjective living conditions measures generate
almost identical results: the impact of elections on subjective living conditions is positive and significant. This leads us to reject
the first hypothesis. The individual perceived living conditions increases by 0.015 standard deviation when competitive elections
are approaching.
Figure 1 shows the proximity to the closest country election on the x-axis and the predicted subjective living conditions on the
y-axis by competitiveness of national elections.9 Results are documented in two groups: high competitiveness, in which the
median of the electoral margin is less than the sample median of 36 percentage points and less competitiveness (landslide
elections), in which the electoral margin is more than the median sample. There are 10 countries where the elections are
competitive and the proximity is lower than 10 months. The relationship is clearly evident. Landslide elections—even when in
the proximity of elections—yield a lower subjective living conditions compared to relatively higher competitive elections. A
subjective living condition becomes higher as time nears the elections. In other words, elections have a positive impact on
subjective living conditions of individuals in SSA, but only where elections are meaningful contests for political power.
Tables 3 and 4 report the tests of the first hypothesis under three specifications. All specifications include country-fixed
effects; weigh each observation by 1/(number of observations from that country) to weigh each country survey round equally.10
The first columns in Tables 3 and 4 suggest mixed results. In the case of ‘your living conditions’, the proximity of the survey to a
presidential election (in months, absolute value) on average decreases the living conditions of the individual. However, the
competitiveness of that election (the margin of victory) has a positive impact on subjective living conditions. ‘Your living
conditions compared to others’ has exactly opposite results; the competitiveness of the election decreases the living conditions

Figure 1: Electoral proximity and subjective living conditions by competitiveness of elections

© 2017 The Authors. African Development Review © 2017 African Development Bank
552 T. Zeydanli

Table 3: Your living conditions and electoral cycle


Dependent variable: Your living conditions
Electoral proximity 0.0048 0.0173 0.0172
(0.0007) (0.0031) (0.003)
Electoral competitiveness 0.833 1.64 1.61
(0.0545) (0.123) (0.12)
Proximity competitiveness – 0.0288 0.027
– (0.004) (0.0038)
Individual characteristics No No Yes
Country rounds Yes Yes Yes
R2 0.1068 0.1082 0.1804
N 41,186 41,186 41,186
Notes:  ,  ,  indicate the 10%, 5%, and 1% significance levels, respectively. Standard errors are reported in parentheses. Regressions are controlled for
individual characteristics, country dummies, and round dummies. Weights are 1/(number of observations of that country). Competitiveness variable is calculated
from vote margin as 1 (Vote Margin). Larger numbers indicate increasing competitiveness. Proximity is 1 abs(months to/from nearest election), so that
larger numbers imply increasing proximity.

level, while living conditions are increased as the election draws nearer. Specifically, 6 out of 12 countries experience a different
election. The dataset allows analysis of ‘your living conditions’ in three rounds, so in that case the variation becomes higher
when all countries have experienced at least one different election. Given these concerns, it is wise to pay attention to interaction
terms between proximity and competitiveness. After adding the interaction term to the regression (the second column of Tables 3
and 4), positive and statistically significant coefficients of election variables among all dependent variables are obtained. More
specifically, every month closer a country gets to a competitive election, on average, individual-level subjective living
conditions increases 0.014–0.017 in standard deviation depending on the dependent variable of living conditions. Moreover, the
higher the competitiveness of the election, the greater the subjective living conditions of the individual. The results are confirmed
in Column 3, which controls for individual-level characteristics such as age, gender, education, economic conditions, and urban
or rural residence. Among all dependent variables, the results are quite consistent with each other, and they are identical between
the last two specifications.
Our first hypothesis was expected to reveal a fall in the individual living conditions due to the intense environment of
competitive elections. However, the results suggest the opposite. In the following sections, the paper tests what could be the
underlying mechanism of these results. Potentially, there could be three leading reasons: (a) people would like to be in a union,
where they support the same ideology and fight for it; in our context union might be ‘ethnic identification’; (b) people would like

Table 4: Your living conditions compared to others and electoral cycle


Dependent variable: Your living conditions compared to others
Electoral proximity 0.0068 0.015 0.014
(0.0009) (0.0093) (0.0009)
Electoral competitiveness 0.721 0.378 0.357
(0.08) (0.029) (0.0276)
Proximity Competitiveness – 0.0178 0.017
– (0.0011) (0.0011)
Individual characteristics No No Yes
Country rounds Yes Yes Yes
R2 0.0681 0.0770 0.1767
N 28,844 28,844 28,844
Notes:  ,  ,  indicate the 10%, 5%, and 1% significance levels, respectively. Standard errors are reported in parentheses. Regressions are controlled for
individual characteristics, country dummies, and round dummies. Weights are 1/(number of observations of that country). Competitiveness variable is calculated
from vote margin as 1 (Vote Margin). Larger numbers indicate increasing competitiveness. Proximity is 1 abs(months to/from nearest election), so that
larger numbers imply increasing proximity.

© 2017 The Authors. African Development Review © 2017 African Development Bank
Elections and Subjective Living Conditions 553

to feel happy if they observed their supported party is close to winning the elections; (c) or simply participating in politics in any
way might give a feeling of contributing to their country’s political and economic affairs.

3.2 Background Mechanisms of Elections and Subjective Living Conditions

It has been clearly documented that elections increase people’s living conditions. This finding requires more research to
understand the underlying mechanism of the positive relationship. There might be several channels, but the most pronounced
ones are ethnic identification, winning elections, ex-ante and ex-post impact of elections, the effects of public expenditure on
education, health, and defence, the effects of having actively participated in politics, and trust in the national electoral
commission. This part of the paper explains these leading background relationships.

Effects of Ethnic Identification


Some political scientists and economists identify ethnicity as a cause of instability and poor economic growth. There is a
considerable literature documenting an inverse relationship between social heterogeneity and economic growth (Easterly and
Levine, 1997; Montalvo and Reynal-Querol, 2005). A high level of ethnic diversification tends to affect economic growth and
development processes such as trust and transaction costs (Leigh, 2006; Knack and Keefer, 1997), provision of public good
(Kimenyi, 2006; Fosu et al., 2006), contact and contracts (Bates, 2000), and the level of investment (Mauro, 1995). Easterly and
Levine’s (1997) famous ‘growth tragedy’ is primarily based on the strong link between ethnic heterogeneity and slow growth in
SSA. They have documented that moving from an ethnically homogeneous country to one with a diversity of ethnic communities
corresponds to a decrease in annual economic growth rates of more than 2 per cent. These findings have been applied to Africa
due to the high ethnic diversification of these countries. However, these studies have been criticized for employing ethnic
fractionalization—known as ELF (ethno-linguistic fractionalization)—as a measure of ethnic diversity. The main criticism
pertains to the assumption upon which ELF is built (Posner, 2004; Alesina et al., 2003; Fearon, 2003; Roeder, 2001). The paper
analyses ethnicity with a different approach; it tries to estimate the relationship between living conditions and ethnicity after
controlling for electoral cycle variables.
Table 5 documents the importance of personal identification on living conditions. The regressions account for individual
characteristics, country and time dummies, and election variables.11 Having ethnicity as a personal identification yields mixed
results, however having religion as a personal identification has a positive and statistically significant impact on subjective living
conditions of the individual, controlling for electoral cycle variables. Gender also plays a role in that purpose, but this result
should be evaluated with caution since a low percentage of individuals identified themselves with their gender.12 When the
individual evaluates their living conditions based on time preferences, which is ‘your living conditions’, this is negatively related

Table 5: Personal identification and subjective living conditions


Dependent variable Your living conditions Your living condition compared to others

Language/Ethnic/Tribe Group 0.043 0.043


(0.017) (0.018)
Religion 0.07 0.149
(0.019) (0.0209)
Gender 0.038 0.104
(0.034) (0.038)
Other 0.015 0.016
(0.020) (0.023)
R2 0.1500 0.1182
N 20,123 17,554
Notes: Occupation/Class is taken as a reference point. Regressions are controlled for individual characteristics, country dummies, and election variables.  ,  ,

indicate the 10%, 5%, and 1% significance levels, respectively. Standard errors are reported in parentheses. Weights are 1/(number of observations of that
country).

© 2017 The Authors. African Development Review © 2017 African Development Bank
554 T. Zeydanli

to ethnic identification. However, the other dependent variable is positively related. There may be several mechanisms behind
this relationship. Several studies argue that there are cultural differences between the ethnic groups regarding subjective living
conditions. Ethnic groups have different conceptions of well-being and that different factors influence their subjective living
conditions (Neff, 2007; Agyemang et al., 2013). For instance, Addai and Pokimica (2010) report that ethnicity is an important
determinant of perceived living conditions of individuals in Ghana. In their analysis, ethnicity tends to have both negative and
positive effects on living conditions among different ethnic groups and different sub-samples.
Identifying oneself in an ethnic group may be linked to belonging to a community, which in turn increases well-being. When
the analysis is conducted without electoral cycle variables, the coefficient of ethnic identification is smaller compared to
controlling for electoral cycle variables. This suggests that the more pronounced the ethnic identification, the higher is the
subjective living conditions of the individual. Unlike general expectations about high ethnic diversification creating lower living
conditions, individuals who identify themselves as their ethnic group report higher living conditions. Social heterogeneity in
SSA can make use of this information for policy. It may be the case that politicians who play the ethnic card strategy in their
election campaign stimulate the well-being of individuals.
Religion identification yields positive relationship with all dependent variables and has statistically significance in ‘your
living conditions’ and ‘your living conditions compared to others’. Individuals with strong religious beliefs report a higher level
of life satisfaction and greater personal happiness (Ellison, 1991; Ellison et al., 2001). In addition to that religion is also a key
determinant in Ghana for perceived living conditions (Pokimica et al., 2012).

Effects of Winning Elections

The model is tested to determine whether winning the election has an influence on subjective living conditions. The dummy
variable is created for that purpose utilizing the following question: ‘Do you feel close to any particular political party or political
organization? If so, which party or organization is that?’ Since the winning party of this election is known, if the winning party is
the same as the answer from the individual, it is scored as a one; otherwise it is scored as zero. Using the same model, the results
show that an individual supporting the winning party in a competitive election increases individual-level subjective living
conditions (Table 6).
Individuals might think that winning the election brings economic privilege, employment opportunity, protection from
possible threats in the future such as civil war, ethnic clashes and so on, and easy access to health and education services. These
opportunities increase the expectations and aspirations of individuals and lead to higher individual subjective living conditions.
Moreover, almost all countries in the dataset except for Malawi and Mali have had the same incumbent for at least three
presidential elections. Winners might have perceived retaining the presidency as maintaining the status quo of ongoing policies.

Asymmetrical Effects of Elections


An individual’s expectations and aspirations may be higher before rather than after the election and these may form one’s level of
living conditions. There is a strong possibility that subjective living conditions increase as election day approaches but then starts
to fall post-election. The model is designed for testing the symmetric effects of elections. In order to observe asymmetrical
effects, especially before the election, a dummy variable is created for countries in which round surveys would be completed

Table 6: Winning the election


Dependent variable Your living conditions Your condition compared to others

Winning Dummy 0.0401 0.030


(0.0108) (0.0126)
R2 0.1825 0.1784
N 41,186 28,844
Notes:  ,  ,  indicate the 10%, 5%, and 1% significance levels, respectively. Standard errors are reported in parentheses. Regressions are controlled for
individual characteristics, country dummies, round dummies, and election variables. Weights are 1/(number of observations of that country).

© 2017 The Authors. African Development Review © 2017 African Development Bank
Elections and Subjective Living Conditions 555

Table 7: Before the election


Dependent variable Your living conditions Your living condition compared to others

Before election 1.58 1.44


(0.105) (0.248)
Before Election Electoral proximity 0.042 0.044
(0.003) (0.0097)
Before election Electoral competitiveness 1.78 1.26
(0.095) (0.0207)
Before election Proximity Competitiveness 0.051 0.039
(0.0043) (0.012)
R2 0.1883 0.1727
N 41,186 28,844
Notes:  ,  ,  indicate the 10%, 5%, and 1% significance levels, respectively. Standard errors are reported in parentheses. Regressions are controlled for
individual characteristics, country dummies, and round dummies. Weights are 1/(number of observations of that country). Competitiveness variable is calculated
from vote margin as 1 (Vote Margin). Larger numbers indicate increasing competitiveness. Proximity is- 1 abs(months to/from nearest election), so that
larger numbers imply increasing proximity. Before the election is 1 if round survey would be before the nearest election.

prior to the nearest election. If the election to month variable is above (below) zero, it is referred to as ‘after (before) election’ and
carries a value of 1. The ex-ante and ex-post effects have a positive and statistically significant relationship with individual living
conditions.13 The impact of the electoral cycle on subjective living conditions is more powerful before the election than after the
election (Table 7).
During election campaigns, the general tendency is to conduct populist policies such as expansionary fiscal policies—cut
taxes, increase government spending, and subsidize small and medium-sized enterprises—by the incumbent, providing food,
water or other necessities that the people need, and gifts to entertain society. Block (2002) analyses a number of fiscal and
monetary variables in SSA during and after elections and concludes that governmental spending shifts toward more visible,
current expenditures, and away from public investment. This temporary help may increase individual living conditions. In
addition, individuals want to believe that something will change within their country with the upcoming election; this hope may
yield higher subjective living conditions. The leading reason for observing positive living conditions after an election may be due
to a decrease in the intensity of the environment. Individuals can attain relief since the uncertainty deriving from the election is
over.

Effects of Public Expenditure


There might be a potential link between public expenditure and voter’s living conditions. The individual may perceive that their
living conditions increase during the period of the incumbent party due to an increase in public welfare. Moreover, it is shown in
the literature that increased public investment as a percentage of gross domestic product decreases the probability of the
incumbent losing (Hausken and Ncube, 2014). Thereby, the paper analyses whether a change in public expenditure on education,
health, and defence can have an impact on the living conditions of individual. The paper focuses on changes in public
expenditures by services such as education, health, and defence, instead of total change in public expenditure. This is because
politicians may change the composition of expenditure in an election year, without increasing the overall budget. The paper
analyses the change in public expenditure during the election year. For example, during Botswana’s 1999 election, the paper
accounts for the change in public expenditure from 1998 to 1999. It utilizes the change in ‘government expenditure on education’
for education and ‘government expenditure on health’ for health, and defence is based on ‘government expenditure on
military’.14
Among three types of public expenditure, the impact of defence expenditure on individual living conditions is the highest and
statistically significant for satisfaction with your living conditions and satisfaction with your living conditions compared to
others (Table 8). The military spending is an important issue for SSA, which has been through considerable turmoil, with high
levels of conflict in the region and within country that especially increases during the election period. Individuals may feel secure
when the government expenditure is higher on the military and this feeling of security may lead to an increase in their living
© 2017 The Authors. African Development Review © 2017 African Development Bank
556 T. Zeydanli

Table 8: Public expenditures


Dependent variable Your living conditions Your condition compared to others

PE on education 0.0758 0.0581


(0.230) (0.135)
PE on health 0.0648 0.00692
(0.00337) (0.216)
PE on defence 0.0322 0.0179
(0.0158) (0.00885)
R2 0.1825 0.1784
N 41,186 28,844
Notes:  ,  ,  indicate the 10%, 5%, and 1% significance levels, respectively. Standard errors are reported in parentheses. Regressions are controlled for
individual characteristics, country dummies, round dummies, and election variables. Weights are 1/(number of observations of that country).

conditions. The relationship between health expenditure and individual living conditions is only positive and statistically
significant in satisfaction with your living conditions. Regarding education expenses, it is positively related with living
conditions but is not statistically significant.

Effects of Participating in Politics


The Afrobarometer permits analysing this mechanism in various ways. The main way is to look into whether discussing
politics15 and being interested in public affairs16 increase individual living conditions (Table 9). The mean of discussing politics
is 0.91 and a standard deviation of 0.72. The mean of interested in public affairs is 1.78 and a standard deviation of 1.10. In line
with each other, they are positively related with individual subjective living conditions.17 The paper also checks the impact of
active participation such as attending a demonstration or protest march. The final way may be that the interaction with political
party official increases living conditions of the individual. The individual can think that being in contact with the party official
can increase their living conditions due to money or food aid. Moreover, the individual can believe that when he/she discusses
topics related to the country with someone in power, this gives some hope to enhance living conditions. Both have no significant
impact on any dependent variable. However, the number of responses to the questions are low; more than 50 per cent of the
sample never attended a demonstration, and less than 10 per cent of the sample have been in contact with a political party official.
Individuals participating in politics in low-income countries may feel particularly valuable during the election period. These
individuals make a fundamental contribution to democratic governance in their country, which can change the future of the
country and their living conditions. Moreover, when the electoral process is competitive and candidates or parties are forced to
expose their records and future intentions to popular scrutiny, more discussion and interest in politics arises.

Effects of Illegitimacy of Elections


To gain a sense of the illegitimacy of elections, the paper controls for ‘trust national electoral commission’, which has a mean of
1.56 and a standard deviation of 1.1. The trust variable is positively and significantly correlated with subjective living conditions

Table 9: Interested in public affairs


Dependent variable Your living conditions Your condition compared to others

Interested in public affairs 0.0285 0.0280


(0.0069) (0.0082)
R2 0.1817 0.1769
N 41,186 28,844
Notes:  ,  ,  indicate the 10%, 5%, and 1% significance levels, respectively. Standard errors are reported in parentheses. Regressions are controlled for
individual characteristics, country dummies, round dummies, and election variables. Weights are 1/(number of observations of that country).

© 2017 The Authors. African Development Review © 2017 African Development Bank
Elections and Subjective Living Conditions 557

Table 10: Illegitimacy of elections


Dependent variable Your living conditions Your condition compared to others

Trust national Electoral Commission 0.0410 0.0252


(0.00474) (0.00542)
R2 0.1783 0.1764
N 41,186 28,844
Notes:  ,  ,  indicate the 10%, 5%, and 1% significance levels, respectively. Standard errors are reported in parentheses. Regressions are controlled for
individual characteristics, country dummies, round dummies, and election variables. Weights are 1/(number of observations of that country).

(Table 10). The higher the legitimacy in elections, the higher is the living conditions of the individual. The legitimacy of
elections also presents transparent political environment and democracy. Even in 1980s Africa, Odedokun (1993) documents
that the political democracy promotes economic growth. Williams (2016) shows the importance of the democratic institutions on
financial development, in addition to that more transparent countries attract foreign direct investment, which affects the
individual living conditions (Mijiyawa, 2015). Allowing people to freely choose from different alternatives in competitive
elections increases political trust as well as creating the democratic environment and those increases lead to greater subjective
living conditions.
In SSA, vote buying and ballot fraud are serious problems during elections. The only variable to control for that purpose is
‘trust national electoral commission’. However, this variable should be evaluated carefully as conditional correlations exist. An
individual who thinks that elections are free and fair, probably voted for the one who wins the election: s/he already feels content
about the election result. On the other hand, the loser starts blaming the commission since s/he is not happy with the result.

4. Sensitivity Analyses

4.1 The Salience of Ethnicity


This subsection conducts the same exercise of Eifert et al. (2010) with our dataset. The dataset differs from their dataset in terms
of sample countries—Ghana and Lesotho are included—and some individual characteristics such as the economic conditions.
The main motivation is to observe whether the ethnic identification is more pronounced during competitive elections with our
dataset. A multinomial logit model and OLS model are utilized and control for individual characteristics and time and country
dummies. The results in Table 11 are in line with Eifert et al. (2010). Every month closer to competitive elections, survey
respondents are on average 1.8 percentage points more likely to identify themselves in ethnic terms.

Table 11: The salience of ethnic identification


Language/Tribe/Ethnic Group OLS Logit

Electoral proximity 0.012 0.018


(0.002) (0.14)
Electoral competitiveness 0.63 0.643
(0.746) (0.425)
Proximity Competitiveness 0.010 0.091
(0.0034) (0.022)
R2 0.0746 0.0700
N 20,735 20,735
Notes:  ,  ,  indicate the 10%, 5%, and 1% significance levels, respectively. Standard errors are reported in parentheses. Regressions are controlled for
individual characteristics, country dummies and round dummies. Weights are 1/(number of observations of that country).

© 2017 The Authors. African Development Review © 2017 African Development Bank
558 T. Zeydanli

Table 12: Ordered probit model


Dependent variable Your living conditions Your condition compared to others

Electoral proximity 0.022 0.019


(0.002) (0.14)
Electoral competitiveness 1.67 0.463
(0.13) (0.025)
Proximity Competitiveness 0.015 0.011
(0.0034) (0.0022)
R2 0.0746 0.0700
N 20,735 20,735
Notes:  ,  ,  indicate the 10%, 5%, and 1% significance levels, respectively. Standard errors are reported in parentheses. Regressions are controlled for
individual characteristics, country dummies, and round dummies. Weights are 1/(number of observations of that country). Competitiveness variable is calculated
from vote margin as 1 (Vote Margin). Larger numbers indicate increasing competitiveness. Proximity is 1 abs(months to/from nearest election), so that
larger numbers imply increasing proximity.

Since 1990, the banning of ethnic parties has become the norm in SSA. In our dataset the ethnic banning countries are
Tanzania and Uganda. For instance, Tanzania has used the education system and redistribution of resources to develop a sense of
national as opposed to ethnic identity. The studies that show the impact of ethnic banning in parties conclude that these laws have
only marginally influenced the character of the political parties (Moroff, 2010). Ethnic banning may alter the origin of parties,
resulting in ethnic-free parties. This affects voting behaviour and subjective living conditions. Ethnic banning can also influence
the salience of ethnic identification. Tanzania has among the lowest degree of ethnic identity salience in one of the
Afrobarometer survey rounds, at just 3 per cent. Eifert et al. (2010) also show Tanzania’s outlier status. The presidential election
has little impact on the share of the population that identifies themselves in ethnic terms. In Figure 1, one can also observe that the
impact of the proximity of elections on subjective living conditions is less in Tanzania compared to other countries. Tanzania’s
situation is proof of the strength of ethnic identification in politics.18 Miguel (2004) examines the success of nation-building
policies in Tanzania, which have had a beneficial long-run impact on country’s political stability and economic development.

4.2 Ordered Probit Model


Our baseline model is re-estimated by utilizing ordered probit model since our dependent variables are ordinal. The results of the
ordered probit model (see Table 12) are in line with the OLS model. After controlling for individual characteristics, country, and
time dummies ‘your living conditions’ increases by 0.0173 standard deviation, ‘your living conditions compared to others’ by
0.014 standard deviation, and ‘mental health’ rises by 0.012 standard deviation when competitive elections are near. The
magnitudes are slightly higher than it is in the baseline regression. These results put us in a position that the OLS result does not
suffer from any model issue.

5. Concluding Remarks and Discussion


As Eifert et al. (2010) stated, the source of ethnic salience comes from political competition: in other words, proximity to
competitive elections increases the strength of ethnic attachments. A general consensus exists about the negative relationship
between economic development and social heterogeneity based on both cross-country regressions and individual country
studies. This paper brings a different angle to the discussion of election, ethnic identification, and individual perceived living
conditions. It explores these phenomena under the umbrella of living conditions by asking: ‘How do competitive elections affect
individual-level subjective living conditions?’ The results show that for every month closer a country is to a competitive election,
on average individual-level subjective living conditions has a 0.015 standard deviation increase.
These findings point to the background mechanism of this question: ‘Why do elections make people happier given that
competitive elections in this region have a very intense environment?’ The paper looks into five possible mechanisms:
winning the election, the effects of forthcoming election, the impact of public expenditure on education, health, and defense,
© 2017 The Authors. African Development Review © 2017 African Development Bank
Elections and Subjective Living Conditions 559

the environment of free and fair elections, and participation in politics. These mechanisms have a positive impact on the
subjective living conditions. Moreover, if individuals identify themselves ethnically higher, this is positively correlated with
individual-level subjective living conditions. As stated in Habyarimana et al. (2007), enforcing cooperation among
individuals, in this context the same ethnic group, would make policies more effective. Thereby, combination of these results
might enable us to draw a conclusion such that policies may be implemented in an ethnic group-level rather than a country-
level. These findings may be important for designing policies to increase social welfare in SSA. However, these positive
living conditions’ effects should be treated cautiously. Because these effects occur only when elections are proximate. In the
short run, these positive externalities might boost economic growth, but the long-run implications are ambiguous. Moreover,
there might be some events during competitive elections, which could possibly alter the living conditions, and these may
create some bias in the results.
Apart from the empirical findings, this paper has three good features for analysing hypotheses. It creates indices of living
conditions questions to aggregate same outcomes across domains such as time and comparison. The paper has grouped two
living conditions questions that evaluate living conditions regarding time and comparisons of respondents’ lives. A second
methodological contribution is to make use of repeated country-level observations with micro-individual survey data. Since the
data have been collected at multiple points in time for the same countries, it allows for variation in key parameters of interest such
as the proximity of the survey to the nearest election and the competitiveness of that contest. Moreover, living conditions of
individuals is affected mostly by the characteristics of the social and political environment in which he or she lives. Using the
feature of data, the paper employs a country-fixed effect model to overcome country-level characteristics. Thirdly, these results
are drawn from cross-national survey data rather than case studies and anecdotal evidence, which allows for generalized cross
settings and creates a much stronger position.

Notes
1. Afrobarometer surveys are conducted in 35 African countries and are repeated on a regular cycle. It measures social,
political, and economic atmosphere of countries. See Section 4.2 for more information.
2. The paper makes use of the countries whose data are present in at least three rounds of survey so as to gain variation in
election variables.
3. Number of observations in the regressions varies depending on the availability of the variable in rounds.
4. See Table 2 for detailed information of countries and survey years.
5. All regressions were also tested separately using the five living condition questions on the list. The results (not shown) are
parallel with the indices results. They are available upon request.
6. Results are robust even if dropping a single country in the dataset.
7. It can be three depending on the dependent variable.
8. See Table 1 for individual-level variables.
9. Since there is a robust result among dependent variables, this analysis is conducted with only the ‘your living conditions’
variable for the figure.
10. The results are robust to a non-parametric bootstrap of the standard errors using resampling at the country level, which is
theoretically more appropriate given the relatively small number of countries.
11. The personal identification question is derived from the specific question, which is available only in Rounds 1 and 2. Given
the availability of the dependent variable, it is possible to run the regression only for Round 2.
12. The same regressions have been run without election variables (not shown), which is in line with the previous one. The
results are available upon request.
13. See Table 7. Only ex-ante effects of the electoral cycle are represented. The results of ex-post effects are available upon
request.
© 2017 The Authors. African Development Review © 2017 African Development Bank
560 T. Zeydanli

14. Data is taken from World Development Indicators.


15. The exact wording of ‘discuss politics’ is ‘Here is a list of actions that people sometimes take as citizens. For each of these,
please tell me whether you, personally, have done any of these things during the past year. If not, would you do this if you
had the chance: Discussed politics with friends or neighbors?’ It is a three-point based scale; 0 refers to ‘No, would never do
this’; and 2 is ‘Yes, frequently’. It is available only in Round 3 and 4.
16. The formal question of ‘interested in public affairs’ is ‘How interested are you in public affairs?’ It is a four-point based
scale; 0 refers to ‘Not interested’; and 3 refers to ‘Very interested’. It is available in all rounds.
17. These variables are tested separately. Only the result of ‘interested in public affairs’ is presented.
18. However, subtracting these countries in the regression of ethnic salience and the impact of elections to the subjective living
conditions does not alter the results. The analyses based in Tanzania and Uganda cannot be conducted due to high
collinearity of election variables.

References
Addai, I. and J. Pokimica (2010), ‘Ethnicity and Economic Well-Being: The Case of Ghana’, Social Indicators Research, Vol.
99, No. 3, pp. 487–510.
Agyemang, C., C. Airhihenbuwa and A. Aikins (2013), Ethnicity: Theories, International Perspectives and Challenges, Nova
Science Pub Incorporated.
Ajayi, J. A. (2000), Tradition and Change in Africa, Africa World Press, Trenton, NJ and Asmara (Eritrea).
Alesina, A., A. Devleeschauwer, W. Easterly, S. Kurlat and R. Wacziarg (2003), ‘Fractionalization’, Journal of Economic
Growth, Vol. 8, No. 2, pp. 155–94.
Bates, R. H. (1983), ‘Modernization, Ethnic Competition, and the Rationality of Politics in Contemporary Africa’, in D.
Rothchild and V. A. Olunsorola (eds.), State versus Ethnic Claims: African Policy Dilemmas, Westview Press, Boulder, CO.
Bates, R. H. (2000), ‘Ethnicity and Development in Africa: A Reappraisal’, American Economic Review, Vol. 90, No. 2, pp.
131–34.
Block, S. A. (2002), ‘Political Business Cycles, Democratization, and Economic Reform: The Case of Africa’, Journal of
Development Economics, Vol. 67, No. 1, pp. 205–28.
Easterly, W. and R. Levine (1997), ‘Africa’s Growth Tragedy: Policies and Ethnic Divisions’, The Quarterly Journal of
Economics, Vol. 112, No. 4, pp. 1203–50.
Eifert, B., E. Miguel and D. Posner (2010), ‘Political Competition and Ethnic Identification in Africa’, American Journal of
Political Science, Vol. 54, No. 2, pp. 494–510.
Ellison, C. (1991), ‘Religious Involvement and Subjective Well-Being’, Journal of Health and Social Behavior, Vol. 32, No. 1,
pp. 80–99.
Ellison, C. G., J. D. Boardman, D. R. Williams and J. S. Jackson (2001), ‘Religious Involvement, Stress, and Mental Health:
Findings from the 1995 Detroit Area Study’, Social Forces, Vol. 80, No. 1, pp. 215–49.
Fearon, J. D. (2003), ‘Ethnic and Cultural Diversity by Country’, Journal of Economic Growth, Vol. 8, No. 2, pp. 195–222.
Ferrer-i-Carbonell, A. and P. Frijters (2004), ‘How Important is Methodology For the Estimates of the Determinants of
Happiness?’, Economic Journal, Vol. 114, No. 497, pp. 641–59.
Fishman, J. (1996), Ethnicity as Being, Doing and Knowing, in J. Hutchinson and A. Smith (eds.), Ethnicity, Oxford University
Press, Oxford.
Fosu, A., R. Bates and A. Hoeffler (2006), ‘Institutions, Governance and Economic Development in Africa: An Overview’,
Journal of African Economies, Vol. 15, No. 1, pp. 1–9.
© 2017 The Authors. African Development Review © 2017 African Development Bank
Elections and Subjective Living Conditions 561

Geertz, C. (1993), The Interpretation of Cultures, Fountain Press, London.


Habyarimana, J., M. Humphreys, D. Posner and J. Weinstein (2007), ‘Why Does Ethnic Diversity Undermine Public Goods
Provision?’, American Political Science Review, Vol. 101, No. 4, pp. 709–25.
Hausken, K. and M. Ncube (2014), ‘Determinants of Election Outcomes: New Evidence from Africa’, African Development
Review, Vol. 26, No. 4, pp. 610–30.
Horowitz, D. L. (1985), Ethnic Groups in Conflict, University of California Press, Berkeley and Los Angeles.
Kahneman, D., E. Diener and N. Schwarz (1999), Well-Being: The Foundation of Hedonic Psychology, Russell Sage
Foundation, New York.
Kimenyi, M. S. (2006), ‘Ethnicity, Governance and the Provision of Public Goods’, Journal of African Economies, Vol. 15, No.
1, pp. 62–99.
Kling, J. R., J. B. Liebman and L. F. Katz (2007), ‘Experimental Analysis of Neighborhood Effects’, Econometrica, Vol. 75, No.
1, pp. 83–119.
Knack, S. and P. Keefer (1997), ‘Does Social Capital Have an Economic Payoff? A Cross-country Investigation’, The Quarterly
Journal of Economics, Vol. 112, No. 4, pp. 1251–88.
Leigh, A. (2006), ‘Trust, Inequality and Ethnic Heterogeneity’, The Economic Record, Vol. 82, No. 258, pp. 268–80.
Mauro, P. (1995), ‘Corruption and Growth’, The Quarterly Journal of Economics, Vol. 110, No. 3, pp. 681–712.
Miguel, E. (2004), ‘Tribe or Nation? Nation Building and Public Goods in Kenya versus Tanzania’, World Politics, Vol. 56, pp.
328–62.
Mijiyawa, A. G. (2015), ‘What Drives Foreign Direct Investment in Africa? An Empirical Investigation with Panel Data’,
African Development Review, Vol. 27, No. 4, pp. 392–402.
Montalvo, J. G. and M. Reynal-Querol (2005), ‘Ethnic Diversity and Economic Development’, Journal of Development
Economics, Vol. 76, No. 2, pp. 293–323.
Moroff, A. (2010), ‘Ethnic Party Bans in East Africa from a Comparative Perspective’, GIGA Working Papers, 129.
Neff, D. (2007), ‘Subjective Well-Being, Poverty and Ethnicity in South Africa: Insights from an Exploratory Analysis’, Social
Indicators Research, Vol. 80, No. 2, pp. 313–41.
Odedokun, M. (1993), ‘Factors Responsible for the Poor Economic Growth Performance of Africa in the 1970s and 1980s:
Cross-sectional Evidence from 42 Countries’, African Development Review, Vol. 5, No. 1, pp. 32–61.
Pokimica, J., I. Addai and B. Takyi (2012), ‘Religion and Subjective Well-Being in Ghana’, Social Indicators Research, Vol.
106, No. 1, pp. 61–79.
Posner, D. (2004), ‘Measuring Ethnic Fractionalization in Africa’, American Journal of Political Science, Vol. 48, No. 4, pp.
849–63.
Roeder, P. G. (2001), Ethnolinguistic Fractionalization (elf) Indices, 1961 and 1985, University of California, San Diego.
Stutzer, A. and B. Frey (2006), ‘Political Participation and Procedural Utility: An Empirical Study’, European Journal of
Political Research, Vol. 45, pp. 391–418.
Williams, K. (2016), ‘Remittances and Financial Development: Evidence from Sub-Saharan Africa’, African Development
Review, Vol. 28, No. 3, pp. 357–67.
Young, C. (1976), The Politics of Cultural Pluralism, University of Wisconsin Press, Madison.

© 2017 The Authors. African Development Review © 2017 African Development Bank
African Development Review, Vol. 29, No. 4, 2017, 562–574

Oil and the Naira: A Markov Switching Perspective

Idowu Oluwasayo Ayodeji

Abstract: Using a Markov switching regression model, the study investigated the relationship between oil price shocks and
exchange rates movements in an oil-exporting economy. On the strength of theoretical and empirical evidence, Nigerian naira
exchange rates to US dollars were decomposed into two segmented time trends of depreciation (state 1) and appreciation
(state 2). Empirical results provided evidence that exchange rate returns exhibited an asymmetric response to oil price shocks
during depreciation and appreciation. We observed that, while the system is in depreciation, nominal oil prices had a positive
influence on nominal USD/NGN, whereas nominal oil price shocks had no significant effect on nominal exchange rate
movements during appreciation. In addition, real effective exchange rates appreciate with decreasing real oil prices but its
depreciation does not depend significantly on real oil prices. Given that a decrease in nominal oil prices induces rapid
naira depreciation and topples exchange rates stability, the study suggested that the government works towards a sustainable
diversification of the economy to avoid continued over-dependence on oil exports. Also, to rigorously pursue a sustainable fiscal
policy regime in order to boost local productions, encourage non-oil commodities exportation, and consequently increase
avenues for influx of foreign exchange.

1. Introduction
Oil-exporting economies may experience exchange rate appreciation when oil prices rise and depreciation when they fall. This
was the submission of early influential papers such as Golub (1983) and Corden (1984). Results from subsequent studies,
however, are divided on this submission: While some agreed that an increase in oil price leads to appreciation of domestic
currency (e.g. Amano and van Noden, 1998; Chen and Chen, 2007; Beckmann and Czudaj, 2012), some others have reported a
negative relationship (e.g. Huang and Guo, 2007; Ghosh, 2011). Some even claimed that no relationship exists between the two
variables (e.g. Akram and Holter, 1996; Buetzer et al., 2012). Figure 1 provides support for the latter group: the scatter plot and
the fitted regression line indicate an insignificant relationship between naira-dollar exchange rates and Brent prices in nominal
and real terms; meaning that oil price changes do not in any significant way contribute to the movements in exchange rates of the
US dollar to Nigerian naira.
However, is it intuitively realistic that oil prices do not contribute to naira-dollar movements in Nigeria? It is a fact that Nigeria
is a mono-product (crude oil) economy. Her annual federal budget is designed based on projections made on global oil prices and
the quantity of oil sold. Also, it is known that oil is denominated in US dollars. Hence, it is expected that any increase or decrease
in the global price of crude oil would affect the supply of foreign exchange, value of naira (relative to dollar), the Nigerian foreign
reserve, excess crude oil account, inflation rates, level of production, rates of unemployment and the economy. Thus, from the
theoretical point of view, a study of the nexus between oil prices and USD/NGN rates is key to exchange rates and economic
stability. Judging from the current experience in Nigeria, it could also help to avert potential spike in inflation and unemployment
rates; and more importantly, currency crises and economic recession.
From the empirical perspective, the plots of NGN/USD exchange rates and Brent crude oil prices displayed in Figure 2 show
clearly that the linear regression modelling depicted in Figure 1 does not reflect the true relationship between oil prices and
exchange rates in Nigeria in either nominal or real terms. It is evident that the two major depreciation episodes recorded in
2008M11–2012M01 and 2014M10–2015M03 in Figure 2a were preceded by some substantial fall in oil prices in 2008M07–
2008M12 and 2014M06–2015M03. Similarly, the major depreciation periods of 2008M11–2009M01 and 2014M10–2015M02
in Figure 2b were preceded by some substantial fall in oil prices in 2008M07–2008M12 and 2014M06–2015M01. However, the


Department of Mathematics, Obafemi Awolowo University, 220005, Ile-Ife, Nigeria; e-mail: idowu.sayo@yahoo.com, olusayo11@gmail.com
© 2017 The Author. African Development Review © 2017 African Development Bank. Published by Blackwell Publishing Ltd,
9600 Garsington Road, Oxford OX4 2DQ, UK and 350 Main Street, Malden, MA 02148, USA. 562
Oil and the Naira 563

Figure 1: Scatter plots of log-differenced USD/NGN and Brent for 2004M01–2016M05 in nominal and
real terms

same cannot be said about appreciation periods in both figures. It appears that an increase in oil prices did not result in any
substantial appreciation of naira against the dollar either in the real or nominal term, whereas the value of naira seems to
depreciate in the nominal and the real as oil price reduces. It is noteworthy that real effective exchange rate (REER) generally
appreciates with decreasing oil prices between 2011M06 and 2014M10, after which it briefly went into depreciation for some
four months, then it recovered. In other words, there seems to be some asymmetry in the response of exchange rates to the
changes in oil prices when naira is in appreciation and depreciation.
Now, why does asymmetry matter in oil prices-exchange rates relationship? From the foregoing discussion, it is obvious that
in order to obtain a realistic and accurate empirical relationship between exchange rates and oil price movements, a researcher
must take into cognizance the dual nature of the exchange rates; otherwise, one may obtain wrong estimates, and consequently,
misleading inferences.
Previous empirical studies investigating this relationship in the matured markets have employed some form of linear models
such as the vector error correction (VEC) (e.g. Indjehagopian et al., 2000; Zalduendo, 2006), cointegration tests (e.g. Amano and
van Noden, 1998; Coudert et al., 2008), vector autoregressions (VAR) (e.g. Huang and Guo, 2007; Zhang et al., 2008), and an
ARCH-type model specification which assumed symmetric response from exchange rates to oil prices in the mean equation

Figure 2: Plots of USD/NGN and Brent for 2004M01–2016M05 in nominal and real terms

© 2017 The Author. African Development Review © 2017 African Development Bank
564 I. O. Ayodeji

models (e.g. Narayan et al., 2008; Cifarelli and Paladino, 2010). All these models assumed (though unrealistically) that the
response from exchange rates to changes in oil prices is symmetric.1
To show the significance of oil and exchange rates across regions, there have also been studies in Africa on exchange rates
(Igwe and Ogunleye, 2014), its policies (Eregha et al., 2016) and its volatility (Ozturk and Kalyoncu, 2009); also on
the oil wealth and its effect on governance (Anyanwu and Erhijakpor, 2014). In particular, there have been a few studies on the
oil-exchange rate link in Africa, such as Turhan et al. (1991) and Pershin et al. (2015). These studies, like those on matured
markets, also ignored the dual nature of exchange rates while measuring the contributions of oil price shocks to exchange rate
movements.
However, there is evidence for asymmetries in exchange rates that challenge the assumptions of the conventional linear
models. Enders and Dibooglu (2001), for instance, documented several reasons that exchange rates may respond asymmetrically
to oil price shocks. Hamilton (1989) proposed the Markov switching model as an alternative to linear models. The Markov
switching model (MSM) may be used to characterize exchange rates into different states that are consistent with the reality.
MSM incorporates the Markov process into an autoregressive model such that the mean exchange rate may vary across
unobserved states that evolve according to an nth (usually, 1st) order Markov transition process. Engel and Hamilton (1990),
applied a two-state MSM to divide exchange rates into segmented time trends. They found convincing evidence to support that
exchange rates go through two distinct states of appreciation and depreciation, exhibiting asymmetric features in those states.
Some other studies have also confirmed the appropriateness of Markov switching specification in modelling exchange rates. See,
for instance, Kaminsky (1993), Evans and Lewis (1995), Dewachter (1997). All these provide motivation and justification for
modelling the oil–exchange rates link with a state-dependent process.
This study investigates whether oil price changes possess the relevant dynamics to drive the dichotomous exchange rate
system; or at least explain exchange rate movements at certain periods. Accordingly, we incorporate oil price shocks in the
conditional mean equation of the Markov switching process. We employed both the nominal and real effective exchange rates
together with the nominal and real prices of oil. This enables us to effectively scrutinize the relationship between the two
macroeconomic variables with respect to the Nigerian economy alone, and also other economies. As a direct consequence, it is
possible to draw fairly general inferences that are not limited to the Nigerian economy. In addition to measuring the impact of oil
price changes on exchange rates, our specification also identifies the periods that are mainly explained by oil price movements.
Information on the relationship between oil prices and exchange rates is very important for a country like Nigeria, as it would
enhance the exchange rates determination process, which is very significant if naira would indeed be stabilized (Emenike, 2016).
Thus, results from this study are very useful to monetary authorities, since it is always their goal to achieve exchange rate
stability; traders, since most commodities in Nigeria are denominated in dollars; and researchers alike. It also has important
implications for government revenue, rates of unemployment and inflation, economic growth, import structure, and other
economic factors.
The overview of the study is as follows: Section 2 describes briefly some policy issues in the Nigerian foreign exchange
market. Section 3 discusses the data and the Markov switching model to be employed. Section 4 presents empirical results while
Section 5 concludes.

2. Overview of the Nigerian Foreign Exchange Market


According to the Central Bank of Nigeria (CBN), the main objectives of exchange rate policy in Nigeria are to (i) preserve the
value of naira; (ii) maintain a favourable external reserves position; and (iii) ensure external balance with due regard to internal
balance and the main goal of macroeconomic stability. These objectives have guided the formulation of policies since the
inception of the Nigeria Foreign Exchange Market (FEM) to date.
Following enhanced official foreign exchange receipts from crude oil export in the 1970s, the Nigerian exchange rate market
experienced a boom; then the need arose to manage effectively the accrued resources in order to avoid shortage. Subsequently,
the Second-tier Foreign Exchange Market (SFEM) was introduced in September 1986. During this period, until 1993, a dollar
exchanged for 17. Shortly thereafter, oil prices fell below $20 per barrel. Consequently, naira depreciated so rapidly and
the economy got worse. There was not enough dollars in the Nigerian coffer to satisfy the daily need of end-users,
hence the mainstreaming of the parallel market, otherwise known as the ‘black’ market. The black market rate ranged from 80
to 90 to $1 while the official market maintained the illusion of 22 per $1. In 1995, CBN improved its services by introducing
the Autonomous Foreign Exchange Market (AFEM) which allowed it to sell foreign exchange to end-users directly through
© 2017 The Author. African Development Review © 2017 African Development Bank
Oil and the Naira 565

certain authorized dealers. The Inter-bank Foreign Exchange Market (IFEM) was introduced, so also was the Dutch Auction
System (DAS). We note that in addition to low global oil prices, Nigeria also had some $33 trillion foreign debt to settle at the
time. Thus, due to the pressure on naira, the IFEM had to be suspended for 6 months.
From late 2003 to the middle of 2008, oil prices rose steadily in the global market from US$30 to US$140 per barrel.
Consequently, Nigerian foreign reserves received a huge boost. According to the dailies, Nigeria had, in addition to the
foreign reserves, the Excess Crude Account which contained more than $20 billion sometime in 2008. Even CBN
reported that naira gained up to 20 per cent against the dollar at this period without conscious effort from any quarters to
strengthen it.
Sadly in late 2008, the era of the oil boom ended; oil prices had fallen to $50 per barrel at the end of that year. As expected,
CBN responded in the usual manner: Governor Soludo banned IFEM and DAS. By 2009, the value of naira had gone down to
147 to $1 from 127 in 2004. Soon enough, however, oil prices began to recover. Then came the Sanusi Lamido Sanusi era in
June 2009: he restored IFEM that Soludo had previously banned. It is noteworthy that oil prices remained high all through from
2009 until February 2014. In addition to the oil boom, Sanusi relaxed the one-year restriction on foreign investors who wanted to
buy government bonds, then dollars came pouring in like heavy rain. The Nigerian market was kept liquid, and consequently it
was included in the famous JP Morgan Index.
Emefiele took office after Sanusi. Global Brent oil prices again plunged to $62 at the close of 2014 from $112 it was in June of
the same year (Breul, 2015). As did Soludo, Emefiele also closed DAS on 18 February 2015 and quoted an exchange rate of
197–199 to $1. This continued until June 2016 when CBN was forced to revert back to the market-driven foreign exchange
system. The Nigerian naira immediately plunged to 261 to a dollar overnight.

3. Materials and Methods


Datasets consist of monthly exchange rates and Brent crude oil prices over the period January 2004–May 2016. Brent blend was
chosen because it is the benchmark for Nigerian bonny light. Nominal exchange rates were obtained from the website of the Central
Bank of Nigeria while nominal Brent prices were downloaded from the website of the Energy Information Administration. Real oil
prices were the nominal prices deflated by US Consumer Price Index (CPI). Real effective exchange rates (REER) were calculated
as the trade-weighted average of bilateral real exchange rates of the domestic currency against its trade partners. The detail of this
variable and its methodology are indicated in the open data portal of the National Bureau of Statistics (NBS).
   
Denote yt ¼ log et1et
and xt ¼ log ct1
ct
the returns on exchange rates et and crude oil prices ct , respectively. Consider the
following system
9
yt ¼ mSt þ et ; >
>
  =
et  i:i:d: N 0; s 2St ; : ð1Þ
>
>
;
S t ¼ 1; 2;    ; k; t ¼ 12;    ; T

System (1) represents the simplest model with switching features. For k ¼ 2 for instance, the intercept m takes two different
values representing the expectations m in the two different states, so also the variance V ðet Þ ¼ s 2 . S t is the unobservable
Markov-switching variable which evolves according to transition probabilities P:
!
P11 P12
P¼ ; ð2Þ
P21 P22

where Pij ¼ PrðS t ¼ jjS t1 ¼ iÞ and

X
2
Pij ¼ 1; j ¼ 1; 2; and for all i ð3Þ
j¼1

© 2017 The Author. African Development Review © 2017 African Development Bank
566 I. O. Ayodeji

System (1) involving covariate xt is expressed as


9
yt ¼ mSt þ bSt xt þ et ; >
>
  =
et  i:i:d: N 0; s 2St ; ð4Þ
>
>
;
S t ¼ 1; 2;    ; k; t ¼ 1, 2;    ; T

The log likelihood function (l) corresponding to Model (4) is


0 1
!2
XT
B 1 yt  mSt  bSt xt C
l¼ log@qffiffiffiffiffiffiffiffiffiffiffiffi exp  pffiffiffi A ð5Þ
t¼1 2ps 2St 2s St

If the values of S t are known, estimation of the relevant parameters by maximum likelihood estimation is direct. However, in
Markov switching where the states S t are not known, the log likelihood equation may not be easily maximized. Hence, following
Hamilton (1994), we modify Equation (5) as follows:
0 1
!2
X T X 2
B 1 y  mS  bSt xt C
l¼ log @qffiffiffiffiffiffiffiffiffiffiffiffi exp  t pffiffitffi PðS t jqt1 ÞA ð6Þ
t¼1 S t ¼1
2
2ps St 2 s S t

where qt1 is the information available up to time ðt  1Þ. Hamilton (1994) described in detail a suitable process, popularly
referred to as ‘the Hamilton filter’ which can be used to compute the estimates of the probabilities PðS t Þ for each state.
^ ðS t ¼ jjqT Þ of the smoothing probabilities for the full sample T may be obtained through a smoothing
Further, the estimates P
iterative process (Kim, 1994). In the same vein, the expected duration E½D of a regime is calculated as (Kim and Nelson, 1999)

X
1
1
E½D ¼ jPðD ¼ jÞ  ; ð7Þ
j¼1
1  Pjj

where D is the duration of state j. More details of the estimation process can be found in Hamilton (1994) and Engel and Hamilton
(1990), among others.
As a closing remark, we note that the number of states can be extended from 2 to k ð> 2Þ. An example was demonstrated in
Ayodeji (2016) in the case of Nigeria when MSM for k ¼ 3 states of appreciation, depreciation, and stagnation were developed.
There is, however, no hard and fast rules in the choice of optimal number of states. The choice of k generally depends on the
exchange rates to be modelled, objectives of study, computational demand and mathematical tractability (Lee and Chen, 2006).
Here, when System (1) is extended to include covariates xt as in Model (4), the computational complexity increases. Besides, our
objective of study is to determine the movements of exchange rates in response to changes in the oil prices during appreciation
and depreciation periods. Clearly, the specification k ¼ 2 suffices to achieve this objective of study. Thus, we could have two
broad states of appreciation and depreciation, and study the intended relationship under them, without loss of generality.

4. Results and Discussion


 
 for linear regression, yt ¼ m þ bxt þ et ; et  N 0; s and the
2
Tables 1 and 2 contain the maximum likelihood estimates
switching regression, yt ¼ mSt þ bSt xt þ et ; et  N 0; s 2St results for both nominal and real terms.

4.1 Linear Regression Results


From the linear regression results, we observed, on average, that naira fell by 0.26 per cent in nominal terms while it rose by
0.40 per cent in the real. It should be noted that the real effective exchange rates is a measure of international competitiveness
© 2017 The Author. African Development Review © 2017 African Development Bank
Oil and the Naira 567

Table 1: Table of maximum likelihood estimates in the nominal term


Switching Regression
Parameter Linear Regression 1 2

m 0.00257 0.006266 4.08  105


(0.00159) (0.003228) (2.12  105)
b 0.02553 0.080438 4.23  105
(0.01702) (0.040034) (0.000185 )
s 0.01934 0.02644672 0.00017226
NA (0.0022504) (1.59  105)
Notes: Standard errors in parentheses.  not significant at a ¼ 0:05. NA ¼ not applicable.

(Loretan, 2005). Thus it can be said that, on average, the Nigerian naira experienced depreciation locally but appreciation
internationally from the period January 2004 to May 2016.
Whereas two distinct states of depreciation and appreciation could be observed in Figure 2, the linear regression model was
only able to identify one of them — depreciation in nominal and appreciation in the real. A very important inference is that when
exchange rate modelling is carried out without regard for all the possible states, the linear regression model only captures the
state with the longer swing. However, inferences from other possible states are also required to understand the dynamics of the
exchange rate. Policymakers, for instance, must have access to all available information to guide them in making the right
decisions while the currency value is either falling or rising.
The aggregate relationship between oil prices and the exchange rates, as depicted by b, is insignificant in the nominal and real
terms; meaning that change in oil prices, on average, does not affect exchange rate movements in any form. This position clearly
disagrees with Figure 2, in which the naira depreciates with decreasing oil prices and (seems to) appreciate with increasing oil
prices in the nominal. Also in the case of the real, the same pattern as in the nominal could be observed up until 2011 after which
REER, more or less, began to rise though real oil price was falling. It is noteworthy that the two measures of exchange rates differ
in the direction of the aggregate relationship: While b is positive in the nominal, it is negative in the real. In summary, from
January 2004 to May 2016, nominal exchange rates decreases, on average, with decreasing nominal oil prices while real
effective exchange rates increases, on average, with decreasing real oil prices.

4.2 Switching Regression Results


It is evident that the switching regression model gave values that are consistent with the exchange rate movements plotted in
Figure 2. Two states were identified, the depreciation (state 1) and appreciation (state 2). In the nominal, State 1 is associated with
a fall of 0.6 per cent on average, while state 2 corresponds to a monthly average appreciation rate of 0.004 per cent. In real
effective rates, naira (on average) fell by 2.01 per cent and rose by 0.63 per cent. Considering s, state 1 was more volatile than 2 in

Table 2: Table of maximum likelihood estimates in the real term


Switching Regression
Parameter Linear Regression 1 2

m 0.00396 0.02013 0.00626


(0.00180) (0.01915 ) (0.00134)
b 0.01050 0.08207 0.04740
(0.01702 ) (0.11857 ) (0.01600)
s 0.02192 0.05303 0.01420
NA (0.01381) (0.00102)
Notes: Standard errors in parentheses.  not significant at a ¼ 0:05. NA ¼ not applicable.

© 2017 The Author. African Development Review © 2017 African Development Bank
568 I. O. Ayodeji

both real and nominal terms. This implies that whether in real or nominal, Nigerian naira with respect to dollars, experienced
more volatility while in depreciation than when in appreciation.
For nominal, the estimates of b in the depreciation and appreciation states were 0:08ðp ¼ 0:04Þ and 4:23  105 ðp ¼ 0:82Þ,
respectively. The fact that b had a positive sign means that as oil prices increase, more US dollars would be required to buy 1,
and vice versa, while the system is in depreciation. This implies that positive changes in oil prices would lead to higher
depreciation for dollar and less depreciation for naira. However, whether oil prices are increasing or decreasing does not impact
on exchange rates while the systemis in appreciation. 
We note, in passing, that b ^ 2 ¼ 0:08; 4:23  105 is consistent with Figure 2a: while a nominal oil price decrease
^ 1; b
from $132 to $40 between 2008M06 and 2008M12 led to a corresponding 0.19 per cent depreciation in nominal NGN/USD
in 2008M11–2009M09, an increase in nominal oil prices from $40 to $123 in 2008M12–2011M04 did not yield
an appreciation of corresponding magnitude; NGN/USD only fluctuated between 0.0063 and 0.0066 from 2009M09 to
2012M03.
Taking into account Nigeria’s trading partners, the relationship between naira/dollar and oil price shocks is regime dependent:
it is positive in depreciation but negative in appreciation. In particular, b1 ¼ 0:082ðp ¼ 0:12Þ suggests that REER depreciates
less in value when oil prices rise. In the same vein, b2 ¼ 4:74  102 ðp ¼ 0:016Þ implies that while the system is in
appreciation, naira, relative to other relevant currencies, increases with decreasing oil prices. It is easily deduced that the
magnitude and direction of the relationship are similar for both the nominal and real terms while the exchange rate system is in
depreciation, though the relationship is insignificant in the real. Watching Figure 2b closely, we observe that the prevailing state
between 2004M01 and 2011M04 in REER is depreciation. We observe further that REER generally decreased with decreasing
real oil prices, in this prevailing state. In the same vein, it is obvious that REER generally appreciates in the periods following
2011M04. It is also obvious that the period between 2011M06 and 2016M05 is a clear departure from what was obtained earlier:
REER increased with decreasing   real oil prices. Hence,
 the nexus between real oil price and real effective exchange rates is
state-dependent, and b ^ 2 ¼ 0:082; 4:74  102 truly reflected Figure 2b more than the b estimate provided earlier by the
^ 1; b
linear model.
A reconstruction of the depreciation episodes for nominal rates and real effective rates are presented in the bottom panels of
Figures 3a and 3b, respectively. The top panels of the figures displayed their corresponding exchange rates returns. As is usually
the practice, we reckoned that switching has occurred when the smoothing probability PðS T ¼ jjqT Þ > 0:5. It is easily seen that
the Markov switching model augmented with oil price shocks provided a matching description of the movements in exchange
rates in both nominal and real terms. In the nominal, the depreciation episodes (i.e. the peaks) 2008M11–2012M10 and
2014M10–2015M02 were clearly identified. Recall that earlier in Section 2, we documented that global oil prices fell twice
between 2008 and 2014; precisely late 2008 and late 2014. Notice that those periods of oil price decrease correspond to the peaks
found in Figure 3a. These, in addition to b estimates obtained earlier, provide strong evidence that nominal oil price shocks drive
the dichotomous nominal naira exchange rates, hence should be included in the nominal exchange rates determination. In the

Figure 3: Exchange rate returns and smoothed probability plots of USD/NGN for 2004M01–2016M05 in
nominal and real terms

© 2017 The Author. African Development Review © 2017 African Development Bank
Oil and the Naira 569

same vein, the periods of ‘appreciation’ (the troughs) were also well extracted in Figure 3a. These were mostly the times the naira
was pegged against dollar. It can be inferred that there was more or less no natural appreciation of naira against the dollar from
2004 to 2016 despite the fact that the oil price increased continuously from 2009 to late 2014.
Similarly, in the real smoothing plot, the peaks were also well identified as 2005M06–2005M07, 2008M12–2009M01 and
2014M11–2015M03. We note that in addition to the documented periods (late) 2008 and 2014, the real smoothing plot captured
the incidence of 2005 when oil sold globally for as low as $50 for a long time.
   
Lastly, the estimated transition probability matrices for the nominal P ^ n and the real P ^ r terms were P ^n ¼
! !
0:888156 0:111844 0:527662 0:472338
^
and P r ¼ . First and foremost, both estimated transition probabilities
0:110663 0:889337 0:036947 0:963053
indicate that (i) none of the two regimes is permanent, since all the estimated transition probabilities are less than one; and (ii)
REER had more appreciation episodes than the nominal exchange rates since 0:472 > 0:112. In addition, two other important points
can be inferred from the matrix P ^ n : (i) The nominal exchange rate system is almost equally likely to remain in either states;
^ 11  P
P ^ 22  0:89: This is true since aggregate duration of the peaks and those of the troughs depicted in bottom panel of Figure 3a are
almost equivalent. Accordingly, the estimated expected length of state 1 and 2 are approximately nine months. (ii) Movement
from state 1 to 2, and 2 to 1 are less likely, only a little chance of 0.112; whereas, with P ^ 22  0:96 in the P ^ r , REER has a
higher likelihood of remaining in the appreciation state than in the depreciation. This can be deduced from the smoothing plot
in Figure 3b since the longest duration of depreciation was only 4 months, that is, from 2014M11 to 2015M02 compared to
that of appreciation. Consequently, the movement from depreciation to appreciation is more likely with P ^ 12 ¼ 0:47 than the
^
converse, P 21 ¼ 0:0369.

4.3 Hypothesis Testing

In line with our objectives of study, the following research questions must be answered: (i) Was the Markov switching model
adequate for modelling the relation between oil price shocks and exchange rates movements in nominal and real terms? If this
were true, then there should be some significant changes in the values of m; s and b from regime to regime in both terms. For this
reason, we tested the following null hypotheses:

H 0m : m1 ¼ m2 ;
H 0b : b1 ¼ b2 ; ð8Þ
H 0s : s 1 ¼ s 2 :

Since the parameter vector u ¼ ðm; b; s Þ and the transition probabilities were all identified under the null hypotheses (8), we
may therefore test using the standard theory of distribution. The first segment of Tables 3 and 4 report Wald statistic (W) values
for the nominal and real terms. In the case of the nominal, we observed that all statistic values were significant at a ¼ 0:05
except, of course, for W m which was not too convincing. For the avoidance of doubts, the joint hypothesis

H 0 : m1 ¼ m2 ; b1 ¼ b2 ; s 1 ¼ s 2 ð9Þ

was also tested. The corresponding statistic W was highly significant with p ¼ 0:00.
Following Ayodeji (2016), an alternative test of MSM adequacy may be achieved by testing the following null hypothesis

H 0p : P11 þ P22 ¼ 1; ð10Þ

noting that if no switching


! took place, then P11 þ P22 ¼ 1 and the transition matrix in Equation (2) reduces to
P11 1  P11
P¼ . The appropriate Wald statistic is
P11 1  P11
© 2017 The Author. African Development Review © 2017 African Development Bank
570 I. O. Ayodeji

Table 3: Hypotheses testing and model diagnostics — nominal


Segment Statistic Linear regression Switching regression

1 Wm NA 3.817511
(0.0507)
Wb NA 4.032747
(0.0446)
Ws NA 1644.785
(0.0000)
W NA 1652.427
(0.0000)
Wp NA 24.2964
(4.29  107)
2 LogL 374.95880 654.1552
AIC 5.03998 8.731826
BIC 4.99948 8.569815
HIC 5.02353 8.666002
3 Qð10Þ 10.380 1.4470
(0.408) (0.999)
Q2 ð10Þ 1.4310 0.1472
(0.999) (1.000)
Notes: p-values in parentheses. NA ¼ not applicable. Where AIC ¼ 2logL þ 2p; BIC ¼ 2logL þ plogT and HIC ¼ 2logL þ 2plogðlogT Þ; L is the
likelihood function defined earlier in Equation (6) and p represents the number of parameters included in the model.

Table 4: Hypotheses testing and model diagnostics — real


Segment Statistic Linear regression Switching regression

1 Wm NA 1.880474
(0.1703)
Wb NA 1.150189
(0.2835)
Ws NA 26.34564
(0.0000)
W NA 29.66608
(0.0000)
Wp NA 0.495099
(0.443)
2 LogL 356.3934 390.8641
AIC 4.789100 5.173839
BIC 4.748597 5.011827
HIC 4.772643 5.108014
3 Q(10) 17.389 14.790
(0.066) (0.140)
Q2(10) 23.101 3.3473
(0.010) (0.972)
Notes: p-values in parentheses. NA ¼ not applicable. Where AIC ¼ 2logL þ 2p; BIC ¼ 2logL þ plogT and HIC ¼ 2logL þ 2plogðlogT Þ; L is the
likelihood function defined earlier in Equation (6) and p represents the number of parameters included in the model.

© 2017 The Author. African Development Review © 2017 African Development Bank
Oil and the Naira 571

P11 þ P22  1
Wp ¼  x2ð1Þ : ð11Þ
V ar½P11  þ V ar½P22  þ 2Cov½P11 ; P22 

As can be seen in Table 3, W p ¼ 24:2964 is highly significant with p ¼ 0:00000043: Therefore, based on these statistics,
we reject the null hypothesis of equality of the tested parameters in the two regimes and conclude that the Markov switching
model employed was adequate for modelling the relation between nominal oil price shocks and nominal exchange rates
movements.
In the case of REER, W m and W s were not significant at a ¼ 0:05 since their p-values were p ¼ 0:17 and p ¼ 0:28,
respectively. W p was also not significant at p ¼ 0:495. Of course, this result could be explained using the smoothing probability
plot 3b. We observed that the switches from appreciation to depreciation occurred only thrice within the period of study. We
observed further that the maximum duration of those switches was only 5 months, that is, from 2014M11 to 2015M03. More
evidently, the probability for the REER system to remain in appreciation was computed as P ^ 22 ¼ 0:96  1. Also, the expected
duration of appreciation was 27 months compared to depreciation which was only 2 months. Therefore, in line with Klaassen
(2005), we conclude that the period 2004M01–2016M05 gave too few observations to differentiate the few depreciation periods
from appreciation.
(ii) Did exchange rate return exhibit some asymmetries in its response to shocks? If this were so, then b1 6¼ b2 statistically. So
we test the hypothesis

H 0 : b1 ¼ b2 : ð12Þ

Again, for the nominal series, the Wald statistic W b ¼ 4:03ðp ¼ 0:04Þ showed that the response to shocks during depreciation
was different from that of appreciation. As a matter of fact, the test of H 0 : b1 ¼ 0 yielded W ¼ 4:04ðp ¼ 0:0445Þ while
H 0 : b2 ¼ 0 yielded W ¼ 0:052ðp ¼ 0:819Þ (see Table 1). Thus we can conclude that while nominal oil price shocks had no
significant effect on nominal exchange rate movements during appreciation, it significantly drives nominal exchange rate
movements when the system is in depreciation. Negative changes in nominal oil prices would lead to more depreciation; positive
changes in nominal oil prices reduce the rate of depreciation.
In contrast to the nominal series, the Wald statistic for the real terms W b ¼ 1:50ðp ¼ 0:28Þ is insignificant. Thus the seeming
regime-dependent relationship reported earlier in b estimates does not hold significantly.

4.4 Specification Testing


More confirmation on model adequacy can be achieved by checking the status of the specified model. This section addresses
concerns on the specification of the Markov switching model. The section is divided into two. The first part compares
specification measures from the linear and switching models. These include the log likelihood function, Akaike information
criteria (AIC), Bayesian information criteria (BIC) and the Hannan-Quinn information criteria (HIC); while the second part
checks model diagnostics. Results are placed in the second and third segments of Tables 3 and 4.
Usually in practice, the model with the higher log likelihood value and lower information criteria values is preferred. It is
evident that the switching regression model outperformed its linear counterpart in both nominal and real series with higher
likelihood values and lower AIC, BIC and HIC values.
In the same vein, misspecification of time series models has important implications for financial returns (Nelson,
1991). Accordingly, we tested for serial correlation in the standardized residuals and excess heteroscedasticity in the
squared standardized residuals at lags one through ten. The third segment of Tables 3 and 4 reported the Ljung-Box Q-
statistic with 10 lags (denoted Qð10Þ and Q2 ð10Þ), and their associated p-values. Though at a ¼ 0:05, evidence of serial
correlation was not present in the residuals of the linear models corresponding to the nominal and the real terms, the
Markov switching models performed better with higher probability values 0.99 and 0.14; respectively.
In addition, Table 4 revealed that the real effective exchange rates remained strongly heteroscedastic ðp ¼ 0:01Þ after linear
modelling. However, the excess heteroscedasticity was removed at p ¼ 0:97 after the Markov switching model was applied.
Indicating that the exchange rates under the switching models did not exhibit any more serial correlation nor ARCH effects;
giving justification for the specified models.

© 2017 The Author. African Development Review © 2017 African Development Bank
572 I. O. Ayodeji

4.5 A ‘Post-Mortem’ Analysis


Throughout Sections 4.1–4.4, the exchange rates were taken as the dependent variable, the oil prices being the predictor. However,
some strand of studies opined that causality may also run from exchange rates to oil prices. An example is Akram (2009) who
explained that since oil is denominated in dollars, an appreciation of the local currency against the dollar would lower the price of oil
measured in terms of the domestic currency, which increases demand and may result in a general rise of oil prices. On the basis of
this, we ran the usual Granger causality test. We also experimented with specifying Brent prices as the dependent variable in a
Markov switching model, while exchange rates were allowed to predict. The results are presented in what follows.
A ‘post-mortem’ analysis using Granger causality (results not presented here but are available on request) revealed that the
causality pattern is unidirectional, running from Brent prices to USD/NGN rates. We also experimented with the specification
9
xt ¼ mSt þ aSt yt þ et ; >
>
  =
et  i:i:d: N 0; s 2St ; ð13Þ
>
>
;
S t ¼ 1; 2;    ; k; t ¼ 1; 2;    ; T:

where xt and yt represent the returns on Brent prices and exchange rates, respectively. The results (not reported here but are
available on request) showed that the Markov switching approach is also appropriate for modelling the upward and downward
movements in oil prices. Two states were identified: the ‘fall’ (State 1) and the ‘rise’ (State 2). However, exchange rates did not
explain the movements in oil prices either in the real or the nominal at any of the two states within the specified period.

5. Conclusion
Following Engel and Hamilton (1990) that exchange rates are better modelled using regime switching models rather than
random walk, this study employed the Markov switching model to study the effects of oil price shocks on exchange rate
movements in nominal and real terms. The mathematical approach adopted was to incorporate oil price changes in the
conditional mean equations in order to measure how much oil price shocks drive the dichotomous exchange rates. The outputs
from MSM were compared with those from linear regression models.
Empirical results from linear regression erroneously made us think that oil price shocks had no effect whatsoever on exchange
rates movements either in the nominal or real. However, with the dual nature of the exchange rate system taken into account,
nominal exchange rate returns exhibited an asymmetric response to nominal oil price changes during depreciation and
appreciation: we observed that negative changes in oil prices would lead to higher depreciation in naira while positive changes
would yield less depreciation; whereas oil price shocks had no significant effect on exchange rate movements when the exchange
rates system is in appreciation. We observed further that the nexus between real effective exchange rates (REER) and real oil
price is regime-dependent, though insignificant. In addition, smoothing probabilities indicated that the oil price shocks
adequately accounted for the depreciation episodes of 2008M11–2012M01 and 2014M10–2015M03 in the nominal series; and
2005M06–2005M07, 2008M12–2009M01 and 2014M11–2015M02 in the real effective series.
Results from this study have shown that a long-term relation exists between oil prices and exchange rates, especially while the
system is in depreciation. This implies that the value of naira would continue to fall so long as oil prices continue to fall. Thus, the
policy of pegging that the CBN usually adopts whenever oil prices fall is not a solution to Nigeria’s exchange rate instability as it
would only encourage dollar illiquidity and give control of the foreign exchange movements to speculators, currency traffickers
and perpetrators of arbitrage. Rather, in the meantime, CBN’s intervention should be to inject foreign exchange into the system
in order to ensure sustainable liquidity in the market; to identify genuine end users and ensure accessibility and affordability of
foreign exchange to them. This would bring some relief to local manufacturers, traders and other local users of US dollars while
waiting for crude oil prices to improve globally. In the long run, the government may work towards a sustainable diversification
of the economy to avoid continued over-dependence on oil exports; rigorously pursue a sustainable fiscal policy regime in order
to boost local productions. All these are necessary to establish a stable exchange rate system with a flourishing foreign exchange
market to manage supply and demand of foreign currencies, reduce inflation and unemployment rates.
The inclusion of the real effective exchange rates in our modelling mix gives a picture of the overall performance of the naira.
REER appreciates with decreasing real oil prices though nominal naira exchange rates generally depreciate against the dollar from
© 2017 The Author. African Development Review © 2017 African Development Bank
Oil and the Naira 573

January 2004 to May 2016. This suggests that the nominal appreciation of the dollar against the naira may be due to some temporary
weakness in the naira owing to the current status of oil prices in the global market, in view of the fact that oil is denominated in dollars.
Thus, if the current crisis is properly managed, the domestic value and overall competitiveness of the naira may be quickly restored.
The present study considers oil prices as the only independent variable. This is because, our intent here is to examine the direct
influence of oil prices on USD/NGN exchange rates. Besides, influential studies such as Amano and van Norden (1998) and Chaudhuri
and Daniel (1998) provided theoretical and empirical evidence that oil price shocks are the most important factor determining real
exchange rates in the long run. Further studies may, however, be carried out to determine the influence of other relevant macroeconomic
variables such as inflation rates, interest rates and GDP on the relationship between oil prices and exchange rates.

Note
1. A notable exception however is Beckmann and Czudaj (2012) who employed a Markov switching vector error correction model
(MS-VEC) to analyse the cointegration pattern between oil prices and effective dollar exchange rates in nominal and real terms.

References
Akram, Q. F. (2009), ‘Commodity Prices, Interest Rates and the Dollar’, Energy Economics, Vol. 31, No. 6, pp. 838-51.
Akram, Q. and J. Holter (1996), ‘The Effect of the Dollar Exchange Rate on Oil Prices: An Empirical Analysis’, Penger og
Kreditt, 1996/3.
Amano, R. and S. van Norden (1998), ‘Exchange Rates and Oil Prices’, Review of International Economics, Vol. 6, No. 4, pp. 683–94.
Anyanwu, J. and A. Erhijakpor (2014), ‘Does Oil Wealth Affect Democracy in Nigeria?’, African Development Review, Vol. 26,
No. 1, pp. 15–37.
Ayodeji, I. (2016), ‘A Three-state Markov-modulated Switching Model for Exchange Rates’, Journal of Applied Mathematics,
pp. 1–9. https://doi.org/10.1155/2016/5061749.
Beckmann, J. and R. Czudaj (2012), ‘Oil Price and U.S. Dollar Exchange Rates Dynamics’, University of Duisburg-
Essen, Department of Economics. Available at: https://editorialexpress.com/cgibin/conference/download.cgi?
dbname=MMF2012&paperid=128.
Breul, H. (2015), ‘Crude Oil Prices Down Sharply in Fourth Quarter of 2014’, US Energy Information Administration.
Available at: https://www.eia.gov/todayinenergy
Buetzer, S., M. Habib and L. Stracca (2012), ‘Global Exchange Rate Configurations: Do Oil Shocks Matter?’, Working Paper,
European Central Bank.
Chaudhuri, K., and B. Daniel (1998), ‘Long-run Equilibrium Real Exchange Rates and Oil Prices’, Economics Letters, Vol. 58,
No. 2, pp. 231–38.
Chen, S. and H. Chen (2007), ‘Oil Prices and Real Exchange Rates’, Energy Economics, Vol. 29, pp. 390–404.
Cifarelli, G. and G. Paladino (2010), ‘Oil Price Dynamics and Speculation: A Multivariate Financial Approach’, Energy
Economics, Vol. 32, No. 2, pp. 363–72.
Corden, W. (1984), ‘Booming Sector and Dutch Disease Economics: Survey and Consolidations’, Oxford Economic Papers,
Vol. 36, pp. 359–80.
Coudert, V., V. Mignon and A. Penot (2008), ‘Oil Price and the Dollar’, Energy Studies Review, Vol. 15, No. 2, pp. 45–58.
Dewachter, H. (1997), ‘Sign Predictions of Exchange Rate Changes: Charts as Proxies for Bayesian inferences’,
Weltwirtschaftliches Archiv, Vol. 133, pp. 39–55.
Emenike, K. (2016), ‘Comparative Analysis of Bureaux de Change and Official Exchange Rates Volatility in Nigeria’,
Intellectual Economics, Vol. 10, pp. 28–37.

© 2017 The Author. African Development Review © 2017 African Development Bank
574 I. O. Ayodeji

Enders, W. and S. Dibooglu (2001), ‘Long-run Purchasing Power Parity with Asymmetric Adjustment’, Southern Economic
Journal, Vol. 68, No. 2, pp. 433–45.
Engel, C. and J. Hamilton (1990), ‘Long Swings in the Dollar: Are They in the Data and do Market Know It?’, American
Economic Review, Vol. 80, pp. 687–713.
Eregha, P., A. Ndoricampa, S. Olakojo, M. Nchake, O. Nyang’oro and E. Togba (2016), ‘Nigeria: Should the Government Oat or
Devalue the Naira?’, African Development Review, Vol. 28, No. 3, pp. 247–63.
Evans, M. and K. Lewis (1995), ‘Do Long-term Swings in the Dollar Affect Estimates of the Risk Premia?’, Review of Financial
Studies, Vol. 8, pp. 709–42.
Ghosh, S. (2011), ‘Examining Crude Oil Price Exchange Rate Nexus for India during the Period of Extreme Oil Price Volatility’,
Applied Energy, Vol. 88, pp. 1886–89.
Golub, S. (1983), ‘Oil Prices and Exchange Rates’, Economic Journal, Vol. 93, pp. 576–93.
Hamilton, J. (1989), ‘A New Approach to the Economic Analysis of Nonstationary Time Series and the Business Cycle’,
Econometrica, Vol. 57, pp. 357–84.
Hamilton, J. (1994), Time Series Analysis, Princeton, NJ: Princeton University Press.
Huang, Y. and F. Guo (2007), ‘The Role of Oil Price Shocks on China’s Real Exchange Rate’, China Economic Review, Vol. 18,
pp. 403–16.
Igwe, N. and T. Ogunleye (2014), ‘Impact of Real Exchange Rate on Trade Balance in Nigeria’, African Development Review,
Vol. 26, No. 2, pp. 347–58.
Indjehagopian, J., F. Lantz and V. Simon (2000), ‘Dynamics of Heating Oil Market Prices in Europe’, Energy Economics,
Vol. 22, No. 2, pp. 225–52.
Kaminsky, G. (1993), ‘Is There a Peso Problem? Evidence from the Dollar/Pound Exchange Rate, 1976–1987’, American
Economic Review, Vol. 83, pp. 450–72.
Kim, J. (1994), ‘Dynamic Linear Models with Markov-switching’, Journal of Econometrics, Vol. 60, pp. 1–22.
Kim, J. and Nelson, C. (1999), State-Space Models with Regime Switching, Cambridge, MA: MIT Press.
Klaassen, F. (2005), ‘Long Swings in Exchange Rates: Are They Really in the Data?’, Journal of Business and Economic
Statistics, Vol. 23, pp. 87–95.
Lee, H. and S. Chen (2006), ‘Why Use Markov-switching Models in Exchange Rate Prediction?’, Economic Modeling, Vol. 23,
No. 4, pp. 662–68.
Loretan, M. (2005), ‘Indexes of the Foreign Exchange Value of the Dollar’, Federal Reserve Bulletin, pp. 1–8.
Narayan, P. K. S. Narayan and A. Prasad (2008), ‘Understanding the Oil Price-Exchange Rate Nexus for the Fiji Islands’, Energy
Economics, Vol. 30, pp. 2686–96.
Nelson, D. (1991), ‘Conditional Heteroskedasticity in Asset Returns: A New Approach’, Econometrica, Vol. 59, pp. 347–70.
Ozturk, L. and H. Kalyoncu (2009), ‘Exchange Rate Volatility and Trade: An Empirical Investigation from Cross-country
Comparison’, African Development Review, Vol. 21, No. 3, pp. 499–513.
Pershin, V., J. Molero and F. Gracia (2015), ‘Exploring the Oil Prices and Exchange Rates Nexus in Some African Economies’,
Journal of Policy Modeling, available at: https://doi.org/10.1016/j.jpolmod.2015.11.001.
Turhan, I., E. Hacihasanoglu and U. Soytas (1991), ‘Oil Prices and Emerging Market Exchange Rates’, Emerging Markets
Finance and Trade, Vol. 49, Sup. 1, pp. 21–36.
Zalduendo, J. (2006), ‘Determinants of Venezuelas Equilibrium Real Exchange Rate’, IMF Working Paper, WP-0674.
Zhang, Y., Y. Fan, H. Tsai and Y. Wei (2008), ‘Spillover Effect of US Dollar Exchange Rate on Oil Prices’, Journal of Policy
Modeling, Vol. 30, No. 6, pp. 973–91.

© 2017 The Author. African Development Review © 2017 African Development Bank
African Development Review, Vol. 29, No. 4, 2017, 575–588

Foreign Direct Investment in Ghana: The Role of Infrastructural


Development and Natural Resources

Ebenezer Bugri Anarfo, Abel Mawuko Agoba and Robert Abebreseh

Abstract: This paper examines the role of infrastructural development and natural resources on FDI inflows in Ghana. We
examine the empirical relations using the Prais–Winsten regression estimation procedure which is meant to overcome
autocorrelation and heteroscedasticity in the error terms in the model, often for regressions applied to series. The data was
obtained from World Development Indicators (WDI) spanning from 1975 to 2014. Our findings suggest that infrastructural
development and natural resources are drivers of FDI inflows in Ghana. The other variables that significantly influence FDI
inflows in Ghana include the lending interest rate, market size and GDP growth rate. The policy implications of this study is that,
while Ghana seeks to expand its infrastructure and natural resources to enhance FDI inflows and economic growth, this must be
done in tandem with creating an enabling environment to ensure macroeconomic stability.

1. Introduction
Foreign direct investment (FDI) has been defined by Graham and Krugman (1982) as the source of acquisition of managerial
control by a business enterprise of a foreign country over a business activity in a host country. FDI, according to the World
Bank, are the net inflows of investments made to acquire a lasting management interest, which is usually 10 per cent or more
of voting rights, in a business functioning in an economy and not that of the investor. It is arrived at by summing equity
capital, reinvestment of earnings, other long-term capital, and short-term capital, all depicted in the balance of payments. In
the past two decades, there has been an upward surge in FDI inflows across most regions of the world. This has led to a
resurgence in academia, of the debate regarding the benefits of FDI and it determinants (Omri et al., 2014). Benefits of FDI
include direct capital financing, generation of positive externalities, and the subsequent improvement in economic growth as
a result of technology transfer, spillover effects, productivity gains, and the introduction of new processes and managerial
skills (Lee, 2013).
The World Bank (2015) has witnessed a change in the direction of flow of FDI from developed to developing countries. Net
capital flows from FDI for instance, in 124 low- and middle-income countries as at 2015 was $1.2 trillion. Developing and
transition economies also received greater Greenfield investments than developed countries between 2008 and 2009 (World
Bank, 2010). Among the top 20 economies ranked by FDI inflows are developing countries and transition economies, who from
2011 to 2013 recorded FDI inflows between $725 billion and $778 billion, representing an increase from 43 per cent to 54 per
cent of global FDI inflows (World Bank, 2014). While on a yearly average basis Africa’s share of FDI inflows has been low and it
is far behind other regions of the world, Ghana on the average has experienced a significant rise in FDI inflows relative to other
countries in percentage terms. This has drawn attention as to what might have accounted for this in the face of massive
infrastructural development, oil discovery and challenges with power generation in recent times.
FDI acts as a rich source of capital, technology and skill transfer. For the home economy, FDI helps it reap the benefit of
an expanding market, lowering costs factors and other tariff measures (World Bank, 2010). In this regard, subsequent
governments in Ghana have aimed at improving the infrastructure and natural resource explorations in a bid to attract
increased FDI. In the last decade, Ghana has experienced an enormous investment in transport, health, education and


Ebenezer Bugri Anarfo (corresponding author), GIMPA Business School, Ghana Institute of Management and Public Administration, PO Box AH 50,
Achimota, Accra, Ghana; tel: þ233 21 4010681, fax: þ233 21 421 622, e-mail: ebugri@yahoo.com. Abel Mawuko Agoba, Department of Banking and
Finance, Central University, PO Box 2305, Tema, Ghana; e-mail: aagoba@central.edu.gh. Robert Abebreseh, GIMPA Business School, Ghana
Institute of Management and Public Administration, PO Box AH 50, Achimota, Accra, Ghana; e-mail: rabebreseh@yahoo.com
© 2017 The Authors. African Development Review © 2017 African Development Bank. Published by Blackwell Publishing Ltd,
9600 Garsington Road, Oxford OX4 2DQ, UK and 350 Main Street, Malden, MA 02148, USA. 575
576 E. B. Anarfo et al.

electricity infrastructure. There have been discoveries of oil in many parts of the country in addition to the gold and cocoa
natural resource base of the country. These two events are expected to attract resource-seeking FDI and efficiency seeking
FDI into the country. However, in the past year, for example, FDI reduced by about US$3 billion due to the erratic power
supply situation that the country has experienced in spite of the massive investments in the electricity, oil and gas sectors.
Factors that significantly influence FDI in one country may not be significant in other countries (Ietto-Gillies, 2005). Even
though Asiedu (2002) examines the determinants of FDI in sub-Saharan Africa (SSA), where infrastructural development
was considered as a determinant, the paper did not find infrastructural development significant in attracting FDI inflows in
SSA but found it significant in determining FDI inflows in non-SSA countries. Asiedu’s paper was a cross-country study
and therefore there is the need to examine the FDI, infrastructural development and natural resource nexus specific to a
particular country due to differences in political structure, the rule of law, FDI and capital repatriation regulations.
Countries in SSA do not have a homogeneous economic structures and hence doing a cross-country study on FDI
determinants will not be informative. Due to the heterogeneous nature of countries in SSA, this paper examines the
FDI, infrastructural development and natural resource nexus in Ghana. Ghana has a robust rule of law, political stability,
high flexibility of capital repatriation and fewer restrictions of FDI in to the extractive industry. Due to the lesser
restrictions of FDI into the extractive industry, this paper also considers the role of natural resources as a determinant of
FDI in Ghana.
Considering the expansion in infrastructure, oil discovery and electricity challenges in Ghana, the inflows of FDI has not been
consistent. With these investments, there is an expected increase in FDI inflow that is not forthcoming. Among many other
determinants of FDI, the study seeks to consider the impact of massive investments in infrastructural development and natural
resources on Ghana’s FDI inflows. Our paper contributes in a number ways to the FDI literature. This paper contributes to the
literature on FDI by being the first to examine the role of infrastructural development and natural resources in attracting FDI
specific to one country in sub-Saharan Africa.
Our paper extends the frontier of knowledge in economics by presenting new evidence showing that infrastructural
development and natural resources significantly influence FDI in Ghana. We also add to those recent studies that confirm a causal
link among FDI, infrastructural development and natural resources (see Asiedu, 2002).
Lastly, from a theoretical point of view, this paper will lay the foundation for the development of theory on the nexus of FDI,
infrastructural development and natural resources. Uncovering the theoretical underpinnings of the nexus of FDI–infrastructural
development–natural resources will help developing countries’ policymakers to devise policies that will let them get the most out
of FDI, infrastructural development and natural resources for socio-economic development. The study seeks to answer the
following research question: Does infrastructural development and natural resorces play a role in attracting FDI inflows in
Ghana?
Using data spanning from 1975 to 2014 we find that infrastructural development and natural resources positively and
significantly influence FDI in Ghana.
The rest of the study is organized as follows. In Section 2 we present an overview of Ghana’s FDI and infrastructural
development. Section 3 presents a literature review on FDI. Section 4 discusses the empirical models and methodology. Section
5 describes the data and presents the estimation results. Section 6 presents the issues and implications of our findings.

2. Overview of FDI, Infrastructure and Natural Resources in Ghana


FDI inflows for Ghana have increased from US$855 million in 2007 to US$3.2 billion in 2012, accounting for around 20 per cent
of total FDI inflows to the ECOWAS region (African Economic Outlook, 2014). This is on the backdrop of fluctuating trends of
FDI to Ghana in recent times. In spite of efforts by the Ghana Investment Promotion Council (GIPC), to attract investments into
the country. Between 1980 and 2013, FDI has risen in nominal terms from $15.6 million to $3,227 million. During the military
rule regimes, net inflows of FDI into Ghana as a percentage of GDP averaged 0.1 per cent between 1980 and 1984; 0.14 per cent
between 1985 and 1989 and 0.3 per cent between 1990 and 1991 (African Economic Outlook, 2014). On the assumption of
democracy in 1992, this figure has varied on average at 3.93 per cent between 1992 and 1994; 1.98 per cent between 1995 and
2006; and with the discovery of oil in 2007, jumped to 8.1 per cent between 2007 and 2009 and 7.96 per cent between 2010 and
2013 (World Bank, 2015). Most of the FDIs have been directed towards oil and gas, transport and logistics, telecommunications,
financial services and food and beverages subsectors (African Economic Outlook, 2014). FDI inflows to Ghana are significantly
sizeable, due to the fact that the share of FDI to gross fixed capital formation averages around 40 per cent (African Economic

© 2017 The Authors. African Development Review © 2017 African Development Bank
Foreign Direct Investment in Ghana 577

Figure 1: FDI inflows

Source: World Bank data.

Outlook, 2014). As shown in Figure 1, data on the flow of FDI globally shows that Ghana’s FDI as a ratio of its GDP over the past
decade has significantly been far above that of the world and the sub-Saharan region.
African Economic Outlook (2014) gives an extensive overview of Ghana’s infrastructural development drive. Unlike many
other African countries, Ghana’s infrastructure spans the length and breadth of the country, helping to integrate the different
regions (Africa Economic Outlook, 2014). Two road corridors linking the northern and southern sectors, a national electricity
and power production and distribution grid, and an ICT network, connect major population centres. According to the African
Economic Outlook (2014), domestic air transport has increased significantly over the years with the modernization of air fields
and introduction of alternative domestic carriers. Recent data suggests passenger numbers have surged upwards from 540,000 to
780,000 between 2012 and 2013; representing a 43.3 per cent increment. The distribution of electricity, mobile
telecommunication and road construction has increased significantly. As at 2008 ending, the total number of fixed and
mobile lines was 11,714,330. Compared to other low-income countries, Ghana’s ICT penetration is better and above the 1.1 per
cent average for SSA (World Bank, 2015) which all goes to enhance investments. In 2009, Ghana was ranked as the most
preferred business destination in the sub-region for Business Process Outsourcing (BPO) in the AT Kearney Global Services
Location Index.
By African standards, Ghana also has a wide water resource infrastructure and arable lands made possible by the building of
dams for agricultural purposes in the north. However, in the natural resources endowed regions, though infrastructural
development is poor, Ghana’s major revenue sources, until recent times, has been from cocoa and gold. Ghana is currently the
second largest producer of cocoa in the world (GIPC, 2016). The discovery of oil introduced huge investments in the oil and gas
sector. This is subsequent to the increase in global demand for energy and a consequential upsurge in profits in the extractive
industry. Tullow reported a 100 per cent success rate in its oil explorations in Ghana in 2008 and a subsequent 88 per cent success
in 2009 (Asiedu, 2013). Within the same period, Tullow’s investments in oil in the SSA region including Ghana, went up by 230
per cent, with the expenditure in Africa alone being 70 per cent of the company’s total production expenditure in 2008 and 88 per
cent in 2009. Exxon Mobil Corp also agreed to buy a $4 billion stake in an oil field off the coast of Ghana (Asiedu, 2013). In the
midst of all these investments in infrastructure and natural resources, Ghana has not experienced the expected inflow of FDI from
these investments.

3. Literature Review
Studies on FDI have centred on the fundamental theories of FDI, the effect of various macroeconomic factors on FDI, the role
economic integration plays in the direction of flow of FDI and the advantages and disadvantages of FDI in the host country
(Inekwe, 2013; Yasin, 2005; Cheng and Kwan, 2000; Barros et al., 2014). Dunning’s (1980) eclectic paradigm drew a
framework to assess the reasons why and where foreign enterprises invest overseas. It further grouped the determinants of FDI
into factors specific to particular industries and those that affect the entire economy. According to the theory, the determinants of

© 2017 The Authors. African Development Review © 2017 African Development Bank
578 E. B. Anarfo et al.

FDI include the benefits of ownership, location and internalization that come to multinational corporations (MNCs) from foreign
companies. FDI takes place when these benefits exceed the associated costs.
Studies have also shown that the inflow of foreign capital significantly increases domestic savings, leading to investment
in the host country (Ajayi et al., 2016), improved technology transfers and spillovers (Carkovic and Levine, 2002). Since FDI
is market-driven, it increases productivity in the sectors they go into, and the economy as a whole, because of increased
competitiveness in the local market. This engenders an increase in demand for goods and services (Ajayi et al., 2016).

3.1 Determinants of FDI


Economic growth, market size, human capital, infrastructure, exchange rate, openness, political stability, government
expenditure, labour force and export, have been identified by several researchers as the main determinants of FDI (Anyanwu and
Erhijakpor, 2004; Asiedu, 2006; Bevan and Estrin, 2000; Khadaroo and Seetanah, 2010; Kyereboah-Coleman and Agyire-
Tettey, 2008; Prasai, 2014; Musila and Sigue, 2006; Sekkat and Veganzones-Varoudakis, 2007; Zenasni and Benhabib, 2013;
Zhang, 2011; Zheng, 2009). Obwona (1997) notes that the decision by foreigners to invest in a country is based on the
availability and cost of natural and human resources; adequacy of infrastructure and support facilities; market size; trade policies
and other policies that affect macroeconomic stability; economic growth, level of development; and political stability.

3.2 Economic Growth


Economic growth could influence FDI based on the market seeking objective of foreign companies. Economic growth is
expected to enlarge the demand for goods and services due to an increase in per capita income of the populace. Most studies have
examined whether increasing FDI inflows leads to higher economic growth (e.g., Anwar and Nguyen, 2010; Batten and Vo,
2009; Ekanayake and Vogel, 2003; Hermes and Lensink, 2003; Nguyen and Nguyen, 2007; Obwona, 2001; Tsang and Yip,
2007). Some of these studies have found a reverse causality between FDI and GDP growth whiles others found a unidirectional
relationship between FDI and GDP growth (Anwar and Nguyen, 2010).
Omri et al. (2014) found a bidirectional causality between FDI inflows and economic growth. Basu et al. (2003) also found a
bidirectional causality between economic growth and FDI in 23 developing countries between 1978 and 1996. They argued that
for economies that are relatively open, causality runs in both directions, while for relatively closed economies long-run causality
mainly runs from growth to FDI; thus providing supporting for the unidirectional hypothesis. Abbes et al. (2015) note that in
theory, the causal relation between FDI and GDP growth can be bidirectional. Tsai (1994) shows that while FDI promotes
economic growth, economic growth in turn, is viewed as a mechanism of attracting FDI, a finding consistent with the feedback
hypothesis. On the one hand, according to the FDI-led growth hypothesis, FDI inflows can lead to growth for the host country
through increased capital stock, creation of new employment opportunities, and easing the transfer of technology (Borensztein
et al., 1998; De Gregorio, 2003; De Mello, 1997). The market size hypothesis posits that a rapid GDP growth creates new
investment opportunities in the host country, leading to larger inflows of FDI (Mah, 2010; Rodriguez and Rodrik, 2000). The
neutrality hypothesis, however, states a non-existent causal relationship between FDI and economic growth (Mah, 2010).

3.3 Market Size


Market size is the proportion of the population of a country in demands for goods and services. An increase in market size
increases the prospect of the producers since the opportunity to increase their capacity as well as their revenue earning ability
arises. This should have a positive influence on inflow of FDI into host countries. Bevan and Estrin (2000) find that host countries
having large market sizes experience greater FDI due to greater market opportunities for investors. Agarwal (1980) also found
the size of a host country markets best explains the country’s propensity to attract FDI, particularly when considering FDI flows
to developing countries. Chakrabarti (2001) finds a robust correlation between FDI and market size. Other studies have also
supported the view that market size positively influences FDI (Zhang, 2011; Asiedu, 2006; Anyanwu, 2012; Prasai, 2014).
Kahouli and Maktouf’s (2015) findings, consistent with Medvedev (2012), Hejazi (2009) and Kahouli and Kadhraoui (2012),
was that there is a positive and significant effect of market size on FDI. Also, in large markets, there are acquisitions and
independent trade barriers, due to the attractiveness the monopoly power from the acquisition presents. Where there is high

© 2017 The Authors. African Development Review © 2017 African Development Bank
Foreign Direct Investment in Ghana 579

competition, FDI became the main input mode, thereby getting MNCs to enjoy their property benefit (Agarwal and Ramaswami,
1992). Shah and Ahmed (2003) also showed a positive relationship between market size and inward FDI flow.

3.4 Infrastructure

Infrastructural development is able to motivate FDI as it facilitates access to resources and markets in line with resource seeking
and market seeking motivations for FDI. It also helps efficiency-seeking multinationals whose costs of production can be
reduced significantly by the availability of road networks, internet and telecommunication. Asiedu (2006), Erdal and Mahmut
(2008) and Khadaroo and Seetanah (2010) all argue that infrastructure significantly improves FDI inflows. Infrastructure
consists of road networks, railway networks, telecommunication networks and all tools, machines and materials which improve
the production and service turnaround cycle time (Musila and Sigue, 2006). Studies by Musila and Sigue (2006) find that FDI in
Africa is dependent on improving infrastructure which attracts foreign investors. Increased levels of internet users increase the
level of FDI (Mateev and Tsekov, 2013). According to Kahouli and Maktouf (2015), investors choose a country with greater
availability of the internet. Their result confirms the findings of Choi (2003) who also find that the number of internet users
increases and promotes inward FDI.

3.5 Exchange Rate


Many studies have found the exchange rate to impact inwards FDI (Bouoiyour and Rey, 2005). Kyereboah-Coleman and Agyire-
Tettey (2008), Masayuki and Ivohasina (2005) studies show that fluctuations in the exchange rate reduce the inflow of FDI. Thus,
an appreciating exchange rate attracts FDI as this exchange rate signals a stable, strong and sound economy. In contrast, Khan
and Nawaz (2010) asserts that when exchange rates deteriorate, it makes it cheaper for foreign investors to engage in FDI as their
home country’s currency becomes relatively stronger, purchasing more goods and services. Findings of Chakrabarti (2001), Di
Giovanni (2005) and Bhavan et al. (2011) also show that the depreciation of currencies of investment partners against the
currency of the host country encourages inward FDI in the host countries. Kahouli and Maktouf (2015) further show that the
depreciation of currencies of the home country against the currency of the host country encourages FDI inflows. This is in
contradiction of findings of Hejazi and Safarian (2001) and Tahir and Larimo (2005) who argue that a country can attract FDI by
depreciating its currency because of the benefits the investor will gain from the weak currency of the host country. Froot and
Stein (1991) and Klein and Rosengren (1994) emphasized that when the currency of a country devalues, it leads to a reduction in
the cost of production when measured in the foreign currency which will result in increased inflow of FDI, since it will lead to
growth in the wealth of foreign investors.

3.6 Inflation

Inflation is the persistent and appreciable rise in the general price level of goods and services (Erdal and Mahmut, 2008).
Countries confronted with higher inflation require appropriate higher return to reward the risk of high inflation. High inflation
usually signals macroeconomic instability of a country. In investigating the determinants of foreign direct investment flows to
developing countries while employing a cross-sectional analysis, Erdal and Mahmut (2008) found that low inflation rates have
significant positive influence on FDI. As inflation reduces, FDI into the receiving countries increases. Other studies show that
high inflation rate may also discourage investors and reduce FDI in the host country (Mina, 2012; Ngouhouo, 2013). Kahouli and
Maktouf (2015) and Cuong (2013) find that while high inflation rates in source countries decrease FDI in receiving countries, in
the home country it decreases FDI. When the price index increases and there is an increased demand, then investors can increase
investment overseas, with lower inflation rates to maintain competitiveness in prices (Kahouli and Maktouf, 2015; Kleinert and
Toubal, 2010; Medvedev, 2012).

3.7 Trade Openness


A country’s trade openness is the extent to which its business regulatory environment enhances or discourages businesses
in investing, creating employment and increasing productivity (Asiedu 2006). Asiedu (2006), found that trade openness
© 2017 The Authors. African Development Review © 2017 African Development Bank
580 E. B. Anarfo et al.

increases FDI inflows into the economy. According to Chakrabarti (2001), openness to trade, which was proxied by
exports plus imports to GDP is positively correlated with FDI. Asiedu (2002), using the same proxy for openness, comes
to a similar conclusion when the paper distinguishes between sub-Saharan host countries from other host countries in
other regions. She finds that Africa differs significantly from non-African sample countries with regard to other FDI
determinants, whereas the promotional effect of openness to trade on FDI is found to be only slightly weaker in Africa.
Gastanaga et al. (1998), in addressing the tariff jumping hypothesis in the context of a panel analysis on the effects of
host country reforms on FDI, find that while cross-section results suggest that FDI flows were encouraged more strongly
by tariff jumping than by potential exports, the effects of import tariffs on FDI were negative in a time-series context.
Gastanaga et al. (1998) conclude that in individual countries, trade liberalization is a more important motive for FDI
overtime. Eicher and Kang (2005), in examining how optimal entry of multinationals in foreign markets is dependent on
the market size, FDI fixed costs and tariffs, showed that low trade barriers encourage the entry of multinational
companies.

3.8 Natural Resources


The discussions of the effect of natural resources on FDI has centred on two hypothesis. The Hecksher–Ohlin model
suggests that resource endowments affect resource FDI, but deter non-resource FDI. The other hypothesis gives a central
role to the surrounding market potential and surrounding FDI, which enables one to distinguish whether FDI is horizontal,
vertical export-platform or vertically fragmented (Poelbekk and van der Ploeg, 2010). In their study of the effect of natural
resources on FDI, Poelbekk and van der Ploeg (2010) find that while subsoil assets boost resource FDI but crowd out non-
resource FDI, the huge effect on non-resource FDI results in aggregate FDI being low in less resource-rich countries. Asiedu
(2004) finds that there is no significant impact of natural resource availability on multinational employment in sub-Saharan
Africa. Asiedu (2006) also finds that countries that are endowed with natural resources attract more FDI. Asiedu (2013)
further finds evidence of the FDI-resource curse in which case the abundance of natural resources adversely affects the
inflow of FDI.
Therefore, it is certain to say that in examining the determinants of FDI in Ghana, it is prudent to specify and estimate a model
linking them; with particular focus on infrastructural development and natural resources.

4. Methodology
Prais–Winsten regression estimation is used to estimate the coefficients of the model of this study. Prais–Winsten estimation is
used in this study because its procedure is meant to overcome autocorrelation of type AR(1) in a linear model and
heteroscedasticity in the error terms in the model, often for regressions applied to time series data. It was devised by Sigbert Prais
and Christopher Winsten in 1954. It is a modification of Cochrane–Orcutt estimation in the sense that it does not lose the first
observation, which leads to more efficiency as a result and makes it a special case of feasible generalized least squares.

4.1 Data and Data Sources

The study used secondary data obtained from World Development Indicators (WDI). The data spans from 1975 to 2014. In
examining the effect of infrastructural development and natural resources on FDI inflows in Ghana, the study control for the
following variables which include: GDP growth rate, inflation rate, exchange rate, market size, trade openness and lending
interest rate while FDI is the dependent variable. Table 1 presents a brief description of the variables and how they are measured.

4.2 Model Specification


The main independent variables used in this study are infrastructural development (INFRAS) and natural resources (NATRES).
However, there are a number of other factors that influence FDI, which serve as controls in our study. The control variables used
includes: GDP growth rate, inflation rate, exchange rate, market size, trade openness and lending interest rate. The model is
therefore specified as:
© 2017 The Authors. African Development Review © 2017 African Development Bank
Foreign Direct Investment in Ghana 581

Table 1: Description of variables


Variable Notation Description

Foreign direct investment FDI Foreign direct investment, net inflows (% of GDP)
Infrastructure INFRAS Fixed telephone subscriptions
Natural resources NATRES Total natural resources rents (% of GDP)
GDP growth rate GDPGR GDP growth rate (annual %)
Inflation INFLRA Inflation, consumer prices (annual %)
Exchange rate EXCHRATE Official exchange rate (LCU per US$, period average)
Market size MARKSIZE Log of GDP
Trade openness TRADE Exports plus imports as a % of GDP
Lending interest rate LEINRA Lending interest rate (%)

X
Y t ¼ b0 þ b X t þ et ð1Þ

with the subscript t representing the time series dimension. The left-hand side variable represents the dependent variable in the
model which is FDI, X t represents the independent variables in the estimation model, b0 is the constant term and et is the error
term.
We specify the general equation as:

X
n X
n
Y t ¼ bo þ bt X t þ bt Z t þ et ð2Þ
i¼1 i¼1

where Yit is the dependent variable FDI; Xt are the main independent variables, that is, INFRAS and NATRES; Zt are the
controlled variables, that is, GDPGR, INFLRA, EXCHRATE, MARKSIZE, TRADE and LEINRA; and eit is the error term,
assumed to be normally distributed with zero mean and constant variance.
The equation above can subsequently be estimated as follows:

FDIPGDt ¼ b0 þ b1 INFRAS t þ b2 NATRES t þ b3 LEINRAt þ b4 GDPGRt þ b5 INFLRAt þ b6 EX CHRATEt


þb7 MARKSIZEt þ b8 TRADEt þ eit

5. Results and Discussion

5.1 Unit Roots Test

The study employed STATA 11.2 package to carry out two unit root test (Augmented Dickey-Fuller and Phillips-Perron) in
order to determine whether the variables used in the study are stationary. All the variables were integrated of order one I(1)
except GDP growth rate and inflation rate that are stationary at levels and hence they are integrated of order zero I(0) stochastic
process. The results of the unit-root test are shown in Table 2.

5.2 Descriptive Statistics


Table 3 shows the descriptive statistic of all the variables used in the study. The mean value of FDI as a percentage of GDP is 2.48
per cent in Ghana. This indicates that FDI contributes about 2.48 per cent on average to Ghana’s GDP and this is quite significant
relative to other African countries. The average number of fixed telephone subscriptions which is used to proxy infrastructure in
this study is 137886.075. The results also indicate that on average, Ghana’s natural resources rent as a percentage of GDP is 9.32
per cent. The GDP growth rate for the period under study averages 3.78 per cent and this is quite low, partly due to the recent

© 2017 The Authors. African Development Review © 2017 African Development Bank
582 E. B. Anarfo et al.

Table 2: Unit root test


Variable Augmented Dickey–Fuller Phillips–Perron

Foreign direct investment I(1) I(I)


Infrastructure I(2) I(I)
Natural resources I(1) I(I)
GDP growth rate I(0) I(0)
Inflation rate I(0) I(0)
Exchange rate I(1) I(I)
Market size I(1) I(I)
Trade openness I(1) I(I)
Lending interest rate I(1) I(I)

Table 3: Descriptive statistics of variables


FDI INFRAS NATRES LEINRA GDPGR INFLRA EXCHRATE TRADE

Mean 2.481 137,886.075 9.323 20.674 3.776 33.228 0.427 58.246


Median 1.459 56,537.000 8.252 19.000 4.651 24.915 0.080 59.722
Maximum 9.517 376,509.000 21.128 25.583 14.046 122.875 1.954 116.048
Minimum 0.660 31,259.000 3.127 19.000 12.432 8.727 0.000 6.320
Std. Dev. 3.022 118,180.417 4.968 2.552 4.752 29.251 0.576 30.914
Skewness 1.195 0.615 0.740 1.277 1.289 1.899 1.181 0.010
Kurtosis 3.038 1.717 2.790 3.034 5.609 6.017 3.185 1.847
Jarque-Bera 9.530 5.266 3.720 2.991 22.416 39.213 9.360 2.218
Probability 0.009 0.072 0.156 0.224 0.000 0.000 0.009 0.330
Observations 40.000 40.000 40.000 11.000 40.000 40.000 40.000 40.000

global financial crunch and the energy crises that plagued the country for the past six years. The average lending interest in Ghana
is 20.67 per cent. The mean values of inflation rate and exchange rate are 33.23 per cent and 0.427 per cent respectively.

5.3 Diagnostics Tests


Correlation Analysis
One of the problems in time series regression estimates is the presence of linear relationship among the regressors. Due to the
problem of multicollinearity among the variables, a correlation matrix of the variables used in the regression is presented in
Table 4. Based on the table, the variables are not highly correlated except trade openness (TRADE) which is correlating with the
lending interest (LEINRA) and the market size (MARKSIZE) variables. It is for this reason that we drop trade openness (TRADE)
in the second regression model (2) presented in Table 5.

Test for Heteroscedasticity


Heteroscedasticity is the assumption that the variance of the residuals from regression estimates are not constant, otherwise
they are homoscedastic. Heteroscedasticity is a violation of the classical linear regression assumption of homoscedastic
residual or error. The Breusch-Pagan test for heteroscedasticity suggests that there is a problem of heteroscedasticity in the
residuals as shown below. It is for this reason that the standard errors used in the regression model are robust; that is, robust
standard errors were used when estimating the coefficients of the regression model to correct for heteroscedasticity in the
residuals.
© 2017 The Authors. African Development Review © 2017 African Development Bank
Foreign Direct Investment in Ghana 583

Table 4: Correlation matrix


FDI INFRAS NATRES LEINRA GDPGR INFLRA EXCHRATE MARKSIZE TRADE

FDI 1.00 –0.27 0.60 –0.34 –0.33 0.15 –0.36 –0.30 –0.52
INFRAS –0.27 1.00 –0.08 0.57 0.32 –0.47 0.30 0.55 0.36
NATRES 0.60 –0.08 1.00 –0.31 –0.32 0.22 –0.35 –0.23 –0.54
LEINRA –0.34 0.57 –0.31 1.00 0.49 –0.40 0.35 0.53 0.84
GDPGR –0.33 0.32 –0.32 0.49 1.00 –0.41 0.47 0.37 0.59
INFLRA 0.15 –0.47 0.22 –0.40 –0.41 1.00 –0.43 –0.48 –0.47
EXCHRATE –0.36 0.30 –0.35 0.35 0.47 –0.43 1.00 0.49 0.30
MARKSIZE –0.30 0.55 –0.23 0.53 0.37 –0.48 0.49 1.00 0.80
TRADE –0.52 0.36 –0.54 0.84 0.59 –0.47 0.30 0.80 1.00

Breusch-Pagan / Cook-Weisberg test for heteroscedasticity


Ho: Constant variance
Variables: fitted values of FDI
Chi2(1) ¼ 0.01
Prob > chi2 ¼ 0.9436

Test for Serial Correlation

The Breusch-Godfrey LM test for autocorrelation in time series data suggests that there is no problem of serial correlation in the
residuals. This implies that successive residuals are not correlated with each other and hence the classical linear regression
assumption of the absence of correlation between successive residuals or errors is not violated.
Breusch-Godfrey LM test for autocorrelation
–––––––––––––––––––––––––––––––––––––
lags(p) | chi2 df Prob > chi2
–––––––––––––––––––––––––––––––––––––
1 | 0.173 1 0.6777
–––––––––––––––––––––––––––––––––––––
H0: no serial correlation

5.4 Regression Results and Discussions


Table 5 presents the regression results of the impact of infrastructural development and natural resources on FDI in Ghana. There
is a strong confirmation that infrastructural development and natural resources are drivers of FDI inflows in Ghana. This
evidence is shown in Table 5. There is strong evidence of a positive impact of infrastructural development and natural resources
on FDI inflows in Ghana that is evidenced by the statistical significance as well as the positive coefficient of infrastructural
development and natural resources in both model 1 and model 2 of the regression table. The explanation we offer for this finding
is that infrastructural development positively influences FDI in Ghana and this is consistent with other studies that found it to
have a positive impact on FDI (Asiedu, 2002; Erdal and Mahmut, 2008; and Khadaroo and Seetanah, 2010). However, natural
resources is also statistically significant and its coefficient is positive as expected. This means that natural resources is an
important variable in determining FDI in Ghana and this is consistent with other empirical studies that examine the relationship
between FDI and natural resources in other jurisdictions (Asiedu, 2006).
Market size, which represents the size of the Ghanaian market, is also statistically significant at 5 per cent and it has a positive
coefficient as expected. This implies that the larger the market size, the more FDI Ghana will attract and this is consistent with
previous empirical studies (Zhang, 2011; Asiedu, 2006; Anyanwu, 2012; Prasai, 2014). The lending interest rate and GDP
growth rate are also statistically significant; however, trade openness and inflation rate are not statistically significant but they
carry the expected signs. The exchange rate is also statistically significant in model 2.

© 2017 The Authors. African Development Review © 2017 African Development Bank
584 E. B. Anarfo et al.

Table 5: Regression results


Model(1) FDI Model(2) FDI

INFRAS 11.36 11.90


(7.79) (10.14)
NATRES 0.0836 0.0940
(5.61) (8.73)
LEINRA –0.194 –0.168
(–3.87) (–6.44)
GDPGR 0.0495 0.0528
(4.74) (6.15)
INFLRA 0.000128 0.000628
(0.10) (0.69)
EXCHRATE –15.23 –63.43
(–0.26) (–3.52)
MARKSIZED1 5.161 5.347
(5.59) (6.33)
TRADED1 0.00774
(0.85)
_cons –115.7 –121.9
(–7.67) (–10.24)
N 11 11
R-sq 0.975 0.963
adj. R-sq 0.873 0.875
t statistics in parentheses.

p < 0.01,  p < 0.05,  p < 0.10.

6. Summary, Conclusion and Policy Implications


This paper examines FDI in Ghana, the role of infrastructural development and natural resources. The study finds that Ghana’s
infrastructural development and natural resources are drivers of FDI inflows. This implies that the new discovery of oil and other
natural resources in Ghana is expected to drive up FDI inflows. Ghana needs to put policies in place that will reduce its lending
interest rate since its reduction will enhance FDI inflows into the country. A reduction in interest rate will reduce the cost of
borrowing leading to an increase in domestic investment and FDI. The policy implication of this study is that Ghana needs to
create an enabling environment to ensure macroeconomic stability to be able to attract more FDI inflows since economic growth,
exchange rate and the lending interest rate are drivers of FDI inflows in Ghana. In improving its infrastructure, natural resources,
reducing lending interest rate and policies to promote economic growth, Ghana will create an enabling economic environment
for FDI inflows.

References
Abbes, S. M., B. Mostefa, G. M. Seghir and G. Y. Zakarya (2015), ‘Causal Interactions between FDI, and Economic Growth:
Evidence from Dynamic Panel Co-Integration’, Procedia Economics and Finance, Vol. 23, pp. 276–90.
African Economic Outlook (2014), Available at: www.africaneconomicoutlook.org/files (accessed 20 September 2016).
Agarwal, J. P. (1980), ‘Determinants of Foreign Direct Investment: A Survey’, Review of World Economics, Vol. 116, No. 4, pp.
739–73.
Agarwal, S. and S. N. Ramaswami (1992), ‘Choice of Foreign Market Entry Mode: Impact of Ownership, Location and
Internalization Factors’, Journal of International Business Studies, Vol. 23, No. 1, pp. 1–27.

© 2017 The Authors. African Development Review © 2017 African Development Bank
Foreign Direct Investment in Ghana 585

Ajayi, O. E., T. O. Akinbobola, S. B. Okposin and O. Ola-David (2016), ‘Interactive Effects of Exchange Rate Volatility and
Foreign Capital Inflows on Economic Growth in Nigeria’, in 3rd International Conference on African Development Issues (CU-
ICADI), 9-11 May, Covenant University, Ota, Nigeria, pp. 139–45.
Anwar, S., and L. P. Nguyen (2010), ‘Foreign Direct Investment and Economic Growth in Vietnam. Asia Pacific’, Business
Review, Vol. 16, pp. 83–202.
Anyanwu, J. C. (2012), ‘Why Does Foreign Direct Investment Go Where it Goes? New Evidence From African Countries’,
Annals of Economics & Finance, Vol. 13, No. 2, pp. 425–62.
Anyanwu, J. C. and A. E. O. Erhijakpor (2004), ‘Trends and Determinants of Foreign Direct Investment in Africa’, West African
Journal of Monetary and Economic Integration, Second Half, pp. 21–44.
Asiedu, E. (2002), ‘On the Determinants of Foreign Direct Investment to Developing Countries: Is Africa Different?’ World
Development, Vol. 30, No. 1, pp. 107–19.
Asiedu, E. (2004), ‘The Determinants of Employment of Affiliates of US Multinational Enterprises in Africa’, Development
Policy Review, Vol. 22, No. 4, pp. 371–39.
Asiedu, E. (2006), ‘Foreign Direct Investment in Africa: The Role of Natural Resources, Market Size, Government Policy,
Institutions and Political Instability’, UNU-WIDER Working Paper 24/2005.
Asiedu, E. (2013), ‘Foreign Direct Investment, Natural Resources and Institutions. International Growth Centre’, London
School of Economics Working Paper.
Barros, C. P., B. Damasio and J. R. Faria (2014), ‘Reverse FDI in Europe: An Analysis of Angola’s FDI in Portugal’, African
Development Review, Vol. 26, No. 1, pp. 160–71.
Basu, P., C. Chakraborty and D. Reagle (2003), ‘Liberalization, FDI, and Growth in Developing Countries: A Panel Co-
Integration Approach’, Economic Inquiry, Vol. 51, pp. 510–16.
Batten, J. A. and X. V. Vo (2009), ‘An Analysis of the Relationship between Foreign Direct Investment and Economic Growth’,
Applied Economics, Vol. 41, No. 13, pp. 1621–41.
Bevan, A. A., and S. Estrin (2000), ‘The Determinants of Foreign Direct Investment into European Transition Economies’,
Journal of Comparative Economics, Vol. 32, No. 4, pp. 775–87.
Bhavan, T., C. Xu and C. Zhong (2011), ‘The Relationship between Foreign Aid and FDI in South Asian Economies’,
International Journal of Economics and Finance, Vol. 3, pp. 143–49.
Borensztein, E., J. De Gregorio and J. W. Lee (1998), ‘How Does Foreign Investment Affect Growth?’, Journal of International
Economics, Vol. 45, pp. 115–35.
Bouoiyour, J. and S. Rey (2005), ‘Exchange Rate Regime, Real Exchange Rate, Trade Flows and Foreign Direct Investments:
The Case of Morocco’, African Development Review, Vol. 17, No. 2, pp. 302–34.
Carkovic, M. and R. Levine (2002), ‘Does Foreign Direct Investment Accelerate Economic Growth?’ in H. Moran and E. M.
Graham (eds.), Does Foreign Direct Investment Promote Development? Institute for International Economics, Washington, DC.
Chakrabarti, A. (2001), ‘The Determinants of Foreign Direct Investment: Sensitivity Analyses of Cross-Country Regressions’,
International Review for Social Sciences, Vol. 54, pp. 89–114.
Cheng, L. K. and Y. K. Kwan (2000), ‘What Are the Determinants of the Location of Foreign Direct Investment? The Chinese
Experience’, Journal of International Economics, Vol. 51, No. 2, pp. 379–400.
Choi, C. (2003), ‘Does the Internet Stimulate Inward Foreign Direct Investment?’, Journal of Policy Modeling, Vol. 25, No. 4,
pp. 319–26.
Cuong, H. C. (2013), ‘The Impact of the WTO Regime on FDI Flows to Vietnam: A Gravity Model Approach’, Journal of
Modern Accounting and Auditing, Vol. 9, pp. 961–87.

© 2017 The Authors. African Development Review © 2017 African Development Bank
586 E. B. Anarfo et al.

De Gregorio, J. (2003), ‘The Role of Foreign Direct Investment and Natural Resources in Economic Development’, Working
Paper No. 196. Central Bank of Chile, Santiago.
De Mello Jr, L. R. (1997), ‘Foreign Direct Investment in Developing Countries and Growth: A Selective Survey’, Journal of
Development Studies, Vol. 34, No. 1, pp. 1–34.
Di Giovanni, J. (2005), ‘What Drives Capital Flows? The Case of Cross-Border Merger and Acquisition Activity and Financial
Deepening’, Journal of International Economics, Vol. 65, pp. 127–49.
Dunning, J. H. (1980), ‘Toward an Eclectic Theory of International Production: Some Empirical Tests’, Journal of International
Business Studies’, Vol. 11, pp. 9–31.
Eicher, T. and J. W. Kang (2005), ‘Trade, Foreign Direct Investment or Acquisition: Optimal Entry Modes for Multinationals’,
Journal of Development Economics, Vol. 77, pp. 207–28.
Ekanayake, E. M. and R. Vogel (2003), ‘Openness and Economic Growth: Empirical Evidence on the Relationship between
Output, FDI, and Trade’, Journal of Business Strategy, Vol. 20, pp. 59–72.
Erdal, D. and M. Mahmut (2008), ‘Determinants of Foreign Direct Investment Flows to Developing Countries: A Cross-
Sectional Analysis’, Prague Economic Papers, No. 4, pp. 356–69.
Froot, K. A. and J. C. Stein (1991), ‘Exchange Rate and Foreign Direct Investment: An Imperfect Capital Market Approach’,
Quarterly Journal of Economics, Vol. 1066, No. 4, pp. 119–217.
Gastanaga, V. M., J. B. Nugent and B. Pashamova (1998), ‘Host Country Reforms and FDI Inflows: How Much Difference Do
They Make?’, World Development, Vol. 26, No. 7, pp. 1299–314.
GIPC (2016), ‘About Ghana Investment Promotion Centre’, available at: http://www.gipcghana.com/.
Graham, E. M. and P. R. Krugman (1982), ‘The Surge Foreign Direct Investment in the 1980sþ’, in K. A. Froot (ed.), Foreign
Direct Investment, University of Chicago Press, Chicago, pp. 13–36.
Hejazi, W. (2009), ‘Does China Receive More Regional FDI Than Gravity Would Suggest?’ European Management Journal,
Vol. 27, pp. 327–35.
Hejazi, W. and A. E. Safarian (2001), ‘The Complementarity between U.S Foreign Direct Investment Stock and Trade’, Atlantic
Economic Journal, Vol. 29, pp. 420–37.
Hermes, N. and R. Lensink (2003), ‘Foreign Direct Investment, Financial Growth and Economic Growth’, Journal of Growth
Studies, Vol. 40, pp. 142–63.
Ietto-Gillies, G. (2005), ‘Dunning’s Eclectic Frame Work’, in G. Ietto-Gillies (ed.), Transnational Corporations and
International Production, Edward Elgar, Cheltenham.
Inekwe, J. N. (2013), ‘FDI, Employment and Economic Growth in Nigeria’, African Development Review, Vol. 25, No. 4, pp.
421–33.
Kahouli, B. and N. Kadhraoui (2012), ‘Consolidation of Regional Groupings and Economic Growth: Empirical Investigation by
Panel Data’, International Journal of Euro-Mediterranean Studies, Vol. 5, No. 1, pp. 71–92.
Kahouli, B. and S. Maktouf (2015), ‘The Determinants of FDI and the Impact of the Economic Crisis on the Implementation of
RTAs: A Static and Dynamic Gravity Model’, International Business Review, Vol. 24, pp. 518–29.
Khan, R. E. A. and M. A. Nawaz (2010), ‘Economic Determinants of Foreign Direct Investment in Pakistan’, Journal of
Economics, Vol. 1, No. 2, pp. 99–104.
Khadaroo, A. J. and B. Seetanah (2010), ‘Transport Infrastructure and Foreign Direct Investment’, Journal of International
Development, Vol. 22, No. 1, pp. 103–23.
Klein, M. W. and R. Rosengren (1994), ‘The Real Exchange Rate and the Foreign Direct Investment in the United States:
Relative Wealth vs Relative Wage Effects’, Journal of International Economics, Vol. 36, pp. 373–89.
Kleinert, J. and F. Toubal (2010), ‘Gravity for FDI’, Review of International Economics, Vol. 18, pp. 1–13.

© 2017 The Authors. African Development Review © 2017 African Development Bank
Foreign Direct Investment in Ghana 587

Kyereboah-Coleman, A. and K. F. Agyire-Tettey (2008), ‘Effect of Exchange-rate Volatility on Foreign Direct Investment in
Sub-Saharan Africa: The Case of Ghana’, Journal of Risk Finance, Vol. 9, No. 1, pp. 52–70.
Lee, J. W. (2013), ‘The Contribution of Foreign Direct Investment to Clean Energy Use, Carbon Emissions and Economic
Growth’, Energy Policy, Vol. 55, pp. 483–89.
Mah, J. S. (2010), ‘Foreign Direct Investment Inflows and Economic Growth of China’, Journal of Policy Modeling, Vol. 32,
No. 1, pp. 155–58.
Masayuki, H. and F. Ivohasina (2005), ‘The Determinants of Foreign Direct Investment into Japan’, Kobe University Economic
Review, Vol. 51, pp. 21–34.
Mateev, M. I. and I. Tsekov (2013), ‘Do Central and Eastern European Countries Possess FDI Advantages to More Developed
Western Countries?’ paper presented to the 2013 Annual Meeting of the Midwest Finance Association.
Medvedev, D. (2012), ‘Beyond Trade: The Impact of Preferential Trade Agreements on FDI Inflows’, World Development, Vol.
40, pp. 49–61.
Mina, W. M. (2012), ‘The Institutional Reforms Debate and FDI Inflows to the MENA Region: The “Best” Ensemble’, World
Development, Vol. 40, pp. 1798–809.
Musila, J. W. and S. P. Sigue (2006), ‘Accelerating Foreign Direct Investment Flow to Africa: From Policy Statements to
Successful Strategies’, Managerial Finance, Vol. 32, No. 7, pp. 577–93.
Ngouhouo, I. (2013), ‘Multidimensional Determinants of Foreign Direct Investment in Central Africa: A Modified Gravity
GMM Panel Approach’, Mediterranean Journal of Social Sciences, Vol. 4, pp. 575–86.
Nguyen, A. N. and T. Nguyen (2007), ‘Foreign Direct Investment in Vietnam: An Overview and Analysis of the Determination
of Spatial Distribution’, Working Paper, Development and Polices Research Center, Hanoi, Vietnam.
Obwona, M. B. (1997), ‘Determinants of FDI and Their Impact on Economic Growth in Uganda’, Economic Policy Research
Centre Working Paper.
Obwona, M. B. (2001), ‘Determinants of FDI and Their Impact on Economic Growth in Uganda’, African Development Review,
Vol. 13, pp. 46–81.
Omri, A., D. K. Nguyen and C. Rault (2014), ‘Causal Interactions between CO2 Emissions, FDI, and Economic Growth:
Evidence from Dynamic Simultaneous-Equation Models’, Economic Modelling, Vol. 42, pp. 382–89.
Poelbekk, S. and R. Van Der Ploeg (2010), ‘Do Natural Resources Attract FDI? Evidence from Non-Stationary Sector Level
Data’, Nederlandsche Bank Working Paper No. 266/2010.
Prasai, L. P. (2014), ‘Foreign Trade Pattern of Nepal: Gravity Model Approach’, NRB Economic Review, Vol. 26, No. 1, pp.
24–43.
Rodriguez, F. and D. Rodrik (2000), ‘Trade Policy and Economic Growth: A Skeptic’s Guide to the Cross-national Evidence’, in
B. Bernanke and K. Rogoff (eds.), NBER Macroeconomics Annual 2000, MIT Press, Cambridge, MA.
Sekkat, K. and M.-A. Veganzones-Varoudakis (2007), ‘Openness, Investment Climate, and FDI in Developing Countries’,
Review of Development Economics, Vol. 11, No. 4, pp. 607–20.
Shah, Z. and Q. M. Ahmed (2003), ‘The Determinants of Foreign Direct Investment in Pakistan: An Empirical Investigation’,
The Pakistan Development Review, Vol. 42, No. 4.
Tahir, R. and J. Larimo (2005), ‘Understanding the Strategic Motivations of Finnish Manufacturing FDIs in Emerging Asian
Economies’, Asian Business And Management, Vol. 4, pp. 293–313.
Tsai, P. (1994), ‘Determinants of Foreign Direct Investment and its Impact on Economic Growth’, Journal of Economic
Development, Vol. 19, No. 1, pp. 137–63.
Tsang, E. W. K. and P. S. L. Yip (2007), ‘Economic Distance and the Survival of Foreign Direct Investments’, Academic
Management Journal, Vol. 50, pp. 1156–68.

© 2017 The Authors. African Development Review © 2017 African Development Bank
588 E. B. Anarfo et al.

World Bank (2010), World Investment Report 2010: Investing in a Low-Carbon Economy, World Bank, Washington DC.
World Bank (2014), World Investment Report 2014: Investing in the SDGs: An Action Plan, World Bank, Washington DC.
World Bank (2015), World Investment Report 2015: Reforming International Investment Governance, World Bank, Washington DC.
Yasin, M. (2005), ‘Official Development Assistance and Foreign Direct Investment Flows to Sub-Saharan Africa’, African
Development Review, Vol. 17, No. 1, pp. 23–40.
Zenasni, S. and A. Benhabib (2013), ‘The Determinants of Foreign Direct Investment and their Impact on Growth: Panel Data
Analysis for AMU Countries’, International Journal of Innovation and Applied Studies, Vol. 2, No. 3, pp. 300–13.
Zhang, Y. J. (2011), ‘The Impact of Financial Growth on Carbon Emissions: An Empirical Analysis in China’, Energy Policy,
Vol. 39, pp. 2197–203.
Zheng, P. (2009), ‘A Comparison of FDI Determinants in China and India’, Thunderbird International Business Review, Vol.
51, No. 3, pp. 263–79.

© 2017 The Authors. African Development Review © 2017 African Development Bank
African Development Review, Vol. 29, No. 4, 2017, 589–600

FDI Determinants in Least Recipient Regions: The Case of Sub-Saharan


Africa and MENA

Godwin Okafor, Jenifer Piesse and Allan Webster

Abstract: This paper explores the determinants of foreign direct investment (FDI) into FDI least recipient regions. Panel
data for 20 sub-Saharan Africa (SSA) and 11 Middle East and North Africa (MENA) countries are used for the period 2000–12.
Findings of the fixed effects estimations suggest that FDI inflows into these regions are influenced by GDP per capita,
infrastructure development, trade openness, and control of corruption. Conversely, inflation negatively affects FDI inflows, and
rents from natural resources do not significantly influence FDI. Furthermore, the findings show that marginal benefits from any
increase in the quantity of FDI determinants (with the exception of control of corruption) will be less for SSA countries. The
paper concludes with important policy implications deduced from the findings.

1. Introduction
Sub-Saharan Africa (SSA) and the Middle East and North Africa (MENA) receive the lowest levels of foreign direct investment
(FDI) inflows in the world — around 2 per cent and 5 per cent respectively of all global FDI inflows (World Bank, 2015a). The
poor record of FDI flows into these regions is surprising considering the quality and quantity of natural resources and their
strategic location. It is often argued that the high levels of instability and corruption, weak governance, and poor quality
infrastructure account for their inability to attract FDI (Kandiero and Chitiga, 2006). However, over the last couple of years
efforts have been focused on attracting more FDI. For example, in the late 1980s the MENA countries began a significant shift
toward trade and FDI openness and the creation of an environment that is more favourable to FDI and exports (AbuAl-Foul and
Soliman, 2008). In SSA, structural adjustment programmes were also introduced to attract investment after years of policies that
deterred foreign investment due to fears that this would result in a loss of political sovereignty, a negative impact on domestic
firms and economic degradation with respect to the natural resource sectors (Dupasquier and Osakwe, 2006).
Tables 1 and 2 report levels of FDI inflows across different periods. In Table 1 it is clear that both SSA and the MENA regions
received very little FDI prior to the 1980s. However, the 1980s have seen a significant shift in flows particularly for the MENA
region. In Table 2 (Panel A) it is also clear that the countries in SSA have received by far the lowest amount of inward investment
in the latter period, followed by the MENA countries. Interestingly, the coefficient of variation for all regions, with the exception
of Europe and Central Asia, is very similar, suggesting that the dispersion of foreign investment activity is uniform. Panel B in
the table shows some encouraging growth in inward FDI for both regions in the present study although SSA lags behind the
MENA region to a considerable extent.
FDI can play a critical role in providing capital for investment, high quality managerial skills and technology transfer. It also
creates employment, competition and productivity, transfer of modern technology, increases exports and enhances opportunities
for growth and development, particularly in developing and emerging countries (Asiedu, 2002; Akinlo, 2004; Anyanwu and
Yameogo, 2015a). Hence, it is important that SSA and the MENA region attract sustained foreign investment that can be used to
assist in their development programmes and achieve higher levels of growth. Regrettably, as the data suggest, the regions are still
at the lower end of the distribution of FDI inflows, which suggests that the reforms over the last couple of decades to attract FDI
are still inadequate and have not attracted sufficient foreign investors.


Godwin Okafor (corresponding author), School of Business, Law and Communications, Southampton Solent University, UK; e-mail:
godwin.okafor@solent.ac.uk. Jenifer Piesse, Business School, Bournemouth University, UK and Stellenbosch University, South Africa; e-mail:
jpiesse@bournemouth.ac.uk. Allan Webster, Business School, Bournemouth University, UK; e-mail: awebster@bournemouth.ac.uk
© 2017 The Authors. African Development Review © 2017 African Development Bank. Published by Blackwell Publishing Ltd,
9600 Garsington Road, Oxford OX4 2DQ, UK and 350 Main Street, Malden, MA 02148, USA. 589
590 G. Okafor et al.

Table 1: FDI inflows ($ billions) to SSA and MENA regions (1970–99)


Regions Mean Std. dev. Minimum Maximum

Panel A. FDI inflows (1970–79)


SSA 0.821 0.203 0.574 1.200
MENA 0.335 1.476 3.025 2.658
Panel B. FDI inflows (1980–99)
SSA 2.883 2.625 0.252 9.105
MENA 4.209 3.444 3.077 11.674
Source: World Bank (2015a).

This paper is largely motivated by Asiedu (2002), who examined the determinants of inward FDI to developing countries and
questioned whether Africa is different from other potential investment destinations. However, the present study differs in a
number of ways. Firstly, it focuses specifically on the determinants of FDI into the two least recipient regions. Both SSA and the
MENA region have embarked upon trade liberalization and reforms (adjustment programmes) at almost the same time, which
makes them more suitable for this study than the common approach of using a sample of developing countries with a huge
disparity in characteristics, levels of trade liberalization and reforms.1 Secondly, it incorporates an investigation of the
differences between the regions that arise from structural and behavioural factors. In addition, the paper departs from much of
the literature by using FDI per capita as the dependent variable as this allows for country size in a way that is not possible in the
standard FDI inflows as a percentage of GDP. The use of FDI per capita will also avoid any bias in estimates that might arise due
to the dominant importance of some large FDI recipient countries. Finally, comparing FDI flows between these regions
contributes to the literature as while there are several shared characteristics some factors differ, which presents a platform for
further incentives, reforms and complementarities.
Panel data estimation (fixed effects) was applied to a sample of 20 SSA and 11 MENA countries to determine the factors that
influence FDI inflows. Findings suggest that trade openness, infrastructure development, and control of corruption positively
influence FDI inflows. Surprisingly, rents from natural resources do not significantly influence FDI while inflation negatively
affects FDI. In addition, the null hypothesis that both regions are not behaviourally and structurally different in terms of FDI
determinants was rejected. When considered separately, SSA performed poorly compared with the MENA countries, with the
latter group attracting more FDI inflows.
The paper is organized as follows. Section 2 reviews the theoretical and empirical literature on the determinants of FDI.
Section 3 develops the hypotheses to be tested. Section 4 describes the variables and presents the preliminary data analysis.

Table 2: FDI inflows to developing and emerging regions ($ billions)


Panel A. FDI inflows (2000–12)
Regions Mean Std. dev. Coeff. of var. Minimum Maximum

SSA 5.92 9.11 1.54 0.74 28.70


East Asia and Pacific 66.30 104.00 1.57 1.39 328.00
Latin America and Caribbean 30.60 42.90 1.40 0.61 122.0
MENA 13.00 27.10 1.40 0.02 87.50
Europe and Central Asia 190.00 286.00 2.08 4.31 852.00
Panel B. Growth in FDI inflows for SSA and MENA regions
Regions 2000–2002 2003–2006 2007–2012

SSA 11.040 15.524 31.736


MENA 9.295 45.759 87.886
Source: World Bank (2015a).

© 2017 The Authors. African Development Review © 2017 African Development Bank
FDI Determinants in Least Recipient Regions 591

Section 5 specifies the models and reports the results followed by a discussion of the implications. The final section
concludes.

2. Determinants of FDI

2.1 A Brief Review of the Theoretical Literature


Braunerhjelm and Svensson (1996) note the complex nature of the theoretical foundation of FDI and the literature is now
fragmented across different areas of economics and international business. The earliest explanation of FDI inflows was from a
neoclassical trade theory perspective. The Heckscher–Ohlin model assumed that since commodities vary in relative factor
intensities and countries vary in relative factor abundance, capital will move to those countries where the return to capital is
higher and the return to labour is lower (Jones, 1957; Hodd, 1967). Aliber (1970) extended the discussion of why capital moves
across borders to include differences in the premium associated with exchange rate risk. Multinational firms in countries with
stronger currencies have an advantage over local firms in countries with weaker currencies since they can borrow capital with a
lower exchange rate risk premium (Harvey, 1990).
The neoclassical approach was criticized because of its inability to clarify the nature of FDI flows and was replaced with the
concept of oligopoly by Kindleberger (1969) and Hymer (1976) to provide a better explanation of why firms move across borders
(Faeth, 2009). Thus, firms will only operate internationally when they possess certain advantages over local firms and where the
market to explore these advantages is imperfect (Denisia, 2010). Buckley and Casson (1976) formulated a theory of
multinational enterprise within a broad-based intellectual framework defined as internationalization. This theory suggests that
firms internalize markets by bringing the activities linked by the market under common ownership and control and move abroad
if the expected benefits exceed the expected costs. Dunning (1979) combined these two concepts to create the eclectic paradigm,
which is a combination of traditional trade economics and internalization theory, which assumes that the likelihood of a firm
investing abroad is based on three main factors: the degree to which a firm owns an asset that its competitors do not; whether the
firm can benefit from selling or leasing these assets to other firms; and the level of rents that can be earned by exploiting these
assets. In all cases, the locational characteristics of the host country are important, where these include market size/market
growth, skilled labour, labour costs, synergistic/knowledge-related assets, availability/quality of infrastructure and natural
resources (Dunning, 1998; Sun et al., 2002; Dunning, 1980). It is the locational aspects of the eclectic paradigm that separates
this theory of FDI from the earlier market structure approaches based on oligopoly and monopoly (Faeth, 2009).
In addition, national policies have had an impact on the determinants of FDI and these have tended to concentrate on attracting
investment from abroad rather than emphasize differences in market structure. Hence, FDI can be regarded as a game between
the multinational firm and the host government, complicated by the competition between host countries for inward FDI, and
various inducements and incentives are frequently offered with the intention of influencing the decision of the firm to invest in a
particular location (Faeth, 2009). Exchange rates, tariffs and other trade barriers, taxes and the ease with which capital can
be repatriated are some of the ways through which host governments influence FDI activity (Lim, 2002). In terms of negative
influences, host governments that neglect to ensure a stable environment can deter investment as political risk is a disincentive
for firms wishing to undertake FDI (Khrawish and Siam, 2010).

2.2 Empirical Studies of FDI Determinants

There is a vast empirical literature on FDI that includes developed and developing countries with interests on various sectors and
for different time periods. However, the papers reviewed here focus solely on developing countries and regions as this is the
context of the present study. The topics specific to developing countries tend to concentrate on the impact of corruption, rate of
return, trade openness and natural resources with mixed findings on their relationship with FDI. Most emphasis has been on
market size, education and economic growth. For example, Tsen (2005) attribute the positive significance of human capital to
FDI to the fact that foreign investment does not only seek to reduce costs but also acquire access to technology and innovative
capacity. Conversely, Oke et al. (2012) find an insignificant relationship between education and FDI because of a lack of training
and integration in the pool of human capital in their sample. Akin (2009) argued that their findings that FDI is not related to GDP
per capita suggests that the small size of the market in low income countries is not an important determinant in the decision to
© 2017 The Authors. African Development Review © 2017 African Development Bank
592 G. Okafor et al.

invest internationally, although again this is sample specific. Sanfilippo (2010) argued that the significant relationship between
FDI and gross national income shows that their study supports the market size hypothesis. Srinivasan (2011) claimed that the
efforts by governments to increase economic growth and GDP per capita are successfully attracting market seeking FDI. In slight
contrast, Anyanwu and Yameogo (2015a) found a U-shaped relationship between GDP per capita and FDI. They argued that for
FDI to be positively related to GDP per capita, certain thresholds of GDP per capita should be attained.
With respect to infrastructure variables, there is further controversy. Adefeso and Agboola (2012), and Soremekun and
Malgwi (2012) find that the positive and significant relationship between infrastructure development (mobile users) and FDI
inflows is due to the fast penetration and adoption of mobile phones in the sample of developing countries they studied. However,
Wadhwa and Sudhakara (2011) used internet access as a measure of infrastructure and found a negative relationship to FDI. This
was justified by the fact that the developing countries in their sample have started using internet services extensively only in the
last few years and hence are yet to have a positive influence on FDI.
Finally, governance measures have been used extensively in FDI studies, and in particular with developing country samples.
Woo and Heo (2009) find a negative relationship between FDI and corruption in a sample of developing Asian countries and
suggested this was due to weak economic reforms, monopolistic power and rent-seeking behaviours of government officials, all
of which deter investors. Political instability was found to have a significant and negative impact on FDI in a study by Buthe and
Milner (2008). This is explained by increases in the uncertainty of the political environment that heightens the risk of policy
change and thus discourages FDI. Basemera et al. (2012) argue that the influence of free trade has been responsible for increased
levels of FDI in a sample of sub-regional governments. A similar line of argument was also adopted by Anyanwu and Yameogo
(2015b) following the positive relationship between FDI and trade openness in their study.

3. Hypothesis Development
The framework for the hypotheses was developed mainly according to the ownership, location and internalization (OLI)
paradigm although with emphasis on locational factors. Country-level studies can only explore the locational aspect of the OLI
paradigm. The literature on the location-specific variables of FDI suggests that infrastructure, human capital, natural resources,
market size, inflation, corruption, and trade openness influence the patterns of FDI inflows (Tsen, 2005; Mijiyawa, 2015).

H1: Larger market size/growth is positively associated with FDI inflows.

The size of the market can be measured by GDP growth rate or GDP per capita. It is expected that a positive relationship will
exist between market size and FDI inflows especially if FDI is motivated by market-seeking activities (Ranjan and Agrawal,
2011). However, while the growth rate of GDP or growth rate of per capita GDP is often argued to be a poor indicator for market
seeking FDI activity in developing countries due to the fact that it is difficult to differentiate in the data the strategic imperative
behind FDI, this study nevertheless hypothesizes that a positive relationship with FDI will be found (Akin, 2009).

H2: FDI is positively related to rents from natural resources.

Natural resources have been found to be important in attracting FDI, particularly in the African continent (Asiedu, 2006;
Nsiah and Wu, 2014). The regions under review in this study are rich in natural resources and this is the sector that has
historically attracted large amounts of FDI, especially the mineral and oil sectors. This study uses rents from natural resources as
a percentage of GDP to capture the availability of these resources.

H3: Infrastructure development stimulates FDI inflows.

Available infrastructure increases productivity and thus the return on investment. Therefore a positive relationship between
infrastructure and FDI is expected (Asiedu, 2002; Akin, 2009). However, the quality of infrastructure in these countries is highly
variable and a quality adjusted measure would be preferred. Unfortunately, data constraints limit the construction of this variable
and in common with the literature, infrastructure availability and or development is used. This is proxied by per capita mobile
phone users, as is established in similar studies.

H4: Human capital has a positive impact on FDI inflows.


© 2017 The Authors. African Development Review © 2017 African Development Bank
FDI Determinants in Least Recipient Regions 593

An educated workforce has been recognized as an important determinant of FDI especially when firms are efficiency seeking.
Srinivasan (2011) notes that a higher level of education can impact positively on FDI. The measure of human capital this study
uses is the number of technical education students per capita. For some of the countries employed in this study, FDI is also
attracted in technically oriented industries and not just in labour-intensive countries. The use of technical education students is
therefore justified because multinationals often seek to improve their technical efficiency particularly in the face of increased
competitive pressures. Furthermore, technical knowledge available in a country can positively impact on FDI flows because
innovation creates new demands for raw materials leading to FDI in their extraction and production. Such a technical knowledge
base can be supported through investments in skills acquisition. For example, large investments in education and training
increased the stock of skilled labour in some Asia-Pacific countries and, thus, helped increase their share of global FDI (Addison
and Heshmati, 2003).

H5: Trade openness has a positive impact on FDI inflows.

Countries with greater levels of trade openness and with more links to the world economy attract foreign capital and welcome
overseas investment (Srinivasan, 2011; Owusu-Antwi, 2012). Using the established measure of openness (exports plus imports
as a share of GDP), the study hypothesizes a positive relationship with FDI. Evidence of this has been provided by numerous
empirical studies for the regions under review. This is particularly important because both SSA and MENA have embarked on
adjustment programmes and trade liberalization over the past two decades and few barriers to trade remain in these regions.

H6: There is a positive relationship between control of corruption and FDI inflows.

Corruption can create a considerable barrier to investment. Corruption impedes investment directly and indirectly (Habib and
Zurawicki, 2002; Al-Sadig, 2009). Although several countries in this study are not known for their high levels of control of
corruption (Owusu-Antwi, 2012), this study nevertheless hypothesizes there is a positive relationship between control of
corruption and FDI as the latter can reduce uncertainty in investment activities.

H7: Foreign investors are less likely to invest in countries with high levels of inflation.

One of the indicators of stable macroeconomic environment is price stability. Foreign investors are often deterred by high
inflation because it erodes their return on investment, raises uncertainty and shows the inability of host government to implement
sound macroeconomic policies that are conducive to business activities (Azam, 2010).

4. Sample and Data

4.1 Sample Countries


Table 3 shows the sample of countries used in the analysis. The initial sample included all SSA and MENA countries but due to
missing data or because some of the values were outliers that would bias the estimates, a few countries were removed. This also
guided the chosen time period for the study. Regarding the outliers, preliminary regression plots of the standardized residuals
against the fitted values confirms that Bahrain and Qatar are outliers and thus were excluded from the sample in the subsequent
analysis. Bahrain and Qatar are likely to be outliers because of their high GDP per capita and thus they do not fit with the
developing country profile of the remainder of the sample.

Table 3: Sample countries


MENA region Algeria, Bahrain , Egypt, Iran, Jordan, Kuwait, Libya, Morocco, Qatar , Saudi Arabia, Syria, Tunisia and Yemen


SSA region Angola, Botswana, Burkina Faso, Burundi, Chad, Ethiopia, Ghana, Guinea, Kenya, Lesotho, Mali, Mauritania,
Mozambique, Niger, Rwanda, Senegal, South Africa, Sudan and Uganda
Note:  These countries are outliers and excluded from the analysis.

© 2017 The Authors. African Development Review © 2017 African Development Bank
594 G. Okafor et al.

Table 4: Variable definitions


Variable Definition

FDI Inflow per Capita FDI inflows by country divided by the total host country population ($) (UNCTAD, 2015)
% Population in Vocational or Technical Percentage of population enrolled in technical and vocational education (World Bank,
Education 2015a)
Resource Rent Total natural resources rent are the sum of oil rents, natural gas rents, coal rents, mineral
rents and forest rents as a percentage of GDP (World Bank, 2015a)
% Population of Mobile Phone Users Percentage of population using mobile telephones either on a post-paid or prepaid basis,
proxies infrastructure (World Bank, 2015a)
Trade Openness Sum of imports plus exports as a percentage of GDP, proxies the degree of liberalization, as
in Srinivasan (2011)
Control of Corruption Measures the extent to which public power for personal gain is controlled (World Bank,
2015b). Ranges from 0 (lowest) to 100 (highest) rank.
Inflation Annual percentage change in the cost of consumer goods and services (World Bank, 2015a)
GDP per Capita GDP per capita is gross domestic product divided by mid-year population (World Bank, 2015a)

4.2 Variable Description

Given the widely different sizes of the countries under review it is important that the variables used take account of
population size in order that comparisons are valid and useful. In addition, levels of development are not constant and some
countries have higher income levels than others. Thus, the majority of variables in the modelling are considered on the basis
of percentage of total population or values per capita. Data on FDI inflows, pupils in technical education, and mobile users
are expressed in per capita terms. The data were obtained from the World Development Indicators (World Bank, 2015a),
UNCTAD (2015), and the World Bank Governance Indicators (World Bank, 2015b). Data definitions and sources are in
Table 4.

4.3 Preliminary Data Analysis


Table 5 reports the descriptive statistics for the variables used in the estimation. It is clear that the MENA region has a higher
level of development at the mean, with many values greater than SSA. In particular, the extent of FDI, human capital,
infrastructure development, resource rents, and GDP per capita are greater in the MENA sample. The mean trade openness is
similar, although the SSA sample has a much higher dispersion. The institutional governance variables — that is, control of
corruption — in this study is higher in the MENA region although the differences are not great. At the mean, inflation is lower in

Table 5: Descriptive statistics


Sample countries
(total) MENA SSA
Variables Mean Std. Dev. Min Max Mean Std. Dev. Mean Std. Dev.

FDI Inflow per Capita ($ US) 79.402 170.693 331.306 1458.000 151.736 251.108 39.618 78.575
GDP per Capita 3350.042 7233.623 108.015 56366.570 7275.122 10921.180 1191.248 1624.291
Resource Rent (% of GDP) 19.402 17.599 0.256 71.605 27.878 19.744 14.740 14.336
Infrastructure Development 37.079 41.447 0.019 193.453 57.731 50.423 25.720 30.122
(% of Mobile Users)
% of Population in Technical Education 0.546 0.798 0.012 4.599 1.102 1.081 0.241 0.291
Trade Openness 74.641 34.157 19.356 202.850 76.786 22.025 73.461 39.255
Control of Corruption 37.434 21.222 0.957 85.854 41.208 19.992 35.358 21.627
Inflation 9.375 20.108 9.798 324.997 5.835 6.405 11.322 24.380

© 2017 The Authors. African Development Review © 2017 African Development Bank
FDI Determinants in Least Recipient Regions 595

Table 6: Correlation matrix


1 2 3 4 5 6 7 8

1 FDI Inflow per Capita ($ US) 1.000


2 GDP per Capita 0.484 1.000
3 Resource Rent (% of GDP) 0.259 0.482 1.000
4 Infrastructure Development (% of Mobile Users) 0.600 0.590 0.233 1.000
5 % of Population in Technical Education 0.121 0.108 0.317 0.188 1.000
6 Trade Openness 0.234 0.119 0.103 0.156 0.028 1.000
7 Control of Corruption 0.244 0.324 0.334 0.276 0.069 0.215 1.000
8 Inflation 0.013 0.078 0.200 0.096 0.050 0.159 0.219 1.000

the MENA region including the variation from the mean. Correlation coefficients are listed in Table 6. The coefficients show no
high collinearity between the variables.

5. Models, Estimation and Results

5.1 Panel Specification

The models use a balanced panel of 20 SSA and 11 MENA countries. The data are annual for the period 2000–12. As already
identified above, this was mainly due to data availability for some of the variables. Fixed effects estimation was used as
the random effects estimator was rejected based on the Hausman test. Panel models are valuable for a number of reasons.
Firstly, panel data allow both the cross-section and the time series aspects of the data to contribute to the parameter estimates.
Secondly, panel data suggest that countries are heterogeneous. Time series and cross-section studies not controlling for this
heterogeneity run the risk of obtaining biased results. Not accounting for country-specific differences in economic or
behavioural assumptions, such as countries operating under different political systems or more or less restrictive regulations, can
cause serious mis-specification in the models. Thirdly, it may be important to incorporate dynamic effects and these models
provide a means to study the dynamics of adjustment (Greene, 1997).
Given the differences between the regions as highlighted by the descriptive statistics (summarized in Table 5), it is important
that the models take into consideration possible heterogeneity across countries in order to reduce the risk of obtaining biased
estimates. The fixed effects model data used also allows the intercept to vary for each individual country but still assumes that the
slope coefficients are constant across the sample. The estimating equation can be expressed

yit ¼ ai þ bX it þ mi þ vit ð1Þ

where y is FDI inflows per capita in country i at time t, X is a matrix of independent variables and a and b are coefficients to be
estimated. mi and vit represent the decomposed disturbance term where mit are country specific effects and vit are random errors
distributed (Gujarati, 2004).
Equation (1) was first estimated with and without the SSA dummy. With respect to statistical tests, the Chow Test showed that
SSA and the MENA countries are behaviourally and structurally different based on the F test and critical values. Given these
statistical differences, the significant variables were interacted2 with the SSA dummy to establish any differences in marginal
effects. These behavioural and structural differences could also have accounted for the differences at the means and variations
already presented in the preliminary data analysis above.
In the specification tests, all models are acceptable. A Breusch-Pagan/Cook-Weisberg Test for heteroscedasticity indicated
the presence of heteroscedasticity and therefore robust standard errors were used to relax the assumptions that the errors were
both independent and identically distributed. The GMM results were not reported in this study as the estimates and
instruments were inefficient and inconsistent. The inconsistency and inefficiency were not surprising considering that GMM
fits better for panels with large numbers of cross-section (N) and small time-series (T). However, the fixed effects technique
© 2017 The Authors. African Development Review © 2017 African Development Bank
596 G. Okafor et al.

used is known to control for possible heterogeneity. Tests also revealed no statistical mis-specification of the model and no
omitted variable bias. Thus, the conclusion can be drawn that the results obtained from the fixed effects estimation are
consistent and not spurious.

5.2 Results and Discussion


The results are in Tables 7 and 8. The SSA dummy is negative and significant. This implies that all things being equal, the SSA
region on average receives less FDI compared to the MENA region. H1 tested the importance of market size. The GDP per capita
is positive and significant. Coefficient of the interaction between SSA dummy and GDP per capita was negative and significant.
This shows that the marginal effect of GDP per capita on FDI inflows is less in SSA than the MENA region. To further support

Table 7: Fixed effects estimations (robust standard errors)


Fixed Fixed Fixed Fixed Fixed Fixed Fixed
FDI Inflow per Capita ($ US) Effects Effects Effects Effects Effects Effects Effects

Independent Variables Model 1 Model 2 Model 3 Model 4 Model 5 Model 6 Model 7

GDP per Capita 0.014 0.014 0.014 0.013 0.014 0.013 0.014
(0.007) (0.007) (0.007) (0.008) (0.008) (0.008) (0.007)
Resource Rent (% of GDP) –0.137 –0.137 0.119 0.294 –0.554 –0.099 –0.334
(0.965) (0.965) (0.987) (1.150) (0.893) (0.968) (1.048)
Infrastructure Development 2.199 2.199 2.379 2.284 1.795 2.203 2.249
(% of Mobile Users) (0.687) (0.687) (0.687) (0.715) (0.661) (0.681) (0.686)
% of Population in Technical Education –37.269 –37.269 –16.384 –27.983 –44.698 –36.631 –42.864
(30.798) (30.798) (29.431) (34.563) (32.431) (30.222) (32.913)
Trade Openness 1.969 1.969 1.628 1.893 6.052 1.974 1.996
(0.479) (0.479) (0.497) (0.489) (1.302) (0.475) (0.480)
Control of Corruption 66.477 66.477 68.859 69.932 42.325 58.554 62.978
(33.589) (33.589) (33.564) (34.208) (30.529) (77.767) (35.235)
Inflation –0.606 –0.606 –0.446 –0.553 –0.517 –0.610 –0.899
(0.229) (0.229) (0.227) (0.232) (0.236) (0.229) (0.325)
SSA Dummy –199.706
(67.821)

GDP per Capita SSA –0.031
(0.014)
Infrastructure Development (% of Mobile –0.513
Users)  SSA
(0.579)

Trade Openness SSA –4.944
(1.342)

Control of Corruption SSA 12.212
(74.842)

Inflation SSA 0.368
(0.384)
Cons 47.14627 –53.569 –91.259 –82.554 –292.237 –60.238 –16.061
29.12621 (64.803) (67.713) (77.546) (99.915) (71.414) (79.383)
No. of Obs. 403 403 403 403 403 403 403
F Stat 13.25 7.77 7.71 7.62 8.52 7.61 7.64
Prob. > F 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000 0.0000
R-Squared 0.4097 0.6650 0.6713 0.6664 0.6865 0.6651 0.6655
Notes: Robust standard errors are in parentheses;  Significance at the 10% level; 
Significance at the 5% level; 
Significance at the 1% level.

© 2017 The Authors. African Development Review © 2017 African Development Bank
FDI Determinants in Least Recipient Regions 597

Table 8: Marginal effects of the significant variables


Variables MENA SSA

GDP per Capita 0.014 0.017


(0.007) (0.015)
Infrastructure Development (% of Mobile Users) 2.284 1.771
(0.715) (0.713)
Trade Openness 6.052 1.108
(1.302) (0.401)
Control of Corruption 58.554 70.765
(77.767) (21.009)
Inflation 0.899 0.530
(0.325) (0.257)
Notes: Robust standard errors are in parentheses;  Significance at the 10% level; 
Significance at the 5% level; 
Significance at the 1% level.

the result of the interaction, the estimated partial coefficient of GDP per capita in SSA was insignificant compared to the positive
and significant effect in MENA. These findings most likely reflect that the higher levels of disposable income in the MENA
region attract FDI for market seeking opportunities. H2 tested the impact of rents from natural resources on FDI inflows.
Surprisingly, this was insignificantly related to FDI. These findings were similar to those of Asiedu and Lien (2011) and can be
justified with similar arguments. Huge rents generated from natural resources can lead to the appreciation of local currency and
thus can diminish the competitiveness of exports. This results in the crowding out of investments in non-natural resource tradable
sectors. Some of the countries in the sample often attract huge FDI inflows into their resource sectors and thus, while the
exploration of natural resources initially comes with high capital outlay, continued operations within that sector are usually
accompanied by smaller cash flows. Also, countries with a significant share of natural resources in total merchandise exports are
more prone to external shocks since they are weak in trade diversification. These shocks create macroeconomic instability and
can lead to the decline in FDI.
Infrastructure development is clearly important as a determinant of FDI inflows and it is positive and significant although SSA
has a lower elasticity overall than the MENA region. The interaction between infrastructure development and the SSA dummy
was negative and the estimated partial coefficient was higher for MENA countries compared to SSA countries. The findings are
expected and support the literature on FDI, particularly on the relationship between infrastructure and investment from
developed to developing countries. The impact of human capital as a predictor of FDI was tested in H4. The results showed that
technical education has an insignificant effect on FDI. The results suggest that human capital in these regions has not yet reached
the required threshold in technical education to stimulate efficiency and attract skill-seeking FDI. A test of H5 showed that trade
openness is positive and significant although SSA has a lower elasticity. That is, the result of the interaction between the SSA
dummy and trade openness was negative and significant. These findings nevertheless, demonstrate the importance of trade
openness in attracting FDI.
Results also showed that control of corruption has a positive influence on FDI inflows, however the marginal effect of
corruption control on FDI is significantly higher in SSA countries compared to MENA countries. The result of the interaction
between control of corruption and the SSA dummy is positive and the partial coefficient is positive but insignificant for the
MENA countries. These findings support the empirical evidence that controlling for corruption can be a means through which
FDI can be positively influenced. Foreign investors perceive SSA countries to be very corrupt and thus, genuine efforts against
corruption will have the most impact on investment in SSA. The impact of inflation was negative and significant. The partial
coefficient shows that the negative impact of inflation will be higher in the MENA region compared to the SSA region. Inflation
creates macroeconomic instability, reduces buying power, and erodes the return on investment.

6. Conclusion and Policy Implications


This study investigates the determinants of FDI inflows into two of the least recipient regions, SSA and MENA. The findings
reveal that infrastructure development, trade openness and control of corruption encourage FDI into these regions. On the other
© 2017 The Authors. African Development Review © 2017 African Development Bank
598 G. Okafor et al.

hand, inflation deters FDI while rents from natural resources do not significantly attract FDI. Some of these findings are not
consistent with the stated hypotheses. The analyses also considered whether the regions are behaviourally and structurally
different and if so, how they compare in their FDI determinants. The results confirmed differences between the two regions and
that the marginal benefits from increases in the quality of FDI determinants will be less for SSA countries compared to the
MENA countries.
A number of policy implications follow from these findings. First, trade openness and control of corruption are very important
determinants of FDI and, thus, efforts targeted at reducing corruption and improving trade policies should be seriously pursued.
Second, since rent from resources does not significantly influence FDI, the SSA and MENA regions should encourage more trade
diversification by pursuing policies that will increase the competitiveness of the non-resource tradable sectors. Third, serious
attention should be paid to technical education because countries with high levels of low-skilled labour are less likely to be
attractive to foreign investors focusing on high value-added industries or FDI inflows motivated by efficiency seeking. This is
also important because of the spillovers that flow to host country firms from FDI in high-skilled sectors that contribute more
value added than those from low-skilled sectors. Finally, all things being equal, SSA countries will attract less FDI compared to
MENA countries and thus it is crucial that countries and sub-regional blocs in SSA direct their efforts towards programmes that
improve their image as international partners. Therefore they should introduce credible policies targeted at restoring confidence
and maintaining global relationships, thus countering the negative perception of SSA as a region.

Notes
1. For further details that African countries embarked on adjustment programmes in the 1980s see Nsouli and Zulu (1985) and
Campbell and Loxley (1989).
2. An interaction is formed as of a product off two (or more) variables. An important application of the interaction variables is
that it allows for differences in the slopes of two regression lines. For further reading, see Dielman (2005).

References
AbuAl-Foul, B. M. and M. Soliman (2008), ‘Foreign Direct Investment and LDC Exports: Evidence MENA Region’, Emerging
Markets Finance & Trade, Vol. 44, No. 2, pp. 4–14.
Addison, T. and A. Heshmati (2003), ‘The New Global Determinants of FDI Flows to Developing Countries: The Importance of
ICT and Democratisation (No. 2003/45)’, WIDER Discussion Papers//World Institute for Development Economics
(UNU-WIDER)
Adefeso, H. A. and A. A. Agboola (2012), ‘Determinants of Foreign Direct Investment Inflows in Nigeria: An Empirical
Investigation’, International Business Management, Vol. 6, No. 1, pp. 83–89.
Akin, M. S. (2009), ‘How Is the Market Size Relevant as a Determinant of FDI in Developing Countries? A Research on
Population and the Cohort Size’, paper presented at the International Symposium on Sustainable Development, 9–10 June,
Sarajevo, Bosnia and Herzegovina.
Akinlo, A. E. (2004), ‘Foreign Direct Investment and Growth in Nigeria: An Empirical Investigation’, Journal of Policy
Modeling, Vol. 26, No. 5, pp. 627–39.
Aliber, R. Z. (1970), ‘A Theory of Direct Foreign Investment’, The International Corporation, Vol. 4, pp. 17–37.
Al-Sadig, A. (2009), ‘The Effects of Corruption on FDI Inflows’, Cato Journal, Vol. 29, No. 2, pp. 267–94.
Anyanwu, J. C. and N. D. Yameogo (2015a), ‘What Drives Foreign Direct into West Africa? An Empirical Investigation’,
African Development Review, Vol. 27, No. 3, pp. 199–215.
Anyanwu, J. C. and N. D. Yameogo (2015b), ‘Regional Comparison of Foreign Direct Investment to Africa: Empirical
Analysis’, African Development Review, Vol. 27, No. 4, pp. 345–63.
Asiedu, E. (2002), ‘On the Determinants of Foreign Direct Investment to Developing Countries: Is Africa Different?’, World
Development, Vol. 30, No. 1, pp. 107–19.
© 2017 The Authors. African Development Review © 2017 African Development Bank
FDI Determinants in Least Recipient Regions 599

Asiedu, E. (2006), ‘Foreign Direct Investment in Africa: The Role of Natural Resources, Market Size, Government Policy,
Institutions and Political Instability’, The World Economy, Vol. 29, No. 1, pp. 63–77.
Asiedu, E. and D. Lien (2011), ‘Democracy, Foreign Direct Investment and Natural Resources’, Journal of International
Economics, Vol. 84, No. 1, pp. 99–111.
Azam, M. (2010), ‘Economic Determinants of Foreign Direct Investment in Armenia, Kyrgyz Republic and Turkmenistan:
Theory and Evidence’, Eurasian Journal of Business and Economics, Vol. 3, No. 6, pp. 27–40.
Basemera, S., J. Mutenyo, E. Hisali and E. Bbaale (2012), ‘Foreign Direct Investment Inflows to East Africa: Do Institutions
Matter?’, Journal of Business and Management and Applied Economics, Vol. 1, No. 5, pp. 49–71.
Braunerhjelm, P. and R. Svensson (1996), ‘Host Country Characteristics and Agglomeration in Foreign Direct Investment’,
Applied Economics, Vol. 28, No. 7, pp. 833–40.
Buckley, P. J. and M. Casson (1976), The Future of the Multinational Enterprise, Macmillan, London.
Buthe, T. and H. V. Milner (2008), ‘The Politics of Foreign Direct Investment into Developing Countries: Increasing FDI
through International Trade Agreements?’, American Journal of Political Science, Vol. 52, No. 4, pp. 741–62.
Campbell, B., and J. Loxley (eds.) (1989), Structural Adjustment in Africa, Palgrave Macmillan, London.
Denisia, V. (2010), ‘Foreign Direct Investment Theories: An Overview of the Main FDI Theories’, European Journal of
Interdisciplinary Studies, Vol. 2, No. 2, pp. 104–10.
Dielman, T. E. (2005), Applied Regression Analysis: A Second Course in Business and Economic Statistics, Brooks/Cole
Thomson Learning, Belmont, CA.
Dunning, J. H. (1979), ‘Explaining Changing Patterns of International Production: In Defence of the Eclectic Theory’, Oxford
Bulletin of Economics and Statistics, Vol. 41, No. 4, pp. 269–95.
Dunning, J. H. (1980), ‘Toward an Eclectic Theory of International Production: Some Empirical Tests’, Journal of International
Business Studies, Vol. 11, No. 1, pp. 9–31.
Dunning, J. H. (1998), ‘Location and the Multinational Enterprise: A Neglected Factor?’, Journal of International Business
Studies, Vol. 29, No. 1, pp. 45–66.
Dupasquier, C. and P. Osakwe (2006), ‘Foreign Direct Investment in Africa: Performance, Challenges and Responsibilities’,
Journal of Asian Economics, Vol. 17, No. 2, pp. 241–60.
Faeth, I. (2009), ‘Determinants of Foreign Direct Investment – A Tale of Nine Theoretical Models’, Journal of Economic
Surveys, Vol. 23, No. 1, pp. 165–96.
Greene, W. H. (1997), Econometric Analysis, Prentice-Hall International, London.
Gujarati, D. N. (2004), Basic Econometrics, 4th edn, McGraw-Hill, New York.
Habib, M. and L. Zurawicki (2002), ‘Corruption and Foreign Direct Investment’, Journal of International Business Studies,
Vol. 33, No. 2, pp. 291–307.
Harvey, J. T. (1990), ‘The Determinants of Direct Foreign Investment’, Journal of Post Keynesian Economics, Vol. 12, No. 2,
pp. 260–72.
Hodd, M. (1967), ‘An Empirical Investigation of the Heckscher-Ohlin Theory’, Economica, Vol. 34, No. 133, pp. 20–29.
Hymer, S. H. (1976), ‘The International Operations of National Firms: A Study of Direct Investment’, MIT Press,
Cambridge, MA.
Jones, R. W. (1957), ‘Factor Proportions and the Heckscher-Ohlin Theorem’, The Review of Economics Studies, Vol. 24, No. 1,
pp. 1–10.
Kandiero, T. and M. Chitiga (2006), ‘Trade Openness and Foreign Direct Investment in Africa’, South Africa Journal of
Economic and Management Sciences, Vol. 9, No. 3, pp. 355–70.
© 2017 The Authors. African Development Review © 2017 African Development Bank
600 G. Okafor et al.

Khrawish, H. and W. Z. Siam (2010), ‘Determinants of Direct Foreign Investment: Evidence from Jordan’, Business and
Economic Horizons, Vol. 1, No. 1, pp. 67–75.
Kindleberger, C. P. (1969), American Business Abroad: Six Lectures on Direct Investment, Yale University Press,
New Haven, CT.
Lim, S. H. (2002). ‘How Investment Promotion Affects Attracting Foreign Direct Investment: Analytical Argument and
Empirical Analyses’, International Business Review, Vol. 17, No. 1, pp. 39–53.
Mijiyawa, A. G. (2015), ‘What Drives Foreign Direct Investment in Africa? An Empirical Investigation with Panel Data’,
African Development Review, Vol. 27, No. 4, pp. 392–402.
Nsiah, C. and C. Wu (2014), ‘The Role of Spatial Dynamics in the Determination of Foreign Direct Investment Inflows to
Africa’, African Development Review, Vol. 26, No. 3, pp. 494–507.
Nsouli, M. S. M. and J. B. Zulu (1985), ‘Adjustment Programs in Africa: The Recent Experience’, International Monetary Fund
Occasional Paper No. 34.
Oke, B. O., J. E. Ezike and S. O. Ojogbo (2012), ‘Locational Determinants of Foreign Direct Investment in Nigeria’,
International Business Research, Vol. 5, No. 4, pp. 103–11.
Owusu-Antwi, G. (2012), ‘Determinants of Foreign Direct Investment: Is it a Better Prescription for Economic Growth in
Africa?’, International Business and Economic Research Journal, Vol. 11, No. 7, pp. 757–70.
Ranjan, V. and G. Agrawal (2011), ‘FDI Inflow Determinants in BRIC Countries: A Panel Data Analysis’, International
Business Research’, Vol. 4, No. 4, pp. 255–63.
Sanfilippo, M. (2010), ‘Chinese FDI to Africa: What is the Nexus with Foreign Economic Cooperation?’, African Development
Review, Vol. 22, No. 1, pp. 599–614.
Soremekun, O. and C. Malgwi (2012), ‘Exploring the Relationship Between Foreign Direct Investment and Mobile Technology
in Africa: An Application of Directed Acyclic Graphs’, Case Studies in Business, Industry and Government Statistics, Vol. 5,
No. 1, pp. 58–66.
Srinivasan, P. (2011), ‘Determinants of Foreign Direct Investment in SAARC Nations: An Econometric Investigation’, The IUP
Journal of Managerial Economics, Vol. 9, No. 3, pp. 26–42.
Sun, Q., W. Tong and Q. Yu (2002), ‘Determinants of Foreign Direct Investment across China’, Journal of International Money
and Finance, Vol. 21, No. 1, pp. 79–113.
Tsen, W. H. (2005), ‘The Determinants of Foreign Direct Investment in the Manufacturing Industry of Malaysia’, Journal of
Economic Cooperation, Vol. 26, No. 2, pp. 91–110.
UNCTAD (2015), World Investment Report 2015, United Nations, Geneva.
Wadhwa, K. and R. S. Sudhakara (2011), ‘Foreign Direct Investment into Developing Asia Countries: The Role of Market
Seeking, Resource Seeking and Efficiency Seeking Factors’, International Journal of Business and Management, Vol. 6, No. 11,
pp. 219–26.
Woo, J. and U. Heo (2009), ‘Corruption and Foreign Direct Investment Attractiveness in Asia’, Asian Politics and Policy, Vol. 1,
No. 2, pp. 223–38.
World Bank (2015a), ‘World Bank Development Indicators’, World Bank, Washington DC.
World Bank (2015b), ‘World Bank Governance Indicators’, World Bank, Washington DC.

© 2017 The Authors. African Development Review © 2017 African Development Bank
African Development Review, Vol. 29, No. 4, 2017, 601–614

Quantifying the Road Influence Zone on Socio-economic Developments


in Rural Tigray, Ethiopia

Fredu Nega Tegebu and Edris Hussein Seid

Abstract: The road influence zone has been used to analyze the influence of roads on ecological and environmental effects
but not on socio-economic conditions. In a first study of its kind, we measured the extent and type of relationship underlying the
road influence zone of rural roads on socio-economic outcomes. Four indicators — namely, trip per capita, use of fertilizers,
motorized transport and commercial activities — and three indicators — namely, road dust, flooding and erosion — have been
used to analyze the positive and negative effects of roads respectively. Four rural roads and 529 households from four tabias
were surveyed and we used piecewise and linear regressions to determine delineation of road influence zones. Except for trips
per capita and erosion, threshold effects were observed for the rest of the outcome variables. The road influence zone ranges from
240 meters on both sides of the road in the case of road dust to about 2.6 kilometers in the case of motorized transport. The study
results suggest that socio-economic impacts of roads differ not only on socio-economic and wealth differences of households but
also on distance to road.

1. Introduction
In most African countries, road transport network and road density measured per person and per square kilometer of land area is
very low compared to the global average. Yet the road transport sector is the dominant means of transport in the continent,
carrying around 80 to 90 percent of passenger and freight traffic; and most rural communities are accessible only through roads
(Gwilliam, 2011). Recently countries in the continent are making huge progress in allocating funds and building road
infrastructure, but still rural road accessibility remains one of the challenges to eradicate extreme poverty and achieve the
sustainable development goals (SDGs).
Cognizant of the importance of roads, the Government of Ethiopia has embarked on massive investment on road
construction and formulated a large-scale infrastructure development program, the Road Sector Development Program
(RSDP) in 1997. Since then, the RSDP has been implemented in four separate phases, and as part of the fourth RSDP, a
Universal Rural Roads Access Program (URRAP), envisaged to connect rural kebeles by standard rural roads, has been set
out and implemented.
Previous empirical works conducted in various countries show that rural roads infrastructure development reduced poverty
and improved the quality of life, especially for the poor, and narrowed down the income gap between citizens (Calderon and
Serven, 2014). In their study in rural Ghana, Menash et al. (2014) found that infrastructure access has important but differential
impacts on household welfare, indicating that better infrastructure such as roads are not sufficient but necessary conditions to
benefit the poor. The ability of the poor to make significant economic use of a road depends on their asset base and the
entitlements to resources and opportunities that they can command.
Besides socio-economic conditions of the poor, distance of location of households from road also affects their ability to
harness the direct and indirect benefits of roads. In the rural areas where settlement of rural households is highly spread, some
households may travel hours on foot to get a road. Households located far from a road, for example, may not have similar
motivation to use motorized transport compared to households located close to a road due to long distance they have to walk
before they get to a road. Thus the socio-economic impact of a road on both sides may become different as we go outwards from


Fredu Nega Tegebu (corresponding author), Horn Economic and Social Policy Institute (HESPI), PO Box 2692 Code 1250, Addis Ababa,
Ethiopia; e-mail: fredu.nega@hespi.org. Edris Hussein Seid, Horn Economic and Social Policy Institute (HESPI); e-mail: eddrissa@gmail.com
© 2017 The Authors. African Development Review © 2017 African Development Bank. Published by Blackwell Publishing Ltd,
9600 Garsington Road, Oxford OX4 2DQ, UK and 350 Main Street, Malden, MA 02148, USA. 601
602 F. N. Tegebu and E. H. Seid

the road. If the change follows some pattern, it is possible to delineate the socio-economic impact of roads in an attempt to
formulate a road influence zone.
The road influence zone, defined as the area over which significant ecological, environmental, socio-economic effects
extend outward from a road (Forman and Deblinger, 2000), is widely used to assess ecological and environmental
effects of roads. However, this has not been used to analyze the socio-economic influence of roads. This study intends to
use the methodology to quantify the socio-economic influence zone of rural roads in two woredas1 in northern
Ethiopia. Four rural roads were selected for the study — two of the roads connect rural villages to main road and the
other two roads connect two rural woredas. The study focuses on mobility and other socio-economic impacts of roads
including some roadside effects. It is the first attempt to quantify the influence zone of roads on socio-economic
outcomes.
The rest of the paper is divided into five sections. Section 2 reviews relevant literature followed by Section 3 which describes
the data source and methodology. Results and discussion are presented in Section 4 and Section 5 concludes.

2. Empirical Evidence
Infrastructure investments, especially in rural road development, enhance access to markets for inputs such as fertilizers and
improved seeds and enables farmers to sell their produce to nearby markets through a reduction in transport fare and time
(Raballand et al., 2010). On the other hand, lack of road network can lead to increased transaction cost in rural areas which results
in limited market access for farmers (Key et al., 2000). In developing countries like Africa, transport costs constitute a significant
proportion of marketing costs. For example, in Benin, Madagascar and Malawi, transport cost constitutes more than half of the
marketing costs (Fafchamps et al., 2005). Hence such roads are vital in improving agricultural productivity and raising living
standards in poor rural areas (Gannon and Liu, 1997), enhance intra-regional trade (Akpan, 2014) and increase competitiveness
(Yogo and Verdier-Chouchane, 2015).
Rural roads also allow farmers to achieve additional non-farm employment opportunities, leading to a rise in income and
reduce rural poverty (Ali and Pernia, 2003). Rural roads improve mobility which in turn facilitates access. In Uganda, Fan and
Zhang (2008) found that the impact of low-grade roads such as feeder roads is larger than that of high-grade roads such as
murram and tarmac roads.
Using Generalized Methods of Moments and controlling for household fixed effects, Dercon et al. (2008) found that access to
all-weather roads reduces poverty by 6.9 percentage points and increases consumption growth by 16.3 percent in Ethiopia.
Improvement in road infrastructure resulting from a large-scale public investment program like RSDP also contributed
positively to the size and structure of the manufacturing sector in Ethiopia (Admasu et al., 2012). Bryceson et al. (2008) found
that in extremely remote areas, road improvements may catalyze the expansion of social-service provision, as evidenced in
Ethiopia.
Although the impact of roads on socio-economic development as evidenced in the literature review above and elsewhere is
not disputable, the extent to which such influences extend on both sides of a road is not clear. Most empirical evidences use
quasi-experimental methods and wealth differentiated approaches to analyse impact of roads on socio-economic developments.
Attempts to delineate the road influence on the basis of distance to road are scanty at best. This study is the first attempt to
quantify and map out the influence zone of roads on socio-economic outcomes.

3. Data Source and Methodology

3.1 Data Source

To account for geographical context, two woredas namely Raya Azebo and Kilte Awlaelo in Tigray regional state in northern
Ethiopia were selected. The two woredas have diverse topography and economic characteristics. Kilte Awlaelo is situated at
higher altitude, features mountainous terrain and is relatively intensively irrigated. Raya Azebo is lower, more arid and includes
pastoralist activities. Two types of rural roads and four tabias2 — namely, Hade Alga, Werebaye, Lailay Adiksanded and Mai
Quiha — were selected from the two woredas. Figure 1 shows the location of the study sites. Our study sites are representative in
terms of their connectivity and distance to localities of economic importance through a network of rural roads. The roads selected

© 2017 The Authors. African Development Review © 2017 African Development Bank
Quantifying the Road Influence Zone 603

Figure 1: Location of study sites

in the study sites include the two types of rural roads in the area — rural roads that connect village centres to regional highways
and those that connect two woredas. In two of the study villages — Adiksanded and Worebaye — the roads connect the village
centers to regional highways, whereas, in the remaining two villages, rural roads that connect two woredas cross the village
centers.
A total of 529 sample households were selected from the four study tabias along the four roads. Sample households were
distributed proportionately to each tabia and in each tabia, sample households were further proportionately distributed to each
sub-village locally called kushet. Once sample households were proportionately distributed to each kushet, simple random
sampling was used to select sample households from each kushet. List of households was obtained from each kushet and sample
households were selected at regular intervals with a random start.
A multi-purpose questionnaire was used to gather information on household demographics, mobility, household activities and
income; and impact of roads on selected welfare indicators. The survey questionnaire was administered to the 529 sample
households selected from the four tabias.

3.2 Methodology

Both descriptive and statistical analysis were used to summarize the influence of roads on the socio-economic conditions of rural
households. Tables and figures were used for the descriptive analysis. For the statistical analysis of the road influence zone on
selected welfare indicators and roadside effects, we built a generalized linear regression and, when appropriate, a generalized
linear piecewise regression model for each indicator. Because the distribution of the response varied among the welfare and

© 2017 The Authors. African Development Review © 2017 African Development Bank
604 F. N. Tegebu and E. H. Seid

roadside effects, we used a Gaussian, Poisson, Binomial or Negative Binomial link function as appropriate. We used the glm
function for the regressions.
A piecewise linear regression with one breakpoint (Muggeo, 2003 and Toms and Lesperrence, 2003, cited in Eigenbrod et al.,
2009) is expressed as:

Y ¼ a þ b1 X 1 þ . . . þ bj X j þ dD where D  breakpoint T

Y ¼ a þ b1 X 1 þ . . . þ bj X j þ dD þ ðd e ÞðD  T Þwhere D > T

where Y is the response, a is the intercept; b1 ; . . . ; bj are the regression coefficients for all X j local predictors; d is the slope of
distance to the road (D) in the piecewise regression model to the left of the breakpoint (T), and e is ‘difference in slope parameter’
so that d þ e is the slope of the line segment to the right of the breakpoint.

4. Results and Discussion

4.1 Descriptive Result


The economic and social benefits of a road to individual households depends among others on how far or close a household is
located from the road. In this section, we discuss how the influence of rural roads varies with distance of households to road and
whether or not a threshold effect exists.

Distance to Road and Travel Patterns


Trips are defined as travel both inside and outside the village by any means for the following purposes: economic activities,
social services, social activities, and community association. For our analysis, we aggregated all types of trips to report travel
patterns. Table 1 summarizes travel patterns by distance to road categorized into close, medium and long. The categorization is
made in such a way that there are comparable number of households in each category. The first category includes households
located up to 0.3 km (300 m) from a road. The medium category lies in the range of 0.31 to 1.5 km, while that of the long distance
covers households located 1.51 km and further from the road.
The table indicates that the number of trips household members make correlate with distance to road. The closer households
are to the road, the higher is the number of trips they make. Households located close to the road made on average 79 trips per
month while households in the medium and long distance ranges made 77 and 73 trips per month respectively and the difference
is statistically significant. This means households located close to the road made on average five more trips per month than
households located more than 1.5 km from the road. On a per capita basis, that is, when total number of trips in a household is

Table 1: Travel patterns by distance from road


Distance from road
Travel pattern Close Medium Long F-value

Average no. of trips per month 78.78 77.30 73.09 4.08


Average no. of trips per month per capita 14.92 13.48 12.82 5.35
Distance per capita (km) 58.98 50.53 54.91 4.44
Walking per capita 10.43 10.82 10.48 0.64
Walking 73.43 72.19 69.15 0.46
Animal transport per capita 0.27 0.21 0.36 2.02
Animal transport 0.77 0.45 0.79 1.16
Motorized transport per capita 2.65 2.27 1.76 9.51
Motorized transport 9.53 7.78 6.16 4.06

© 2017 The Authors. African Development Review © 2017 African Development Bank
Quantifying the Road Influence Zone 605

divided by household members, a person located close to the road made 2 more trips per month than a person located more than
1.5 km from the road.
In a similar way the per capita distance travelled in kilometers corresponds with distance to road. Households located close to
a road not only travel frequently but also travel a relatively long distance per capita than households located far from road.
Households located up to 300 m on both sides of the road travel on average 59 km per month compared to 51 km for households
located 1.51 km or more on both sides of the road. This could be due to the fact that households located close to a road have better
access to motorized transport and possibly travel relatively long distances.
As far as travel by mode of transportation (walking, animal transport and motorized transport) is concerned, there are no
significant or consistent differences across all modes of transport between households located at different distances from
the road except for motorized transport. As expected, walking is the frequent mode of transportation. Regardless of how
far or close to a road rural households are located, the frequency of trips made on walking is comparable. On the other
hand, transportation by pack animals, although very low, seems to be relatively frequently used by households at the far
end from the main road than households close to the road although statistically insignificant. Finally, motorized transport
which is the second most frequently used by the sample households after walking, shows clear association with distance to
road. Households located close to the road use motorized transport for an average of 10 trips per month compared to an
average of 8 and 6 trips per month for households located in the medium and long distance ranges from the road
respectively and the difference is statistically significant at the 1 percent level of significance. This means households
close to the road travel on average four more trips per month, a 67 percent increase, than households located 1.5 km or
more from the road.

Distance to Road and Access to Agricultural Extension and Agricultural Inputs


Agricultural extension service has been widely used as a policy instrument worldwide to support rural development by
providing information and advisory services to farmers and other actors in agro-food systems. Such extension services
contribute to agricultural sustainability, livelihood improvement and well-being of the rural populations especially in the
developing world (Mbo’o-Tchouawou and Colverson, 2014). In Ethiopia, access to agricultural extension services has led to
adoption of improved technologies and farming practices which has contributed to poverty reduction (Dercon et al., 2008). In
the survey, respondents were asked about their access to agricultural extension and utilization of modern inputs — fertilizers
and selected seeds. Table 2 provides a summary of access to agricultural extension and utilization of agricultural inputs by
distance to road.
Following large investments by the Ethiopian government towards improving agricultural extension in the country, results
show that access to agricultural extension is high, but application of modern inputs is still low. The average use of fertilizers per
timad (a quarter of a hectare) is 34 kg and that of selected seeds is 7 kg per household.
In columns 4 to 6, access to agricultural extension and application of modern inputs by distance to road are indicated. Location
of households from road is significantly correlated with application of modern inputs. Households close to road applied fertilizer
on average more than two and half times per timad than households located more than 1.5 km from road. Households were
further asked what would have happened to their purchase of fertilizers and improved seeds if roads were not constructed.
Table 3 summarizes the response.

Table 2: Access to agricultural extension and input utilization by distance to road


Distance to road
Sample Close Medium Long
Input application Unit average (0–0.3 km) (0.31–1.5 km) (>1.5 km) F-value

Access to agricultural extension Yes¼1 0.849 0.886 0.824 0.835 1.52


Visits by agricultural or health extension workers Yes¼1 0.974 0.983 0.960 0.977 0.95
Average fertilizer use per household Kg 88.53 109.23 97.94 58.40 28.06
Average fertilizer use per timad Kg 33.78 47.86 36.22 18.07 35.83
Average selected seed use Kg 6.73 7.61 8.53 4.04 3.34

© 2017 The Authors. African Development Review © 2017 African Development Bank
606 F. N. Tegebu and E. H. Seid

Table 3: Purchase of fertilizers and seeds without a road


Purchase of fertilizers without a road Freq. Percent

Cannot purchase inputs 124 24.85


Purchase smaller amount 159 31.86
Purchase same amount but higher transportation cost 155 31.06
Price would be higher 8 1.6
purchase and transporting inputs remain unchanged 53 10.62

A quarter of the respondents said that they would not have purchased inputs if roads were not constructed. Similarly, a third of
the respondents (32 percent) said they would have responded by purchasing smaller amount of inputs. Close to a third of the
respondents (31 percent) said they would continue to purchase the same amount of inputs they are purchasing now but
transportation cost would have been higher. Only 11 percent of the respondents believe that both the purchase and transportation
cost of inputs would have remained unchanged.

Non-farm Activities and Distance to Road

The survey asked questions on different sources of income of households and types of commercial activities households were
engaged in. Table 4 indicates contributions of different sources of income and percentage of respondents who reported changes
in income following construction of rural roads.
As agricultural communities, it is understandable that most of the income source (more than three-fourths) comes from
agriculture followed by wage income contributing 12 percent of the total income. Income from own business and transfer income
(both aid and remittance) contribute about 5 percent each. Livestock income which contributes only 2 percent of the income
source includes sale of livestock products. Sale of live animals and livestock products used for own-consumption were not
included.
Survey respondents were further asked how income from the different sources has changed after construction of rural roads.
The last three columns in Table 4 summarize the result. A reasonably good percentage of respondents have said that income from
the different categories has risen after roads were constructed. The percentage varies among income types, the highest being in
agriculture with 48 percent of the respondents who earned agricultural income reporting a rise in income, followed by wage
earning respondents where 41 percent indicated their wage income had increased. The least being in business income where only
29 percent reported a rise in business income after construction of a rural road. The relatively low percentage response of
business income earners could be partly due to the fact that most business activities operated only after roads were constructed.

Table 4: Household income source and changes in income after road construction
% of households who
responded changes in income
% of households Share of
No. of with income Mean household Standard total
Income source observations source income (in Birr) deviation income Increased Decreased Unchanged

Agricultural income 518 98 12501 19906 76 48 3 49


Wage income 241 46 2067 3365 12 41 2 57
Business income 66 13 885 3879 5 29 71
Transfer income 101 19 757 2696 5 30 1 69
Livestock income 272 52 336 727 2 37 2 61

© 2017 The Authors. African Development Review © 2017 African Development Bank
Quantifying the Road Influence Zone 607

Table 5: Attribution of changes in income to road construction


Changes in income attributed to road construction (% of households)
Income type No. of observations All of the change Part of the change None of the change

Agricultural income 260 30 66 4


Livestock income 140 33 63 4
Wage income 102 32 65 3
Transfer income 32 25 53 22
Business income 24 17 83 –

The perceived increase in income could be due to the fact that road construction eases access to markets and technology, which in
turn expands farm and nonfarm production through increased availability of relevant inputs and lower input costs as well as
growth of rural enterprises.
The changes in income discussed above indicate only possible association. The reason(s) for changes in income could be due
to other factors other than road construction. To indicate how much of the changes in income could be due to rural road
construction, respondents were asked how much of the change in income they attribute to road construction. The three-point
scale response of households is summarized in Table 5. Most respondents believe that only part of the change in income is
attributable to the construction of a road. However, a good percentage of respondents (ranging from 17 to 33 percent) also believe
that the change in income is fully attributable to construction of a road. Except in the case of transfer income, where close to 22
percent of the respondents believe road construction had no influence on income change, in all the remaining cases, only a small
fraction — not more than 4 percent — believe road construction contributed nothing to income change.

Commercial Activities and Distance to Road

Although communities in the study sites mainly depend on agriculture for their living, with the expansion of infrastructure and
rural towns, it is expected that non-farm activities, especially small business activities, play a role in diversifying income sources
and supporting the lives of rural households. In the survey, rural households were asked whether they engage in commercial
activities and type of commercial activities they undertake, and the result is summarized in Table 6.
The commonest type of commercial activity in the rural area is trading of live animals, grain and other agricultural products.
Close to 36 percent of the commercial activities is constituted by trading of livestock and other agricultural products, followed by
small shops which constitute close to a third of the business activities. The third dominant form of business activity is selling
local drinks. The remaining commercial activities — selling food, tea and coffee; and hair dressing — constitute only 8 percent
and 4 percent of the business activities respectively.
To relate the commercial activities to construction of rural roads, survey respondents were asked whether they started to
engage in the business after/before construction of rural roads. As indicated in Table 7, most commercial activities were opened
only after construction of a road (65 percent). If we add to this those that were operating before construction of a road but were
closed for some time and reopened after roads were constructed, it increases to 77 percent.

Table 6: Types of commercial activities


Type of commercial activity Frequency %

Small shops 16 33
Barber/Women’s hairdresser 2 4
Selling local drinks 9 19
Selling food, tea, coffee 4 8
Trader (grain, livestock, etc.) 17 36

© 2017 The Authors. African Development Review © 2017 African Development Bank
608 F. N. Tegebu and E. H. Seid

Table 7: Operation of commercial activities


Frequency %

Operating before construction of road 11 23


Reopened after construction of road 6 12
Opened after construction of road 31 65

Moreover, most of the newly opened or reopened businesses are concentrated in areas close to the rural road (Figure 2). As
indicated in the figure, the number of newly opened and reopened businesses in areas close to a road is almost three times that of
the newly opened and reopened businesses in the areas far from road.

Negative Effects of Roads (Roadside Effects)


Besides mobility and positive impacts on welfare of households, roads can have unwanted consequences. Fertile land can be
used for road construction and hence local communities can lose an important source of income. Moreover, roads can have
environmental effects such as flooding, erosion, water logging, sediment deposition and dust.
In the survey, households were asked about effects of roads on the local environment. Table 8 summarizes the percentage
of households who reported that roads have an effect on the local environment and the perceived amount of decrease in crop
production. The three most common consequences cited in the order of number of respondents were dust, flooding and
erosion. Close to 44 percent of the respondents said that the occurrence of dust had increased after roads were constructed.
The second most occurrence reported by 34 percent of the respondents is flooding, followed by erosion where 29 percent said
it had increased after roads were constructed.
However, all households who reported environmental effects of roads may not be directly affected by the problem. The
fourth column of Table 8 indicates the percentage of households directly affected through reduction in crop production as a
result of the various effects of roads. Close to 11 percent of the sample households faced a decline in crop production due to
dust lifted up by trucks from roads. There are a number of evidences that dust can have both a physical and chemical impact
on crops. Dust on plants can smother the leaves, block stomata and obstruct photosynthetic activities (Rahul and Jain, 2014).
Moreover, dust particles can interfere with the mechanisms of stomatal pores. The dust accumulation on leaf surfaces causes
conditions similar to water stress, such as a reduction of stomata conductance, photosynthesis and transpiration and increased
leaf temperature (Zia-Zia-Khan et al., 2015). Following dust, flooding and erosion were the other two factors that negatively
affected crop production of nearly 8 percent and 7 percent of the sampled households respectively.
Respondents were further asked the magnitude of reduction in crop production due to effects of the road on the local
environment. Results indicate that the loss in output reached up to 50 percent of crop production. However, for most
households the loss ranges from 10 to 30 percent of crop output. Such a loss calls for additional research and interventions to

Figure 2: Distribution of commercial activities by distance to road

© 2017 The Authors. African Development Review © 2017 African Development Bank
Quantifying the Road Influence Zone 609

Table 8: Effects of roads on flooding, erosion, water logging, sediment deposition and dust
Households
reported effect of
road Amount of decrease in crop production
Percentage of households with crop Between Between
Effect of road Frequency % production affected Less than 10% 10–30% 30–50%

Increased flooding 179 34 8 10 (24%) 24 (59%) 7 (17%)


Water logging 61 12 2 1 (11%) 6 (67%) 2 (22%)
Erosion 153 29 7 15 (44%) 15 (44%) 4 (12%)
Sediment deposition 95 18 4 8 (33%) 12 (57%) 2 (10%)
Dust clouds 229 44 11 14 (29%) 27 (55%) 8 (16%)
Occurrence of weeds 47 10 2 5 (63%) 3 (37%)

be undertaken to turn negative effects of roads into productive purposes. One such intervention is roadside water harvesting
which not only reduces erosion and sediment deposition, but also harvests water for productive purposes (see, e.g., Kifle
et al., 2015). As pointed out in Valipour (2015, 2016), adoption of an efficient water management system along roads to
minimize the possible soil erosion and conserve the limited water resources could contribute to the poverty reduction efforts
of governments.

4.2 Statistical Results

In the previous section we saw how different socio-economic and local environmental effects are associated with distance to
road. In this section, we present statistical analysis in order to identify if a threshold type of relationship can be established
between the socio-economic and local environmental outcome variables and distance to road. The statistical analysis is made for
seven outcome variables — trips per capita, access to motorized transport, use of fertilizer, commercial activities, dust, flooding,
and erosion — using linear regression and piecewise regression models.
To quantify the magnitude of the effect of road on the different welfare indicators, we compared the Akaike Information
Criterion (AIC) values for the simple linear models, the piecewise regression model, and of an additional simple log linear
model, which was identical to the first, except that we log-transformed distance to road.
A stepwise regression model was used to determine the variables to include on the regression models. We included all
household characteristics, income variables and some asset variables along with distance to road. The variables that were
retained through the stepwise selection (see Table 9) were used in the piecewise and linear regression models.

Table 9: Variables included in both the linear and piecewise regressions models
Response Variables

Trip per capita Sex of hh head, no. of children, marital status and hh size
Motorized transport Sex of hh head, no. of children, marital status, hh size and income per capita
Use of fertilizer Sex of hh head, age of hh head, land holding and hh size
Commercial activities Sex of hh head, age of hh head and education of hh head
Dust Age of hh head, level of education of hh head and income per capita
Flooding Income per capita and no. of adults
Erosion Income per capita and no. of children
Note: hh ¼ household.

© 2017 The Authors. African Development Review © 2017 African Development Bank
610 F. N. Tegebu and E. H. Seid

Table 10: Relationship between outcome variables and distance to road as measured by linear, log-
linear and piecewise regressions
Simple linear model Log linear model Piecewise regression model
Slope: Slope:
Distance log(distance Breakpoint Slope before Difference in
Response AICs to road AIC to road) AIC (km) breakpoint slope parameter

Trip per capita 6.4239 –0.0006357 6.4244 –0.5248077 6.4213 1.162871 –1.63418 1.667077
0.0002 0.1746 0.589226 1.1299 1.184257
Motorize trans. 3.926561 –0.0002057 3.932637 –0.1537283 4.132065 2.582799 –0.6803487 0.8057044
0.0000475 0.0392254 0.5999455 0.1315623 0.2423928
Fertilizer 10.84331 –0.0002489 10.86748 –0.1762703 11.12777 1.827203 –25.4246 12.49647
0.0000343 0.0320652 0.7809998 5.830958 7.562724
Commerce 0.5814599 –0.0001687 0.578345 –0.1728623 0.575394 1.509743 –0.7504323 1.178707
0.0001486 0.0975074 0.022683 0.5296078 0.5754481
Dust 0.5653387 –0.0022705 0.562908 –0.7454171 0.554017 0.2380927 –8.35692 7.097131
0.0004803 0.1040662 0.0429969 3.104761 3.138872
Flooding 0.5145403 –0.6977623 0.499878 –0.4607678 0.520029 0.6984002 –2.127998 2.091812
0.2355912 0.1020981 0.0270316 1.112379 1.155207
Erosion 0.4568031 –0.491879 0.450038 –0.3546718 0.462087 1.745005 –0.8228435 0.7570253
0.2239143 0.1099543 1.041801 0.4713619 0.647741
  
Significant at 10%; significant at 5%; significant at 1%.

Regression results are shown in Table 10. We tested the significance of the slope of the relationship between distance to road
and the response in the linear models. For the piecewise regression model, we tested the significance of the slope of the
relationship between distance to road before the breakpoint, and the significance of the ‘difference in slope parameter’. We
compared overall model fit of the linear and piecewise regression models for each response using the AIC.
We found a statistically significant difference in slope parameter and that the piecewise regressions model gave at least as good a
fit to the data as the linear models for all the response variables except for trips per capita and effect of erosion on households.
The breakpoint occurred approximately at 2 to 3 km from the road for access to motorized transport which includes
transportation by bus, car, bajaj (three wheel drive) and so on, 1 to 2.6 km from the road for use of fertilizers, 1.5 to 1.53 km for
commercial activities, 0.2 to 0.24 km for effects of dust, and 0.67 to 0.74 km for flooding effect. For all the five response variables
— access to motorized transport, use of fertilizer, commercial activities, dust, and flooding — there was also a statistically
significant relationship with distance to road up to the threshold distance (Table 10). For trip per capita and effects of erosion,
there was no significant change in slope in the piecewise regression models and the linear regression model gave a better fit than
the piecewise regression model.
Table 11 shows the best fitting models for the statistical analysis. Three of the seven outcome variables were fitted using
simple linear regression, two using log linear regression and the remaining two using piecewise regression.

Table 11: Best fitting model


Type of effect Best fitting regression model

Trips per capita Simple linear


Motorized transport Simple linear
Use of fertilizer Simple linear
Commercial activities Piecewise
Dust Piecewise
Flooding Log linear
Erosion Log linear

© 2017 The Authors. African Development Review © 2017 African Development Bank
Quantifying the Road Influence Zone 611

4.3 Discussion
Our results show that road influence zone delineated by threshold exists for access to motorized transport, use of
fertilizers, commercial activities, dust and flooding, and that the influence extends from up to 0.24 km in the case of road
dust to about 2.6 km in the case of access to motorized transport from the road (Table 12). The other two responses, trips
per capita and erosion showed a significant linear negative response to distance from road, but not threshold type
relationship.
Our findings indicate that comparing linear and piecewise regressions is a useful method to quantify the shape of the distance
response of the welfare and effects of roads. However, for the statistically significant piecewise regression models, we found
more than one estimate of the location of the breakpoint. This is likely a result of there being a zone in which threshold effects
exist, rather than a sharp break.
There is no threshold effect observed for trips per capita. This is mainly due to the fact that most trips in rural areas of Ethiopia
including in the study sites are made on foot. For walking, roads are not required and that is why a threshold effect could not be
observed in trips per capita.
For the responses in which a threshold type relationship was observed, the threshold effects occurred at different distances to
road. Considering the number of times household members use motorized transport such as cars, buses, bajaj and so on, in a
typical month as a proxy for access to motorized transport, we observe that a threshold for access to motorized transport exists at
a distance of about 2.6 km from both sides of the road. Although different other factors such as topography of the land can affect
people’s ability to access motorized transport, the idea of the above threshold is that households up to 2.6 km on both sides of the
road have comparable patterns in their access to motorized transport.
Similar to access to motorized transport, threshold type relationships were observed in application of fertilizer
and commercial activities. For fertilizer application, threshold occurs at a distance of about 1.83 km from both sides of
the road indicating that the influence of rural roads on use of fertilizer is well perceived in a radius of 1.83 km from
the road.
Similarly, commercial activities in rural areas are influenced by road. The influence of rural roads on commercial activities is
well perceived in the range of 1.5 km from both sides of a road. After all, most of the commercial activities in the study sites were
opened only after the roads were constructed.
Finally, roads can also have negative consequences. The two typical consequences reported in this case are road dust and
flooding. A threshold type relationship has been observed between distance to road and the roadside effects. The effect of dust
that is lifted up when cars, especially heavy trucks, use rural roads is typically felt in a radius of 240 m from both sides of the road.
In a similar way, floods coming out of the road especially from the road culverts, influence households in the range of 0.7 km on
both sides of the road.
Figure 3 shows the influence map for the selected indicators for each of the four study sites. Depending on distance of
settlement of households from road, the road influence zone is different for the different study sites. In the case of the lowlands
(Hade Alga and Werebaye), the settlement is so disperse that there are a good number of households that lie outside the influence
zone of the different indicators including motorized transport. On the other hand, in the highland tabias, mainly in Adi Kisanded
where most households reside close to the road, the influence of the road on the community is large even in the case of road dust,
where the influence dimension is the smallest of all.

Table 12: Threshold distance for road influence zone


Response Breakpoint (km)

Trips per capita NA


Motorized transport 2.58  0.6
Fertilizer 1.83  0.78
Commercial activities 1.51  0.02
Dust 0.24  0.04
Flooding 0.7  0.03
Erosion NA

© 2017 The Authors. African Development Review © 2017 African Development Bank
612 F. N. Tegebu and E. H. Seid

Figure 3: Map of road influence zone

© 2017 The Authors. African Development Review © 2017 African Development Bank
Quantifying the Road Influence Zone 613

5. Conclusion
This study investigated the influence zone of rural roads on socio-economic development in rural Tigray using four rural roads
and a sample of 529 households. Analysis was made on rural mobility, selected welfare indicators and the local environment.
Results indicate that rural roads contribute to the socio-economic development of rural livelihoods. However, roads can also
have unwanted environmental consequences. Dust, flooding and erosion were the commonest environmental consequences of
rural roads negatively affecting rural livelihoods.
One of the direct effects of rural roads is that it enhances rural mobility. What is abundantly clear is that when roads
enhance mobility they do so in association with wheeled or motorized transport easing people’s movement and making
them faster and capable of achieving longer distance. This can have influence on the socio-economic development when the
time saved and distances bridged provide either greater economic opportunities or better access to social services or useful
social contacts. Households in the study areas traveled on average 14 trips per month per person. However, more than
80 percent of these trips were made by ‘walking’. The use of motorized transport is limited, indicating the fact that road
improvement alone may not enhance the mobility of the rural poor unless appropriate and affordable means of transport is
available.
Besides wealth, location of households also plays an important role to benefit from roads. Classifying distance of location
of households from road into three — close, medium and far — produces results that indicate that households located close to
roads are more mobile both in terms of number of trips and average distance traveled, and better use motorized transport than
households located far from the road. Given the differential impact of roads on households located at different distance to a
road, attempts were made to see whether there exists a threshold pattern that would enable us to delineate the influence zone
of a road. Our results indicate that threshold patterns were observed on use of motorized transport, use of fertilizers,
commercial activities that the rural households engaged in and on two environmental effects of roads  dust and flooding.
However, threshold patterns were not observed on mobility. The fact that the study sites are overwhelmingly a walking
world, and since walking in the rural areas does not require roads, a threshold effect was not observed on mobility proxied by
number of trips.

Notes
1. Woreda is the second administrative unit above tabia.
2. Tabia is the smallest unit of local government in rural communities in present day Tigray and consists of three to four kushets.

References
Admasu S., M. S€oderbom, S. Eyerusalem and A. Getnet (2012), ‘Road Infrastructure and Enterprise Development in Ethiopia’,
IGC Working Paper, C-32001-ETH-1.
Akpan, U. (2014), ‘Impact of Regional Road Infrastructure Improvement on Intra-Regional trade in ECOWAS’, African
Development Review, Vol. 26, No. s1, pp. 64–76.
Ali, I. and E. M. Pernia (2003), ‘Infrastructure and Poverty Reduction: What is the Connection?’ ERD Policy Brief Series,
Economics and Research Department, No. 13.
Bryceson, D. F., A. Bradbury and T. Bradbury (2008), ‘Roads to Poverty Reduction? Exploring Rural Roads’ Impact on
Mobility in Africa and Asia’, Development Policy Review, Vol. 26, No. 4, pp. 459–82.
on, C. and L. Serven (2014), ‘Infrastructure, Growth, and Inequality: An Overview’, World Bank Policy Research
Calder
Working Paper, No. 7034.
Dercon, S., D. O. Gilligan, J. Hoddinott and W. Tassew (2008), ‘The Impact of Agricultural Extension and Roads on Poverty and
Consumption Growth in Fifteen Ethiopian Villages’, IFPRI Discussion Paper, 00840.
Eigenbrod, F., S. J. Hecnar and L. Fahrig (2009), ‘Quantifying the Road Effect Zone: Threshold Effects of a Motorway on
Anuran Populations in Ontario, Canada’, Ecology and Society, Vol. 14, No. 1, p. 24.
© 2017 The Authors. African Development Review © 2017 African Development Bank
614 F. N. Tegebu and E. H. Seid

Fafchamps, M., B. Minten and G. Eleni (2005), ‘Increasing Returns and Market Efficiency in Agricultural Trade’, Journal of
Development Economics, Vol. 78, No. 2, pp. 406–42.
Fan, S. and X. Zhang (2008), ‘Public Expenditure, Growth and Poverty Reduction in Rural Uganda’, African Development
Review, Vol. 2, No. 3, pp. 466–96.
Forman, R. T. T. and R. D. Deblinger (2000), ‘The Ecological Road-effect Zone of a Massachusetts (USA) Suburban Highway’,
Conservation Biology, Vol. 14, pp. 36–46.
Gannon, C. and Z. Liu (1997), ‘Poverty and Transport’, TWU Discussion Papers, TWU-30, World Bank, Washington, DC.
Gwilliam, K. (2011), Africa’s Transport Infrastructure, Mainstreaming Maintenance and Management, World Bank,
Washington DC.
Key, N., A. de Janvry and E. Sadoulet (2000), ‘Transaction Costs and Agricultural Household Supply Response’, American
Journal of Agricultural Economics, Vol. 82, No. 2, pp. 245–59.
Kifle, W., V. S. Frank, A. P. Marta, G. Berhane and V. P. Martin (2015), ‘Water Harvesting from Roads: Climate Resilience in
Tigray, Ethiopia’, IRF Europe and Central Asia Regional Congress, 15–18 September, Istanbul, Turkey.
Mbo’o-Tchouawou, M. and K. Colverson (2014), ‘Increasing Access to Agricultural Extension and Advisory Services: How
Effective Are New Approaches in Reaching Women Farmers in Rural Areas?’ International Livestock Research Institute (ILRI),
Nairobi, Kenya.
Menash, E. J., M. Huchet-Bourdon and L. Latruffe (2014), ‘Infrastructure Access and Household Welfare in Rural Ghana’,
African Development Review, Vol. 26, No. 3, pp. 508–19.
Raballand, G., P. Macchi and C. Petracco (2010), Rural Road Investment Efficiency: Lessons from Burkina Faso, Cameroon,
and Uganda, World Bank, Washington, DC.
Rahul, J. and M. K. Jain (2014), ‘An Investigation into the Impact of Particulate Matter on Vegetation along the National
Highway: A Review’, Research Journal of Environmental Sciences, Vol. 8, No. 7, pp. 356–72.
Valipour, M. (2015), ‘Land Use Policy and Agricultural Water Management of the Previous Half of Century in Africa’, Applied
Water Science, Vol. 5, pp. 367–95.
Valipour, M. (2016), ‘How Do Different Factors Impact Agricultural Water Management?’ Open Agriculture, Vol. 1, pp.
89–111.
Yogo, U. T. and A. Verdier-Chouchane (2015), ‘Enhancing North Africa’s Infrastructure for Improved Competitiveness’,
African Development Review, Vol. 27, No. 3, pp. 274–87.
Zia-Khan, S., W. Spreer, Y. Pengnin, X. Zhao, H. Othmanli, X. He and J. Muller (2015), ‘Effects of Dust Deposition on Stomatal
Conductance and Leaf Temperature of Cotton in Northwest China’, Water, Vol. 7, pp. 116–31.

© 2017 The Authors. African Development Review © 2017 African Development Bank
African Development Review, Vol. 29, No. 4, 2017, 615–629

Croissance economique et degradation de l’environnement au Cameroun

Edmond Noubissi Domguia and Henri Njangang Ndieupa

Resume : Ce papier a pour objectif d’examiner les effets de la croissance economique sur les emissions de dioxyde de
carbone (CO2) et sur l’evolution de la temperature au Cameroun sur la periode 1972–2010. Le test de cointegration par les retards
echelonnes ou Auto Regressive Distributed Lags (ARDL) indique l’existence d’une relation de long terme sous forme de « N
inverse » entre la croissance economique et les indicateurs de degradation environnementale (emissions de CO2 et evolution de
la temperature) retenus dans cette etude. Ensuite, on constate que la consommation d’energie et les activites industrielles
accroissent les emissions de CO2 mais pas l’evolution des temperatures au Cameroun. Par contre, l’ouverture commerciale
entraine une baisse du niveau de CO2 mais pas de la temperature. Ce resultat laisse entrevoir que le Cameroun exporte ou
transmet sa pollution vers l’exterieur ou vers d’autres pays.

1. Introduction
Les dernieres decennies ont ete marquees par une prise de conscience accrue des problemes environnementaux, comme en
temoigne l’ampleur des debats sur ce sujet. Une des questions centrales soulevees par ces debats est celle de la possibilite d’une
croissance davantage respectueuse de l’environnement, d’un developpement durable qui ne serait pas dommageable aux
generations futures.
L’accumulation de problemes environnementaux engendres par l’activite humaine semble ^etre au cœur d’une telle prise de
conscience. Parmi les degradations de l’environnement naturel, on peut citer la pollution de l’air, des sols et de l’eau, la
deforestation ou la surexploitation des ressources. Ces dommages etant interconnectes, ils ont de surcro^ıt tendance a se renforcer
mutuellement.
Dans les annees 1970, les travaux du club de Rome (Meadows et al., 1972) se sont attardes a repondre a cette question et ont
ainsi permis d’informer et de sensibiliser l’opinion publique sur les dangers que peut representer la croissance economique sur
l’environnement. Les conclusions de ce rapport sont tres pessimistes. Au-dela des travaux passes comme ceux de Malthus
(1798), Ricardo (1817) ou Jevons (1865) qui concluaient qu’une croissance economique nulle etait une fin inevitable en raison
du caractere limite des ressources, ce rapport preconise une croissance zero afin d’eviter une catastrophe economique a venir. En
se focalisant sur la dimension environnementale, le Club de Rome considere que la degradation environnementale engendree par
l’activite economique est ineluctable et conclut ainsi que la seule solution pour assurer la subsistance de l’environnement est de
stopper la croissance economique.
On comprend ainsi mieux pourquoi l’analyse des determinants de la degradation environnementale est devenue un sujet tres
passionnant dans la litterature economique et l’essentiel des travaux s’attele a verifier l’hypothese de la Courbe
Environnementale de Kuznets entre la croissance economique et les indicateurs de degradation environnementale (Grossman
et Krueger, 1995; Panayotou, 1993 et 1995; Shafik et Bandyopadhyay, 1992). L’inter^et de la CEK est qu’elle postule la
possibilite pour les pays pauvres d’ameliorer la qualite environnementale au fur et a mesure qu’ils se developpent, a mesure que
le niveau de vie des individus s’ameliore et favorise l’eclosion d’une conscience environnementale (Banque Mondiale, 1992).
Divers auteurs ont propose une revue detaillee des travaux empiriques sur la relation entre la croissance economique et la qualite
de l’environnement (Dinda, 2004; Nourry, 2007). La diversite des travaux confirme que les problemes environnementaux sont


Edmond Noubissi Domguia (auteur correspondant), Docteur au departement d’Economie Publique et des Ressources Humaines, Faculte des Sciences
Economiques et de Gestion, Universite de Dschang, B.P. 110 Dschang, Cameroun; Tel: (þ237) 675 72 20 91, e-mail: enoubissidomguia@yahoo.fr.
Henri Njangang Ndieupa, Doctorant au departement d’Analyse et Politiques Economiques, Faculte des Sciences Economiques et de Gestion,
Universite de Dschang, B.P. 110 Dschang, Cameroun, Tel: (þ237) 679 35 82 89, e-mail: ndieupahenri@yahoo.fr
© 2017 The Authors. African Development Review © 2017 African Development Bank. Published by Blackwell Publishing Ltd,
9600 Garsington Road, Oxford OX4 2DQ, UK and 350 Main Street, Malden, MA 02148, USA. 615
616 E. Noubissi Domguia et H. Njangang Ndieupa

differents d’une region a l’autre voir d’un pays a l’autre, rendant particulieres les solutions proposees afin de limiter le desastre
environnemental.
Consideres longtemps comme un sujet sans importance dans les pays en developpement, les problemes environnementaux
prennent de plus en plus d’ampleur depuis la tenue du premier sommet de Rio en 1992. Cette prise de conscience est d’autant
pertinente que l’augmentation des emissions de CO2 a ete acceleree par la croissance economique dans les regions en
developpement. En effet, au cours de la decennie 1990–2000, les emissions de CO2 ont augmente de 48% dans ces regions, et de
81% au cours de la decennie suivante (2000–2010), tandis qu’elles diminuaient de 7% et 1% respectivement dans les pays
developpes au cours de la m^eme periode (ONU, 2013, p. 43).
Nous voudrions rappeler qu’il existe bien entendu d’autres faSc ons d’etudier la relation developpement-environnement. La
plupart des etudes empiriques consacrees a l’analyse de cette relation s’appuient sur des modeles dont les variables expliquees
sont des indicateurs de pollution comme les emissions de CO2 ou celles de SO2, il en existe d’autres qui prennent en compte
diverses variables de degradation environnementale comme la deforestation (Van et Azomahou, 2003) ou encore la pollution
organique de l’eau (Clement et Meunie, 2010), mais, une fois encore, les resultats sont contradictoires en ce qui concerne la
forme fonctionnelle de cette relation.
L’une des reproches qui est faite a la courbe environnementale de Kuznets est d’utiliser comme seul indicateur de la qualite de
l’environnement le CO2 par t^ete. En tentant d’identifier les effets des hausses de temperature sur les performances
macroeconomiques des pays, les travaux de Melissa Dell, Benjamin Jones et Benjamin Olken (2012), s’inscrivent d’une part,
dans le debat entourant le r^ole de la temperature dans le developpement economique et, d’autre part, dans le debat autour des
repercussions du rechauffement climatique. Leur etude identifie un important lien causal de la temperature sur le processus de
developpement en mettant en evidence qu’elle n’affecte pas seulement diverses composantes dans une economie, mais influence
aussi fortement la production nationale.
Cette etude se propose de determiner la relation existante entre la croissance economique et deux indicateurs de
determinations de la qualite de l’environnement ou du changement climatique a savoir le niveau d’emission de CO2 par t^ete et la
temperature au Cameroun. Le choix de la temperature se justifie par le fait de sa forte correlation avec les gaz a effet de serre et
donc le CO2.

2. Le debat autour de la courbe Environnementale de Kuznets


L’effet de serre est un phenomene naturel lie a l’absorption des rayonnements infrarouge (IR) renvoyes par la surface terrestre
par des composes presents dans l’atmosphere appeles gaz a effet de serre: dioxyde de carbone (CO2), methane (CH4), eau (H2O),
ozone (O3), protoxyde d’azote (N2O), hexafluorure de soufre (SF6) et halocarbures (dont les CFC). Toutefois, une partie de ce
rayonnement IR est absorbee par la terre et transformee en chaleur. Ce phenomene est important pour la « survie de la planete ».
Sans effet de serre, la temperature moyenne sur Terre chuterait dans un premier temps a 18 °C, mais a cette temperature la glace
s’etendrait sur le globe et l’albedo terrestre augmenterait entra^ınant une chute supplementaire de la temperature moyenne de la
planete qui se stabiliserait vraisemblablement en dessous de 50 °C et « la vie n’existerait peut-^etre pas ». L’augmentation de
l’effet de serre est tres probablement responsable du rechauffement climatique en cours. L’origine anthropique de celui-ci ne fait
aujourd’hui pratiquement plus aucun doute chez les experts (Bessagnet et al., 2009).
L’hypothese de la courbe environnementale de Kuznets (CEK) est tiree d’un modele economique dans lequel il n’existe pas de
retroaction de la qualite de l’environnement sur la croissance economique. La degradation de la qualite de l’environnement est
reconnue avoir des effets pervers sur la qualite de la vie mais pas directement sur les possibilites de production (Stern et al.,
1996). En l’absence de cette retroaction, la croissance peut ^etre une solution pour acceder a une meilleure qualite de vie dans les
pays en voie de developpement lorsque l’hypothese CEK est satisfaite.
Le debat autour de l’hypothese de la CEK a pour origine la controverse de la croissance et les politiques correspondantes
(Dinda, 2004). Les chercheurs avaient, au depart emis l’hypothese qu’un niveau eleve du revenu par t^ete accentuerait la
degradation de la qualite de l’environnement. Puis l’on a emis l’hypothese selon laquelle, des niveaux eleves de revenu peuvent
reduire la degradation de l’environnement (Beckerman, 1992). Ainsi, selon Bhagawati (1993), la croissance economique peut
^etre une condition prealable a une amelioration de la qualite de l’environnement. Ce qui permet a Panayotou (1993) d’affirmer
que la croissance economique peut ^etre un puissant canal pour l’amelioration de la qualite de l’environnement dans les pays en
voie de developpement. Toutefois, la derniere hypothese a fait l’objet de polemiques.

© 2017 The Authors. African Development Review © 2017 African Development Bank
Croissance economique et degradation de l’environnement 617

La principale explication a la forme de la CEK est la notion selon laquelle, lorsqu’une population atteint un niveau de vie
suffisamment eleve, elle accorde une plus grande importance aux equipements environnementaux (Pezzey, 1989; Selden &
Song, 1994; Baldwin, 1995). En effet, apres avoir franchi un seuil particulier du revenu par t^ete, la volonte des populations a
payer pour se procurer un environnement de qualite s’accro^ıt dans une proportion plus grande que celle du revenu (Roca, 2003).
Ceci se traduit generalement par de plus en plus de dons aux organisations de protection de l’environnement, la demande et la
consommation de produits moins polluants. A ce niveau, l’elasticite revenue de la demande d’une qualite de l’environnement est
superieure a l’unite; un environnement de qualite et sa preservation deviennent des biens de luxe.
L’utilisation des emissions de CO2 comme proxy de la degradation environnementale pose souvent un probleme de pertinence
selon les auteurs. En effet, certains auteurs justifient l’absence d’une relation en U inverse entre la croissance et les emissions de
^ de reduction des changements climatiques
CO2 par le fait qu’il n’existe pas d’incitation a la reduction de rejets polluants, le coUt
etant locaux et leurs benefices globaux (Nourry, 2007).
Grossman et Krueger (1994, 1995), Panayotou (1993, 1995), Shafik et Bandyopadhyay (1992) et Selden et Song (1994) sont
les premiers auteurs a tester empiriquement les effets de la croissance economique sur les indicateurs environnementaux (SO2,
NOx, CO2, CO, dechets municipaux, particules en suspension). A  la suite de ces travaux, plusieurs auteurs ont examines
l’hypothese de la CEK en utilisant differentes variables de la degradation environnementale, telles que le dioxyde de carbone
(CO2) (Apergis et Payne, 2009; Lean et Smyth, 2010; Du et al., 2012; Shahbaz et al., 2013; Tiwari et al., 2013), le dioxyde de
soufre (SO2) (Day et Grafton, 2003 et Llorca et Meunie, 2009), l’Oxyde Nitreux (N2O) (Cho et al., 2014), methane (CH4) (Cho
et al., 2014), le gaspillage d’eau (Haisheng et al., 2005).
Grossman et Krueger (1994) sont les premiers a obtenir une courbe en U inverse dans leur Working Paper sur les effets
environnementaux de l’Accord de Libre Echange Nord-Americain. Ils verifient la relation de Kuznets pour la pollution de l’air et
la pollution de l’eau avec des points de retournement fixes a 5000$ et 8000$ respectivement. Cette conclusion est confortee par
les travaux des auteurs tels que Shafik et Bandyopadhyay (1992), Panayotou (1993) et Selden et Song (1994); Jalil et Feridum
(2011) pour la Chine; Saboori et al. (2012) et Shahbaz et al. (2013) pour la Malaysie; Yavuz (2014) pour la Turkey; Shahbaz
et al. (2015) pour le Portugal; Zambrano-Monserrate et al. (2016) pour l’Equateur. Ils obtiennent aussi la courbe
environnementale de Kuznets pour differents indicateurs environnementaux et zones d’etudes. Plus recemment, Aboagye
(2017) a egalement montre a partir d’un modele ADRL l’existence d’une relation en U inverse entre la croissance economique et
la soutenabilite environnementale au Ghana.
Toutefois d’autres investigations (Shukla and Parikh, 1996; Carson et al., 1997; Halicioglu, 2009; Akpan et Chuku, 2011;
Chandran et Tang, 2013; Ozcan, 2013) n’aboutissent pas a la relation en U inverse. Ces auteurs obtiennent diverses formes
alternatives en fonction du degre du modele econometrique retenu. Cependant, la plus part de ces auteurs ont retenu pour la
plupart uniquement la croissance economique comme seule variable determinante de la qualite de l’environnement.
C’est ce qui a justifie par la suite l’introduction des variables supplementaires dans l’analyse des determinants de la qualite
environnementale par un second groupe d’auteurs. Ainsi, Shi (2003); Cole et Neumayer (2004); Halicioglu (2009); Shahbaz
et al. (2010); Ngepah (2011); Wang et al. (2011); Akpan et al. (2012); Ismael and Mawar (2012); Gao et Zhang (2014); Ben Jebli
et al. (2015) entre autres ont obtenu une relation positive entre les emissions de CO2 et un ensemble de variables
macroeconomiques telles que l’ouverture commerciale, la consommation energetique, la population (densite et taux
d’urbanisation).
Cependant, le choix de l’indicateur pour capter la qualite de l’environnement n’a pas ete entierement elucide par ces differents
auteurs. De plus, une telle etude n’a pas encore ete faite pour le cas du Cameroun qui ambitionne pourtant ^etre emergent a
l’horizon 2035. La nouveaute de cette etude reside notamment dans le fait qu’elle utilise un indicateur nouveau pour capter la
qualite de l’environnement (la temperature) et qu’elle le fait pour le cas particulier du Cameroun.

3. Methodologie et donnees
Depuis que le debat sur l’hypothese CEK a ete lance, de nombreuses etudes ont prouve l’existence de la relation en forme de U
inverse entre la pollution et le revenu par t^ete. Toutefois, d’autres etudes ont trouve diverses formes a cette relation. Dans cette
section, nous presentons le modele et determinons pour le Cameroun la forme de la relation entre les emissions de CO2 par
habitant ou l’evolution de la temperature et le revenu par t^ete.
L’Hypothese CEK telle que proposee par Grossman et Krueger (1991) est empiriquement testee en estimant l’equation (1)
definie ci-apres.

© 2017 The Authors. African Development Review © 2017 African Development Bank
618 E. Noubissi Domguia et H. Njangang Ndieupa

yt ¼ a0 þ a1 xt þ a2 x2t þ a3 x3t þ a4 zt þ et ð1Þ

Ou y represente la variable environnementale, x represente le revenu par t^ete et z designe toute autre variable qui peut
influencer la degradation de l’environnement. L’indice t designe le temps. a0 est une constante et a1, a2, a3 et a4 sont les
coefficients des variables explicatives. L’equation (1) permet de tester toutes les formes possibles de la relation entre le revenu
par t^ete et la pollution. Sept differentes formes de cette relation peuvent ^etre obtenues en fonction des valeurs prises par les
coefficients.
a) Si a1 ¼ a2 ¼ a3 ¼ 0, alors il n’existe aucune relation entre la variable economique x et la variable environnementale y.
b) Si a1 > 0 et a2 ¼ a3 ¼ 0, alors il existe une relation lineaire positive entre le revenu x et la pollution y.
c) Si a1 < 0 et a2¼ a3 ¼ 0, alors il existe entre le revenu x et la pollution y une relation lineaire negative.
d) Si a1 > 0; a2 < 0 et a3 ¼ 0, alors la forme de la relation entre le revenu x et la pollution y est en « U » inverse. L’hypothese
CEK est dans ce cas verifiee.
e) Si a1 < 0; a2 > 0 et a3 ¼ 0, alors la relation est en forme de « U » entre le revenu x et la pollution y.
f) Si a1 > 0; a2 < 0 et a3 < 0, alors la relation entre le revenu x et la pollution y a une forme cubique polynomiale ou en forme
de « N ».
g) Si a1 < 0; a2 > 0 et a3 < 0, alors la relation prend la forme d’un « N » inverse.

L’hypothese CEK correspond au cas (d), il faut que a1 > 0 et a2 < 0. Si c’est le cas l’on pourra aussi calculer le point de
retournement dont la formule est:
 
a1
x ¼ exp 
2a2

L’estimation du modele (1) permet de detecter et de definir la forme de la relation entre le developpement economique et
l’environnement certes, toutefois elle peut permettre de repondre a d’autres questions. Par exemple, y a-t-il une influence
d’autres variables ? Si oui, comment affectent-elles la forme de la relation environnement-developpement economique ? Notre
objectif est de determiner la trajectoire des emissions de CO2 par habitant ainsi que l’evolution de la temperature au Cameroun et
de capter l’influence d’autres variables telles que la consommation d’energie fossile, la valeur ajoutee de l’industrie, le taux de
croissance economique, la densite de la population et l’ouverture commerciale. Le but est d’ameliorer la specification de base du
modele en y incluant des variables de contr^ole susceptibles d’influencer directement ou indirectement les emissions de CO2
d’une economie.

lnðCO2Þt ¼ a0 þ a1 lnðPIBÞt þ a2 ðln PIBÞ2t þ a3 ðln PIBÞ3t þ a4 lnðENERGÞt þ a5 lnðDPOPÞt þ a6 lnðOUV Þt þ a7 lnðV AINDÞt þ et

ð2Þ

lnðTEMPÞt ¼ a0 þ a1 lnðPIBÞt þ a2 ðln PIBÞ2t þ a3 ðln PIBÞ3t þ a4 lnðENERGÞt þ a5 lnðDPOPÞt þ a6 lnðOUV Þt þ a7 lnðV AINDÞt þ et

ð3Þ

Ou la variable CO2 et TEMP representent respectivement les emissions de dioxyde de carbone par t^ete et l’evolution de la
temperature. La variable PIB est le revenu par t^ete il capture l’impact du niveau de developpement sur l’environnement. ENERG
est la consommation d’energie fossile en % de la consommation energetique totale, DPOP est la densite de la population, VAIND
est la valeur ajoutee du secteur industriel en % du produit interieur brut. Le terme a0 represente l’effet specifique inobserve, et est
le terme d’erreur, a1, a2, a3, a4, a5, a6 et a7 sont les coefficients a estimer, t ¼ 1. . ., T designe les periodes. OUV represente
l’ouverture commerciale, il permet de verifier l’hypothese de « havres de pollution » telle que developpe par Birdsall et Wheeler
(1992). En effet les pays en developpement attirent les activites dommageables a l’environnement gr^ace a une reglementation
environnementale moins stricte, ce qui reduit les coUts ^ de production et peut donc favoriser les delocalisations (Low et Yeats,
1992). Ces indicateurs sont extraits du World Development Indicators (World Bank, 2013). Les statistiques descriptives des
© 2017 The Authors. African Development Review © 2017 African Development Bank
Croissance economique et degradation de l’environnement 619

differentes variables quantitatives sont synthetisees dans le tableau 1 tandis que les tableaux 2 et 3 recensent les differentes
correlations entre les variables. Les matrices de correlation des variables des deux modeles laissent entrevoir une forte
correlation entre certaines variables.
Les tests d’Engel et Granger (1987) et le test de Johansen and Juselius (1990) sont les plus utilises pour identifier une relation de
cointegration (relation de long terme) entre plusieurs variables. Ces methodes necessitent que toutes les variables soient stationnaires
en difference premiere. Toutefois, ces methodes presentent des limites dans le cas des echantillons de petite taille (Chaudhry et
Choudhary, 2006). Pour pallier a ces limites, Pesaran et al. (2001) ont developpe le test de cointegration par les retards echelonnes, en
Anglais Auto Regressive Distributed Lags (ADRL) approach to cointegration popularisee par Pesaran et al. (2001); Pesaran et
Pesaran (1997). Cette methode a ceci de particuliere qu’elle ne necessite pas que toutes les variables soient integrees du m^eme ordre,
c’est-a-dire I(1). Cette methode est valable aussi bien pour les variable I(0), I(1) ou les deux (Pesaran et Pesaran, 1997).

Tableau 1: Statistique descriptives des variables


Variables Obs Moy Mediane Etendue Ecart type Min Max Iqr Skewness Kurtosis

lnCO2 40 –1.4 –1.53 2.02 0.522 –2.41 –0.394 0.516 0.441 2.6
lnPIB 40 7.04 7.03 0.677 0.175 6.75 7.42 0.203 0.456 2.58
lnPIB2 40 49.6 49.4 9.59 2.48 45.5 55.1 2.85 0.506 2.61
lnPIB3 40 350 348 102 26.4 307 409 30.1 0.555 2.64
lnENERG 40 2.8 2.8 1.28 0.295 2.2 3.48 0.206 0.0187 3.65
lnDPOP 40 3.24 3.25 1.09 0.331 2.69 3.77 0.577 –0.0633 1.76
lnOUV 40 –1.08 –1.05 0.71 0.199 –1.49 –0.782 0.345 –0.478 2.19
lnVAIND 40 3.32 3.4 0.854 0.233 2.77 3.62 0.222 –1.03 2.68
lnTEMP 40 3.17 3.17 0.0982 0.0204 3.1 3.2 0.0187 –1.35 5.14

Tableau 2: Matrice de correlation entre variables du modele 2


lnCO2 lnPIB lnPIB2 lnPIB3 lnENERG lnDPOP lnOUV lnVAIND

lnCO2 1.0000
lnPIB 0.5461 1.0000
lnPIB2 0.5439 0.9999 1.0000
lnPIB3 0.5415 0.9996 0.9999 1.0000
lnENERG 0.5260 0.5463 0.5399 0.5334 1.0000
lnDPOP 0.1542 0.0378 0.0294 0.0211 0.6993 1.0000
lnOUV 0.1525 0.0873 0.0823 0.0773 0.5977 0.8894 1.0000
lnVAIND 0.5079 0.4903 0.4875 0.4847 0.6508 0.6763 0.7563 1.0000

Tableau 3: Matrice de correlation entre variables du modele 3


lnTEMP lnPIB lnPIB2 lnPIB3 lnENERG lnDPOP lnOUV lnVAIND

lnTEMP 1.0000
lnPIB 0.0149 1.0000
lnPIB2 0.0156 0.9999 1.0000
lnPIB3 0.0163 0.9996 0.9999 1.0000
lnENERG 0.2953 0.5463 0.5399 0.5334 1.0000
lnDPOP 0.1507 0.0378 0.0294 0.0211 0.6993 1.0000
lnOUV 0.1561 0.0873 0.0823 0.0773 0.5977 0.8894 1.0000
lnVAIND 0.0123 0.4903 0.4875 0.4847 0.6508 0.6763 0.7563 1.0000

© 2017 The Authors. African Development Review © 2017 African Development Bank
620 E. Noubissi Domguia et H. Njangang Ndieupa

C’est sur cette base que nous effectuerons le test de cointegration a partir des equations suivantes:

X
q X
q X
q
DlnðCO2t Þ ¼ a0 þ a1 D lnðCO2ti Þ þ a2 D lnðPIBti Þ þ a3 D lnðENERGti Þ
t¼0 t¼0 t¼0
X
q X
q X
q
þ a4 D lnðDPOPti Þ þ a5 D lnðOUV ti Þ þ a6 D lnðV AINDti Þ ð4Þ
t¼0 t¼0 t¼0
þ a7 D lnðCO2t1 Þ þ a8 D lnðPIBt1 Þ þ a9 D lnðENERGt1 Þ þ a10 D lnðDPOPt1 Þ
þ a11 D lnðOUV t1 Þ þ a12 D lnðV AINDt1 Þ þ et

X
q X
q X
q
DlnðTEMPt Þ ¼ a0 þ a1 DlnðTEMPti Þ þ a2 D lnðPIBti Þ þ þ a3 D lnðENERGti Þ
i¼0 i¼0 i¼0
X
q X
q X
q
þ a4 D lnðDPOPti Þ þ a5 D lnðOUV ti Þ þ a6 D lnðV AINDti Þ þ a7 lnðTEMPti Þ ð5Þ
i¼0 i¼0 i¼0
þ a8 lnðPIBt1 Þ þ a9 D lnðENERGt1 Þ þ a10 D lnðDPOPt1 Þ þ a11 D lnðOUV t1 Þ
þ a12 D lnðV AINDt1 Þ þ et

L’implementation du test de cointegration par les retards echelonnes necessite qu’on effectue un test de racine unitaire pour
s’assurer que les variables ne sont pas I(2), auquel cas cette methode n’est plus valide. Dans l’optique de determiner la relation de
long terme donnee par les equations (4) et (5), nous effectuerons un test de cointegration au sens de Pesaran et al. (2001) en
utilisant la valeur de la statistique F. L’hypothese nulle stipule l’absence de cointegration. Ainsi lorsque la valeur du F est
superieure a la borne superieure on rejette H0. Si la valeur du F est inferieure a la borne inferieure on accepte H0. Si F est
comprise entre les deux bornes, le test est juge non concluant. A la suite de ce test, le retard optimal a ete selectionne a travers les
criteres d’information d’Akaike (AIC) et de Schwarz (SBC). La methode ADRL s’acheve avec l’estimation de la dynamique de
court terme (MCE) donne par les equations (6) et (7) suivantes:

X
q1 X
q2 X
q3 X
q4
D lnðCO2t Þ ¼ a0 þ a1i DlnðCO2ti Þ þ a2i D lnðPIBti Þ þ a2i D lnðENERGti Þþ a3i D lnðDPOPti Þ
t¼0 t¼0 t¼0 t¼0
X
q5 X
q6
þ a4i D lnðV AINDti Þ þ a5i D lnðOUV ti Þ þ lECT t1 þ U t ð6Þ
t¼0 t¼0

X
q1 X
q2 X
q3 X
q4
D lnðTEMPt Þ ¼ a0 þ a1i D lnðTEMPti Þ þ a2i D lnðPIBti Þ þ a2i D lnðENERGti Þþ a3i D lnðDPOPti Þ
t¼0 t¼0 t¼0 t¼0
X
q5 X
q6
þ a4i D lnðV AINDti Þ þ a5i D lnðOUV ti Þ þ lECT t1 þ U t ð7Þ
t¼0 t¼0

Ou qi (i ¼ 1,2,. . .6) represente les retards optimaux.

4. Resultats des estimations et discussion

4.1 Test de stationnarite


Les tests de stationnarite de Dickey–Fuller Augmente (DFA; 1979) et Phillips–Perron (PP; 1988) ont ete effectues afin de se
rassurer qu’aucune variable n’est integree a un ordre superieur a 1, condition sous laquelle le test de cointegration par les retards
echelonnes propose par Pesaran et al. (Pesaran et Shin, 1999; Pesaran et al., 2001) cesse d’^etre valable. Ces tests indiquent que

© 2017 The Authors. African Development Review © 2017 African Development Bank
Croissance economique et degradation de l’environnement 621

Tableau 4: Resultat des tests de stationnarite de DFA, DF-GLS et de PP


ADF test DF-GLS test PP test
Variables SIC retards t-Stat Valeur critique a 5% SIC retards t-Stat Valeur critique a 5% t-Stat Valeur critique a 5%

lnCO2 1 –6.53 –2.94 1 –6.63 –1.94 –8.49 –2.94


lnPIB 0 –4.015 –2.94 0 –4.06 –1.94 –4.018 –2.94
lnENERG 0 –6.69 –2.94 0 –6.78 –1.94 –6.67 –2.94
lnDPOP 0 –6.60 –3.53 1 –2.40 –1.95 –18.21 –1.94
lnVAIND 0 –5.17 –2.94 0 –5.15 –2.62 –5.16 –2.94
lnOUV 1 –5.96 –2.94 1 –5.90 –2.62 –8.09 –2.94
lnTEMP 0 –4.059 –2.93 0 –7.93 –1.94 –4.09 –2.93

toutes les variables respectent les normes d’application de l’ARDL, l’ordre d’integration maximal des variables etant 1 (tableau
4). De m^eme, ce resultat est confirme par le test de racine unitaire avec une rupture (tableau 5).

4.2 Test de cointegration


Le tableau 6 presente les resultats du test de cointegration de Pesaran.
Ces resultats rejettent l’hypothese d’absence de cointegration au seuil de 1% pour l’ensemble des modeles. Ouattara (2004) et
Akpan et Chuku (2011) affirment que l’une des conditions d’application du test de cointegration par les Retards Echelonnes de
Pesaran et al. (2001) est qu’aucune serie ne doit ^etre integree a un ordre superieur a 1.

Tableau 5: Resultats du Test de racine unitaire avec ruptures (une rupture)


Perron (1997)(a) IO Model Perron (1997)a AO Model
Variables t-stat Valeur critique a 5% Date de rupture t-stat Valeur critique a 5% Date de rupture Decisions

lnCO2 –5.02 –4.44 1989 –6.55 –4.44 1988 Stationnaire


lnPIB –5.03 –4.44 1986 –9.57 –4.44 1992 Stationnaire
lnENERG –5.55 –4.44 2006 –6.93 –4.44 2000 Stationnaire
lnDPOP –6.90 –4.44 2007 –6.24 –4.44 2001 Stationnaire
lnVAIND –5.60 –4.44 1984 –5.39 –4.44 1989 Stationnaire
lnOUV –6.97 –4.44 1984 –6.11 –4.44 1995 Stationnaire
lnTEMP –6.90 –4.44 2007 –6.24 –4.44 2006 Stationnaire
a
Suppose l’absence de rupture a l’hypothese nulle de racine unitaire.
Dans le modele IO (Inovational Outlier model) les changements sont supposes se produire progressivement, permettant la rupture aussi bien dans la constante
que sur la pente. Dans le modele AO (Additive Outlier), les changements se produisent rapidement, permettant la rupture uniquement dans la pente.

Tableau 6: Estimation du modele optimal et resultat du test de cointegration


Retard optimal F-Statistique Decision
Modele 2 2 9.3731 Cointegration
Modele 3 2 8.2149 Cointegration
Valeurs critiques de Pesaran 10% 5% 1%
2.03 3.13 2.32 3.5 2.96 4.26
Notes: Les retards optimaux de chaque modele sont obtenus a partir des criteres d’information BIC. La borne superieure de Pesaran est lue dans la table CI(ii),
Case II: restricted intercept and no trend, k ¼ 6.

© 2017 The Authors. African Development Review © 2017 African Development Bank
622 E. Noubissi Domguia et H. Njangang Ndieupa

4.3 Resultats de l’estimation du modele de long terme


Le tableau 7 presente les resultats de l’estimation des effets de la croissance economique et de ses determinants sur les emissions
de CO2 et la temperature au Cameroun. Plusieurs tests a l’instar du test de correlation serielle (LM test), le test de normalite des
termes d’erreurs (JB) et le test d’heteroscedasticite (Test ARCH) ont ete effectues sur le modele de long terme. Les resultats
montrent que ces tests sont concluants. En effet, il n’existe pas de correlation serielle et les termes d’erreurs sont normalement
distribues. Il n’y a pas de preuve d’une heteroscedasticite traditionnelle autoregressive. Par consequent, le modele de long terme
est bien specifie.
De ce tableau, on constate que la croissance economique (PIB) a des effets negatifs et significatifs sur les emissions de CO2 de
m^eme que sur l’evolution des temperatures.
Le modele 2 laisse presager une relation en N inverse entre la croissance economique et le niveau de CO2. En effet, a1 < 0;
a2 > 0 et a3 < 0, alors la relation entre le revenu (PIB) et la pollution CO2 a une forme cubique polynomiale ou en forme de « N
inverse ». De plus, les valeurs prises par ces parametres sont significatifs. Ce resultat est conforme a celui trouve par BaSs ar et
Temurlenk (2007) pour la Turkey. Par ailleurs, la consommation d’energies fossiles, la valeur ajoutee des industries et
l’ouverture commerciale constituent les principaux determinants de la pollution atmospherique dans ce modele. Ce modele nous
montre egalement que l’utilisation des energies fossiles contribue a accentuer la degradation de l’environnement au vu de la
valeur de son parametre positif et significatif. Ce resultat est conforme a celui de Halicioglu (2009), Akpan et al. (2012) qui
constatent qu’une augmentation de la consommation d’energie se traduit par la hausse des emissions de CO2 en Turquie et au
Nigeria respectivement. On constate egalement que le secteur industriel contribue de maniere significative aux emissions de CO2
au Cameroun. Ce resultat peut s’expliquer, comme a long terme par la forte dependance de ce pays aux activites extractives

Tableau 7: Resultats des relations de long terme


Modele 2 ayant lnCO2 comme variable Modele 3 ayant lnTEMP comme variable
dependante de capture de la qualite de dependante de capture de la qualite de
Variables l’environnement l’environnement

DLnPIB –3868.173914 –129.859092


(–2.639168) (–2.470142)
D(LnPIB2) 543.701554 18.378424
(2.636241) (2.482676)
D(LnPIB3) –25.459563 –0.866407
(–2.633225) (–2.494472)
DLnDPOP 2.433023 1.109816
(0.186685) (1.297321)
DLnENERG 0.800377 –0.042701
(2.346905) (–1.859442)
DLnOUV –2.729462 –0.030667
(–3.522011) (–1.039365)
DLnVAIND 1.366347 0.006197
(2.498772) (0.239584)
C –0.038349 –0.030668
(–0.105288) (–1.274005)
Nombre d’observations 40 40
Test de normalite des residus (Jarque Berra) 0.2479 1.3031
LM test 5.095 8.7467
(0.0783)a (0.0677)a
Test ARCH 0.433 3.7803
(0.8051)a (0.1510)a
Test de stabilite (CUSUM ET CUSUMQ) Les lignes bleues evoluent a Les lignes bleues evoluent a
l’interieur des bornes l’interieur des bornes
Notes: (.) est le t de students, (.)a est la probabilite  ,  
, representent respectivement la significativite a 1%, 5% et 10%.

© 2017 The Authors. African Development Review © 2017 African Development Bank
Croissance economique et degradation de l’environnement 623

(Bois, Petrole et minerais) qui contribuent de plus en plus a la croissance economique. Par contre, l’ouverture commerciale
semble ne pas contribuer a la degradation de l’environnement au Cameroun tout au contraire, elle semble ^etre benefique pour
l’environnement. Ceci peut se justifier par le fait qu’au Cameroun, les exportations occupent une place preponderante dans les
transactions commerciales du Cameroun avec l’exterieur. De m^eme, les importations ont ete longtemps contr^olees et limitees par
les differentes barrieres douanieres. De plus, les activites menees par la plupart des menages sont des activites agricoles
traditionnelles et le tissu industriel est faiblement developpe. Ce modele accepte donc l’hypothese du havre de transmission car
la valeur de son parametre est significative.
Toutefois, pour le modele 3 bien que la valeur des parametres des variables explicatives soit moins importante, elle laisse
presager une relation en N inverse entre l’evolution de la temperature et la croissance economique. Nous sommes en effet dans le
cas ou a1 < 0; a2 > 0 et a3 < 0 donc la forme de la relation entre le revenu (PIB) et l’evolution des temperatures est en « N »
inverse. Ce resultat vient confirmer celui trouve par Nordhaus (2008) qui semble confirmer l’existence d’une correlation
negative entre la temperature et la croissance economique. En outre, la consommation d’energies fossiles constitue le principal
determinant de la pollution atmospherique dans ce modele. L’accroissement de la population entraine l’augmentation des
temperatures ou la deterioration de la qualite de l’environnement. Ce resultat vient confirmer les resultats trouves par Brajer et al.
(2007) qui montrent que l’augmentation de la densite de la population est associee a un accroissement des emissions de dioxyde

Tableau 8: Resultats de l’estimation de modele de court terme


Modele 2 Modele 3
Variables Retards Estimations Retards Estimations

D(lnCO2) 1 0.823051
(4.835272)
D(lnTEMP) 1 0.341723
(2.835001)
D(lnPIB) 0 –2480.885046 0 –226.117248
(–1.021056) (–2.562347)
1 4598.681991
(1.747573)
D(lnPIB2) 0 335.873651 0 31.992813
(0.982745) (2.573108)
1 –650.014766
(–1.747083)
D(lnPIB3) 0 –15.099137 0 –1.508225
(–0.942547) (–2.583789)
1 30.618582
(1.746833)
D(lnDPOP) 0 6.440274 1 1.931947
(0.186575) (1.262347)
D(lnENERG) 0 0.769281 0 0.043252
(1.288701) (1.771288)
D(lnOUV) 0 –3.451543 0 0.033839
(–3.071996) (1.001135)
1 2.730081
(2.806866)
D(lnVAIND) 0 2.012114 0 –0.039112
(1.894521) (–1.226613)
1 1.0920034
(1.118563)
ECT(-1) 1 –2.647026 –1.740781
(–8.247710) (–8.491169)

© 2017 The Authors. African Development Review © 2017 African Development Bank
624 E. Noubissi Domguia et H. Njangang Ndieupa

Figure 1: Test de stabilite des parametres du modele 2

Figure 2: Test de stabilite des parametres du modele 3

© 2017 The Authors. African Development Review © 2017 African Development Bank
Croissance economique et degradation de l’environnement 625

de souffre en Chine. Enfin, on constate que le commerce international ne serait pas un determinant de l’evolution de la
temperature au Cameroun. Contrairement a la prediction theorique, ce resultat indique que la liberalisation commerciale ne se
traduit pas necessairement par la migration des entreprises polluantes des pays developpes en direction des pays en
developpement, moins intransigeants en matiere de protection environnementale.

4.4 Resultats de l’estimation du modele de court terme


Le tableau 8 presente les resultats de l’estimation de la dynamique de court terme des effets de la croissance economique aussi
bien sur le CO2 que sur l’evolution de la temperature au Cameroun.
Du mecanisme a correction d’erreur, il ressort que le terme de correction d’erreur (ECT) a un coefficient negatif et significatif
a 1%. Ce qui confirme l’existence d’une relation de cointegration. Le coefficient du terme ECT en valeur absolue est 2,647026 et
1,740781 pour les modeles 2 et 3 respectivement. L’existence de l’ECT implique des variations de la variable dependante. Ces
variations sont fonction des niveaux de desequilibre dans la relation de cointegration d’une part et des variations des autres
variables explicatives d’autre part. Ceux-ci indiquent l’allure de la deviation de la variable dependante a court terme vers sa
trajectoire de long terme. On peut constater que dans le modele 2, l’ouverture commerciale impacte negativement les emissions
de CO2 tandis que le secteur industriel affecte positivement ces emissions de CO2. Par contre dans le modele 3, la consommation
d’energie fossile impacte positivement et significativement l’evolution de la temperature. Les tests de stabilite sont utilises pour
explorer la stabilite des parametres. Pour ce fait, nous avons effectue le test de CUSUM et le test de CUSUMQ (Figures 1 et 2).
De ce modele a court terme, on constate que le modele 2 prend la forme de N alors que le modele 3 prend la forme d’un N
inverse.

5. Conclusion et recommandations
L’objectif de ce papier etait d’examiner les effets de la croissance economique sur les emissions de CO2 et l’evolution de la
temperature au Cameroun. A partir des series temporelles sur une periode allant de 1972 a 2010, il a ete applique le test de
cointegration par les retards echelonnes developpe par Pesaran et al. (Pesaran et Pesaran, 1997; Pesaran et al., 2001). Afin de
contourner les limites des tests de cointegration traditionnels. Globalement, il apparait que la relation qui lie les emissions de
CO2 avec le PIB rev^et la forme de « N » inversee de m^eme, le lien entre l’evolution de la temperature par rapport au PIB est de la
m^eme nature. Les variables explicatives retenues ont des effets differents suivant les modeles. Par exemple, la consommation
d’energie et les activites industrielles seraient les principaux determinants des emissions de CO2 ce qui n’est pas le cas pour la
determination de l’evolution de la temperature au Cameroun.
Au plan politique, ces resultats suggerent que les objectifs de croissance soient accompagnes des mesures d’adaptation. Dans
ce cas, il faut incorporer des programmes d’adaptation aux strategies de developpement, telle que l’initiative ethiopienne qui
prevoit des limites aux emissions, une productivite accrue et un meilleur rendement des ressources.
Par ailleurs, le Cameroun doit promouvoir une Croissance Inclusive Verte, qui passera necessairement par la lutte contre les
inegalites d’opportunites, les investissements en Recherche et Developpement, la sensibilisation des populations sur les risques
environnementaux et enfin, la collecte et le suivi des indicateurs environnementaux.

References
Aboagye, S. (2017), ‘Economic Expansion and Environmental Sustainability Nexus in Ghana’, African Development Review,
Vol. 29, pp. 155–68.
Akpan, U. F. et C. A. Chuku (2011), ‘Economic Growth and Environmental Degradation in Nigeria: Beyond the Environmental
Kuznets Curve’, paper presented at the 4th NAEE/IAEE International Conference, Abuja, April.
Akpan, S. B., V. Ini-mfon, E. Glory, and E. Daniel (2012), ‘Agricultural Productivity and Macro-economic Variable Fluctuation
in Nigeria’, International Journal of Economics and Finance, Vol. 4, pp. 114–25.
Apergis, N. et J. E. Payne (2009), ‘CO2 Emissions, Energy Usage, and Output in Central America’, Energy Policy, Vol. 37,
pp. 3282–86.

© 2017 The Authors. African Development Review © 2017 African Development Bank
626 E. Noubissi Domguia et H. Njangang Ndieupa

Baldwin, R. (1995), ‘Does Sustainability Require Growth?’, in I. Goldin and L. A. Winters (eds.), The Economics of Sustainable
Development, Cambridge University Press, Cambridge, pp. 19–47.
Banque Mondiale (1992), Le developpement et l'environnement, rapport sur le developpement dans le monde, Oxford
University Press, Oxford.
BaSs ar, S. et M. S. Temurlenk (2007), ‘Environmental Kuznets Curve: An Empirical Analysis for Turkey’, Journal of Economics
and Administrative Sciences, Vol. 21, No. 1, pp. 1–12 (in Turkish).
Beckerman, W. (1992), ‘Economic Growth and the Environment: Whose Growth? Whose Environment?’, World Development,
Vol. 20, pp. 481–96.
Ben Jebli, M., S. Ben Youssef et I. Ozturk (2015), ‘The Role of Renewable Energy Consumption and Trade: Environmental
Kuznets Curve Analysis for Sub-Saharan Africa Countries’, African Development Review, Vol. 27, pp. 288–300.
Bessagnet, B., J. M. Brignon, A. C. Le Gall, F. Meleux, S. Schucht et L. Rou€ıl (2009), ‘Politiques combinees de gestion de la
qualite de l’air et du changement climatique (partie 1): enjeux, synergies et antagonismes’, Rapport d'etude /2009 No. DRC-09-
103681-02123A PDF 91 pages, INERIS, realise pour MEEDDAT/DGEC.
Bhagawati, J. (1993), ‘The Case for Free Trade’, Scientific American, pp. 42–49.
Birdsall, N. et D. Wheeler (1992), ‘Trade Policy and Industrial Pollution in Latin America: Where are the Pollution Havens?’, in
P. Low (ed.), International Trade and the Environment, World Bank Discussion Papers No.159, World Bank, Washington DC,
pp. 159–69.
Brajer, V., R. W. Mead et F. Xiao (2007), ‘Health Benefits of Tunneling through the Chinese Environmental Kuznets Curve
(EKC)’, Ecological Economics, Vol. 66, No. 4, pp. 674–86.
Carson, R. T., Y. Jeon et D. McCubbin (1997), ‘The Relationship between Air Pollution Emission and Income: US Data’,
Environmental and Development Economics, Vol. 2, pp. 433–50.
Chandran, V. G. R. et C. F. Tang (2013), ‘The Impacts of Transport Energy Consumption, Foreign Direct Investment and
Income on CO2 Emissions in ASEAN-5 Economies’, Renewable and Sustainable Energy Reviews, Vol. 24, pp. 445–53.
Chaudhry, M. A. et M. A. S. Choudhary (2006), ‘Why the State Bank of Pakistan Should not Adopt Inflation Targeting’, SBP
Research Bulletin, Vol. 2, pp. 195–209.
Cho, C. H., Y. P. Chu et H. Y. Yang (2014), ‘An Environment Kuznets Curve for GHG Emissions: A Panel Cointegration
Analysis’, Energy Sources, Part B: Economics, Planning, and Policy, Vol. 9, pp. 120–29.
Clement, M. et A. Meunie (2010), ‘Inegalites, developpement et qualite de l’environnement: mecanismes et application
empirique’, Mondes en developpement, No. 151, pp. 67–82.
Cole, M. A. et E. Neumayer (2004), ‘Examining the Impact of Demographic Factors on Air Pollution’, Population and
Environment, Vol. 26, No. 1, pp. 5–21.
Day, K. M. et R. Q. Grafton (2003), ‘Growth and the Environment in Canada: An Empirical Analysis’, Canadian Journal of
Agricultural Economics, Vol. 51, pp. 197–216.
Dell, M., B. Jones et B. Olken (2012), ‘Temperature Shocks and Economic Growth: Evidence from the Last Half Century’,
American Economic Journal: Macroeconomics, Vol. 4, No. 3, pp. 66–95.
Dickey, D. A. et W. A. Fuller (1979), ‘Distribution of the Estimators for Autoregressive Time Series with a Root’, Journal of the
American Statistical Association, Vol. 74, pp. 427–31.
Dinda, S. (2004), ‘Environmental Kuznets Curve Hypothesis: A Survey’, Ecological Economics, Vol. 49, pp. 431–55.
Du, Y. L, Z. C. Liu and C. R. Hao (2012), ‘Study on the Relationship between Economic Development and Ambient Air
Pollution Levels of Jinan’, Journal of Shandong Jianzhu University, Vol. 27, pp. 20–24.
Engle, R. F. et C. W. J. Granger (1987), ‘Cointegration and Error-Correction: Representation, Estimation and Testing’,
Econometrica, Vol. 64, pp. 813–36.

© 2017 The Authors. African Development Review © 2017 African Development Bank
Croissance economique et degradation de l’environnement 627

Gao, J. et L. Zhang (2014), ‘Electricity Consumption-Economic Growth-CO2 Emissions Nexus in Sub-Saharan Africa:
Evidence from Panel Cointegration’, African Development Review, Vol. 26, pp. 359–71.
Grossman, G. M. et A. B. Krueger (1991), ‘Environmental Impacts of the North American Free Trade Agreement’, NBER
Working Paper 3914.
Grossman, G. M. et A. B. Krueger (1994), ‘Economic Growth and the Environment’, National Bureau of Economic Research,
Working Paper No. 4634.
Grossman, G. M. et A. B. Krueger (1995), ‘Economic Growth and the Environment’, Quarterly Journal of Economics, Vol. 110,
No. 2, pp. 353–77.
Haisheng, Y., J. Jia, Z. Yongzhang et W. Shugong (2005), ‘The Impact on Environmental Kuznets Curve by Trade and Foreign
Direct Investment in China’, Chinese Journal of Population Resource and Environment, Vol. 3, pp. 14–19.
Halicioglu, F. (2009), ‘An Econometric Study of CO2 Emissions, Energy Consumption, Income and Foreign Trade in Turkey’,
Energy Policy, Vol. 39, pp. 1156–64.
Ismael, M. A. and M. Y. Mawar (2012), ‘Energy Use, Emissions, Economic Growth and Trade: A Granger Non-causality
Evidence for Malaysia’, MPRA Paper No. 38473.
Jalil, A. et M. Feridun (2011), ‘The Impact of Growth, Energy and Financial Development on the Environment in China:
A Cointegration Analysis’, Energy Economics, Vol. 33, pp. 284–91.
Jevons, W. S. (1865), The Coal Question, Macmillan and Co., London.
Johansen, S. and K. Juselius (1990), ‘Maximum Likelihood Estimation and Inference on Cointegratio–With Application to the
Demand for Money’, Oxford Bulletin of Economics and Statistics, Vol. 52, pp. 169–210.
Lean, H. H. et R. Smyth (2010), ‘CO2 Emissions, Electricity Consumption and Output in ASEAN’, Applied Energy, Vol. 87,
pp. 1858–64.
Llorca, M. et A. Meunie (2009), ‘SO2 Emissions and the Environmental Kuznets Curve: The Case of Chinese Provinces’,
Journal of Chinese Economic and Business Studies, Vol. 7, pp. 1–16.
Low, P. et A. Yeats (1992), ‘Do “Dirty” Industries Migrate?’, in P. Low (ed.), International Trade and the Environment, World
Bank Discussion Paper No. 159, World Bank, Washington DC, pp. 89–104.
Malthus, T. R. (1798), An Essay on the Principle of Population, St. Paul’s Church-Yard, London.
Meadows, D. H. et Club of Rome (1972), ‘The Limits to Growth’, A Report for the Club of Rome’s Project on the Predicament of
Mankind, Universe Books, New York.
Ngepah, N. (2011), ‘Exploring the Impact of Energy Sources on Production, Inequality and Poverty in Simultaneous Equations
Models for South Africa’, African Development Review, Vol. 23, No. 3, pp. 335–51.
Nordhaus, W. (2008), A Question of Balance: Weighing the Options on Global Warming Policies, Yale University Press, New
Haven, CT.
Nourry, M. (2007), ‘La croissance economique est-elle un moyen de lutte contre la pollution ?’, Revue franSc aise d’economie,
Vol. 21 No. 3, pp. 137–76.
ONU (2013), Objectifs du Millenaire pour le developpement, Rapport 2013, Nations Unies, New York.
Ouattara, B. (2004), ‘The Impact of Project Aid and Program Aid on Domestic Savings: A Case Study of C^ote d’Ivoire’, Centre
for the Study of African Economies Conference on Growth, Poverty Reduction and Human Development in Africa, April.
Ozcan, B. (2013), ‘The Nexus between Carbon Emissions, Energy Consumption and Economic Growth in Middle East
Countries: A Panel Data Analysis’, Energy Policy, Vol. 62, pp. 1138–47.
Panayotou, T. (1993), ‘Empirical Tests and Policy Analysis of Environmental Degradation at Different Stages of Economic
Development’, Working Paper WP238 Technology and Employment Programme, International Labor Office, Geneva.

© 2017 The Authors. African Development Review © 2017 African Development Bank
628 E. Noubissi Domguia et H. Njangang Ndieupa

Panayotou, T. (1995), ‘Environmental Degradation at Different Stages of Economic Development’, in I. Ahmed and J. A.
Doeleman (eds.), Beyond Rio: The Environmental Crises and Sustainable Livelihoods in the Third World, Macmillan Press,
London.
Pesaran, M. H. et B. Pesaran (1997), Working with Microfit 4.0: Interactive Econometic Analysis, Oxford University Press,
Oxford.
Pesaran, M. H. et Y. Shin (1999), ‘An Autoregressive Distributed Lag Modeling Approach to Cointegration Analysis’, Chapter
11 in S. Strom (ed.), Econometrics and Economic Theory in the 20th Century: The Ragnar Frisch Centennial Symposium,
Cambridge University Press, Cambridge.
Pesaran, M. H., Y. Shin et R. J. Smith (2001), ‘Bounds Testing Approaches to the Analysis of Level Relationships’, Journal of
Applied Econometrics, Vol. 16, pp. 289–326.
Pezzey, J. C. V. (1989), ‘Economic Analysis of Sustainable Growth and Sustainable Development’, Environment Department
Working Paper, 15, World Bank.
Phillips, P. C. and P. Perron (1988), ‘Testing for a Unit Root in Time Series Regression’, Biometrika, Vol. 75, pp. 335–46.
Ricardo, D. (1817), ‘On the Principles of Political Economy and Taxation’, London.
Roca, J. (2003), ‘Do Individual Preferences Explain Environmental Kuznets Curve?’, Ecological Economics, Vol. 45, pp. 3–10.
Saboori, B., J. Sulaiman et S. Mohd (2012), ‘Economic Growth and CO2 Emissions in Malaysia: A Cointegration Analysis of the
Environmental Kuznets Curve’, Energy Policy, Vol. 51, pp. 184–91.
Selden, T. M. et D. Song (1994), ‘Environmental Quality and Development: Is There a Kuznets Curve for Air Pollution
Emissions?’, Journal of Environmental Economics and Management, Vol. 27, No. 2, pp. 147–62.
Shafik, N. et S. Bandyopadhyay (1992), ‘Economic Growth and Environmental Quality: Time Series and Cross-country
Evidence’, Background Paper for the World Development Report 1992, World Bank, Washington, DC.
Shahbaz, M., A. Jalil et S. Dube (2010), ‘Environmental Kuznets Curve (EKC): Times Series Evidence from Portugal’, MPRA.
Shahbaz, M., S. A. Solarin, H. Mahmood and M. Arouri (2013), ‘Does Financial Development Reduce CO2 Emissions in
Malaysian Economy? A Time Series Analysis’, Economic Modeling, Vol. 35, pp. 145–52.
Shahbaz, M., S. Dube and A. Jalil (2015), ‘Testing the Environmental Kuznets Curve Hypothesis in Portugal’, International
Journal of Energy Economics and Policy, Vol. 5, No. 2, pp. 475–81.
Shi, A. (2003), ‘The Impact of Population Pressure on Global Carbon Dioxide Emissions: Evidence from Pooled Cross-country
Data’, Ecological Economics, Vol. 44, No. 1, pp. 29–42.
Shukla, V. and K. Parikh (1996), ‘The Environmental Consequences of Urban Growth: Cross-national Perspectives on
Economic Development, Air Pollution and City Size’, in V. Shukla (ed.), Urbanisation and Economic Growth, Oxford
University Press, Delhi, Ch. 15, pp. 361–95.
Stern, D. I., M. S. Common and E. B. Barbier (1996), ‘Economic Growth and Environmental Degradation: The Environmental
Kuznets Curve and Sustainable Development’, World Development, Vol. 24, pp. 1151–60.
Tiwari, A. K., M. Shahbaz and Q. M. A. Hye (2013), ‘The Environmental Kuznets Curve and the Role of Coal Consumption in
India: Cointegration and Causality Analysis in an Open Economy’, Renewable and Sustainable Energy Review, Vol. 18,
pp. 519–27.
Van, P. N. et T. T. Azomahou (2003), ‘Deforestation, croissance economique et population: Une etude sur donnees de panel’,
Revue economique, Vol. 54, pp. 835–55.
Wang, S. S., D. Q. Zhou, P. Zhou and Q. W. Wang (2011), ‘CO2 Emissions, Energy Consumption and Economic Growth in
China: A Panel Data Analysis’, Energy Policy, Vol. 39, pp. 4870–75.
World Bank (2013), World Development Indicators, World Bank, Washington DC.

© 2017 The Authors. African Development Review © 2017 African Development Bank
Croissance economique et degradation de l’environnement 629

Yavuz, N. C.S (2014), ‘CO2 Emission, Energy Consumption, and Economic Growth for Turkey: Evidence from a
Cointegration Test with a Structural Break’, Energy Sources, Part B: Economics, Planning, and Policy, Vol. 9, No. 3, pp.
229–35.
Zambrano-Monserrate, M., F. Garcia-Alban and K. Henk-Vera (2016), ‘Bounds Testing Approach to Analyze the
Existence of an Environmental Kuznets in Ecuador’, International Journal of Energy Economics and Policy, Vol. 6, No.
2, pp. 159–66.

© 2017 The Authors. African Development Review © 2017 African Development Bank
African Development Review, Vol. 29, No. 4, 2017, 630–647

Inequity in Maternal and Child Health Care Utilization in Nigeria

Rifkatu Nghargbu and Olanrewaju Olaniyan

Abstract: The study assesses the degree and patterns of horizontal inequity in maternal and child health care utilization in
Nigeria. Horizontal inequity was estimated using concentration curves and a standardized concentration index, which was
decomposed into need and non-need variables to capture the major drivers of inequity in utilization. Maternal and child health
care utilization was measured by antenatal visits, skilled delivery and immunization utilization. Four rounds of Nigerian
Demographic and Health Survey (DHS) data (1999, 2003, 2008, 2013) were used. The results show a positive horizontal
inequity index from 1999 to 2013; which indicates that there is pro-rich inequity in utilization of antenatal, skilled delivery and
immunization, respectively. The study concludes that there is higher and increasing inequity in maternal health care while
inequity in utilization of child health care decreased over time with wealth status and education as the major non-need drivers of
inequity in utilization. Interventions to enhance women’s wealth status and education could significantly improve equity in
skilled delivery and antenatal care utilization. Hence wealth creation, women empowerment and education should be given more
attention to improve the deteriorating equity in maternal health care utilization.

1. Introduction
The health and survival of mothers and children constitute a major challenge to governments and health policymakers in
developing countries. In developing countries 830 women and 6,000 children die daily from preventable causes mostly related to
pregnancy and child birth; this accounts for 99 per cent of global maternal and child deaths (WHO, 2016). In Nigeria, 576 of
every 100,000 live births results in the mother’s death; this constitutes 14 per cent of the world maternal mortality burden
(UNICEF, 2009). Although the under-five mortality rate decreased from 201 deaths per 1,000 live births in 2003 to 109 in 2015,
Nigeria could not achieve the Millennium Development Goal (MDG) target of 64 in 2015. The high rate of maternal and child
mortality constitutes a major economic problem due to its impact on productivity, poverty, human capital formation and
economic growth.
The Nigerian Demographic and Health Survey (DHS) reports (National Population Commission, 1990–2013) show few
statistical evidences of socioeconomic inequality in maternal and child health care utilization. However, these evidences are not
sufficient for policy that will deal with the challenge of maternal and child mortality in Nigeria. The primary objective of this
study is to estimate horizontal inequity in maternal and child health care utilization in Nigeria. Further, we investigate the
contributions of need and non-need variables in explaining the drivers of inequity in maternal and child health care utilization in
Nigeria.
In recent times, there has been increased growth of interest on equity/inequity in health care utilization as a research topic on
the part of policymakers, donors and non-governmental organizations due to the essential role it plays in resource allocation and
general well-being of a nation (O’Donnell et al., 2008; Whitehead, 1985). Equity in use of health care is defined according to two
major classifications; vertical and horizontal equity. Vertical equity means providing unequal treatment of health care need to
individuals with unequal health care need (Wagstaff and van Doorslaer, 2000; Culyer and Wagstaff, 1993). Horizontal equity is
the notion that persons with similar health conditions and needs should be given equal treatment irrespective of their income, and
socioeconomic status (Wagstaff and van Doorslaer, 1998, 2000; Culyer and Wagstaff, 1993). Inequity, on the other hand, simply
means when there is no equity; that is, health care resources are unfairly distributed and utilized based on income and
socioeconomic status. Inequity is also categorized into vertical and horizontal. Vertical inequity in health care resource


Rifkatu Nghargbu, Department of Economics, Usmanu Danfodiyo University Sokoto, Nigeria; e-mail: rifnghargbu@gmail.com. Olanrewaju
Olaniyan Department of Economics, University of Ibadan, Nigeria; e-mail: lanreolaniyan@yahoo.co.uk
© 2017 The Authors. African Development Review © 2017 African Development Bank. Published by Blackwell Publishing Ltd,
9600 Garsington Road, Oxford OX4 2DQ, UK and 350 Main Street, Malden, MA 02148, USA. 630
Maternal and Child Health Care 631

allocation occurs when greater treatment of health care need is not provided to persons with greater health care need while
horizontal inequity occurs when persons with similar health care needs do not receive similar treatments for their health care
need, rather health care is utilized based on income and socioeconomic status.
Horizontal inequity also means when utilization of health care is affected by non-need variables so that individuals with the
same level of needs consume different amounts of care (Wagstaff and van Doorslaer, 1998; Gravelle et al., 2006). Horizontal
inequity is identified by comparing the deviation of the actual health care utilization from that which would arise if utilization
were driven by need alone (Gravelle et al., 2006). Need variables are those factors that predispose an individual to use health
care, this can be ill health, age and disability (Culyer, 2001; Culyer and Wagstaff, 1993; Gravelle et al., 2006) while non-need
variables are factors that influence the use of health care outside the need variables. The non-need variables are socioeconomic
factors such as wealth, education, region, residence, employment, insurance and marital status. Inequity is synonymous to
inequality but they are not the same. Inequality in health care represents absolute differences in health care utilization between
individual populations, while inequity in health care utilization represents inequalities that are considered unfair or unjust
between different social groups in relation to their need for the health care (Fleurbaey, 2006). Horizontal inequity in utilization is,
therefore, part of inequality, that is considered unjustified/unfair (O’Donnell et al., 2008; Fleurbaey and Schokkaert, 2009;
Wagstaff et al., n.d.).
Inequity in this study is important for policymakers because the study provides components of the causes of inequity in
maternal and child health care utilization that will give policymakers ideas on what to focus on in tackling the problem of high
maternal and child mortality. Following this introduction, the rest of the paper is organized into literature review, methodology,
results, discussions, conclusions and policy recommendations.

2. Empirical Literature
Empirical studies on inequality, whether on health, health care or nutrition, use the decomposition method of estimation
(Abalo and Agbodji, 2014; Ssewanyana and Kasirye, 2012). Most studies on inequity/equity in health care utilization estimated
horizontal inequity by standardizing for differences in need and non-need variables on out-patient, in-patient, dental, specialist,
hospital care and contraceptives. They proxy need using age, sex, self-reported health problems, symptoms of chronic illness and
family planning desire, while non-need variables were mostly measured by socioeconomic and demographic factors (Ong et al.,
2009; Allin, 2006; Bonfruer et al., 2012; Gravelle et al., 2006; Van de Poel et al., 2011; Fleurbaey and Schokkaert, 2009;
Hotchkiss and Maido, 2011). Some studies have attempted to estimate inequity and inequality using regression, descriptive, and
concentration index methods in maternal and child health care utilization without standardization for differences in need and
non-need variables (Houweling et al., 2007; Bonfruer et al., 2012; Zere et al., 2011; Antai 2011; Alam et al., 2015; Målqvist
et al., 2012).
Houweling et al. (2007) estimated pro-rich inequality in use of skilled delivery and antenatal care for 45 developing countries
and then compared it with child health care. They observed pro-rich inequality in skilled delivery which was higher than
antenatal care, child immunization and treatment of respiratory infection. Bonfruer et al. (2012) examined whether health care
utilization matches needs in 18 sub-Saharan Africa countries using a concentration index to estimate inequality in antenatal and
skilled delivery. Inequality was observed in all countries. The study fails to standardize to obtain horizontal inequity even though
homogenous need across sample was observed. Similarly, Zere et al. (2011) observed high level of socioeconomic inequity
with pro-rich bias in antenatal and skilled delivery in the private health facilities in Namibia. They found that women with high
wealth status and education have a 70 per cent utilization level compared to the poorest with 47 per cent. Alam et al. (2015) found
that inequality in maternal health care utilization in six sub-Saharan African countries (Ethiopia, Madagaska, Uganda,
Cameroon, Zambia and Zimbabwe) have declined over time in the countries making progress towards reducing maternal
mortality although wealth-related inequalities remained high in all countries. Målqvist et al. (2012) observed that influence of
informal fees and testimonies of discrimination and negative attitudes from health staff towards women and ethnic minorities
were responsible for the low utilization level. However, these studies did not estimate horizontal inequity.
There have been attempts to estimate inequality using the concentration index in Bangladesh, Pakistan, Ethiopia and five
other African countries with a focus on standardizing the concentration index (Hossain, 2010; Mahmood and Bashir, 2012;
Memirie et al., 2016). Hossain (2010) measured the horizontal inequity index in skilled delivery using unstandardized
and need-standardized inequity. He observed that the difference between the approaches is marginal. Horizontal inequity for
skilled delivery was found to be positive and decreasing over time. The pattern of inequalities in access in Pakistan was also
© 2017 The Authors. African Development Review © 2017 African Development Bank
632 R. Nghargbu and O. Olaniyan

examined by Mahmood and Bashir (2012) using Pakistan DHS 1990–91 and 2006–2007. Findings indicate that socioeconomic
status and wealth were significant; this was also similar to findings by Ahmed and Mahran (2009). Similarly, Memirie et al.
(2016) observed persistent wealth-related inequality in Ethiopia. Utilization of primary care facilities for selected maternal and
child health interventions showed pro-poor improvement over the period 2005–11. Although there was low coverage of measles
immunization in Ethiopia, it was the most distributed indicator with a horizontal inequity index of 0.08 in 2011.
In the case of Nigeria, several studies have attempted to investigate the determinants of immunization utilization in Nigeria
(Sibeudu et al., 2017; Eboreime et al., 2015; Olorunsaiye and Degge, 2016; Chidiebere et al., 2014). These studies show that
socioeconomic status — such as maternal education, wealth status, child and mother’s age and employment status — determines
immunization utilization. Other variables include lack of information about immunization, fear of side-effects and the
immunization centers being too far away. In terms of inequity studies, there have been previous attempts to study inequity in
child health care utilization in Nigeria (Antai 2011). The study centered on rural-urban inequity in immunization using logistic
regressions at community level. Other dimensions of inequity such as wealth, education and region were not given appropriate
attention in the study. However, these dimensions are important because they constitute some very vital socioeconomic factors
that affect child immunization in Nigeria. Second, the logistic regression is also deficient in the analysis of inequity because it
does not estimate clearly the profile of inequity. This study estimates the extent of horizontal inequity in antenatal, skilled
delivery and immunization utilization and further provides the contributions of need and non-need variables in explaining
inequity in each of the health care areas. This approach is considered appropriate because it provides a true picture of the profile
of inequity due to need and non-need variables. Inequity due to need variables is considered to be justified and does not constitute
a problem since utilization is according to need. On the other hand, inequity due to non-need variables poses a great challenge to
the health sector because individuals who need a particular health care do not have it, owing to factors beyond their control.
Although studies from other SSA countries show that there is wealth-related inequity in antenatal care and skilled delivery,
they have limitations. First, for some of the studies, the concentration index was not standardized for differences in need and
non-need variables as such other driving factors may not be visible. Second, given that Nigeria is the most populous nation in
Africa with about 200 million people, the results from these countries might not apply. This study fills that gap by providing
evidences on horizontal inequity in maternal and child health care among one-fifth of Africa’s population.

3. Methods

3.1 Theoretical Framework

To measure horizontal inequity, utilization of health care must be standardized into need and non-need variables. Standardization
helps to measure and quantify the contributions of need and non-need variables to horizontal inequity. The concentration curves
and index are the most acceptable (Ganle et al., 2014; Wagstaff et al., 1999; O’Donnell et al., 2008; Alaba and Chola, 2014).
The curves show socioeconomic inequality in health care utilization but it does not measure the magnitude. The concentration
index which is directly related to the concentration curve does that. Geometrically, the concentration curve is represented in
Figure 1. It plots the cumulative percentage of health care utilization against the cumulative percentage of the population ranked
by living standards, from poorest to the richest (O’Donnell et al., 2008). Once health care use is standardized for need, horizontal
inequity can be measured by the concentration index. The need standardized concentration index falls above or below the
diagonal of the concentration curve if health care utilization is concentrated among the poor or rich. If it lies on the diagonal, there
is equity. The concentration index is twice the area between the concentration curve (C(p)).
Z 1
CI ¼ 1  2 ½CðpÞdp ð1Þ
0

The CI can be positive or negative depending on where the concentration curve lies. It is positive if utilization is concentrated
among the rich and vice versa. Symbolically, we can use a linear model to illustrate the concept of concentration index and how it
is standardized to obtain the horizontal inequity index. Given a linear model of health care utilization:
X X
yi ¼ a þ bj X ji þ g k Z ki þ ei ð2Þ
j k
Maternal and Child Health Care 633

Figure 1: Geometric representation of the concentration curve

Source: https://www.google.com.ng/search?q=concentration+curve&source=lnms&tbm=isch&sa=X&ved=0ahUKEwj6hcjI5
ILUAhWsK8AKHQyOCdIQ_AUIBigB&biw=1366&bih=665&dpr=1

where the y is a linear function of health care utilization, the Xs are the justifiable determinants or standardizing variables while
the Zs are the unjustifiable determinants, which is termed horizontal inequity. The concentration index for yi can be written as:
X X
CI ¼ j
ðbj X j =mÞC j þ k
ðg k Z k =uÞC k þ GC e =m ð3Þ

The CI is the concentration index which measures total inequality. ðbX =mÞ are the elasticities which indicate the
responsiveness of y to changes in the X and Z. The elasticities of the X and Z as well as their index can take a positive or negative
value. A positive and negative value means that the X or Z and the CI is concentrated among the rich or the poor. However, the CI
must be decomposed and standardized in terms of the need and non-need in Figure 2.
The resulting outcome of direct or indirect standardization of the concentration index forms three parts: the concentration
index CI, which measures the total inequality; the X component of the index, which measures the justified inequality due to need
variable; and the Z, which measures the unjustified inequality also known as the horizontal inequity index HI. The horizontal
inequity derived from concentration index can be defined as;
X
HI ¼ CI  j
ðbj X j =mÞC j ð4Þ

This means horizontal inequity (HI) is total inequality minus need-based inequality. Need-induced inequality is acceptable
because health care utilization should be according to need, while non-need inequality is not acceptable. The non-need inequality

Figure 2: Components of the standardized concentration index

© 2017 The Authors. African Development Review © 2017 African Development Bank
634 R. Nghargbu and O. Olaniyan

can be due to three factors: direct contribution of income, contribution of other variables such as education, health insurance,
residence, marital status and so on, and the contributions of the residuals, which is the unexplained inequality:

HI ¼ ½ðincome contributionÞ þ ðother non-need variablesÞ þ ðresidualsÞ  ½ðneed induced inequityÞ ð5Þ

Note that decomposition of the concentration index holds if the model is linear. For nonlinear models where health care
utilization is measured in binary or count form, decomposition is achieved by a linear approximation of the nonlinear model
(O’Donnell et al., 2008). This can be written as:
X X
yi ¼ a þ bm x þ
i i ji
gmz
k k ki
þ ei ð6Þ

where
 
i ¼
bm dy
dx
and g m
k ¼
dy
dz

The concentration index for the nonlinear model is given as:


X X
CI ¼ ðbm X j =mÞC j þ
j j k
ðg m
k Z k =mÞC k þ GC e =m ð7Þ

Empirically CI is estimated by direct or indirect standardization using STATA or ADePT to obtain the horizontal inequity
index. Indirect standardization is, however, preferred to direct standardization because of its greater accuracy when dealing with
individual level data.

3.2 Data, Variable Measurement and Analysis


Four rounds of the Nigerian DHS (1999, 2003, 2008 and 2013) data sets were used. The NDHS is nationally representative data
collected based on two-stage probability sampling. It provides information on the reproductive health of women aged 15 to 49
and under-five children.
Maternal health care utilization variables were measured by number of antenatal visits during pregnancy and skilled delivery,
antenatal visits range from 1 or more visits. In the survey, women were asked the number of antenatal visits attended during
pregnancy and the assistance they received during delivery. Skilled delivery was represented as deliveries assisted by doctors/
nurses/midwives and auxiliary midwives. This was coded as 1, while assistance by traditional birth attendants, relatives and
friends were categorized as unskilled and coded as 0. Child health care utilization was measured by child having vaccination. In
the survey, mothers were asked to state if their children ‘ever had vaccination’, the responses were binary ‘yes’ or ‘no’.
This paper relied on asset/wealth index or wealth scores as a measure of living standard to construct concentration curve and
index because the NDHS did not provide information on income of the respondents in monetary terms. Wealth status was used as
a proxy for long-run income of household. The scores were derived from ownership of household assets and facilities that depict
a level of standard of living. Wealth index was estimated through Principal Component Analysis (PCA) derived from factor
analysis as five quintiles, namely: poorest, poorer, middle, richer, richest.
Indirect standardization of the non-linear model was used to estimate the concentration index. Since antenatal and skilled
delivery are not conditions of illness, age, pregnancy status and body mass index (BMI) were used as need variables. A woman
within reproductive age 15 and 49 is in need of antenatal and skilled delivery the moment she is pregnant. Respondents with BMI
lower than 18.5 were considered as underweight, 18.5 to 24.99 normal, 25.0 to 29.99 overweight, while 30.00 are obese. In the
survey women were asked to state their age and whether they are currently pregnant, unsure or not pregnant. Child age was
standardized as a need variable because age predisposes the child to be in need of immunization. Most children were immunized
within the age band of 0 to 5. Age for children in the data is measured in months. Non-need variables were respondent’s wealth
status, region, residence, ethnicity, education, partner’s education, health insurance, employment status, and birth order. Child
sex was included as a non-need variable because every child irrespective of sex is to be vaccinated.
The paper used ADePT (Automated DEC Poverty Tables) statistical software developed by World Bank for estimation. The
advantage of ADePT over other software is that it minimizes human errors in programming and makes interpretation of results
Maternal and Child Health Care 635

for policy implication easy. The ADePT software breaks down inequalities into justifiable and unjustifiable, then decomposes
the causes of inequities so that the contribution to inequality from each component of non-need variables can be quantified. Since
ADePT has no data manipulation capability, STATA software was used to clean the data before analyzing in ADePT.

4. Results

4.1 Descriptive Statistics

Table 1 presents the descriptive statistics of the variables that were used in the analysis for the four rounds of NDHS.
Women in the four rounds of survey had an average of 4 to 5 antenatal visits with 50–66 percent attendance. About
90 percent of women did not have skilled delivery, just 6–10 percent were assisted by doctors/nurses/midwives and auxiliary
midwives. An average of 53–70 percent of children received immunization between 1999 and 2013. Women’s mean age
was 34 to 36 years, children were mostly 10 months. About 80–90 percent of women were either not pregnant or unsure,
9–11 percent were pregnant with an average BMI of 23 to 32. In addition, 20–27 percent were within the poorest wealth
index, while 12–20 percent were in the richest category; 70 percent of women were employed; only 2 percent had health
insurance; 50–59 percent of the women and 40–49 percent of their partners had no formal education; and 60–70 percent of
respondents were from rural areas.

4.2 Antenatal Care and Skilled Delivery


Figure 3a shows that inequity exists in antenatal care utilization because the concentration curves fall below the line of equality/
diagonal with positive horizontal inequity index.
The positive values of the horizontal inequity index shows that there is pro-rich inequity in antenatal care utilization. The
values across years consistently increased from 0.262 in 1999 to 0.373 in 2013. However, there was a slight decrease between
0.386 to 0.373 in 2008 and 2013.
In Figure 3b, the components of the index shows that for all periods, non-need variables (wealth status, women’s and partner’s
education) accounted for 60–80 percent of the components that drives inequity in antenatal care utilization in Nigeria. Although
these three non-need variables predominate, region and other non-needed variables also play vital role. The other non-need
variables includes, insurance, birth order, marital and employment status. Need variables accounted for less than 2%, hence,
non-need variables are the major drivers of inequity in antenatal care utilization in Nigeria.
Figures 4a and 4b present the results of analysis for skilled delivery utilization. In Figure 4a, pro-rich inequity in skilled
delivery is identified in all periods because the concentration curves fall below diagonal with the positive horizontal inequity
index. The inequity index consistently increased from 0.353 in 1999 to 0.483 in 2013. Figure 4b shows that wealth status, woman
and her partner’s education accounted for 70–85 percent of the factors driving inequity in skilled delivery utilization in Nigeria.
Contrary to antenatal care results, wealth status and education of the woman and her partner consistently predominates the
components of the non-need variables across all periods. Similarly, need variables accounted for less than 1 percent of the factors
driving skilled delivery utilization for the four periods. The residual error is large as revealed by Table A2 in the appendix
because of factors other than the identified need and non-need variables.
The horizontal inequity index for ‘timing of antenatal visits’ and ‘place of delivery’ were estimated for the purpose of
robustness check and compared with antenatal visit and skilled delivery. The results are presented in the appendix in Figures A1
and A2. The concentration curve for ‘timing of first antenatal visits’ falls above the diagonal line with negative horizontal
inequity index in all periods, while that of ‘place of delivery’ falls below the diagonal with positive horizontal inequity index.
However, the concentration curve for place of delivery is closer to the diagonal with smaller horizontal inequity index compared
to skilled delivery.

4.3 Child Immunization


In Figure 5a, horizontal inequity for child immunization is close to zero for all periods as shown by the concentration curves
and index.
© 2017 The Authors. African Development Review © 2017 African Development Bank
636 R. Nghargbu and O. Olaniyan

Table 1: Utilization variable, need and non-need variables


NDHS 2013 NDHS 2008 NDHS 2003 NDHS 1999
Variable Mean Mean Mean Mean

Health care utilization variables


Antenatal visits 5.32 4.405 5.169 4.746
Antenatal attendance 0.663 0.583 0.656 0.660
Skilled delivery
No skilled delivery 0.90 0.912 0.91 0.94
Had skilled delivery 0.10 0.088 0.09 0.06
Immunization (ever had vaccination)
No 0.29 0.374 0.31 0.47
Yes 0.71 0.626 0.69 0.53
Need standardized variables (Justifiable inequality)
Respondent’s age 35.99 35.69 35.68 34.89
Pregnancy status
No or unsure 0.89 0.901 0.89 0.90
Yes 0.11 0.099 0.11 0.10
Body mass index of respondent
Body mass index of respondent 24.75 24.93 23.09 32.95
Child age 10.17 10.067 10.40 10.10
Non-need standardized variables (unjustifiable inequality)
Wealth status
Poorest 0.240 0.274 0.248 0.201
Poorer 0.229 0.246 0.219 0.203
Middle 0.205 0.202 0.202 0.202
Richer 0.186 0.160 0.186 0.196
Richest 0.14 0.118 0.15 0.20
Employment status
Not employed 0.235 0.293 0.292 0.392
Employed 0.76 0.707 0.71 0.61
Insurance status
No insurance 0.984 0.987
Insured 0.02 0.013
Respondent’s education
No education 0.510 0.559 0.582 0.562
Primary 0.234 0.238 0.237 0.253
Secondary 0.203 0.161 0.143 0.145
Higher 0.05 0.042 0.04 0.04
Partner’s education
No education 0.429 0.475 0.476 0.456
Primary 0.211 0.222 0.247 0.263
Secondary 0.24 0.205 0.177 0.175
Higher 0.12 0.098 0.10 0.11
Birth order
Birth order 3.526 3.565 3.686 3.385
Marital status
Single 0.071 0.067 0.070 0.061
Married 0.93 0.933 0.93 0.94
Ethnicity
Hausa 0.41 0.369
Igbo 0.111 0.106
Ijaw/izon 0.038 0.031
Maternal and Child Health Care 637

Table 1 (Continued)
NDHS 2013 NDHS 2008 NDHS 2003 NDHS 1999
Variable Mean Mean Mean Mean

Kanuri/beriberi 0.015 0.033


Tiv 0.016 0.028
Yoruba 0.113 0.106
Others 0.292 0.326
Region
North Central 0.135 0.178 0.163 0.214
North East 0.202 0.231 0.238 0.190
North West 0.325 0.276 0.288 0.160
South East 0.095 0.088 0.109 0.216
South South 0.122 0.114 0.101
South west 0.12 0.112 0.121 0.22
Residence
Urban 0.326 0.251 0.361 0.291
Rural 0.67 0.749 0.64 0.71
Religion
Islam 0.578 0.560 0.566 0.176
Christianity 0.41 0.418 0.41 0.30
Traditionalist 0.013 0.022 0.025 0.496
Child sex
Female 0.51 0.513 0.52 0.52
Male 0.49 0.487 0.48 0.48

Figure 3a: Concentration curves and horizontal inequity index of antenatal care utilization 1999–2013
1.00
1.00 Line of
Line of 0.90
0.90 equality
equality 0.80
0.80
Cumulative % of antenatal cae utilisation

Antenatal
Cumulative % of antenatal care

0.70 Number of 0.70


0.60 0.60
visits for
antenatal
0.50 0.50 pregnancy
visits during
utilization

0.40 0.40
pregnancy
0.30 0.30
0.20 0.20
0.10
0.10
0.00
0.00
0.00 0.20 0.40 0.60 0.80 1.00
0.00 0.20 0.40 0.60 0.80 1.00

Cumulative % of respondents, ranked from poorest to richest Cumulative % of population, ranked from poorest to richest

Horizontal inequity index = 0.372 and - 0.048 for year 2013 Horizontal inequity index = 0.386 and -0.043 for year 2008

1.00 1.00
Line of Line of
Cumulative % of antenatal care

0.90 0.90
0.80 equality 0.80
equality
Cumulative % of antenatal care

0.70 Antenatal visits 0.70


0.60 Antenatal
uilisation

for pregnancy 0.60


0.50 0.50 visits for
utilisation

0.40 0.40
0.30 pregnancy
0.30
0.20
0.20
0.10
0.10
0.00
0.00
0.00 0.20 0.40 0.60 0.80 1.00
0.00 0.20 0.40 0.60 0.80 1.00
Cumulative % of population, ranked from poorest to richest Cumulative % of population, ranked from poorest to richest

Horizontal inequity index = 0.316 and -0.068 for year 2003 Horizontal inequity index = 0.262 and -0.078 for year 1999

© 2017 The Authors. African Development Review © 2017 African Development Bank
638 R. Nghargbu and O. Olaniyan

Figure 3b: Standardized concentration index of antenatal care utilization for need and non-need variables

Figure 4a: Concentration curves and horizontal inequity index for skilled delivery utilization 1999 to 2013
1.00 1.00
0.90
Line of
0.90
Line of equality

Cumulative % of skilled delivery


Cumulative % of skilled delivery

0.80 0.80
0.70 equality 0.70 Doctor/
0.60 0.60
Doctor/ nurse/midwife/

utilization
0.50 0.50
utilization

nurse/midwife/ 0.40 aux. midwife = 1


0.40
0.30 aux. midwife = 1 0.30
0.20 0.20
0.10 0.10
0.00 0.00

0.00 0.20 0.40 0.60 0.80 1.00 0.00 0.20 0.40 0.60 0.80 1.00
Cumulative % of respondents ranked from poorest to richest Cumulative % of respondents ranked from poorest to richest

Horizontal inequity index = 0.483 for year 2013 Horizontal inequity index = 0.458 for year 2008

1.00 1.00
0.90
Line of 0.90 Line of equality
Cumulative % of skilled delivery

0.80 equality 0.80


Doctor/
Cumulative % of skilled

0.70 0.70
0.60
Doctor/ 0.60 nurse/midwife/
nurse/midwife/ 0.50
aux. midwife = 1
delivery

0.50
utilization

0.40
0.40 aux. midwife = 1
0.30 0.30
0.20 0.20
0.10 0.10
0.00 0.00

0.00 0.20 0.40 0.60 0.80 1.00 0.00 0.20 0.40 0.60 0.80 1.00
Cumulative % of respondents ranked from poorest to richest Cumulative % of respondents ranked from poorest to richest

Horizontal inequity index = 0.433 for year 2003 Horizontal inequity index = 0.353 for year 1999

Figure 4b: Standardized concentration index of skilled delivery utilization for need and non-need variables
Maternal and Child Health Care 639

Figure 5a: Concentration curves and horizontal inequity index for immunization utilization between 1999
and 2013
1.00
Line of
0.90
equality 1.00
0.80 Line of

Cumulative % of utilization of
0.90
Cumulative % of utilisation of immunisation

0.70 Ever had equality


0.80
vaccination Ever had

immunization
0.60 0.70
0.50 0.60 vaccination
0.40 0.50
0.30 0.40
0.20 0.30
0.10 0.20
0.10
0.00
0.00
0.00 0.20 0.40 0.60 0.80 1.00
0.00 0.20 0.40 0.60 0.80 1.00
Cumulative % of respondents ranked from poorest to richest Cumulative % of respondents ranked from poorest to richest

Horizontal inequity index = 0.068 for year 2013 Horizontal inequity index = 0.114 for year 2008

1.00 1.00
Line of

Cumulative % of utilszation of
Line of
Cumulative % of utilization of

0.90 0.90
0.80
equality equality
0.80

immunisation
Ever had
immunization

0.70 0.70 Ever had


0.60 vaccination 0.60 vaccination
0.50 0.50
0.40
0.40
0.30
0.30
0.20
0.20
0.10
0.10
0.00
0.00
0.00 0.20 0.40 0.60 0.80 1.00
Cumulative % of respondents, ranked from poorest to richest 0.00 0.20 0.40 0.60 0.80 1.00
Cumulative % of respondents, ranked from poorest to richest

Horizontal inequity index = 0.069 for year 2003 Horizontal inequity index = 0.216 for 1999

Slight pro-rich inequity in immunization utilization is observed for all periods. The values of the horizontal inequity index
declined consistently from 0.216 to 0.068 in 1999 to 2013. Although an increase was observed from 0.069 to 0.114 in 2003 to
2008. Figure 5b shows that mother’s wealth status, and mother’s and father’s education accounts for 80–90 percent of the
non-need factors driving inequity in immunization utilization in Nigeria. Need variables accounted for less than 0.5 percent of
the inequity for all periods.
The horizontal inequity index was estimated for measles vaccination, medical treatment for fever/cough and diarrhea for the
purpose of robustness check. The results are presented in Figures A3 to A5 in the appendix. Similarly, immunization, measles
vaccination, medical treatment for fever/cough and diarrhea have a positive horizontal inequity index with concentration curves
falling below the diagonal for all periods.

Figure 5b: Standardized concentration index of immunization utilization for need and non-need variables

© 2017 The Authors. African Development Review © 2017 African Development Bank
640 R. Nghargbu and O. Olaniyan

5. Discussion
The results on the concentration index for antenatal, skilled delivery and immunization are in line with the findings of
Houweling et al. (2007), Bonfruer et al. (2012), Zere et al. (2011), Antai (2011), Alam et al. (2015), Målqvist et al. (2012),
Hossain (2010), Mahmood and Bashir (2012), Memirie et al. (2016), Sibeudu et al. (2017), Eboreime et al. (2015),
Olorunsaiye and Degge (2016), Chidiebere et al. (2014) and Antai (2011). This study added value to this literature by
standardizing for differences in need and non-need variables using respondent’s age, pregnancy status, BMI and child age as
well as estimating horizontal inequity for four periods. Standardization revealed that wealth status and education were the
major non-need factors driving inequity in maternal and child health care utilization in Nigeria. Need variables contributed
little to horizontal inequity in utilization. Women with higher economic and educational status had a higher probability of
utilizing antenatal and skilled delivery than those with lower status. This has policy implications. Maternal and child health
care policy on free antenatal and immunization utilization is not effective; this can be attributed to four reasons. First, most
health facilities where free antenatal and skilled delivery care is administered are located in the urban areas where most
women in the middle, richer and richest wealth quintile live. Secondly, very poor women from rural areas may not have
access to information on where free antenatal care is administered even if it is available. Thirdly, there are more private
facilities especially in villages than public ones. In private facilities, out-of-pocket payment is the only way to access any
type of health care services except if the individual obtained insurance. As income of individuals increases, they will prefer
to go to private hospitals to obtain health care services (Anyanwu, 2007). Fourthly, some public health centers request initial
charges which the poor women may not afford to pay. This can affect their health care utilization. Region and other ‘non-
need variables’ contributed largely to horizontal inequity in maternal health care utilization in Nigeria, although it is not the
major driver. The results on region is in contrast with Zere et al. (2011).
Estimation for four periods, which is also a valuable addition, shows that inequity in antenatal and skilled delivery increased
consistently over time with the highest incidence in 2008 and 2013, even with the policy on free maternal health care. These
findings are in contrast with results from other countries especially those that have made substantial progress in achieving the
MDGs of reducing maternal mortality (Hossain, 2010; Mahmood and Bashir, 2012 and Memirie et al., 2016). However, inequity
in immunization decreased over the period.
Inequity is higher in skilled delivery than antenatal care; this was shown by their values of the horizontal inequity index. Poor
women especially from rural areas may go for antenatal care but are found to deliver at home with the assistance of friends,
relatives and traditional midwives/birth attendants. This may be due to the problem of easy accessibility to skilled delivery
especially in rural areas. The accessibility problem poses a great danger to women as they are exposed to infectious diseases,
prolonged labour, hemorrhage and death due to delivery complications. Higher inequity in skilled delivery utilization may be
attributed to distance to health facilities especially in rural areas, cost associated to caesarean sections in some public and private
facilities as well as cultural beliefs. Immunization utilization had lower inequity due to mobile vaccination programs mostly in
rural areas as well as improvement in immunization utilization. The results on the robustness check show that all maternal and
child health care were inequitably utilized either in favour of the rich or poor.
Although the horizontal inequity index does not appear to be significantly different from total inequality, the concentration
index gives a good picture of inequity in health care utilization. The implications for health policy of a low level of justifiable
inequality due to need variables is that there is a serious demand problem in terms of maternal and child health care utilization.
Demand is largely determined by socioeconomic status instead of the need for health care. Health policymakers must work hard
to reduce socioeconomic inequality especially in wealth and education if substantial progress is to be made in reducing maternal
and child mortality.

6. Conclusion and Policy Implications


The study concludes that there is pro-rich inequity in antenatal, skilled delivery and immunization utilization. Socioeconomic
factors such as wealth status and education are the main drivers of inequity in maternal and child health care utilization in
Nigeria. Inequity in utilization is greater in maternal health care than child health care. There is an improvement in the utilization
of child health care especially in immunization over time but inequity in utilization of maternal health care especially on skilled
delivery worsened over time.
Maternal and Child Health Care 641

The policy implication of this study is that maternal health care utilization has to be improved through interventions that can
improve women’s wealth status, their education and that of their partners before substantial progress can be made in reducing
maternal and child mortality.

References
Abalo, K. and A. E. Agbodji (2014), ‘Nutritional Health Inequalities among Children in Togo’, African Development Review,
Vol. 26, No. 4, pp. 584–96.
Ahmed, H. M. M. and H. A. Mahran (2009), ‘Determinants of Demand for Delivery Services in Sudan: An Empirical
Investigation with Reference to Khartoum State, 2004’, African Development Review, Vol. 21, No. 3, pp. 514–40.
Alaba, O. and L. Chola (2014), ‘Socioeconomic Inequalities in Adult Obesity Prevalence in South Africa: A Decomposition
Analysis’, International Journal of Environmental Research and Public Health, Vol. 11, No.3, pp. 3387–406.
Alam, N., M. Hajizadeh, A. Dumont and P. Fournier (2015), ‘Inequalities in Maternal Health Care Utilization in Sub-Saharan
African Countries: A Multiyear and Multi-Country Analysis’, PloS One, Vol. 10, No. 4, pp. 1371.
Allin, S. (2006), ‘Equity in the Use of Health Services in Canada and its Provinces’, Working Paper 3, London School of
Economics and Political Science, ISBN 0753020416.
Antai, D. (2011), ‘Rural-urban Inequalities in Child Hood Immunization in Nigeria: The Role of Community Context’, Africa
Journal of Primary Health Care, Vol. 3, No. 1, article no. 238.
Anyanwu, J. C. (2007), ‘Demand for Health Care Institutions’ Services: Evidence from Malaria Fever Treatment in Nigeria’,
African Development Review, Vol. 19, No. 2, pp. 304–34.
Bonfruer, I., E. Van de Poel, M. Grimm and E. Van Doorslaer (2012), ‘Does Health Care Utilization Match Needs in Africa:
Challenging Conventional Needs Measurement’, Working Paper W2012.02, Erasmus Universeteur Rotterdam Institute of
Health and Policy Management.
Chidiebere, O. D., E. Uchenna and O. S. Kenechi (2014), ‘Maternal Sociodemographic Factors that Influence Full Child
Immunisation Uptake in Nigeria’, South African Journal of Child Health, Vol. 8, No. 4.
Culyer, A. J. (2001), ‘Equity: Some Theory and its Policy Implications’, Journal of Medical Ethics, Vol. 27, No. 10,
pp. 275–83.
Culyer, A. J. and A. Wagstaff (1993), ‘Equity and Equality in Health and Health Care’, Journal of Health Economics, Vol. 12,
No. 4, pp. 431–57.
Eboreime E., S. Abimbola and F. Bozzani (2015), ‘Access to Routine Immunization: A Comparative Analysis of Supply-side
Disparities between Northern and Southern Nigeria’, PLoS One, Vol. 10, No 12, pp. 1371.
Fleurbaey, M. (2006), ‘Health Equity and Social Welfare’, Annales d’Economie et Statistiques, Vol. 83, pp. 24–62.
Fleurbaey, M. and E. Schokkaert (2009), ‘Unfair Inequalities in Health and Health Care’, Journal of Health Economics Vol. 28,
No. 1, pp. 73–90.
Ganle J. K., M. Parker, R. Fitzpatrick and E. Otupiri (2014), ‘Inequities in Accessibility to and Utilisation of Maternal Health
Services in Ghana after User-fee Exemption: A Descriptive Study’, International Journal of Equity in Health, Vol. 13, No. 89.
Gravelle, H., S. Morris and M. Suttun (2006), ‘Economic Studies of Equity in Consumption of Health Care’, in A. M. Jones (ed.),
The Elgar Companion to Health Economics, Cheltenham: Edward Elgar, pp. 193–204.
Hossain, M. I. (2010), ‘Inequality in the Utilization of Maternal Care and the Impact of Macroeconomic Policy: Evidence from
Bangladesh Health Econometrics and Data Group’, HEDG Working Paper 10/08, available at: www.york.ac.uk/res/herc/
hedgwp.
Hotchkiss, R. D. and G. D. Maido (2011), ‘Effect of an Expansion in Private Sector Provision of Contraceptives Supplies on
Horizontal Inequity in Modern Contraceptive Use: Evidence from Africa and Asia’, International Journal for Equity in Health,
Vol. 10, No.1, Art. 33.
© 2017 The Authors. African Development Review © 2017 African Development Bank
642 R. Nghargbu and O. Olaniyan

Houweling, T. A. J., M. R O. Campbell and E. A. Kunst (2007), ‘Huge Pro-rich Inequalities in Maternity Care: An International
Comparative Study of Maternity and Child Care in Developing Countries’, WHO Bulletin, Vol. 85, No. 10, p. 745.
Målqvist, M., D. T. P. H. Hoa and S. Thomsen (2012), ‘Causes and Determinants of Inequity in Maternal and Child Health in
Vietnam, BMC Public Health Vol. 12, No. 641. Available at: http://www.biomedcentral.com/1471-2458/12/641
Mahmood, N. and S. Bashir (2012), ‘Applying an Equity Lens to Maternal Health Care Practices in Pakistan’, Working Paper 83,
Pakistan Institute of Development Economics, Islamabad.
Memirie, S. T., S. Verguet, O. F. Norheim, C. Levin and K. A. Johansson (2016), ‘Inequalities in Utilization of Maternal and
Child Health Services in Ethiopia: The Role of Primary Health care’, BMC Health Services Research, Vol. 16, No. 51.
National Population Commission (1990–2013), Nigeria Demographic and Health Survey Reports, National Population
Commission Abuja, Nigeria.
O’Donnell, O., E. Van Doorslaer, A. Wagstaff and M. Lindelow (2008), ‘Analyzing Health Equity Using Household Survey
Data: A Guide to Techniques and Their Implementation’, World Bank Learning Resource series, Washington DC.
Olorunsaiye, C. Z. and H. Degge (2016), ‘Variations in the Uptake of Routine Immunization in Nigeria: Examining
Determinants of Inequitable Access’, Global Health Communication, Vol. 2, No. 1.
Ong, S. K., M. Kelasher, I. Anderson and R. Carter (2009), ‘A Cost-based Equity Weight for Use in the Economic Evaluation of
Primary Health Care Interventions: Case Study of the Australian Indigenous Population’, International Journal for Equity in
Health, Vol. 8. No. 34, pp. 1–14.
Sibeudu, F. T., B. S. C. Uzochukwu and O. E. Onwujekwe (2017), ‘Investigating Socio- economic Inequity in Access to and
Expenditures on Routine Immunization Services in Anambra State’, BMC Research Notes 10/78. Available at: https://
bmcresnotes.biomedcentral.com/articles/10.1186/s13104-017-2407-1
Ssewanyana, S. and I. Kasirye (2012), ‘Causes of Health Inequalities in Uganda: Evidence from the Demographic and Health
Surveys’, African Development Review, Vol. 24, No. 4, pp. 327–41.
UNICEF (2009), ‘Maternal and Child Health: The Social Protection Dividend’, West and Central Africa Regional Thematic
Report of Four Studis. Available at: http://www.unicef.org/wcaro/wcaro_UNICEF_ODI_4_Health_Social_Protection_
Dividend
Van de Poel, E., E. Van Doorslaer and O. O’Donnel (2011), ‘Measurement of Inequity in Health Care with Heterogeneous
Response of Use to Need’, HEFPA Working Paper 04, Institute of Health Policy and Management.
Wagstaff, A. and E. van Doorslaer (1998), ‘Equity in Health Care Finance and Delivery’, in A. S. Culyer and J. P. Newhouse
(eds.), Handbook of Health Economics, Elsevier, Amsterdam.
Wagstaff, A. and E. van Doorslaer (2000), ‘Measuring and Testing for Inequity in the Delivery of Health Care.’, Journal of
Human Resources. Vol. 35, No. 4, pp. 716–33.
Wagstaff, A., C. Bredenkamp and S. Bales (n.d.), ‘Module 1b: Inequalities and Inequities in Health and Health Care Utilization
— Decomposition, Standardization, and Inequity’, ADePT presentation. Available at: http://siteresources.worldbank.org/
INTPAH/Resources/400378-1301082525331/7823995-1317755202272/Module1bDecompositionstandardizationandinequity.
pdf
Wagstaff, A., P. Paci and E. van Doorslaer (1999), ‘On the Measurement of Inequalities in Health’, Social Science and Medicine,
Vol. 33, pp. 545–57.
Whitehead, M. (1985), ‘The Concepts and Principles of Equity and Health’, European Health for All, Vol. 1, No. 1, pp. 18.
WHO (2016), ‘Maternal Mortality’, Media centre. Available at: http://www.who.int/mediacentre/factsheets/fs348/en/ (accessed
6 September 2016).
Zere, E., D. Oluwole, M. J. Kirigia, N. C. Mwikisa and J. Mbeel (2011), ‘Inequity in Skilled Attendance at Birth in Namibia: A
Decomposition Analysis’, BMC Pregnancy and Childbirth, Vol. 11, No. 1, Art. 34.
Maternal and Child Health Care 643

Appendix

Table A1: Standardized concentration index for antenatal care utilization


NDHS 2013 NDHS 2008 NDHS 2003 NDHS 1999
Variable Index Index Index Index

Need variables
Currently pregnant 0.000<0.001 0.000<0.001 0.000<0.001 0.000<0.001
Age of woman 0.003 0.000<0.001 0.001 0.001
Respondent’s Body Mass Index –0.001 0.000<0.001 0.008 0.001
Subtotal of need variables 0.002 0.000 0.009 0.002
Non-need /controlled variables
Wealth index 0.129 0.146 0.144 0.092
Health insurance –0.001 –0.002
Employment status 0.005 0.005 0.002 0.020
Woman’s educational status 0.044 0.035 0.045 0.039
Partner’s educational level 0.042 0.032 0.022 0.040
Birth order 0.004 0.001 0.014 0.001
Current marital status 0.000 0.000 0.000 –0.001
Ethnicity 0.018 0.031
Region 0.034 0.037 0.019 0.030
Residence 0.019 0.021 0.005 0.020
Religion 0.012 0.006 0.013 0.021
Subtotal of non-need variables 0.309 0.313 0.264 0.261
Residual error 0.067 0.073 0.077 0.001
Residual missing data –0.003 –0.004 –0.003 0.001
Inequality total 0.371 0.382 0.322 0.264
Horizontal inequity index 0.373 0.386 0.316 0.262

Table A2: Standardized concentration index for skilled delivery utilization


NDHS 2013 NDHS 2008 NDHS 2003 NDHS 1999
Variable Index Index Index Index

Need variables
Currently pregnant 0.000<0.001 0.000<0.001 0.000<0.001 0.000
Age of woman –0.015 –0.014 0.001 –0.026
Respondent’s Body Mass Index –0.001 –0.001 0.002 0.001
Subtotal of need variables –0.016 –0.015 0.003 –0.025
Non-need variables
Wealth index 0.097 0.105 0.114 0.091
Health insurance 0.000 –0.000
Employment status 0.000 0.001 0.001 0.014
Woman’s educational 0.066 0.059 0.046 0.073
Partner’s educational status 0.018 0.009 0.015 0.022
Religion 0.024 0.01521 0.017 0.025
Birth order –0.030 –0.023 0.020 –0.011
Current marital status –0.001 0.000 0.000 0.000
Ethnicity 0.010 0.016
Region 0.012 0.015 0.003 0.036
Continued

© 2017 The Authors. African Development Review © 2017 African Development Bank
644 R. Nghargbu and O. Olaniyan

Table A2 (Continued)
NDHS 2013 NDHS 2008 NDHS 2003 NDHS 1999
Variable Index Index Index Index

Residence 0.026 0.017 0.012 0.013


Subtotal of non-need variables 0.221 0.213 0.189 0.262
Residual error 0.261 0.245 0.244 0.091
Residual missing data –0.003 –0.004 –0.012 –0.045
Inequality total 0.463 0.440 0.424 0.283
Horizontal inequity index 0.483 0.458 0.433 0.353

Table A3: Standardized concentration index for immunizations utilization


NDHS 2013 NDHS 2008 NDHS 2003 NDHS 1999
Variable Index Index Index Index

Need variables
Current age of child 0.003 0.007 0.003 0.005
Subtotal 0.003 0.007 0.003 0.005
Non-need/controlled variables
Wealth index of mother 0.043 0.065 0.025 0.119
Educational status of mother 0.015 0.042 0.013 0.059
Religion of mother 0.000 0.008 0.004 0.010
Ethnicity of mother –0.006 0.006
Insurance status of mother 0.001 0.001
Current marital status of mother 0.000 0.000 0.000
Father’s educational level 0.029 0.013 0.017 0.019
Employment status of mother 0.007 0.005 0.001 0.022
Region of mother 0.011 –0.005 0.002 0.014
Residence of mother –0.019 0.001 0.003 0.028
Birth order of child –0.002 –0.002 –0.002
Sex of child 0.000 0.000 0.000 0.000
Subtotal of non-need variable 0.079 0.133 0.064 0.272
Residual regression error –0.011 –0.019 0.006 –0.056
Residual missing data 0.000 –0.001 0.001 –0.002
Inequality total 0.072 0.121 0.073 0.219
Horizontal inequity index 0.068 0.114 0.069 0.216
Maternal and Child Health Care 645

Figure A1: Concentration curves and horizontal inequity index for antenatal visits and timing of antenatal
visits
1.00
Line of 1.00
0.90 Line of
equality 0.90
0.80 equality
Number of 0.80
Cumulative % of utilization variable

0.70 Antenatal

Cumulative % of utilization variable


antenatal 0.70
0.60 0.60 visits for
visits during
0.50 0.50 pregnancy
pregnancy
0.40 0.40 Timing of
Timing of 1st
0.30 0.30 1st
antenatal
0.20
0.20 antenatal
check 0.10
0.10
check
(months) 0.00
0.00
0.00 0.20 0.40 0.60 0.80 1.00
0.00 0.20 0.40 0.60 0.80 1.00
Cumulative % of population, ranked from poorest to richest Cumulative % of population, ranked from poorest to richest

Horizontal inequity index = 0.373 for year 2013 Horizontal inequity index = 0.386 for year 2008

1.00 1.00
0.90
Line of equality Line of equality
0.90
0.80 Antenatal visits 0.80 Antenatal visits
Cumulative % of utilization variable

0.70

Cumulative % of utilization variable


for pregnancy 0.70 for pregnancy
0.60
0.50 Timing of 1st 0.60
Timing of 1st
0.40 antenatal check 0.50
0.30 0.40 antenatal check
0.20 0.30
0.10 0.20
0.00 0.10
0.00 0.20 0.40 0.60 0.80 1.00 0.00
0.00 0.20 0.40 0.60 0.80 1.00
Cumulative % of population, ranked from poorest to richest
Cumulative % of population, ranked from poorest to richest

Horizontal inequity index = 0.316 for year 2003 Horizontal inequity index = 0.26 2 for year 1999

Figure A2: Concentration curves and horizontal inequity index for skilled delivery and place of delivery
1.00
1.00
Line of equality Line of
0.90
0.90
0.80
equality
0.80 Doctor/nurse/
Cumulative % of utilization variable

0.70
0.70 Doctor/nurse/
midwife/aux.
Cumulative % of utilization variable

0.60 midwife/aux.
0.60 midwife = 1 0.50 midwife = 1
0.50
Place of 0.40
0.40 Place of
0.30
delivery 0.30
delivery
0.20 0.20
0.10 0.10
0.00 0.00

0.00 0.20 0.40 0.60 0.80 1.00 0.00 0.20 0.40 0.60 0.80 1.00

Cumulative % of population, ranked from poorest to richest


Cumulative % of population, ranked from poorest to richest

Horizontal inequity index = 0.483 and 0.141 for year 2013 Horizontal inequity index = 0.458 and 0.167 for year 2008

1.00
Line of equality
0.90 1.00
Line of
0.80 Doctor/nurse/ 0.90
equality
Cumulative % of utilization variable

0.70 midwife/aux. 0.80


midwife = 1 Doctor/nurse/
Cumulative % of utilization variable

0.60 0.70
midwife/aux.
0.50 Place of 0.60
midwife = 1
0.40 delivery 0.50
Place of
0.30 0.40
delivery
0.20 0.30
0.10 0.20
0.00 0.10
0.00 0.20 0.40 0.60 0.80 1.00 0.00
Cumulative % of population, ranked from poorest to richest 0.00 0.20 0.40 0.60 0.80 1.00
Cumulative % of population, ranked from poorest to richest

Horizontal inequity index = 0.433 and 0.127 for year 2003 Horizontal inequity index = 0.353 and 0.130 for year 1999

© 2017 The Authors. African Development Review © 2017 African Development Bank
646 R. Nghargbu and O. Olaniyan

Figure A3: Concentration curves and horizontal inequity index for immunization and measles
immunization
1.00 1.00
Line of Line of
0.90 0.90
equality equality

Cumulative % of utilization variable


0.80 0.80
0.70 Ever had Ever had
Cumulative % of utilization variable

0.70
0.60
vaccination 0.60 vaccination
0.50
0.50
0.40 Received Received
0.30 0.40
measles 0.30
measles
0.20
0.10 0.20
0.00 0.10
0.00 0.20 0.40 0.60 0.80 1.00 0.00
0.00 0.20 0.40 0.60 0.80 1.00

Cumulative % of population, ranked from poorest to richest


Cumulative % of population, ranked from poorest to richest

Horizontal inequity index = 0.069 and 0.265 for year 2013 Horizontal inequity index = 0.115 and 0.210 for year 2008

1.00
1.00
Line of
0.90
Line of equality
0.90 0.80
equality Ever had
0.80 0.70
Ever had vaccination
0.70 0.60

Cumulative % of utilization variable


vaccination
Cumulative % of utilization variable

Received
0.60 0.50
Received measles
0.50 0.40
measles
0.40 0.30
0.30 0.20
0.20 0.10
0.10 0.00

0.00 0.00 0.20 0.40 0.60 0.80 1.00

0.00 0.20 0.40 0.60 0.80 1.00 Cumulative % of population, ranked from poorest to richest
Cumulative % of population, ranked from poorest to richest

Horizontal inequity index = 0.070 and 0.186 for year 2003 Horizontal inequity index = 0.216 and 0.216 for year 1999

Figure A4: Concentration curves and horizontal inequity index for fever/cough medical treatment
1.00
Line of
0.90
equality 1.00
0.80 0.90
Line of
0.70 Fever/cough: 0.80
equality
0.60 medical 0.70 Fever/cough:
Cumulative % of utilization variable

0.50 treatment 0.60 medical


Cumulative % of utilization variable

0.50
0.40 treatment
0.40
0.30
0.30
0.20
0.20
0.10
0.10
0.00
0.00
0.00 0.20 0.40 0.60 0.80 1.00
0.00 0.20 0.40 0.60 0.80 1.00

Cumulative % of population, ranked from poorest to richest


Cumulative % of population, ranked from poorest to richest

Horizontal inequity index = 0.132 for year 2013 Horizontal inequity index = 0.132 for year 2008

1.00
Line of 1.00
0.90 Line of
equality 0.90
0.80 equality
0.80
0.70 Fever/cough:
0.70 Fever/cough:
0.60 medical
0.60 medical
Cumulative % of utilization variable

treatment
Cumulative % of utilization variable

0.50
0.50 treatment
0.40
0.40
0.30
0.30
0.20
0.20
0.10 0.10
0.00 0.00
0.00 0.20 0.40 0.60 0.80 1.00 0.00 0.20 0.40 0.60 0.80 1.00

Cumulative % of population, ranked from poorest to richest Cumulative % of population, ranked from poorest to richest

Horizontal inequity index = 0.231 for year 2003 Horizontal inequity index = 0.151 for year 1999
Maternal and Child Health Care 647

Figure A5: Concentration curves and horizontal inequity index for diarrhea medical treatment
1.00
Line of equality
0.90
Diarrhea: 1.00
0.80 Line of
0.90
0.70 medical equality
0.80
treatment
Cumulative % of utilization variable

Cumulative % of utilization variable


0.60 0.70 Diarrhea:
0.50 0.60 medical
0.40 0.50 treatment
0.40
0.30
0.30
0.20 0.20
0.10 0.10
0.00 0.00

0.00 0.20 0.40 0.60 0.80 1.00 0.00 0.20 0.40 0.60 0.80 1.00
Cumulative % of population, ranked from poorest to richest
Cumulative % of population, ranked from poorest to richest

Horizontal inequity index = 0.145 for year 2013 Horizontal inequity index = 0.152 for year 2008

1.00 1.00
0.90
Line of Line of equality
0.90
0.80 equality 0.80
Cumulative % of utilization variable

Cumulative % of utilization variable


0.70 0.70
Diarrhea: medical
Diarrhea:
0.60 0.60 treatment
medical
0.50 0.50
0.40
treatment
0.40
0.30 0.30
0.20 0.20
0.10 0.10
0.00 0.00
0.00 0.20 0.40 0.60 0.80 1.00 0.00 0.20 0.40 0.60 0.80 1.00
Cumulative % of population, ranked from poorest to richest Cumulative % of population, ranked from poorest to richest

Horizontal inequity index = 0.272 for year 2003 Horizontal inequity index = 0.113 for year 1999

© 2017 The Authors. African Development Review © 2017 African Development Bank
African Development Review, Vol. 29, No. 4, 2017, 648–659

Financement Bancaire des Petites et Moyennes Entreprises au


Cameroun

Andre Dumas Tsambou, Christian Zamo Akono, Ludwick Ndokang Esone


et Roger Tsafack Nanfosso

Resume : Le present travail a pour objectif d’identifier les determinants du financement bancaire des PME camerounaises. Il
s’appuie pour cela sur l’enqu^ete « Formulation d’un Plan de Developpement Local des PME au Cameroun » realisee aupres de
413 PME par le ministere des petites et moyennes entreprises, de l’economie sociale et de l’artisanat (MINPMEESA) du
Gouvernement camerounais et l’Agence Japonaise de Cooperation Internationale (JICA). Nous utilisons un modele logit simple
et les methodes d’estimations recentes. L’analyse statistique des donnees revele que les PME autofinancent leurs activites a court
terme et font recours aux institutions financieres a long terme. L’analyse econometrique a montre que les principaux
determinants du financement des PME sont la presence reguliere des effets financiers, la qualite des garanties offertes, la taille de
la PME, la rentabilite et le capital social (appartenance aux reseaux sociaux, participation des etrangers au capital, soutien des
autorites gouvernementales). Cependant, les relations sociales qui sont des indicateurs du degre de confiance des PME
ameliorent significativement la probabilite d’acces au financement bancaire en reduisant le degre d’opacite des PME aux yeux du
banquier.

Abstract: The present work aims to identify the determinants of financing banking Cameroonian SMEs. It is based on the
survey ‘Formulation of Local Development Plan for SMEs in Cameroon’ conducted among 413 SMEs by the Ministry of Small
and Medium Enterprises, Social Economy and Handicraft (MINPMEESA) Government of Cameroon and Japan International
Cooperation Agency (JICA). We use a simple logit model and robust statistical tests. The results of our analysis reveal that SMEs
in the short term self-finance their business while in the long run they use financial institution funds. Moreover, analysis of the
determinants of their access to external finance reveal that the main determinants of SME financing are: regular presence of
financial effects, the cost of financing, quality guarantees, the SME size, its profitability, the skills of the manager and the social
capital. The latter significantly improves SMEs’ conditions of access to credit by reducing their degree of opacity in the eyes of
bankers.

1. Introduction
Depuis plusieurs annees, de nombreux travaux convergent pour souligner l’importance des petites et moyennes entreprises
(PME) dans le tissu economique des pays aussi bien developpes qu’en developpement. Considerees comme des cellules
substantielles de l’economie de marche, les PME ont en effet joue un r^ole important autant dans l’innovation, la creation
d’emplois et la croissance economique des pays industrialises au cours du XXeme siecle (Quiles, 1997). Dans les pays en
developpement, et principalement en Afrique Subsaharienne (ASS), les PME constituent la quasi-totalite de la population des
entreprises, soit environ 99% au Cameroun (INS, 2009), 93% au Maroc, plus de 90% en RDC, et 95% de l’activite
manufacturiere au Nigeria (OCDE, 2006). Malgre ce poids, la contribution des PME au PIB est estimee a moins de 20% dans la
plupart des pays africains, alors qu’elle peut atteindre jusqu’a 60% dans les pays a haut revenu (Admassu, 2009). De plus, les
PME qui operent dans les pays de l’ASS emploient en moyenne moins de 30 % de la main-d’œuvre dans le secteur manufacturier
alors que cette proportion est de 74,4 % dans les pays asiatiques, de 62,1 % dans les pays de l’Amerique latine et Cara€ıbes et de


Andre Dumas Tsambou, Universite de Yaounde II-Soa; e-mail: tsamboudumas@yahoo.fr. Christian Zamo Akono, Universite De Yaounde Ii-Soa; e-
mail: zchristy2@yahoo.fr. Ludwick Ndokang Esone 1er, University of Yaounde II-Soa; e-mail: ndokang2087@gmail.com. Roger Tsafack Nanfosso,
Universite de Dschang; e-mail: rtsafack@gpeyaounde.org
© 2017 The Authors. African Development Review © 2017 African Development Bank. Published by Blackwell Publishing Ltd,
9600 Garsington Road, Oxford OX4 2DQ, UK and 350 Main Street, Malden, MA 02148, USA. 648
Financement Bancaire 649

73,1 % dans les pays de l’OCDE (Ondel’ansek, 2010). Au Cameroun, les PME n’emploient que 19% de la population active et
leur contribution au PIB est estimee a moins de 20% (INS, 2009).
Au-dela de ce contraste, il convient de souligner que dans les pays en developpement comme le Cameroun, les PME
presentent un enorme potentiel en matiere de stimulation de l’esprit d’entreprise et de creation d’un tissu industriel capable de
s’adapter aux besoins des grandes entreprises de leur pays. Aussi admet-on qu’elles pourraient contribuer d’avantage a la
croissance economique des pays de l’ASS si elles parvenaient a trouver de meilleures conditions de leur developpement. Ces
entreprises eprouvent des difficultes pour financer leurs projets d’investissement (Coluzzi et al., 2009). Cela tient notamment a
leur capacite limitee a fournir une information fiable ainsi qu’a leur expertise financiere limitee qui resulte de leurs specificites
financieres. Ang (1992), ces specificites sont liees a la difficulte de separer les ressources financieres de l’entreprise de celles de
son proprietaire, aux problemes de succession qui peuvent rendre le developpement de l’entreprise incertain et a la presence de
contrats implicites. Nombre de travaux affirment a ce sujet que les difficultes d’acces aux financements sont le premier obstacle
au developpement des PME d’ASS, assez loin devant les problemes de corruption, de deficience des infrastructures ou de
fiscalite abusive (Banque Mondiale, 2009). En fait, pour leur bon fonctionnement, les PME ont besoin de ressources financieres
necessaires a leur cycle d’exploitation et a leurs investissements. Or, il apparait que 80 a 90 % d’entre elles connaissent des
difficultes d’acces au marche des financements en ASS (Lefilleur, 2009). Une enqu^ete effectuee au Cameroun en 2009 revele que
pres de 77% des PME camerounaises ont les problemes de financement et que l’indice d’acces au credit bancaire est de 2,625
(MINPMEESA, 2009). Cette fragilite financiere d’une PME peut raccourcir son cycle de vie et une mauvaise conjoncture
engendrerait la defaillance.
Plusieurs syntheses des travaux theoriques et empiriques ont recemment ete consacrees aux PME dans les pays en
developpement, notamment en ASS. Ces travaux mettent principalement l’accent sur le financement interne ou externe des PME
et sur la pertinence des theories concurrentes au regard de l’explication de leur financement. S’inscrivant dans ce contexte
theorique et empirique, le present article a pour objectif d’identifier les leviers sur lesquels il faut agir pour ameliorer l’acces de
ce type d’entreprise au financement bancaire au Cameroun. De maniere plus specifique, il s’agit:
 D’examiner le niveau de financement des PME;
 D’analyser les sources de financement des PME;
 D’identifier les obstacles au financement des PME.

Si bon nombre d’etudes portent sur l’acces au financement des grandes entreprises, nous mettons l’accent sur les PME.
L’analyse d’un echantillon de 413 PME camerounaises nous permet de comparer nos resultats avec ceux d’autres travaux
anterieurs ou contemporains relatifs aux PME.
Considerant que le financement externe est essentiel a la croissance et au developpement de la PME et en admettant que
plusieurs facteurs peuvent expliquer la frilosite des banques a l’egard des PME camerounaises, le reste de ce document est
organise ainsi qu’il suit. La section 2 procede a la revue de la litterature. La section 3 presente la methodologie et les donnees. Les
resultats sont presentes a la section 4 et discutes a la section 5, et la derniere section conclue.

2. Revue de la litterature

2.1 PME et acces au financement


La question du financement des firmes est au cœur de toutes les problematiques et considerations economiques. Mais celui des
PME se caracterise principalement par un acces limite aux financements en raison de leur opacite (Ang, 1991). Cela s’explique
tant par la petite taille des structures considerees qui rend l’investissement necessaire a l’acquisition d’une connaissance de
l’activite a m^eme de lever les problemes d’asymetries d’information difficilement amortissable (Besanko et Kanatas, 1993). Au
vu de ce la, plusieurs etudes (Cassar et Holes, 2003) montrent que la structure financiere des PME peut ^etre expliquee par la
theorie de l’ordre hierarchique qui stipule que pour financer leurs activites, les entreprises privilegieraient, dans l’ordre, le
financement interne au financement externe et, dans ce dernier cas, le financement par dettes a celui par fonds propres (Myers,
1984). Une structure financiere dominee par les dettes pourrait donc ^etre le fait d’une strategie des dirigeants, afin de conserver le
contr^ole de leur entreprise.

© 2017 The Authors. African Development Review © 2017 African Development Bank
650 A. D. Tsambou et al.

Sur le plan du financement bancaire, chaque banque adopte un modele propre d’evaluation de credit dont les criteres et
l’importance accordee a chacun d’eux varie. Par contre, toutes les banques doivent evaluer le risque de defaut des emprunteurs
pour minimiser au maximum le risque de pertes monetaires. Au vue de ces differents risques, la vision classique du marche sur
lequel le mecanisme des prix permet d’ajuster l’offre et la demande, et ainsi induire la satisfaction de l’ensemble des agents,
conna^ıt dans le domaine du credit bancaire d’importantes contradictions. Stiglitz et Weiss (1981) montrent que, dans le cadre
d’une asymetrie d’information entre banquier et emprunteur, le marche de l’intermediation financiere est souvent caracterise par
un equilibre de rationnement. Ce rationnement est attribuable selon eux a deux formes d’asymetrie d’information: le risque
moral et la selection adverse prejudiciables au creancier. Van Pham et al. (2009) montrent sur le cas des PME vietnamiennes et
GICAM-GTZ (2008) et Taka (2009) sur le cas des PME Camerounaises que l’asymetrie d’information constitue un probleme
majeur dans la relation entre les banques et les PME.
En effet, le risque de selection adverse est fortement lie a la meconnaissance de l’emprunteur d’une part, et a la qualite de
l’information fournie dans les demandes de financement d’autre part (Observatoire europeen des PME, 2003). Pour minimiser ce
risque, les banques s’interessent au risque de remboursement en se basant sur les criteres que sont les garanties, les ratios
financiers et la rentabilite anticipee (Rand, 2007; Egbetunde et Akinlo, 2015). Pour ce qui est phenomene de l’alea moral, les
banques sont tres souvent incapables d’evaluer les projets des PME et refusent de s’engager. Selon l’Observatoire Europeen des
PME (2003), le bilan et le compte de resultat sont les documents les plus demandes par les banques europeennes. Or, seulement
60 % des PME fournissent regulierement ce type d’information.
Dans le contexte camerounais, Wamba (2013) montrent que les promoteurs des PME hesitent a devoiler l’ensemble des
informations dont ils disposent concernant leur entreprise, que ce soit pour des raisons fiscales, concurrentielles ou liees aux
^ qu’engendre leur production ou simplement par ignorance ou manque de competences. Tres recemment, certains auteurs
coUts
(Williams, 2016; Tsambou et al., 2015; Nguena et Tsafack, 2014) ont montre que cette insuffisance d’information est due a la
petite taille et le peu de notoriete de la plupart des PME.

2.2 Capital social et acces au financement


Le capital social est donc envisage comme etant une solution a la contrainte de credit. La proposition centrale de la theorie du
capital social repose sur l’argument selon lequel les reseaux de relations sociales constituent une ressource precieuse pour la
conduite des affaires car, ils facilitent l’action economique et permettent aux entrepreneurs d’elargir leur champ d’action,
d’economiser leurs moyens et d’acceder aux ressources et opportunites exclusives (Nahapiet et Ghoshal, 1998). Il est defini
comme l’ensemble des caracteristiques de l’organisation sociale telles que les reseaux, les normes et la confiance, qui facilitent la
coordination et la cooperation, pour un benefice mutuel. Gr^ace a ce dernier, la probabilite de survenance d’asymetrie
d’information sur le marche du credit est significativement reduite (Kobou et al., 2009; Moro et Fink, 2013). Ainsi, une relation
^ d’emprunt, facilite l’acces au credit des PME et
bancaire basee sur la confiance acquise dans les relations sociales reduit le coUt
par la, reduit le montant des garanties demandes aux PME et permet d’eviter les defauts de remboursement des PME (Mazen,
2014).
A cote de ce capital social, la litterature presente plusieurs facteurs susceptibles d’influencer la structure financiere des
firmes. Ainsi, Danso-Abbeam et al. (2014) montre sur le cas des PME au Ghana que le degre d’instruction du
soumissionnaire, la fourniture d’une caution personnelle, la duree dans les affaires, la disposition d’un espace permanent
pour ses operations d’affaires et la taille de menage sont les variables qui influencent de maniere significative la
probabilite d’acces au credit des PME. De m^eme, d’autres auteurs (Olekamma et Tang, 2016; Akhabonje et Namusonge,
2016; Osano and Languitone, 2016) montrent que les ventes, la disponibilite de garantie additionnelle, le revenu
d’affaires, le niveau et la disponibilite de releve du compte bancaire ont une influence significative sur le montant de credit
debourse a la PME.
Par ailleurs, a l’instar des travaux effectues par GICAM-GTZ (2008), Taka (2009), Kobou et al. (2009), Atangana et al. (2013)
et Tsambou et al. (2015) sur les PME au Cameroun, ce travail utilise les donnees plus recentes que ces derniers pour mettre en
exergue les facteurs susceptibles d’expliquer le financement bancaire. Il analyse les determinants du financement bancaire en
fonction de differentes caracteristiques des PME et leur capital social. En s’inscrivant dans le cadre des travaux de Steijvers
(2008), cette demarche montre que les problemes informationnels rencontres par les pr^eteurs se posent differemment en fonction
de la taille et du capital social des entreprises.

© 2017 The Authors. African Development Review © 2017 African Development Bank
Financement Bancaire 651

3. Methodologie
Afin de repondre a la problematique de cette recherche, la presente etude rapproche les theories de financement des organisations
et du Capital Social a l’acces au financement des PME. Les elements methodologiques presentes dans ce travail portent sur la
source des donnees, la specification du modele ainsi que les variables du modele empirique.

3.1 Source de donnees et echantillonnage

Les donnees utilisees pour cet article ont ete recueillies dans le cadre des enqu^etes effectuees en 2009 par le ministere des petites
et moyennes entreprises, de l’economie sociale et de l’artisanat (MINPMEESA) du Gouvernement camerounais et l’Agence
Japonaise de Cooperation Internationale (JICA) au nom du Gouvernement Japonais.1 L’objectif de cette enqu^ete est de mener
une etude de formulation de projet afin de comprendre la situation presente des PME et la politique de promotion des PME dans
le dit pays. Sur les 3193 PME camerounaises identifiees, 500 avaient ete enqu^etees avec un taux de reponse de 82,6% (413 PME).
La methode de sondage par quota a permis de determiner la taille de l’echantillon par region. A cet effet, trois modalites (le
secteur d’activites, la taille et la localisation geographique de la PME) ont servi de base de stratification. Pour le premier, trois
secteurs (secteur industriel 70%, secteur commercial 15% et secteur des services 15%). De m^eme, les exigences de
representativite regionale dans l’echantillon ont conduit a considerer les villes de Douala et de Yaounde comme regions a part,
compte tenu de leur forte densite en entreprises. Enfin, selon le critere de la taille, 20% des entreprises a enqu^eter proviendraient
de micro entreprises et 80% de petites et moyennes entreprises. Le questionnaire administre etait relative aux caracteristiques des
entreprises, a la production et au marche, a l’acces au financement et a la gestion globale de l’entreprise.

3.2 Specification du Modele econometrique

Afin d’enrichir ce travail et d’apporter des elements de reponses a notre problematique de base, une modelisation
econometrique s’avere indispensable. La variable « Acces au financement bancaire » etant dichotomique, il est fait usage du
modele a choix binaire.2 Ainsi, ce modele d’acces au financement bancaire a estimer est tel qu’on observe n individus (PME).
Ce qui conduit de toute evidence a l’observation d’une decision presentee par la variable Yi (acces au financement) qui peut
prendre deux valeurs (0 ou 1). L’acces au financement des PME etant la manifestation visible de la variable latente3 (yi  )
inobservable de la PME, on se ramene a conceptualiser un modele d’analyse de la variance sur cette variable latente. Nous
formulons le probleme en termes de fonction de credibilite (C) de la PME. Pour une PME i de caracteristiques Xi, l’acces au
financement est sous condition d’une certaine credibilite C(1, Xi), alors que le refus du financement est sous condition d’une
credibilite C(0, Xi). On a alors:
(
1 si C ð1; X i Þ > C ð0; X i Þ
Yi ¼ ð1Þ
0 si C ð0; X i Þ > C ð1; X i Þ

La banque choisissant la situation qui lui permet de maximiser le remboursement futur de la somme empruntee, on se ramene
au cas de la variable latente en posant:

yi ¼ C ð1; X i Þ  C ð0; X i Þ ð2Þ


8
< 1 si yi > 0
on a alors : Y i ¼ avec yi ¼ X i b þ ei ð3Þ
: 0 si yi < 0

u X i ¼ ð1 xi1 xi2    xik Þ est le vecteur des variables explicatives; b ¼ ðb0 ; b1 ; b2 ; . . . ; bk Þ le vecteur des parametres a
O
estimer; ei correspond a la realisation des evenements aleatoires. L’acces au financement etant une variable qualitative, la
probabilite que la PME i obtient un financement (Y i ¼ 1) est alors

© 2017 The Authors. African Development Review © 2017 African Development Bank
652 A. D. Tsambou et al.

Pi ¼ PrðY i ¼ 1Þ ¼ Prðei < X i bÞ ¼ F ðX i bÞ ð4Þ

Avec F la fonction de repartition. La solution recherchee vise a trouver la forme fonctionnelle telle que PrðY i ¼ 1Þ respecte
X
N
les proprietes ci-apres: 0 < PrðY i ¼ 1Þ < 1 et PrðY i ¼ 1Þ ¼ 1 tel que lim PrðY i ¼ 1Þ ¼ 1 et lim PrðY i ¼ 1Þ ¼ 0.
X i b!þ1 X i b!1
i¼1
En n’admettant que la fonction de repartition F ðX i bÞ est une representation de la loi logistique (modele LOGIT),4 la
probabilite pour qu’une PME i obtienne un financement bancaire depend d’un ensemble de variables explicatives representees
par le vecteur X i tel que:

expðX i bÞ 1
Pi ¼ PrðY i ¼ 1Þ ¼ FðX i bÞ ¼ ¼ ð5Þ
1 þ expðX i bÞ 1 þ expðX i bÞ

L’estimation de ce modele logit simple est effectuee selon la methode du maximum de vraisemblance.

3.3 Choix de variables

Les variables exogenes sont envisages en fonction de notre objectif. Ces variables sont presentees selon les facteurs de risque
devoiles dans le modele theorique. Pour evaluer le risque d’asymetrie d’information entre les entreprises et les bailleurs de fonds,
nous utiliserons la disponibilite de livres comptables et la qualite des informations qu’ils contiennent (Audet et al., 2009). La
garantie et les informations financieres jouent un r^ole important pour evaluer la capacite de remboursement des entreprises. Les
garanties privilegiees par les bailleurs de fonds au Cameroun sont les biens immobiliers.
Le risque commercial est approxime par la part d’exportation (St-Pierre, 2004). Une PME exportatrice peut facilement
acceder au credit bancaire. Le risque lie aux PME est evalue par la taille (Audet et al., 2009) et l’^age (Zambaldi et al., 2009) de la
PME. La taille de l’entreprise est mesuree par l’effectif de la main d’œuvre alors que l’^age appara^ıt ainsi comme vecteur de la
reputation de l’entreprise. Le risque lie au cas particulier du Cameroun est evalue par les variables suivantes: la localisation des
PME (St-Pierre, 2004), absence d’agence de garantie, financement partiel des banques. Les variables caracteristiques du capital
social portent essentiellement sur: l’appartenance a des reseaux d’affaires et associations (Kim et al., 2009), le soutien des
autorites locales pour obtenir des credits; le poids de la main d’œuvre familiale dans le personnel (Bjornskov, 2006); le poids du
chiffre d’affaires etranger et du chiffre d’affaires familial sur celui de la PME.

3.4 Quelques statistiques descriptives des variables


Cette analyse descriptive montre la forte tendance des entreprises a recourir a l’endettement bancaire. Les tests statistiques
univaries permettent d’avoir une bonne image de l’echantillon et d’identifier les relations significatives entre certaines variables
de risque et l’obtention du financement bancaire. Mais aussi, ils nous renseignent sur le phenomene complexe de financement des
PME dans la globalite et ne permettent pas de demontrer de relations causales des differentes variables qui ont un impact sur la
possibilite d’obtention de financement. C’est dans le but de comprendre les motivations d’un tel comportement que l’analyse
econometrique sera proposee.
En ce qui concerne le risque d’asymetrique d’information, la plupart des PME (86.92%) etudiees presentent regulierement leurs
etats financiers. Celles qui accedent au financement bancaire sont plus susceptibles d’avoir un livre comptable de qualite. Il existe
un differentiel d’information et de communication entre les PME et les etablissements financiers du fait que plus de la moitie (60%)
des entreprises etudiees ont un manque de communication avec les bailleurs de fonds. Le differentiel d’information est d’autant
plus profond que le livret comptable est etabli par une personne non qualifie. Ceci s’explique du fait que 30,77% des PME qui
obtiennent le financement bancaire ont les livrets comptables, alors 96,92% de ces entreprises font recours aux agents comptables
qualifies ou aux cabinets comptables pour une meilleure comptabilite de leur structure. Quant aux risques de remboursement,
95.38% des PME ayant les immobilisations sous forme de garanties obtiennent facilement un financement. Cette garantie exigee
par la banque apparait comme une incitation de l’emprunteur et permet aux banquiers de diminuer le risque de perte monetaire. Au
niveau du risque operationnel et commercial, la majorite des PME financees (70,77%) sont du secteur industriel (confer Tableau 1).

© 2017 The Authors. African Development Review © 2017 African Development Bank
Financement Bancaire 653

Tenant compte du capital social des PME (la localisation, le soutien des autorites locales pour obtenir du financement et la
participation a des associations de PME), les tests pour deux dernieres variables sont tres significatifs. Ce qui demontre une
relation forte entre ces variables et l’obtention de financement bancaire. Parmi les PME qui obtiennent un financement bancaire,
88,46% sont urbaines, 93% obtiennent le soutien des autorites locales et 41,54% sont membres de differentes associations
(Tableau 1).

3.5 Analyse statistique des variables quantitatives

En observant les variables liees aux caracteristiques des PME dans le tableau 2, on se rend compte qu’en moyenne les PME
(58.46%) ont 13 ans d’^age. Les entreprises les plus jeunes ont donc un acces moindre au financement bancaire. Par contre, les
entreprises les plus anciennes obtiennent facilement un financement bancaire du fait de leur reputation, de leur historique
financier et de leur relation de long terme avec les creanciers. Ces PME constituees de tres petites entreprises (0,77%), de petites
entreprises (32,31%) et de moyennes entreprises (66,92%) ont en moyenne un effectif de 39 employes. Les plus grandes
entreprises auraient donc moins de difficulte a obtenir du financement aupres des banques par rapport aux petites entreprises. Par
rapport a la structure de propriete, les PME ayant un dirigeant proprietaire ou appartenant a une seule famille ont une decision de

Tableau 1: Analyse statistique des variables qualitatives (%) et tests statistiques univaries
Variables Designation Financement bancaire Test statistique

Oui (31.5) Non (68.5)


Risque d’asymetrie d’information
Effet comptable 1 ¼ etat financier 30.77 32.16 1.47
2 ¼ occasionnelle 64.62 51.24
3 ¼ aucun 4.62 16.61
Comptable 1 ¼ comptable d’entreprise 96.92 82.33 0.262
2 ¼ cabinet comptable 2.31 12.01
3 ¼ une personne quelconque 0.77 5.65
Risque de remboursement
Garantie 1 ¼ biens immobiliers 95.38 3.53 6.33
0 ¼ aucun 4.62 96.47
Rentabilite 1 ¼ [0; 2000] 44.62 61.13 0.418
2 ¼ [2001; 6000] 8.46 11.31
3 ¼ plus de 6000 46.92 27.56
Risque operationnel et commercial
Motif de la Demande du Credit 1 ¼ fonds d’investissement 70.77 78.45 0.408
0 ¼ fons d’exploitation 29.23 21.55
Exportation 1 ¼ exportation 25.38 27.56 0.111
0 ¼ aucune 74.62 72.44
Particularite du Cameroun
Localisation 1 ¼ urbaine 88.46 92.23 0.436
0 ¼ rurale 11.54 7.77
Attitude des banques 1 ¼ oui 43.85 36.04 0.326
0 ¼ non 56.15 63.96
Absence d’agence de garantie 1 ¼ oui 18.46 16.96 0.103
0 ¼ non 81.54 83.04
Risque lie au capital social de la PME
Soutien des autorites 1 ¼ avoir du soutien 93.08 14.84 4.345
0 ¼ aucun 6.92 85.16
Participation aux Reseaux Sociaux des PME 1 ¼ ^etre membre 41.54 20.49 1.013
0 ¼ aucune 58.46 79.51
  
p < 0.10; p < 0.05; p < 0.01.

© 2017 The Authors. African Development Review © 2017 African Development Bank
654 A. D. Tsambou et al.

Tableau 2: Analyse statistique des variables quantitatives et tests statistiques univaries


Variable Moyenne 
Ecart type Observation Test statistique

Age 13.39952 12.02859 413 0.0307


Taille 39.26392 65.72062 413 0.0124 
Ratio (CA etranger/CA) 10.03148 28.77915 413 0.0075
Ratio (CA familial/CA) 4.513317 17.06549 413 0.0007
Ratio (emplois familial/emplois total) 0.1150809 0.7097499 413 1.805
  
p < 0.10; p < 0.05; p < 0.01; CA ¼ chiffre d’affaires.

financement qui depend du dirigeant ou du chef de famille. Cet indicateur influence negativement l’acces au credit du fait que les
creanciers s’inquietent quant a la gestion de la firme.

4. Resultat
L’analyse multivariee a travers les coefficients de correlations entre variables explicatives permet d’examiner la question de
multi-colinearite d’une part et d’autre part, la liaison entre les differentes variables. En effet, la presence du probleme de multi-
colinearite est signe d’une redondance d’information dans le modele et deteriore sa qualite. Generalement, la solution proposee
est le retrait de la variable responsable du probleme jusqu’a ce que celui-ci disparaisse. Le test de la multi-colinearite nous amene
au constat selon lequel tous les coefficients de correlation partielle qui sont significatifs sont faibles (inferieurs a 0,5). Ceci
conduit a l’absence d’une multi-colinearite entre les variables exogenes. Vu la nature exploratoire de cette etude, nous avons
prefere une regression pas a pas avec les criteres d’entree et sortie des variables selon leur significativite. Nos resultats (tableau 3)
sont robustes selon les standards usuels.
Selon les resultats obtenus, la valeur du ratio de vraisemblance (474,49) est plus grande que la valeur de la statistique de Khi-
deux theorique au seuil de 1% (Prob > chi2 ¼ 0.0000). La valeur du pseudo-R2 de ce modele (0,9223) est assez satisfaisante et le
pourcentage de prediction du modele (97.82%) est largement superieur au seuil (50%) predefini. Ce qui signifie que le modele est
applicable et globalement significatif. L’observation des T de Student et des Prob(z) renseigne sur les variables statistiquement

pertinentes. Etant donne que les parametres du modele estime ne renseignent que sur le degre et le sens de l’evolution de la
probabilite d’acces au financement bancaire, le calcul des Odds Ratio et des effets marginaux (Thomas, 2002, p. 60) permet de
completer les resultats de l’estimation afin de capter l’amplitude de la variation des variables. Suite a ces differents tests, on
obtient les resultats presentes dans le tableau 3.

5. Discussion
Le degre d’opacite de la PME s’est avere comme facteur pertinent limitatif d’acces au credit bancaire. Ce risque d’asymetrie
d’information mesure par la disponibilite et la qualite des livres comptables a un impact positif sur l’obtention du financement
bancaire. En effet, la qualite du livret comptable produit par la PME conditionne son acces au credit bancaire. Les effets comptables
m^eme de maniere occasionnelle influencent positivement la probabilite d’acces au financement. Cette comptabilite constitue une
source importante d’information pour les partenaires externes de l’entreprise et specialement les banquiers. Ces resultats corroborent
avec ceux de Steijvers (2008) relatifs a l’acces au credit bancaire des PME belges. En plus de la qualite des documents comptables
produits par la PME, les caracteristiques de la personne etablissant ces documents jouent sur la decision de la banque. La qualite de ce
document est douteuse lorsque l’entreprise n’a pas un personnel qualifie qui s’occupe de la tenue de sa comptabilite. La plupart des
PME dans le contexte camerounais ont deux rapports financiers (un pour l’entreprise et l’autre pour l’autorite fiscal), ce qui amene les
banquiers a ne plus se fier exclusivement sur les effets comptables pour prendre une decision de credit.
Pour le risque de remboursement, la garantie a une influence positive sur la probabilite d’acces au financement bancaire. Cette
forte significativite de l’octroi de garanties pour l’acces au financement corrobore les resultats empiriques de plusieurs etudes
(Van Pham et al., 2009; Taka, 2009). Offrir un bien en garantie est une condition obligatoire chez plusieurs banquiers et cela s’est
verifie dans le contexte des PME camerounaises. Ainsi, a un seuil de significativite de 1%, une PME presentant des garanties de
remboursement de son credit a plus de chance de voir sa demande de credit acceptee qu’une entreprise sans garanties. Ces

© 2017 The Authors. African Development Review © 2017 African Development Bank
Financement Bancaire 655

Tableau 3: Estimation, odds ratio et effets marginaux du modele


y ¼ Pr (finance bancaire) (predict) ¼ 0.00811309
Variables Coefficient Odds Ratio dy/dx ey/ex

Risque d’asymetrie d’information


Effet comptable 5.225 185.90 0.2231 1.644
(2.687) (499.58) (0.3538) (0.850)
Comptabilite occasionnelle 5.4799 239.82 0.0855 3.0138
(2.682) (643.25) (0.1231) (1.4849)
Comptable de la PME –5.023 0.0066 –0.38765 –4.331
(5.2037) 0.0342) (1.149) (4.483)
Comptable un tiers –4.7863 0.0083 –0.00978 –0.19541
(16.983) (0.1417) (0.0124) (0.6943)
Risque de remboursement
Garantie offerte 11.4427 93219 0.9488 3.682
(2.6024) (242598) (0.0564) (0.8693)
Rentabilite 1 –3.7363 0.0238 –0.0604 –2.0728
(1.5492) (0.0369) (0.0778) (0.868)
Rentabilite 3 –4.0114 0.0181 –0.03001 –1.3391
(1.8027) (0.0326) (0.0354) (0.6094)
Risque lies aux PME
Age 0.0582 39.9 0.00046 0.7737
(0.05151) (69.64) (0.00067) (0.6869)
Taille 0.01888 1.082 0.00015 0.7355
(0.00672) (0.0068) (0.00018) (0.2667)
Risque operationnel et commercial
Motif de demande de pr^et 0.0702 1.073 0.00055 0.0529
(1.322) (1.419) (0.01017) (0.9975)
Exportation –2.3268 0.0976 –0.01356 –0.6202
(1.4108) (0.1377) (0.01721) (0.3789)
Particularite du Cameroun
Localisation –2.680 0.0685 –0.0794 –2.4202
(2.4196) (0.1658) (0.18989) (2.188)
Attitude des banques 1.9910 7.323 0.02329 0.7603
(1.108) (8.11) (0.03096) (0.426)
Agence de garantie –2.1161 0.1205 –0.01026 –0.3659
(1.498) (0.1805) (0.01356) (0.2603)
Risque lie au capital social des PME
Soutien des autorites 8.909 7402.39 0.64244 3.488
(2.4196) (17585.5) (0.2562) (0.9578)
Reseaux sociaux –5.5979 0.0037 –0.03584 –1.506
(1.9707) (0.0073) (0.04137) (0.539)
Ratio (CA etranger/ CA de la PME) 0.06396 1.06 0.00051 0.636
(0.0248) (0.0546) (0.0006) (0.2518)
Ratio (CA familial/CA de la PME) 0.0790 1.019 0.00063 0.354
(0.0357) (0.0068) (0.00076) (0.1623)
Ratio (emplois familial/emplois totale) 3.6865 1.066 0.02966 0.999
(1.7452) (0.0265) (0.03885) (0.4767)
Constante –8.1225
(5.585)
Notes: Logistic regression: Number of obs ¼ 413; LR chi2(19) ¼ s474.49; Prob > chi2 ¼ s0.0000; Log likelihood ¼ s19.9977; Pseudo R2 ¼ s0.9223;

¼ significativite,  p < 0.10;  p < 0.05;  p < 0.01; CA ¼ schiffre d’affaires; Pr(Yi ¼ 0) ¼ reference; (.) ¼ secart type

© 2017 The Authors. African Development Review © 2017 African Development Bank
656 A. D. Tsambou et al.

garanties permettent aux banques de reduire significativement leur risque de pertes monetaires. Ce resultat se confirme par une
correlation positive entre l’acces au credit bancaire et la valeur de la garantie. Au-dela des garanties, la rentabilite influe
significativement au seuil de 5% sur la probabilite d’acces au credit bancaire. Ce resultat s’explique par le fait que, plus une
entreprise est rentable, plus elle s’endettera pour profiter au maximum du principe de deductibilite fiscale des charges des dettes.
De plus, une meilleure rentabilite va de pair avec une probabilite de remboursement des dettes plus elevee. Ce qui, aux yeux des
creanciers, constitue une garantie supplementaire. La significativite negative de cette variable s’explique par le fait que, les
firmes biaisent leurs resultats financiers dans l’optique d’obtenir un financement.
Concernant les risques lies aux PME, il existe une relation positive et significative au seuil de 5% entre la taille des PME
camerounaises et l’acces au financement bancaire. Ce qui montre qu’une PME qui a un grand nombre d’employes a une
probabilite plus grande d’obtenir un financement bancaire. En effet, les PME camerounaises ayant plus de 20 employes ont plus
de chance d’obtenir un financement aupres d’une banque que celles ayant moins de 5 employes. Ce premier constat rejoint les
resultats des etudes empiriques anterieures (Taka, 2009; Zidi et Djelassi, 2016). Cette taille de la PME peut ^etre non seulement
indicative de sa capacite financiere et de la disponibilite d’information, mais aussi, une taille importante entraine une bonne
organisation et une bonne gestion d’un capital important de l’entreprise.
Le risque commercial evalue par les activites d’exportation presente un impact negatif et significatif sur l’acces au financement
des PME. Certes qu’une PME exportatrice est perSc ue comme dynamique par ses partenaires externes; cependant, ses chances
d’acceder au credit bancaire sont moindre. Ceci s’explique par le fait que le circuit des operations d’exportation et la fluctuation des
cours exterieurs peuvent conduire au risque d’insolvabilite susceptible d’hypothequer le respect des engagements de credit.
Quant aux risques lies au contexte particulier du Cameroun, l’attitude passive des banques vis-a-vis des projets PME presente
un impact positif et significatif sur l’acces des PME au credit bancaire. Il est interessant de constater d’ailleurs que les banques
camerounaises ont une attitude passive au regard du financement des PME. Quelque soit la viabilite des projets, les banques ne
financent pas le montant total des besoins de financement requis. Ce montant pour certaines banques est fixe a 80% du montant
total des ressources demandees. En effet, une demande de credit appreciee a 80% par les differentes commissions de la banque a
7 fois plus de chance d’^etre financee qu’une demande ayant une appreciation moindre. En outre, si le taux d’appreciation du
projet de la PME augmente d’une unite, sa probabilite d’obtention de financement augmentera de 76% (tableau 3). En revanche,
l’absence d’un systeme officiel de garantie du credit pour completer l’insuffisance d’hypotheque presente un effet negatif et non
significatif sur l’acces au credit bancaire des PME au Cameroun. Ce resultat s’explique par l’intensification du secteur de la
microfinance, la creation de la banque des PME, la creation d’une agence pour la promotion des PME.
Pour ce qui est du capital social, l’appartenance d’une PME aux reseaux sociaux a un impact significatif au risque de 1% sur
l’acces au financement bancaire. Ainsi, appartenir a un groupe pourrait ^etre un facteur de reduction de l’asymetrie d’information
entre les entreprises et les offreurs de capitaux. Dans un contexte d’information imparfaite comme celui du Cameroun, les reseaux
^
et associations des entreprises rendent moins coUteuses les transactions economiques. Ces structures facilitent la circulation de
l’information, procurent aux banquiers des garanties personnelles plus solides, exercent des mecanismes internes de contr^ole et de
sanction aux membres. Non seulement, l’appartenance aux reseaux d’affaires offre au banquier des garanties plus solides, mais
aussi elle exerce un mecanisme d’auto surveillance sur les membres de l’association et accroit la confiance de la banque.
En outre, le soutien des autorites gouvernementales facilite l’obtention d’un financement bancaire. Cette significativite
positive s’explique par le fait que certaines banques de la place preferent se sacrifier a prendre d’enormes risques pour entretenir
de bonnes relations avec les autorites gouvernementales et poursuivre par la d’autres objectifs non financiers. En plus, le ratio du
chiffre d’affaires etranger, le ratio du chiffre d’affaire familial et le ratio du nombre d’employes familial impactent positivement
sur l’acces au financement bancaire des PME. L’integration des externes dans une PME serait un facteur de reduction de
l’asymetrie d’information vis-a-vis des offreurs de capitaux.
Les relations sociales, qui sont des indicateurs du degre de confiance entre le banquier et l’entrepreneur, ont des effets positifs
sur les conditions de credit ainsi que sur le niveau de risque qu’assument les porteurs de projet. En definitive, le capital social
ameliore significativement la probabilite d’acces de la PME camerounaise au credit bancaire en reduisant son degre d’opacite
aux yeux du banquier.

6. Conclusion
L’objectif de cette recherche etait d’identifier les facteurs explicatifs de l’acces au financement bancaire des PME
camerounaises. Il en ressort que la probabilite d’acces au financement bancaire s’accroit chez les PME etablissant regulierement

© 2017 The Authors. African Development Review © 2017 African Development Bank
Financement Bancaire 657

leur comptabilite et est d’autant plus croissante si ces effets financiers sont « audites » par un comptable qualifie ou par un cabinet
comptable. Ce qui permet de comprendre que les cadres bancaires ont besoin d’une information complete et parfaite pour les
aider a prendre une decision claire et d’evaluer parfaitement les caracteristiques de l’entreprise qui sollicite le credit. Cette
probabilite augmente avec la qualite de garanties offertes.
Le present travail qui contribue a l’amelioration du financement des PME camerounaises suggere conformement aux analyses
theoriques et aux resultats empiriques, les propositions suivantes: Suite au decret de la creation d’une banque specialisee au
financement des PME, il faut egalement la creation d’une institution de renforcement des capacites de la PME et la mise en place
des cabinets de consultation pour mieux renseigner les PME sur leur analyse financiere et leur plan d’investissement. La
necessite de ces propositions trouve sa justification dans les trois raisons suivantes: Premierement, les problemes de financement
de la PME que sont le financement limite, l’absence ou insuffisance d’un fonds de garantie et l’asymetrie d’information ne
peuvent pas ^etre resolus par chaque banque commerciale, car il s’agit d’une affaire de systeme financier. Deuxiemement, les
problemes de financement courant de la PME ne peuvent pas ^etre resolus en laissant jouer les mecanismes du marche comme le
preconisent les classiques. L’intervention et l’appui du gouvernement sont indispensables. Troisiemement, il faut un
developpement du marche financier qui donnera la possibilite aux PME non seulement de se financer directement sur le marche
des capitaux, mais aussi de vendre les actions ou de rechercher de nouveaux actionnaires pour se refinancer.
Les limites de ce travail sont liees a la qualite de donnees qui ne permettent pas d’etudier l’acces a la finance informelle, alors
que l’economie camerounaise est caracterisee par une forte presence des activites informelles dans tous les secteurs. Elles ne
permettent non plus d’evaluer l’acces au credit offert par les Microfinances, alors que la structure de ces dernieres propose des
conditions de credits accessibles aux agents economiques dont le credit bancaire est rationne. Cependant, dans la mesure ou la
microfinance a ete creee pour accompagner l’activite des PME compte tenu des conditions rudes imposees par les banques, quel
serait les facteurs explicatifs de l’acces au financement informel et des Microfinances?

Notes
1. Cf. MINPMEESA, Etude sur la Formulation du Plan Directeur (M/P) pour le Developpement des Petites et Moyennes
Entreprises en Republique du Cameroun (MINPMEESA, 2009), Agence Japonaise de Cooperation Internationale, Unico
International Corporation, IDD- JR-08-069.
2. La base theorique de ce modele a ete donnee par McFadden (1968) a travers une theorie de l’utilite aleatoire.
3. La variable latente est une variable non observable et representative du phenomene etudie.
4. Le choix du modele LOGIT par rapport au modele probit pouvant resoudre la question est que le Logit a l’avantage d’une plus
grande simplicite numerique et aussi parce que les estimateurs obtenus avec le modele LOGIT sont environ p/3 fois plus
grands que ceux obtenus par le modele PROBIT.

Bibliographie
Admassu, T. (2009), ‘Quelles perspectives de financement pour les PME en Afrique?’, La revue de PROPARCO.
Akhabonje, S.M. et G.S. Namusonge (2016), ‘Factors Affecting Access to Credit by Small and Medium Enterprises: A Case of
Kitale Town’, International Journal of Social Sciences and Humanities Invention, Vol. 3, No. 10, pp. 2904–17.
Ang, J. S. (1991), ‘Small Business Uniqueness and the Theory of Financial Management’, Journal of Small Business Finance,
Vol. 1, pp. 1–13.
Ang, J. S. (1992), ‘On the Theory of Finance for Privately Held Firms’, Journal of Entrepreneurial Finance, Vol. 1, No. 3, pp.
185–203.
Atangana, O. H., M. Douzounet, M. Kouty, S. P. Moungou Mbenda, J. M. Nana Djomo, H. Ngoa Tabi, E. Niyonsaba Sebigunda,
P. Ongono, et U. T. Yogo (2013), ‘Capital Social et Entrepreneuriat au Cameroun’, TRUST AFRICA No. 11.
Audet, J., J. St-Pierre et Thi Hong Van Pham (2009), « La vulnerabilite des TPE et des PME dans un environnement mondialise »,
11eme Journees scientifiques du Reseau Entrepreneuriat, 27, 28 et 29 mai 2009, INRPME, Trois-Rivieres, Canada.

© 2017 The Authors. African Development Review © 2017 African Development Bank
658 A. D. Tsambou et al.

Banque Mondiale et de la Societe Financiere Internationale (2009), Doing Business: comparaison des reglementations dans 181
pays, 6eme Edition, World Bank, Washington, DC.
Besanko, D. and A. V. Kanatas (1993), ‘Collaterals and Rationing: Sorting Equilibria in monopolistic and Competitive Credit
Market’, International Economic Review, Vol. 28, No. 3.
Bjornskov, C. (2006), ‘The Multiple Facets of Social Capital’, European Journal of Political Economy, Vol. 22, pp. 22–40.
Cassar, G. et S. Holes (2003), ‘Capital Structure and Financing of SMEs: Australian Evidence’, Accounting and Finance, Vol.
43, No. 2, pp. 123–47.
Coluzzi, C., A. Ferreto et C. Martinez-Carrascal (2009), ‘Financing Obstacles and Growth: An Analysis for Euro Area Non-
financial Corporations’, ECB Working Paper Series, No. 997.
Danso-Abbeam, G., G. K. I. Ansah and S. D. Ehiakpor (2014), ‘Microfinance and Micro-Small-Medium Scale Enterprises
(MSME’s) in Kasoa Municipality, Ghana’, British Journal of Economics, Management & Trade, Vol. 4, No. 12, pp. 1939–56.
Egbetunde, T. and A. Enisan Akinlo (2015), ‘Financial Globalization and Economic Growth in Sub-Saharan Africa: Evidence
from Panel Cointegration Tests’, African Development Review, Vol. 27, No. 3, pp. 187–98.
GICAM-GTZ (2008), “Enquete sur le Climat des Investissements et des Affaires au Cameroun, Principaux resultats”.
INS/RGE (2009), ‘Le recensement general des entreprises au Cameroun’, Rapport final, Ao^ut–Novembre.
Kim, M., J. Surroca and J. A. Tribo (2009), ‘The Effect of Social Capital on Financial Capital’, INDEM - Working Paper
Business Economic Series id-09-02, Instituto para el Desarrollo Empresarial (INDEM).
Kobou G., H. Ngoa Tabi et S. Moungou (2009), « La vulnerabilite des TPE et des PME dans un environnement mondialise »,
11es Journees scientifiques du Reseau Entrepreneuriat, 27, 28 et 29 mai 2009, INRPME, Trois-Rivieres, Canada, Economie
Appliquee.
Lefilleur, J. (2009), ‘Financer les PME dans un contexte de forte asymetrie d’information’, Revue de Proparco, No. 1, pp. 14–16.
McFadden, D. (1968), ‘Regression Based Specification Tests for the Multinomial Logit Model’, Journal of Econometrics, Vol.
34, pp. 63–82.
Mazen, G. (2014), « Le r^ole des liens sociaux et de la confiance sur le financement bancaire des PME: une etude exploratoire »,
These de Doctorat, Universite de Bretagne Occidentale, EA 2652, ed N0 505.
MINPMEESA (2009), ‘La formulation du plan directeur pour le developpement des petites et moyennes entreprises en
republique Camerounaise’, Rapport final, Janvier.
Moro, A. and M. Fink (2013), ‘Loan Managers Trust and Credit Access for SMEs’, Journal of Banking and Finance, Vol. 37, pp.
927–36.
Myers, S. (1984), ‘The Capital Structure Puzzle’, Journal of Finance, Vol. 39, pp. 575–92.
Nahapiet, J. and S. Ghoshal (1998), ‘Social Capital, Intellectual Capital, and the Organizational Advantage’, The Academy of
Management Review, Vol. 23, No. 2, pp. 242–66.
Nguena, C. L. and R. Tsafack Nanfosso (2014), ‘Facteurs Microeconomiques du Deficit de Financement des PME au
Cameroun’, African Development Review, Vol. 26, No. 2.
Observatoire europeen des PME (2003), ‘L’acces au financement pour les P.M.E’, Publications-DG Entreprises, Commission
europeenne, Fevrier, Numero 2.
OCDE (2006), Travaux de Synthese sur Le financement des PME et des entrepreneurs, OCDE, Paris, Fevrier.
Olekamma, K. C. et Z. Tang (2016), ‘The Influence of Entrepreneurial Characteristics on Small and Medium-Sized Enterprise
Accessibility to Debt Finance in Nigeria’, International Journal of Managerial Studies and Research, Vol. 4, No. 10, pp. 83–92.
Ondel’ansek, K. (2010), “Les contraintes de financement des PME en Afrique: le r^ole des registres de credit”, These de Doctorat,
HEC Montreal.

© 2017 The Authors. African Development Review © 2017 African Development Bank
Financement Bancaire 659

Osano Mogaka, H. and H. Languitone (2016), ‘Factors Influencing Access to Finance by SMEs in Mozambique: Case of SMEs in
Maputo Central Business District’, Journal of Innovation and Entrepreneurship, Vol. 5, 13, DOI 10.1186/s13731-016-0041-0.
Quiles, J. (1997), Schumpeter et l’evolution economique: circuit, entrepreneur, capitalisme, Nathan, Paris.
Rand, J. (2007), ‘Credit Constraints and Determinants of the Cost of Capital in Vietnamese Manufacturing’, Small Business
Economics, Vol. 29, No. 1–2, pp. 1–13.
Steijvers, T. (2008), ‘Existence of Credit Rationing for SME’s in the Belgian Corporate Bank Loan Market', Working Paper.
Available at: https://ssrn.com/abstract=495162.
St-Pierre, J. (2004), La gestion du risque, Presses de l’Universite du Quebec, Sainte-Foy.
Stiglitz, J. and A. Weiss (1981), ‘Credit Rationing in Markets with Imperfect Information’, American Economic Review, Vol. 76,
No. 3, pp. 393–410.
Taka, D. (2009), « Pourquoi les banques sont-elles reticentes a finance les PME camerounaise » Revue Camerounaise de
Management, n° 19, janvier-juin 2010.
Thomas, A. (2002), Econometrie des variables qualitative, Dunod, Paris.
Tsambou, A. D., L. Ndokang Esone et O.H. Atangana (2015), ‘Determinants du choix de financement des petites et moyennes
entreprises au Cameroun’, Revue Europeenne du Droit Social, Vol. XXIX, No. 4, pp. 87–115.
Van Pham, T., J. St-Pierre et J. Audet (2009), « La vulnerabilite des TPE et des PME dans un environnement mondialise », 11emes
Journees scientifiques du Reseau Entrepreneuriat, 27, 28 et 29 mai 2009, INRPME, Trois-Rivieres, Canada.
Wamba, H. (2013), ‘Capital social et acces des PME africaines au credit bancaire: le cas du Cameroun’, La Revue des Sciences de
Gestion, No. 259-260, pp. 53–68.
Williams, K. (2016), ‘Remittances and Financial Development: Evidence from Sub-Saharan Africa’, African Development
Review, Vol. 28, No. 3, pp. 357–67.
Zambaldi, F., F. Aranha, H. Lopes and R. Politi (2009), ‘Credit Granting to Small Firms: A Brazilian Case’, Journal of Business
Research, Vol. 64, pp. 309–15.
Zidi, M. et M. Djelassi (2016), « Pouvoir de marche et stabilite financiere: Etude du secteur bancaire Tunisien », African
Development Review, Vol. 28, No. 4, pp. 416–29.

© 2017 The Authors. African Development Review © 2017 African Development Bank
African Development Review, Vol. 29, No. 4, 2017, 660–673

Semi-parametric Regression-based Decomposition Methods: Evidence


from Regional Inequality in Tunisia

Amal Jmaii, Damien Rousseliere and Christophe Daniel

Abstract: This paper examines urban-rural welfare inequality in Tunisia. Founded on the recentred influence function and
quantile regression based counterfactual decomposition, we dissect the gap between the two areas. Results of the present study
suggest that the difference between rural poor households and urban poor households is due essentially to characteristic effects;
while for wealthier households both characteristic and returns to these characteristic effects (for example, efficiency of
educational system) are responsible for this gap. Additionally, the results demonstrate that this is an issue of value, and, more
specifically, an economic development fairness conflict. It is suggested that policymakers should address a positive
discrimination programme in favour of the marginalized region.

1. Introduction
Regional disparities are a central issue in recent literature, and it can be assessed through various approaches. Some of the most
common approaches to measure regional disparity are inequality indices such as coefficient of variation, the Theil index as well
as the generalized entropy index and the Gini coefficient, and so on. These indices are directly used in the studies of regional
inequality. For example, the National Institute of Statistics (INS) used these conventional indices to explore the evolution of
Tunisia’s regional inequality occurring over different temporal scales. Although, inequality can efficiently refer to the overall
trend of regional development, it does not completely explain whether certain families are different in their welfare due to spatial
localization, or due to high social inequality within the regions. In such a case, decomposing inequality, which divides aggregate
inequality to analyse between-regions variability, is required for a positive assessment of inequality and for the purposes of
effective social policies. Moreover, the level of welfare in a country cannot be analysed only through average consumption level
or poverty rate. In fact, quality of life in a country also depends on how consumption is distributed throughout its population. The
analysis of the distribution of consumption, therefore, enables us to better assess the relevance of social policies.
There has been a wide proliferating literature on inequality decomposition measurement between regions and its application
to micro-data surveys in the developing countries context. Several studies have been published addressing the source of
inequality between rural and urban households. Nguyen et al. (2007) discuss the welfare inequality between urban and rural
areas from 1993 to 1998 in Vietnam using Machado-Mata decomposition (2005). In this study, they conclude that inequality
differences between the two regions were due to three factors, namely, education, ethnicity, and age. Mussa (2014) used
disaggregate household expenditure to look at how inequality in household expenditure components affects total inequality and
poverty in Malawi.
Studying in depth the persistent gap between rural and urban sectors can put more emphasis on the principal factors
contributing to this gap and thus improve the decisions required to be undertaken by policymakers to support regional
development. In the spirit of this debate, this paper proposes to use a quantile regression model (Koenker and Bassett, 1978) and
counterfactual decompositions to examine the source of welfare’s inequality across both urban and rural areas. This general
decomposition method involves a semiparametric estimation (quantile regressions) model on log expenditure applied to rural
and urban households, which then will contribute to the building of a counterfactual function for rural log expenditure using


Amal Jmaii (corresponding author), Department of Quantitative Methods, LAREQUAD, Faculty of Economics and Management, University of Tunis
El Manar, Tunisia; e-mail: jemaiamal@yahoo.fr. Damien Rousseliere, Professor, SMART-LERECO, AGROCAMPUS OUEST, INRA, Angers,
France; e-mail: damien.rousseliere@agrocampus-ouest.fr. Christophe Daniel, Associate Professor, GRANEM, Faculty of Law and Economics,
University of Angers, France; e-mail: christophe.daniel@univ-angers.fr
© 2017 The Authors. African Development Review © 2017 African Development Bank. Published by Blackwell Publishing Ltd,
9600 Garsington Road, Oxford OX4 2DQ, UK and 350 Main Street, Malden, MA 02148, USA. 660
Semi-parametric Regression-based Decomposition Methods 661

urban characteristics. The counterfactual distribution aims to assess the function of rural log expenditure, as it would have
existed if the rural households had urban household’s characteristics.
The paper makes an empirical contribution to the literature compared to previous studies on the same topic, namely Hassine
(2015), Chang (2012) and Pieters (2011). We use the Recentered Influence Function (RIF) and the new decomposition method
developed by Chernozhukov et al. (2013), which permit to specify whether expenditure differences between rural and urban
areas are related to the difference in characteristics effects or to returns effects.
In particular, in Tunisia, inequality is still a predominant phenomenon. In fact, an asymmetric allocation of basic services (i.e.
education and employment) attest that growth benefits do not drip equally among regions, which may cause a feeling of
insecurity and injustice, and can lead to a potential social unrest. In this study, we examine and dissect the welfare gap between
urban and rural areas in Tunisia using Chernozhukov’s counterfactual decomposition, which will represent a general form of the
Machado-Mata decomposition (2005), and will decompose regional welfare’s gap to identify the source of this persistent
inequality. In this paper, we use the 2010 National Survey that is considered the most recent available survey in order to assess
the welfare gap between the areas.
The remainder of the paper is organized as follows. Section 2 provides a literature review on inequalities, while Section 3
gives background information on economic reforms in Tunisia. Section 4 provides descriptive statistics of the used data and
presents the details of the decomposition methods. We apply the proposed methodologies to the context of urban-rural welfare
inequality in Tunisia in the fifth section. Finally, we give our final findings and conclude with some recommendations.

2. Brief Overview on Inequalities


Inequality is no longer seen as an inevitable prerequisite for growth, but rather as an obstacle to both economic growth and
poverty reduction (Bourguignon, 2003; Odedokun and Round, 2004; Duclos and Chouchane, 2011). In a quantitative
investigation using semiparametric methods to analyse the relationship between income inequality and economic growth,
Chambers and Krause (2010) found that higher income inequality is harmful for economic growth. Son and Kakwani (2008)
found that initial levels of income inequality and economic development have a positive and significant influence on poverty
reduction. In a study based on a new data set on inequality in the distribution of income, Deininger and Squire (1996) revealed a
systematic link between growth and aggregate inequality variation in parallel with a significant positive relationship between
growth and poverty alleviation.
Income distribution is an important indicator for analysing poverty and economic development in a country. A better
understanding of the pattern and drivers of regional inequality is critical for enhancing social cohesion and inclusive growth in the
region. Considerable work has been undertaken on regional inequalities related to developing countries. In a dynamic analysis of
the patterns of household welfare in Jordan, Mansour (2012) highlighted a slight decline in inequality during 2002–10 mostly
driven by a regional catching-up effect. In a study presenting the pattern of income inequality in the MENA region, Ncube et al.
(2014) found that income inequality reduces economic growth and increases poverty in the region. In an analysis of the urban-rural
gap in North African countries, Boutayeb and Helmert (2011) show that these countries experienced considerable development in
social, economic and health indicators. Unfortunately, regions/socio-economic groups of the same country have not benefited by
these improvements equally. Bibi and Nabli (2009) found that the MENA region has a relatively higher level of income/
expenditure inequality, compared to other regions. Adams and Page (2003) revealed that compared to other regions, the MENA has
a lower income inequality and poverty rate due to public sector employment and international migration/remittances.
The regional gap in education has also been of concern in some studies on Arab countries. In an empirical analysis of
inequality of education opportunities in MENA, Salehi-Isfahani et al. (2014) show that inequality in educational achievements is
mainly explained by inequality of opportunities. Likewise, Krafft and Assaad (2016) highlighted that inequality of opportunity
— unequal resource allocations on the basis of circumstances independent from individuals’ control — can offend people’s
sense of fairness, causing anger and frustration among marginalized groups.

3. Tunisian Background
Tunisia, as a developing country, has implemented since the early years of statehood, several national programmes to reduce
poverty and regional disparities. The country has indeed shown some progress in many areas. The per capita income has risen,
public services have developed, health indicators have improved and the demographic trends have been favourable. In fact, the
© 2017 The Authors. African Development Review © 2017 African Development Bank
662 A. Jmaii et al.

country has opted for an open trade policy since the beginning of the 1990s. This strategy was preceded by the implementation of
a Structural Adjustment Plan. While this policy has led, over the last twenty years, to an increase in growth and income, the gains
from free trade are not equally distributed within the country. Economic restructuring, globalization of capital markets and
structural adjustment are synonymous of drawdown of the number of permanent workers, the subcontracting with resort to
temporary and seasonal work and the reduction of costs through deregulation of the labour market.
The deterioration of living conditions among the rural population, the increase in the unemployment rate and the inflation in the
consumer prices for basic goods associated to climate change and water scarcity problems, all threatened the sustainability of
growth in these areas, which has provoked the uprising of the people in the interior regions leading to the 2011 Tunisian revolution.
Indeed, after the revolution, Tunisia is boldly emerging from the recession considered to be the longest period of economic
downturn since the establishment of statehood. In this paper, we emphasize the failure of agricultural, trade and social policies and
the insufficiency of the measures taken to deal with the vulnerability of the rural population. More than 800,000 people were
unemployed in 2014 and the poverty rate reached 15 per cent, according to data released by the INS. Poverty rate is correlated with
many factors such as unemployment rate, illiteracy, access to health services and especially education (African Development Bank,
2013). Therefore, poverty in Tunisia is mainly concentrated in rural areas and in some regions of the country, particularly the
Central West, a region that first witnessed the first sparkles of the revolution. In addition, the lack of an infrastructure in these areas,
such as roads and communication facilities, may limit poor people’s access to information or to the labour markets.
A strong variation in poverty rates between regions (Figure 1) may be the cause of social instability and population movement.
Households with a higher level of poverty rate are more concentrated in the interior regions of the country than in the inland ones.
In addition, the measurement of poverty at the regional level allows a better definition of the priorities for regional development.
In fact, the decrease in the poverty rate compared to the higher consumption disparities with economic inequalities assert that the
GDP growth is biased towards the non-poor. Up to now, the adopted economic and social development does not correspond to
the good regional governance objectives that Tunisia has to achieve. Hence the disappointment expressed by many Tunisians
because they had better expectations after the revolution. Many Tunisians have expressed feelings of distrust towards public
policies that devote little regard for regional inequalities.

4. Sample Characteristics and Methodology

4.1 The Data

The analysis and the measurement of inequality in developing countries, and in particular in Tunisia, is further impeded by
difficulties in the choice of welfare indicator due to the limitations of the household/individual micro-data available from a
specific population. In fact, there is a wide literature about the choice between consumption expenditures or income to measure
inequality. The ideal is to use the two indicators as advocated by many researchers. In this study, we use welfare measures based
on consumption expenditures because incomes are not observable/biased.
We used data from the 2010 National Survey on Households’ Budget, Consumption and Living Standard. This survey is
conducted by the INS every five years and provides socio-demographic and economic characteristics of households and
individuals. Indeed, for 2010, it takes a representative sample of 11,281 households with 50,371 individuals. The data includes
information concerning the head of the household, composition, area of household’s location and total household’s consumption
expenditures. The choice of explanatory variables is based on the literature (Nguyen et al., 2007; Skoufias and Katayam, 2011)
and is validated by the AIC and BIC criteria.1 Therefore, we use household size, the proportion of children under 15 years old in
each household and the gender of the household head. We divided the age variable into five categories: 16–30, 31–45, 46–60,
61–75 and over 76. As far as the household education and employment characteristics are concerned, we have included the
variable of schooling of household head: illiterate (as reference), primary, secondary and higher level. For the employment
variable, we select four sectors, respectively governmental sector (as reference), private sector, self-employed and agricultural
sector. Since there are frequently monetary transfers from foreign countries, we use a dummy variable depending on whether a
household has received transfers from abroad or not. We also include the west regions as an explanatory variable, for these
regions have the highest poverty rate (Figure 1). Finally, we use log of real per capita expenditure as the dependent variable.
Table 1 and Table 2 show descriptive statistics of the selected variables. First of all, we point out that 85.23 per cent of urban
households are headed by a man against 84 per cent in rural areas. However, in rural zones the number of households headed by a
woman exceeds the urban ones.
© 2017 The Authors. African Development Review © 2017 African Development Bank
Semi-parametric Regression-based Decomposition Methods 663

Figure 1: Poverty rate (%) per region, Tunisia 2010

Source: INS.

Many researchers have highlighted that higher educational level may lead to a potential reduction in poverty by increasing
employment opportunities and the possibility of having higher income (Becker, 1992). Table 2 reveals that household heads in
rural areas are less educated than their counterparts in urban ones. In fact, about 47.68 per cent of rural household heads are
illiterate compared to 24.07 per cent in urban areas. However, the most important statistic is the high level of education, which
was recorded 10 per cent for heads in urban zones against only 1.29 per cent in rural zones. This result shows that there exists a
substantial disparity between rural and urban household heads. This gap is mainly due to geographic isolation and social/
financial norms constraint. In fact, if students in rural-isolated areas want to continue postsecondary education they often need to
move because of insufficient local educational opportunities. Nevertheless, most of the time there is strong preferences to stay
near family because moving and education are expensive. A look at the employment variable enabled us to develop some
interpretations. The share of people in rural regions employed in the governmental sector is only 8 per cent in reference to the
urban areas. For the private sector, we noticed 11.45 per cent in urban zone. However, in the rural zone, the share of this sector is
only 5.32 per cent (half of the urban one).
© 2017 The Authors. African Development Review © 2017 African Development Bank
664 A. Jmaii et al.

Table 1: Quantitative variables for both urban and rural areas


Total Urban Rural
Means Std. dev Means Std. dev Means Std. dev

Consumption expenditure (Tunisian milim dinars) 1,457,248 1,714,634 1,743,354 1,976,768 940,476.5 881,890.7
Log Consumption expenditure 13.843 0.805 14.053 0.758 13.464 0.746
Age 54.321 14.289 54.153 13.652 54.625 15.369
Household size 4.465 1.885 4.353 1.757 4.667 2.081
Percentage of children 0.209 0.231 0.203 0.226 0.219 0.240
Number of observation 11,281 7,261 4,020
Source: Own computing based on INS (National Institute of Statistics) data, Tunisia 2010.

In fact, it is hard to invest in this region because they do not have good infrastructures. The plotted kernel densities for the two
areas (Figure 2), can better visualize the difference between the characteristics of urban and rural areas.

4.2 Methodology
Semi-parametric Regression
This section aims to assess the role of various household characteristics to explain welfare inequality between urban and rural
areas. Most empirical studies focus on means modelling, a model that gives essential but limited information. We choose in this

Table 2: Qualitative variables for both rural and urban areas


Variables Urban (%) Rural (%)

Gender
Man 85.23 84.08
Woman 14.77 15.92
Education
Illiterate 24.07 47.68
Primary Level 36.33 38.59
Secondary Level 29.67 12.96
Superior Level 9.91 1.29
Region
East 71.10 28.9
West 28.89 71.11
Employment
Sector_gov 19.93 8
Sector_private 11.45 5.32
Self 21.29 10.49
Sector_agri 4.8 29.85
Unemployed 18.53 25
Inactive 24 21.43
Housing
Homeowner 84.39 95.17
Tenant 11.51 1.02
Free housing 4.09 3.8
Foreign transfer 1.58 1.21
Source: Own computing based on INS (National Institute of Statistics) data, Tunisia 2010.

© 2017 The Authors. African Development Review © 2017 African Development Bank
Semi-parametric Regression-based Decomposition Methods 665

Figure 2: Kernel density of rural-urban expenditure

paper to deal with quantile regression, which is a semi-parametric approach as it makes no assumptions about the distribution of
the errors. Furthermore, quantiles represent an approximation of poverty lines. Lower quantiles can represent the proportion of
poor in the distribution while higher quantiles represent the richest proportion. We consider the following model, inspired from
Koenker and Bassett (1978):

Qq ðY jX Þ ¼ aq þ X bq ð1Þ

where Y is log expenditure, Qq ðY jX Þ is the qth conditional quantile of Y, aq is the regression intercept and bq are values
parameters for given quantiles. The goal of the estimation is to solve the following program of minimization:
Xn
b q ¼ argminb 1
Q rq ðY i  X i bÞ ð2Þ
n i

This estimation can be used for all quantiles where 0 q  1 and rq (x) ¼ (q  1(x  0))x. Then, there are an infinite number of
possible quantile regressions. Standard in comparable studies, we choose to deal with seven quantiles namely, (5th, 10th, 25th,
50th, 75th, 90th and 95th percentiles). In fact, the estimation of the quantile regression model is relatively straightforward, but
performing the corresponding robust standard errors is often viewed as being more problematic. In this study, we follow
Machado et al. (2013) and Parente and Santos Silva (2016), which present an improvement to the old command of Stata2 to
enable to report standard errors and t statistics that are asymptotically valid. In addition, the new command presents the result of
heteroscedasticity test (Machado-Santos Silva or MSS test) which we can use as a guide in the choice of the appropriate
covariance matrix estimator to use.

Counterfactual Decomposition of the Urban–Rural Gap


After the publication of Oaxaca (1973) and Blinder (1973), which decomposes the wage gap into two components (differences in
characteristics and differences in returns to those characteristics) based on the means regression, several studies have been
developed. Miller (1987) introduced a modified version of the Oaxaca–Blinder (OB) decomposition with ordered probit model
estimation. Neuman and Oaxaca (2004) focus on the selection bias correction. However, all these studies focus on the estimation
of the difference in wage at the mean. Hence, many studies put a stress on a more general counterfactual distribution. Machado
and Mata (2005), and Melly (2005) propose conditional quantile decomposition for the wage gap. These methods are special
case of the general framework of Chernozhukov et al. (2013).
© 2017 The Authors. African Development Review © 2017 African Development Bank
666 A. Jmaii et al.

Table 3: OLS and quantile regression of total consumption expenditure per capita
Quantile regression
5th 10th 25th 50th 75th 90th 95th
Variables Means percentile percentile percentile percentile Percentile Percentile percentile

Intercept 14.632 13.651 13.817 14.165 14.616 15.138 15.510 15.668


(0.043) (0.108) (0.078) (0.036) (0.049) (0.045) (0.100) (0.128)
Gender –0.111 –0.026 –0.032 –0.038 –0.113 –0.172 –0.203 –0.230
(0.015) (0.024) (0.020) (0.019) (0.016) (0.021) (0.038) (0.034)
Age
[31, 45] 0.073 0.089 0.139 0.085 0.099 –0.0003 0.040 0.116
(0.035) (0.069) (0.055) (0.026) (0.060) (0.035) (0.062) (0.080)
[46, 60] 0.121 0.169 0.194 0.160 0.155 0.037 0.049 0.122
(0.035) (0.071) (0.058) (0.033) (0.056) (0.037) (0.056) (0.073)
[61, 75] 0.277 0.238 0.305 0.292 0.308 0.215 0.246 0.350
(0.037) (0.075) (0.062) (0.038) (0.055) (0.038) (0.068) (0.085)
76þ 0.191 0.141 0.236 0.201 0.208 0.016 0.190 0.305
(0.040) (0.090) (0.060) (0.033) (0.057) (0.053) (0.069) (0.102)
Household size –0.209 –0.226 –0.226 –0.218 –0.217 –0.206 –0.200 –0.194
(0.003) (0.009) (0.007) (0.0049) (0.004) (0.004) (0.005) (0.007)
% children –0.681 –0.512 –0.529 –0.592 –0.671 –0.768 –0.815 –0.828
(under 15 (0.030) (0.065) (0.051) (0.036) (0.031) (0.037) (0.058) (0.063)
old)
Tenant of –0.060 –0.083 –0.080 –0.074 –0.083 –0.034 –0.032 –0.043
house
(0.019) (0.038) (0.026) (0.023) (0.030) (0.016) (0.043) (0.060)
Free housing –0.128 –0.131 –0.113 –0.115 –0.141 –0.092 –0.197 –0.168
(0.025) (0.051) (0.062) (0.028) (0.026) (0.024) (0.047) (0.069)
Primary 0.194 0.183 0.197 0.193 0.202 0.189 0.180 0.167
(0.014) (0.037) (0.026) (0.019) (0.017) (0.014) (0.029) (0.040)
Secondary 0.465 0.430 0.445 0.435 0.474 0.483 0.470 0.477
level
(0.016) (0.038) (0.027) (0.022) (0.021) (0.017) (0.035) (0.059)
Higher level 0.880 0.848 0.858 0.874 0.853 0.893 0.892 0.933
(0.024) (0.063) (0.040) (0.025) (0.034) (0.042) (0.044) (0.082)
sector_Gov 0.130 0.154 0.158 0.143 0.145 0.122 0.114 0.098
(0.017) (0.033) (0.030) (0.018) (0.024) (0.015) (0.029) (0.069)
sector_Priv 0.136 0.166 0.148 0.142 0.136 0.127 0.130 0.114
(0.020) (0.041) (0.030) (0.019) (0.020) (0.018) (0.033) (0.068)
sector_Agri 0.020 0.064 0.051 0.031 0.022 0.042 –0.013 –0.005
(0.016) (0.034) (0.026) (0.016) (0.025) (0.025) (0.033) (0.044)
self 0.086 0.084 0.086 0.090 0.086 0.117 0.082 0.089
(0.015) (0.271) (0.024) (0.013) (0.016) (0.020) (0.035) (0.045)
West –0.270 –0.302 –0.301 –0.296 –0.267 –0.248 –0.236 –0.239
(0.010) (0.018) (0.020) (0.016) (0.014) (0.017) (0.022) (0.034)
Rural –0.265 –0.335 –0.301 –0.270 –0.261 –0.234 –0.2236 –0.205
(0.012) (0.030) (0.020) (0.023) (0.014) (0.016) (0.022) (0.027)
Foreign 0.190 0.135 0.262 0.197 0.165 0.169 0.246 0.246
Transfer (0.041) (0.076) (0.136) (0.035) (0.055) (0.068) (0.071) (0.095)
No. of 11,281 11,281 11,281 11,281 11,281 11,281 11,281 11,281
observations
Pseudo R2 0.575 0.358 0.357 0.360 0.362 0.354 0.338 0.327
MSS Test 89.919 78.828 88.969 
114.600 78.828 106.931 59.715 38.325
Chow Test [K, 104.70a
N-2 K]¼
Notes: Quantile regression and OLS estimates of total observation, with standard deviation in parentheses, Significance levels are respectively 1% ( ), 5% ( )
and 10% ( ).

© 2017 The Authors. African Development Review © 2017 African Development Bank
Semi-parametric Regression-based Decomposition Methods 667

a) Recentered Influence Function (RIF): RIF regression (Fortin et al., 2011) is a convenient tool to conduct OB type
decomposition for other methods besides the mean such as quantiles regression. Therefore, when we perform quantile
model, RIF regression will be considered as a rescaled linear model. According to this definition, the rescaling factor relies
on the estimate of the interest quantile density:

q  1fy  Qq g
RIFðy; Qq Þ ¼ Qq þ ð3Þ
f y ðQq Þ

The distributional statistic can be written based on the conditional expectations of its recentered influence function: We,
then, can perform an OB decomposition using the RIF as response variable (dependent variable).
b) The General Chernozhukov-Decomposition: The difference between FY (u|u), the observed distribution functions for the
urban area and FY (r|r), the observed distribution functions for the rural areas could be written:

F Y ðujuÞ  F Y ðrjrÞ ¼ ½F Y ðujuÞ  F Y ðrjuÞ   ½F Y ðrjuÞ  F Y ðrjrÞ  ð4Þ

where FY (r|u) is the counterfactual distribution functions of expenditures that would have prevailed for the urban area had
they faced the expenditure program F Y r jX r :
Z
F Y ðrjuÞ ðyÞ :¼ F Y r jX r ðyjxÞdF X r ðxÞ ð5Þ
xu

5. Empirical Analysis
In this section, the methodology described above is applied within urban and rural subsamples and the entire population. Thus,
three quantile regressions are performed to assess the welfare gap between the two areas.

5.1 Source of Rural-Urban Welfare Inequality


When the models are estimated separately for rural and urban areas, it is possible to see the differences in the factors affecting
consumption expenditures. The rural variable was added (as a dummy variable) in the pooled model, and its coefficient is statistically
significant at a level of 1 per cent with a negative sign. Most variables of both rural and urban models are significant. We present the
result for the whole sample in Table 3. The quantile regression results for both rural and urban areas are given in Table 4 and Table 5.
The finding demonstrates that the gender variable is statistically significant in the urban model for the means regression as well
as for the quantile estimates. From Table 4, we reveal that men’s expenditures are consistently lower than those of women, but
these differences are much greater in the upper quantiles. We note that this variable is not significant in the 10th percentile.
However, for the rural model this variable is not significant in the mean and quantile regression, except in the 25th and
50th percentiles.
Results also show that the variable ‘size of the household’ is found to be significant. Ceteris paribus, with a unit change in the
household size, the level of expenditure of households declines by approximately 23.2 per cent for urban areas against
18.4 per cent for rural areas. The breakdown by age of household heads leads to interesting results in term of social policies. In
general, young people are more exposed to poverty compared to other categories, in both means and quantiles regression. When
looking at the age-class of 31–45 we observe similar results between rural and urban areas. In most quantiles, this category is not
significant. This result confirms many facts related to the economic/social environment in Tunisia. The increase in the
unemployment rate, in the rural area, mainly due to climate change (drought) associated with an absence of the labour market,
forced people to internal migration to urban areas.
In parallel, we witness the urbanization of poverty and unemployment caused by the increasing number of graduates and the
inability of the labour market to absorb them. In addition, results show that category of age 76 and above has no significant effects
or has negative significant effects of welfare in rural area compared to younger categories and compared to urban subgroup. In
fact, for the Tunisian case, the rural population is not homogeneous. There is a two-pronged migratory movement: the departure
© 2017 The Authors. African Development Review © 2017 African Development Bank
668 A. Jmaii et al.

Table 4: Quantile regression of the urban consumption expenditure per capita


Urban
Variables 5th Percentile 10th Percentile 25th Percentile 50th Percentile 75th Percentile 90th Percentile 95th Percentile

Intercept 14.055 14.245 14.507 14.865 15.433 15.795 16.032


(0.102) (0.124) (0.063) (0.068) (0.115) (0.129) (0.138)
Men –0.030 –0.032 –0.065 –0.115 –0.173 –0.220 –0.199
(0.029) (0.035) (0.022) (0.020) (0.030) (0.053) (0.066)
Household size –0.239 –0.236 –0.229 –0.236 –0.229 –0.219 0.217
(0.009) (0.005) (0.005) (0.003) (0.005) (0.009) (0.009)
% of children –0.456 –0.474 –0.509 –0.557 –0.616 –0.764 –0.764
(under 15 years) (0.068) (0.059) (0.046) (0.038) (0.048) (0.089) (0.109)
Age
[31, 45] 0.019 0.024 0.060 0.137 –0.005 0.024 0.115
(0.073) (0.099) (0.038) (0.055) (0.085) (0.062) (0.114)
[46, 60] 0.110 0.080 0.150 0.202 0.041 0.046 0.057
(0.074) (0.097) (0.038) (0.047) (0.079) (0.059) (0.121)
[61, 75] 0.244 0.223 0.294 0.390 0.234 0.291 0.243
(0.080) (0.111) (0.055) (0.052) (0.081) (0.075) (0.131)
76þ 0.217 0.186 0.263 0.345 0.210 0.265 0.311
(0.084) (0.109) (0.061) (0.057) (0.084) (0.077) (0.137)
Homeowner (Ref)
Tenant –0.080 –0.095 –0.083 –0.095 –0.046 –0.022 –0.036
(0.027) (0.015) (0.019) (0.018) (0.024) (0.052) (0.065)
Free Housing –0.109 –0.134 –0.122 –0.117 –0.076 –0.136 –0.107
(0.039) (0.044) (0.032) (0.045) (0.031) (0.047) (0.099)
Illiterate (Ref)
Primary Level 0.233 0.241 0.238 0.242 0.223 0.209 0.193
(0.024) (0.027) (0.031) (0.015) (0.022) (0.035) (0.050)
Secondary Level 0.477 0.478 0.492 0.534 0.523 0.512 0.476
(0.035) (0.030) (0.032) (0.018) (0.023) (0.037) (0.057)
Higher Level 0.888 0.891 0.927 0.888 0.908 0.912 0.948
(0.060) (0.046) (0.047) (0.028) (0.031) (0.044) (0.097)
Sector_Gov 0.155 0.140 0.117 0.128 0.100 0.130 –0.093
(0.046) (0.026) (0.018) (0.024) (0.020) (0.035) (0.063)
Sector_Priv 0.188 0.140 0.127 0.023 0.103 0.184 0.156
(0.026) (0.029) (0.022) (0.114) (0.026) (0.044) (0.066)
Sector Agri 0.030 –0.058 –0.053 –0.023 –0.022 –0.008 –0.087
(0.055) (0.044) (0.046) (0.037) (0.035) (0.061) (0.055)
Self 0.077 0.057 0.055 0.052 0.054 0.084 0.047
(0.025) (0.033) (0.024) (0.022) (0.019) (0.035) (0.053)
West –0.314 –0.296 –0.302 –0.274 –0.252 –0.244 –0.230
(0.025) (0.031) (0.018) (0.013) (0.013) (0.023) (0.046)
Foreign Transfer 0.102 0.280 0.184 0.140 0.104 0.106 –0.060
(0.150) (0.142) (0.032) (0.046) (0.052) (0.100) (0.148)
No. of observations 7,261 7,261 7,261 7,261 7,261 7,261 7,261
Pseudo r2 0.315 0.316 0.332 0.345 0.342 0.324 0.309
MSS Test 37.886 45.657 62.333 69.883 57.430 40.838 39.758
Notes: Significance levels are respectively 1% ( ), 5% ( ) and 10% ( ). Standard deviation in parentheses.

© 2017 The Authors. African Development Review © 2017 African Development Bank
Semi-parametric Regression-based Decomposition Methods 669

Table 5: Quantile regression of the rural consumption expenditure per capita


Rural
Variables 5th percentile 10th percentile 25th percentile 50th percentile 75th percentile 90th percentile 95th percentile

Intercept 13.598 13.783 14.141 14.635 15.044 15.507 15.609


(0.142) (0.106) (0.100) (0.107) (0.064) (0.114) (0.099)
Men –0.038 –0.047 –0.053 –0.124 –0.123 –0.219 –0.235
(0.057) (0.043) (0.039) (0.036) (0.029) (0.061) (0.094)
Household size –0.207 –0.210 –0.195 –0.190 –0.176 –0.177 0.173
(0.010) (0.008) (0.006) (0.007) (0.005) (0.009) (0.011)
% of children –0.709 –0.700 –0.792 –0.858 –0.921 –0.865 0.901
(under 15 years) (0.100) (0.091) (0.063) (0.065) (0.062) (0.083) (0.122)
Age
[31, 45] 0.200 0.192 0.067 0.024 –0.061 0.024 0.171
(0.140) (0.101) (0.052) (0.069) (0.053) (0.091) (0.088)
[46, 60] 0.226 0.241 0.131 0.074 –0.051 –0.007 0.095
(0.132) (0.123) (0.063) (0.079) (0.048) (0.074) (0.062)
[61, 75] 0.195 0.026 0.219 0.173 0.069 0.171 0.399
(0.124) (0.108) (0.062) (0.085) (0.055) (0.099) (0.087)
76þ 0.0004 –0.109 0.084 0.004 –0.044 0.001 0.170
(0.161) (0.127) (0.068) (0.082) (0.066) (0.110) (0.152)
Homeowner (Ref)
Tenant 0.049 –0.101 0.014 0.057 0.079 –0.207 –0.157
(0.098) (0.095) (0.134) (0.119) (0.074) (0.208) (0.194)
Free Housing –0.013 –0.063 –0.070 –0.145 –0.125 –0.228 0.339
(0.144) (0.079) (0.047) (0.038) (0.066) (0.052) (0.115)
Illiterate (Ref)
Primary Level 0.133 0.130 0.175 0.176 0.131 0.111 0.116
(0.054) (0.044) (0.018) (0.021) (0.037) (0.047) (0.060)
Secondary Level 0.329 0.302 0.354 0.367 0.366 0.396 0.410
(0.057) (0.056) (0.040) (0.035) (0.048) (0.087) (0.083)
Higher Level 0.303 0.750 0.809 0.819 0.677 0.702 0.616
(0.361) (0.142) (0.085) (0.087) (0.071) (0.108) (0.128)
Sector_Gov 0.285 0.283 0.198 0.174 0.159 0.154 –0.084
(0.123) (0.067) (0.058) (0.039) (0.047) (0.059) (0.092)
Sector_Priv 0.134 0.133 0.173 0.191 0.148 0.065 0.100
(0.102) (0.046) (0.049) (0.050) (0.052) (0.049) (0.113)
Sector Agri 0.096 0.120 0.083 0.033 0.048 –0.003 –0.013
(0.044) (0.029) (0.020) (0.026) (0.027) (0.051) (0.068)
Self 0.194 0.287 0.245 0.216 0.223 0.209 0.121
(0.099) (0.074) (0.050) (0.053) (0.039) (0.071) (0.071)
West –0.297 –0.270 –0.266 –0.235 –0.222 –0.198 –0.178
(0.028) (0.034) (0.030) (0.020) (0.024) (0.030) (0.047)
Foreign Transfer 0.084 0.080 0.178 0.241 0.549 0.549 –0.697
(0.268) (0.194) (0.074) (0.125) (0.184) (0.172) (0.172)
No. of observation 4020 4020 4020 4020 4020 4020 4020
Pseudo R2 0.272 0.281 0.296 0.291 0.279 0.269 0.266
MSS Test 45.395 51.526 54.473 66.210 57.655 45.419 31.098
Notes: Significance levels are respectively 1% ( ), 5% ( ) and 10% ( ). Standard deviation in parentheses.

© 2017 The Authors. African Development Review © 2017 African Development Bank
670 A. Jmaii et al.

of skilled young people and the arrival of elderly close to retirement (Jelili and Mzali, 1998). This phenomenon of older
age groups having considerable within-cohort inequality is an important acknowledged fact and can be clearly observed in the
Tunisian data. These results confirm the urgency of the issue of rural old-age vulnerability in Tunisia.
Results also show that the agricultural sector has a positively significant coefficient in rural area at all quantiles excluding the
25th and the 90th. For both urban and rural areas, welfare is determined by governmental and private sectors, since they are
statistically significant with a positive sign. In the urban model, the effect of these variables is higher in the upper quantile. In
the rural regression, they are much higher at the bottom of the distribution. As indicated in Tables 4 and 5, returns to education are
statistically significant across all quantiles. The findings show that, for the entire distribution, consumption expenditures for
people with higher education levels are higher than illiterate people and those who have just acquired primary education and
secondary education level. In addition, results show that the west region is more vulnerable and has a lower welfare than East
region. In fact, employments are paying off in the East relatively to the West and probably for the same reasons: the East gets the
lion’s share of the market’s activity in Tunisia (manufacturing and services).

5.2 Counterfactual Decomposition Results


When we estimate the rural model with urban characteristics (counterfactual function), we found a graph very close to the urban
one (Figure 3). This means that the gap will be reduced, but still exist even if it is small, once we attribute the same characteristics
in both urban and rural areas.
In accordance with the results of Figure 3, the overall poverty gap between poor rural and poor urban households may be
explained by the constant term (Table 6) of the coefficient effect (14.8 per cent). This means that, even with urban characteristics,
poor rural households will still suffer from a poverty gap of around 9.36 per cent, compared to urban households. This is due to
the geography effect, which assigns persistent spatial differences in living standards to difference in geographic characteristics
such as roads/transport, water, and so on.
Results reveal that differences in the type of employment and education explain respectively 4.58 per cent and 18.32 per cent
of the poverty gap between rural and urban areas for the poorest households (Table 7). Similarly, if the conditions of the labour
market in rural areas were similar to those in urban areas, the poverty gap could be reduced by three points. Results of
Chernozhukov decomposition revealed that both covariate and returns effects are larger within higher quantiles, resulting in a
wider urban–rural gap. This means that the principal reasons of the difference between the wealthiest rural households and
wealthiest urban households were twofold: due to characteristic effects such as education, employment and household size; and
due to the return effect of these characteristics. Apart from this result, we observe a dominance of characteristics effects at lower
quantile, which attest that differences in characteristics between urban poor households and rural poor households are more
important than differences in returns to those characteristics (Table 7). This may reflect that the poor earn a wage slightly above
the minimum level, therefore urban–rural change in market yields does not matter at the tail of the distribution.

Figure 3: Densities of urban and rural expenditures

© 2017 The Authors. African Development Review © 2017 African Development Bank
Semi-parametric Regression-based Decomposition Methods 671

Table 6: Urban rural gap (RIF-regression)


10th percentile 50th percentile 90th percentile
Estimated std Estimated std Estimated std

Reference group: Urban Coef


Estimated log expenditure gap:
E[RIFq (ln(Expu))]- –0.633 (0.023) –0.564 (0.018) –0.614 (0.025)
E[RIFq(ln(Expr))]
Composition effects attributable to
Age, gender, household size, foreign transfer and logement –0.069 (0.021) –0.071 (0.015) –0.075 (0.021)
West region –0.070 (0.011) –0.077 (0.007) –0.056 (0.011)
Education –0.116 (0.020) –0.117 (0.013) –0.156 (0.020)
Employment –0.029 (0.013) –0.034 (0.009) –0.035 (0.014)
Aggregated characteristics effects –0.298 (0.030) –0.315 (0.021) –0.337 (0.025)
Regional structure effects attributable to
Age, gender, household size, foreign transfer and logement 0.494 (0.319) –0.079 (0.224) –0.036 (0.353)
West region 0.031 (0.013) 0.003 (0.009) 0.017 (0.014)
Education –0.116 (0.020) –0.084 (0.033) –0.148 (0.051)
Employment 0.075 (0.031) 0.041 (0.021) 0.040 (0.033)
Constant –0.094 (0.330) –0.073 (0.232) –0.117 (0.366)
Aggregated coefficient effect –0.335 (0.036) –0.249 (0.024) –0.277 (0.036)
Notes: Standard errors are in parentheses and ( significant at 1%,  significant at 5%, significant at 10%).
Source: Author’s computing from 2010 National Survey on Households’ Budget, Consumption and Standard of Living.

Table 7: Urban rural gap (Chernozhukov method)


Characteristics
Raw gap effects Coefficients effects Residuals effects
  
5th Percentile 0.601823 0.417113 0.151049 0.033661
  
10th Percentile 0.595731 0.396819 0.181029 0.017884
  
25th Percentile 0.581305 0.364286 0.221906 –0.004888
   
50th Percentile 0.571688 0.336543 0.259256 –0.024111
   
75th Percentile 0.585155 0.319868 0.291642 –0.026355
  
90th Percentile 0.619629 0.320544 0.316675 –0.01759
  
95th Percentile 0.647009 0.326744 0.326897 –0.006632

Coefficient is different from 0 at 95%.
Source: Author’s computing from 2010 National Survey on Households’ Budget, Consumption and Standard of Living.

6. Conclusion and Recommendations


The objective of this study was to dissect regional inequalities in Tunisia. For this purpose, we used semi-parametric regression-
based decomposition methods to assess the source of welfare gap between urban and rural areas. Results of the study are in
agreement with the type of policies that have been established in Tunisia. For several decades, the government has implemented
reforms that promote education in the rural area, especially the west region. Nevertheless, it did not take into consideration the
quality of the education programme; as a result, educated youth in these areas do not have the capacity to succeed in their national
arena and are unable to compete with other graduates in the private market. Indeed, the lack of a good educational level may limit
the opportunities for these individuals to find a decent job. This may explain the fact that the gap between the two areas is still
wide. The issue of equitable access to education, especially to the higher level, is probably a real issue that must be addressed as
part of an overall review. On the other hand, the breakdown by age of household heads has confirmed the urgency of the issue of
rural old-age vulnerability in Tunisia. Effective policies must take into consideration the heterogeneous nature of the elderly
group, associating both personal characteristics (sex, education and so on) and the varying contexts in which they live.

© 2017 The Authors. African Development Review © 2017 African Development Bank
672 A. Jmaii et al.

The fight against regional inequalities is not just a matter of equality, but also a question of economic development fairness. We
should know that the development of social groups and deprived regions inevitably involves a strategy to reduce regional
inequality. In addition, the Tunisian state is not only required to build universities and hospitals (etc.), but also to provide the needed
resources to get the rural areas to the level of the urban ones. We recommend a positive discrimination programme in favour of the
marginalized region. For example, the government can start by sending most competent teachers and doctors to work in these areas,
and provide the necessary resources (good infrastructure, access to quality schools, internet, software and so on).
By decomposing the gap between urban and rural areas, this study has shown that the problem is not only about equality but it
is an equity issue. Such equity implies that resources should be allocated equally between areas with regard to the quality.
Although, for a social phenomenon, such as inequality, a dynamic analysis is required. This type of inequality is often known as
income variation/mobility and poverty dynamics and is important, when viewed from the social welfare and political
perspectives, to understand policy changes, an issue left for future research.

Notes
1. Using both AIC (Akaike Information Criterion) and BIC (Bayesian Information Criterion), the most adequately model is the
one that we have chosen.
2. Machado et al. (2013) demonstrated that the qreg2 command on Stata, which is a wrapper for the old qreg command, enables
us to obtain consistent estimators and report robust standard errors.

References
Adams, R. and J. Page (2003), ‘Poverty, Inequality and Growth in Selected Middle East and North Africa Countries’, World
Development, Vol. 31, pp. 2027–48.
African Development Bank (2013), Annual Report, Tunis. Available at: https://www.afdb.org/fileadmin/uploads/afdb/
Documents/Publications/Annual_Report_2013.pdf
Becker, G. S. (1992), ‘Human Capital and the Economy’, American Philosophical Society, Vol. 136, No. 1, pp. 85–92.
Bibi, S. and M. K. Nabli (2009), ‘Income Inequality in the Arab Region: Data and Measurement Patterns and Trends’, Middle
East Development Journal, Vol. 1, No. 2, pp. 275–314.
Blinder, A. S. (1973), ‘Wage Discrimination: Reduced Form and Structural Estimates’, Journal of Human Resources, Vol. 8,
No. 4, pp. 436–55.
Bourguignon, F. (2003), ‘The Growth Elasticity of Poverty Reduction: Explaining Heterogeneity across Countries and Time
Periods’ in T. S. Eicher and S. J. Turnovsky (eds.), Inequality and Growth: Theory and Policy Implications, MIT Press,
Cambridge, MA, pp. 1–26.
Boutayeb, A. and U. Helmert (2011), ‘Social Inequalities, Regional Disparities and Health Inequity in North African Countries’,
International Journal for Equity in Health, Vol. 10, No. 23, pp. 1–9.
Chambers, D. and A. Krause (2010), ‘Is the Relationship between Inequality and Growth Affected by Physical and Human
Capital Accumulation?’, Journal of Economic Inequality, Vol. 8, pp. 153–72.
Chang, H. H. (2012), ‘Consumption Inequality between Farm and Nonfarm Households in Taiwan: A Decomposition Analysis
of Differences in Distribution’, Agricultural Economics, Vol. 43, pp. 487-98.
Chernozhukov, V., I. Fernandez-Val and B. Melly (2013), ‘Inference on Counterfactual Distributions’, Econometrica, Vol. 81,
No. 6, pp. 2205–68.
Deininger, K. and L. Squire (1996), ‘A New Data Set Measuring Income Inequality’, World Bank Economic Review, Vol. 10,
No. 3, pp. 565–91.
Duclos, J.-Y., and V. Chouchane (2011), ‘Analyzing Pro-Poor Growth in Southern Africa: Lessons from Mauritius and South
Africa’, African Development Review, Vol. 23, No. 2, pp. 121–46.

© 2017 The Authors. African Development Review © 2017 African Development Bank
Semi-parametric Regression-based Decomposition Methods 673

Fortin, N., T. Lemieux and S. Firpo (2011), ‘Decomposition Methods in Economics’ in O. Ashfenfelter and D. Card (eds.),
Handbook of Labor Economics, Elsevier, Amsterdam, Vol. 4A, pp. 1–122.
Hassine, N. B. (2015), ‘Economic Inequality in the Arab Region’, World Development, Vol. 66, pp. 532–56.
Jelili, R. B., and H. Mzali (1998), ‘Rural-urban Migration and Self-selection in Tunisia’, Papers in Regional Science, Vol. 77,
No. 4, pp. 347–60.
Koenker, R. and G. Bassett (1978), ‘Regression Quantiles’, Econometrica, Vol. 46, No. 1, pp. 33–50.
Krafft, C. and R. Assaad (2016), ‘Inequality of Opportunity in the Labor Market for Higher Education Graduates in Egypt and
Jordan’, in I. Diwan and A. Galal (eds), International Economic Association Series, Palgrave Macmillan, London.
Machado, J. A. F. and J. Mata (2005), ‘Counterfactual Decomposition of Changes in Wage Distributions Using Quantile
Regression’, Journal of Applied Econometrics, Vol. 20, No. 4, pp. 445–65.
Machado, J. A. F., P. M. D. C. Parente and J. M. C. Santos Silva (2013), ‘qreg2: Stata Module to Perform Quantile Regression
with Robust and Clustered Standard Errors’, Statistical Software Components S457369, Boston College Department of
Economics.
Mansour, W. (2012), ‘The Patterns and Determinants of Household Welfare Growth in Jordan 2002–2010’, World Bank Policy
Research Working Paper 6249.
Melly, B. (2005), ‘Decomposition of Difference in Distribution using Quantile Regression’, Labour Economics, Vol. 12, No. 4,
pp. 577–90.
Miller, P. W. (1987), ‘The Wage Effect of the Occupational Segregation of Women in Britain’, Economic Journal, Vol. 97,
No. 388, pp. 885–96.
Mussa, R., (2014), ‘Household Expenditure Components and the Poverty and Inequality Relationship in Malawi’, African
Development Review, Vol. 26, No. 1, pp. 138–47.
Ncube, M., J. C. Anyanwu and K. Hausken (2014), ‘Inequality, Economic Growth and Poverty in the Middle East and North
Africa (MENA)’, African Development Review, Vol. 26, No. 3, pp. 435–53.
Neuman, S. and R. Oaxaca (2004), Wage Decompositions with Selectivity-Corrected Wage Equations: A Methodological Note,
Journal of Economic Inequality, Vol. 2, No. 1, pp. 3–10.
Nguyen, B. T., J. W. Albrecht, S. B. Vroman and M. D. Westbrook (2007), ‘A Quantile Regression Decomposition of Urban-
Rural Inequality in Vietnam’, Journal of Development Economics, Vol. 83, No. 2, pp. 466–90.
Oaxaca, R. (1973), ‘Male Female Wage Differentials in Urban Labour Markets’, International Economic Review, Vol. 14,
pp. 693–709.
Odedokun, M. O., and J. I. Round (2004), ‘Determinants of Income Inequality and its Effects on Economic Growth: Evidence
from African Countries’, African Development Review, Vol. 16, No. 2, pp. 287–327.
Parente, P. M. and J. Santos Silva (2016), ‘Quantile Regression with Clustered Data’, Journal of Econometric Methods, Vol. 5,
No. 1, pp. 1–15.
Pieters, J. (2011), ‘Education and Household Inequality Change: A Decomposition Analysis for India’, Journal of Development
Studies, Vol. 47, No. 12, pp. 1909-24.
Salehi-Isfahani, D., N. Belhaj Hassine, and R. Assaad (2014). Equality of Opportunity in Educational Achievement in the
Middle East and North Africa. Journal of Economic Inequality, Vol. 12, pp. 489–515.
Skoufias, E. and R. S. Katayam (2011), ‘Sources of Welfare Disparities between and within Regions of Brazil: Evidence from
the 2002–2003 Household Budget Survey (POF)’, Journal of Economic Geography, Vol. 11, pp. 897–918.
Son, H. H., and N. Kakwani (2008), ‘Global Estimates of Pro-poor Growth’, World Development, Vol. 36, pp. 1048–66.

© 2017 The Authors. African Development Review © 2017 African Development Bank
African Development Review, Vol. 29, No. 4, 2017, 674–688

Analyse des effets de la qualite des institutions sur la volatilite des


Investissements Directs Etrangers en Afrique

Bruno Emmanuel Ongo Nkoa and Jacques Simon Song

Resume : Cet article analyse les effets de la qualite des institutions sur la volatilite des Investissements Directs Etrangers
(IDE) en Afrique. Nous utilisons un modele estime en donnees de panel par les methodes des moments generalises en systeme et
des doubles moindres carres sur la periode 1996–2014. Les resultats montrent que la qualite des institutions accroit la volatilite
des entrees d’IDE. L’Afrique en general et les sous-regions Afrique Centrale et Afrique Australe en particulier doivent reformer
leurs institutions.

Abstract: This article analyzes the influence of institutional quality on the volatility of foreign direct investment (FDI) in
Africa. We use an estimated model in dynamic panel data using the generalized method moment in system (GMM-S) and double
least squares (2SLS) over the period 1996–2014. Results show that the quality of institutions increases the volatility of FDI
inflows. In order to stabilize FDI inflows and set development prospects the African authorities must reform their institutions.

1. Introduction
Au cours de la derniere decennie, l’Afrique a enregistre des entrees importantes d’IDE accompagnees d’un volume de plus en
plus croissant des investissements intra africains (Dunne et Masiyandima, 2017). En 1990, les entrees d’IDE en Afrique etaient
evaluees a 2,84 milliards USD. Elles ont atteint 54,08 milliards USD (CNUCED, 2016). Mais cette evolution est caracterisee par
une forte volatilite. En 2000, le continent attire 9,65 milliards USD et en 2015, les IDE se chiffrent a 54,07 milliards USD en 2015
(CNUCED, 2016). En general, la volatilite des IDE suit l’evolution des cours des matieres premieres, notamment le petrole.
Pourtant, l’accroissement recent du prix du baril de petrole contraste avec l’evolution non reguliere des IDE. En decembre 1995,
le baril de petrole se vendait a 14,9 dollars USD, contre 32,62 dollars USD en decembre 2004, et 38,01 dollars en decembre 2015
(INSEE, 2016).
La volatilite des IDE est analysee dans la litterature sous deux angles: ses determinants et son r^ole sur la croissance
economique. S’agissant de l’impact de la volatilite des IDE sur la croissance, les travaux aboutissent a un resultat negatif et
significatif (Eregha, 2015; Lensink et Morrissey, 2006). La litterature sur ses determinants souligne le r^ole des facteurs
institutionnels (Vadlamannati et al., 2009), avec des resultats mitiges (Sane, 2016; Anyanwu et Yameogo, 2015a; Avom et
Ongo, 2013). La volatilite des IDE en Afrique a toujours ete expliquee a travers les variations des cours des matieres premieres.
Cependant, les travaux de Alley et Poloamina (2015) et Broto et al. (2011), ont justifie la volatilite des IDE par des facteurs
politiques.
Une attention portee sur les indicateurs de gouvernance en Afrique entre 1996 et 2015 montre trois groupes selon le
Worldwide Governance Indicators (WGI, 2016). Le premier groupe compose de bons eleves avec un indice moyen fluctuant
entre 0,74 et 0,22 (Ile Maurice et Seychelles). Le deuxieme groupe compose d’eleves moyens avec un indice proche de 0 (Ghana,
Tunisie, Maroc). Le troisieme groupe est constitue de mauvais eleves avec un indice superieur a 1 ou a 2 (Somalie, RDC,


Bruno Emmanuel Ongo Nkoa (auteur correspondant), Departement d’Economie et de Gestion, Faculte des Sciences sociales et de Gestion, Universite
de Buea-Cameroun, BP: 63;Tel: 00237 675 19 40 49, E-mail: ongoema@yahoo.fr, Laboratoire d’Analyse et de Recherche en Economie Appliquee de
l’Universite de Yaounde 2 Soa. Jacques Simon Song, Departement d’Economie publique, Faculte des Sciences Economiques et de Gestion, Universite
de Dschang-Cameroun, Tel: 00237 696 94 58 15, E-mail: jacquessimonsong@gmail.com, Laboratoire d’Analyse et de Recherche en Economie
Appliquee de l’Universite de Yaounde 2 Soa. Les auteurs remercient les trois rapporteurs anonymes de la revue pour leurs critiques et suggestions
pertinentes qui ont permis d’ameliorer substantiellement la premiere version de cet article.
© 2017 The Authors. African Development Review © 2017 African Development Bank. Published by Blackwell Publishing Ltd,
9600 Garsington Road, Oxford OX4 2DQ, UK and 350 Main Street, Malden, MA 02148, USA. 674
Analyse des effets de la qualite des institutions 675

Soudan). Selon une analyse regionale, l’Afrique Australe (0,11) et l’Afrique du Nord (0,57) sont bien classees. L’Afrique de l’Est
(1,03) et l’Afrique Centrale (1,05) sont considerees comme de mauvais eleves. En Afrique Centrale, les IDE ont grandement
fluctue. Ces constats induisent des questionnements sur la volatilite des IDE, et confirment le r^ole des institutions en Afrique.
La prise en compte de la qualite des institutions dans l’analyse de la volatilite des IDE en Afrique est benefique a deux titres.
D’une part, elle permet de se demarquer de l’assertion generale, qui justifie les ressources naturelles comme seul facteur de
volatilite des IDE. D’autre part, elle contribue a enrichir la litterature empirique sur les fluctuations des IDE en Afrique, puis que
l’etude considere deux mesures de la volatilite des IDE (l’ecart type des fluctuations et le filtre de Prescott). Nos resultats
econometriques montrent que la qualite des institutions est responsable de la forte volatilite des IDE en Afrique.
L’article est organise de la maniere suivante apres cette introduction. La deuxieme section presente quelques faits stylises des
IDE et des institutions en Afrique. La troisieme section fait une revue selective de la litterature. La quatrieme section esquisse la
methodologie. La cinquieme section presente et analyse les resultats. La sixieme section conclut et formule quelques
recommandations de politique economique.

2. Les faits stylises des IDE et de la qualite des institutions en Afrique

2.1 Les entrees d’IDE en Afrique restent volatiles et different d’une region 
a une autre

Entre 2002 et 2014, le taux de croissance des entrees d’IDE en Afrique est de 267 %. En 2016, l’Afrique est non seulement une
destination privilegiee d’IDE, mais egalement un fournisseur avec des flux sortants evalues a 13 milliards USD (CNUCED,
2016). Malgre les baisses de 2010 et 2011, les flux entrants d’IDE sont repartis a la hausse en 2014 atteignant 58,3 milliards USD.
Les entrees d’IDE different d’une sous-region a une autre (Graphique 1). Entre 1996-2014, avec une moyenne de 10,26
milliards USD, l’Afrique du Nord est la sous-region la plus attractive (Anyanwu et Yameogo, 2015b). L’Afrique de l’Ouest se
classe au second rang avec une moyenne de 7,76 milliards USD. L’Afrique de l’Est et l’Afrique australe viennent
successivement avec 5,38 milliards USD et 4,83 milliards USD. L’Afrique centrale se classe en derniere position avec une
moyenne d’attractivite des IDE de 3,43 milliards USD. En depit de l’effet de la forte production de produits de base et de la
demande d’exportations, la sous-region fait face aux effets de la guerre civile et du terrorisme.

2.2 Le cadre institutionnel en Afrique est peu reluisant


L’attractivite des IDE en Afrique s’est realisee dans un cadre institutionnel peu performant. En effet, avec un score moyen de
0,11, la sous-region Afrique Australe tire vers le haut l’indice de gouvernance. Parmi les pays les mieux classes, six se
demarquent (Maurice, Botswana, Cap-Vert, Afrique du Sud, Namibie, Seychelles). La faible corruption, le cadre juridique, et la
stabilite politique y ont contribue avec un climat favorable aux initiatives privees. Au rang des pays qualifies de mauvais eleves
figurent la RDC, le Burundi, l’Angola, le Tchad, la Somalie entre autres (Tableau 1).

Graphique 1: Evolution des flux entrants d’IDE en Afrique et dans les sous-regions entre 1996 et 2014
en millions USD

Source: CNUCED (2016) et calculs des auteurs.


© 2017 The Authors. African Development Review © 2017 African Development Bank
676 B. E. Ongo Nkoa et J. S. Song

Tableau 1: Classification de quelques pays africains en fonction de la qualite institutionnelle


Les bons eleves Indices moyens Les eleves moyens Indices moyens Les mauvais eleves Indices moyens

Maurice 0,74 Ghana –0,06 Somalie –2,23


Botswana 0,69 Tunisie –0,12 RDC –1,76
Cap-Vert 0,42 Maroc –0,22 Soudan –1,54
Afrique du Sud 0,33 Egypte –0,50 Burundi –1,33
Namibie 0,31 Angola –1,27
Seychelles 0,22 Tchad –1,20
Source: Auteurs, a partir de WGI (2016).

La periode de 1996–2014 est marquee par une deterioration de la qualite institutionnelle en Afrique. Le score est passe de
0,688 en 1996 pour s’etablir a 0,725 en 2014. De 1996 a 2004, l’indice moyen de gouvernance est passe de 0,688 a 0,659.
Cette periode est caracterisee par la creation des institutions africaines avec l’aide des institutions internationales visant a
ameliorer la qualite de la gouvernance dans les pays. Alors que l’Afrique du Nord progresse dans le classement, l’Afrique
Centrale reste moins bien classee.
En general, il ressort des observations que, l’evolution des institutions ne suit pas toujours la dynamique des entrees des IDE.
L’Afrique Centrale fait preuve d’une forte amelioration de la qualite des institutions passant de 1,164 a 0,982 et d’une forte
variation des IDE. Les bonnes performances institutionnelles en Afrique Australe et en Afrique du Nord contraste avec les
fluctuations des entrees d’IDE. L’Afrique de l’Est et l’Afrique de l’Ouest ont connu des ameliorations bien que relative de leur
qualite des institutions dont les niveaux sont passes respectivement de 1,052 a 1,046 puis de 0,684 a 0,627 entre
1996–2014. Ces contrastes sont aussi expliques dans la litterature.

3. Qualite des institutions et investissements directs etrangers: les enseignements de la


litterature
La litterature sur les IDE s’illustre par trois theories. D’abord, les theories statiques qui englobent les theories du commerce
international, de la concurrence monopolistique, du cycle de vie du produit, de l’internationalisation et du paradigme eclectique
(Dunning, 1980). Ensuite, les theories dynamiques qui englobent la theorie de la multinationalisation et de la nouvelle economie
geographique (Krugman, 1991). Enfin, les theories institutionnelles qui s’illustrent par deux effets. Un effet direct base sur
le postulat qu’un pays h^ote, caracterise par de bonnes institutions, assure la reduction des co^uts de transaction et des discussions
sans tyrannie (North, 1990). Un effet indirect qui explique les IDE par la capacite du pays d’accueil d’eriger des institutions qui
structurent les echanges, assurent la protection des droits de propriete, stabilisent l’environnement economique et stimulent
l’activite entrepreneuriale (Estrin et al., 2013).
Un manque de consensus subsiste quant aux effets des institutions sur l’attractivite des IDE. Wei (2000) et Bokpin (2017)
trouvent que de bonnes institutions favorisent l’attractivite des IDE. Wisniewski et Pathan (2014) concluent que la qualite de la
reglementation accroit l’attractivite des IDE. Aussi, Anyanwu et Yameogo (2015a et 2015b) montrent que la democratie a des
effets positifs sur les flux entrants des IDE en Afrique de l’Ouest. Ce resultat est egalement verifie par Gankou et al. (2016),
Okafor (2015) et Adams et Opoku (2015).
Par contre, Ali et al. (2010) trouvent un effet negatif et significatif des institutions sur les IDE. En effet, la faible qualite
institutionnelle, non seulement augmente le risque de perte des investisseurs mais aussi reduit la credibilite des pays.
Toutefois, la litterature decrit le r^ole des variables institutionnelles sur l’attractivite des IDE. Elle semble occulter leur impact
sur leur volatilite. Pournarakis et Varsakelis (2002) trouvent sur un echantillon de 10 pays d’Europe entre 1997–2000, qu’un
environnement institutionnel credible reduit la volatilite des IDE. Leur resultat est confirme par Buchanan et al. (2012).
Vadlamannati et al. (2009) sur un echantillon de 17 pays d’Asie du Sud-Est entre 1996–2005 montrent que la mauvaise qualite
des institutions accroit la volatilite des IDE. Aussi, Waqas et al. (2015) montrent qu’en plus des variables institutionnelles, un
taux d’inter^et eleve, une depreciation forte de la monnaie, une faible inflation, et un taux de croissance eleve accroissent la
volatilite des entrees d’IDE.
Bien qu’etant des contributions importantes, les etudes passees en revue presentent deux limites principales. D’une part, elles
n’integrent pas tous les indicateurs de gouvernance. Pourtant, chaque indicateur, selon son degre, influe sur les IDE. D’autre part,
© 2017 The Authors. African Development Review © 2017 African Development Bank
Analyse des effets de la qualite des institutions 677

les etudes qui traitent de la volatilite des IDE ne couvrent pas tous les pays Africains, et encore moins ne mettent pas en exergue
les specificites sous regionales. Notre article apporte trois contributions majeures. Premierement, nous traitons de la volatilite
des IDE dans une perspective institutionnelle en prenant en compte les deux principales mesures (Ecart type et filtre de Prescott).
Deuxiemement, nous utilisons les donnees de panel estimees par la methode GMM-Systeme tout en prenant en compte les
problemes d’endogeneite (Sevestre, 2002). Troisiemement, nous etablissons une estimation econometrique par la technique des
doubles moindres carres en considerant des donnees sous regionales.

4. La strategie methodologique

4.1 Le modele
Pour determiner les effets de la qualite des institutions sur la volatilite des IDE en Afrique, nous utilisons un modele empirique
inspire de Mijiyawa (2015), et Asamoah et al. (2016) pour les variables de contr^ole et d’inter^et, de Buchanan et al. (2012) et
Komlan (2016) pour la mesure de la volatilite des IDE. La version compacte du modele est donnee par l’equation (1):

V olIDEit ¼ a0 þ mV olIDEit1 þ bI it þ gX it þ eit ð1Þ

VolIDE capture la volatilite des flux nets d’IDE en pourcentage du PIB. Suivant Broto et al. (2011) et Buchanan et al. (2012),
cette volatilite est mesuree par la variance des flux entrants d’IDE selon la formule:
XT 2
ðIDEit  IDEÞ
V olIDE ¼ t
n
 est la moyenne arithmetique simple des IDE du pays i sur l’ensemble de la periode T. VolIDEit1 la volatilite des entrees
IDE
d’IDE de l’annee precedente. Selon Asiedu et Lien (2011), l’IDE retarde montre un effet de retour positif de l’IDE passe sur
l’IDE actuel. Les investisseurs preferent operer dans un environnement familier. L’IDE passe attire de nouvelles firmes.
Le choix des variables institutionnelles repose sur la disponibilite des donnees (Benassy-Quere et al., 2007). Buchanan et al.
(2012) font usage d’un indice composite. Bien qu’il facilite l’estimation et l’interpretation du resultat, il n’indique pas le r^ole
individuel de chaque indice. La consideration dchaque composante institutionnelle est necessaire, car les progres realises par les
pays dferent d’un indicateur a l’autre (Asamoah et al., 2016). I est une matrice des indicateurs de gouvernance constituee de:
 la stabilite politique qui mesure la perception de la probabilite d’une destabilisation par des moyens non constitutionnels ou
violents;
 la qualite de la reglementation mesure la capacite des pouvoirs publics a elaborer et appliquer de bonnes politiques
favorables au developpement du secteur prive;
 l’etat de droit mesure le degre de confiance qu’ont les citoyens et la maniere dont ils s’y conforment au respect des contrats,
des competences de la police et des tribunaux, la perception de la criminalite, et de la violence;
 la qualite du service public mesure les performances de la fonction publique et son niveau d’independance vis-a-vis des
pressions politiques, la qualite de l’elaboration et de l’application des politiques, la credibilite de l’engagement des pouvoirs
publics a l’egard de ses politiques;
 la liberte d’expression qui mesure la liberte des citoyens face au pouvoir dont les notes des indicateurs reposent sur trois
criteres: pays non libre, partiellement libre ou libre;
 la corruption mesure l’utilisation des pouvoirs publics a des fins d’enrichissement personnel, y compris la grande et la petite
corruption, ainsi que « la prise en otage » de l’Etat  par les elites et les inter^ets prives.

X est la matrice de variables de contr^ole constituee de la Formation Brute du Capital Fixe (FBCF) approximee par
l’accumulation du capital physique (Barrios et al., 2005). Le capital humain est approxime par le taux d’inscription au
secondaire. Il facilite l’absorption de nouvelles technologies, et stimule l’attractivite des IDE (Cleeve et al., 2015). Les
infrastructures sont approximees par le nombre de lignes telephoniques pour 100 habitants (Anyanwu, 2012). L’aide publique au
developpement influence grandement les entrees d’IDE (Yasin, 2005). Le logarithme du PIB par habitant decrit la taille du
© 2017 The Authors. African Development Review © 2017 African Development Bank
678 B. E. Ongo Nkoa et J. S. Song

marche et de la production nationale (Avom et Ongo, 2013). Les ressources naturelles sont approximees par la rente des produits
petroliers, gaziers et mineraux en pourcentage du PIB (Dupasquier et Osakwe, 2006). eit est le terme d’erreur compose de ni qui
capte les effets specifiques pays non observes et de mt qui prend en compte l’effet specifique temporel commun a tous les pays; i
designe les pays de l’echantillon, t represente l’annee, a et b sont les parametres a estimer.

4.2 La technique d’estimation: la Methode des Moments Generalises


Afin de tirer avantage des donnees de panel, nous recourons a la Methode des Moments Generalises (GMM) dont l’estimateur en
difference a ete introduit par Arellano et Bond (1991) et Arellano et Bover (1995). Par la suite, Blundell et Bond (1998) l’ont
perfectionne en developpant l’estimateur des moments generalises en Systeme qui permet de corriger l’endogeneite de la
variable expliquee. Si la Methode des Moments Generalises en systeme semble en theorie plus efficiente que la Methode des
Moments Generalises en difference, elle utilise en revanche plus d’instruments que cette derniere, ce qui la rend inappropriee
lorsque la dimension individuelle est petite. Les mesures des variables institutionnelles utilisees etant objectives, nous ne
pouvons pas exclure les risques d’erreurs de mesure. Les caracteristiques desirables de la methode GMM en systeme permettent
de s’attaquer aux problemes de muticolinearite, d’endogeneite et de biais de variables omises.
L’endogeneite des variables et la double causalite sont en regle generale les deux problemes dans l’estimateur GMM
(Malikane et Chitambara, 2017; Ajide et Osode, 2017). Pour faire face a ces difficultes, nous adoptons la methode des moments
generalises avec variables instrumentales. En effet, Arellano et Bover (1995) ont construit des estimateurs sans biais,
convergents et asymptotiquement distribues. Ils montrent que le probleme d’endogeneite vient soit d’une forte relation entre la
variable dependante et certaines variables independantes; soit d’une multicolinearite entre les variables explicatives. Blundell et
Bond (1998) ont teste cette methode a l’aide des simulations de Monte Carlo. Ces auteurs trouvent que l’estimateur GMM en
systeme est plus performant que celui en difference qui n’exploite que les conditions des moments de l’equation en difference
premiere avec comme instruments des variables retardees en niveau. La validite des instruments est faite par le test de Sargan et
de Hansen (Roodman, 2009).

4.3 Les donnees


Les donnees sont issues de trois bases. Premierement, les variables macroeconomiques sont tirees de la base de la Banque
Mondiale dans le World Development Indicators (WDI, 2016). Deuxiemement, les variables institutionnelles sont issues de la
base de Kaufmann et al. (2010) dans le Worldwide Governance Indicators (WGI, 2016). La variable regime politique est prise
dans la base de Polity IV. L’etude porte sur 53 pays de l’Afrique. La liste complete est presentee dans le tableau 2. Il decrit aussi
l’appartenance geographique de chaque pays.
La periode d’etude va de 1996 a 2014. Les statistiques descriptives et les matrices de correlation des variables sont contenues
dans les tableaux 3 et 4. Les correlations entre les variables d’inter^et ne sont pas elevees pour causer de serieux problemes de
muticolinearite. La volatilite des IDE est positivement correlee aux variables institutionnelles.

Tableau 2: Liste des pays de l’echantillon par sous-region


Afrique centrale Afrique de l’Ouest Afrique australe Afrique du Nord Afrique de l’Est

Angola, Burundi Benin, Burkina-Faso Botswana, Lesotho Algerie Comores


Cameroun, Gabon Cap-Vert, Cote d’Ivoire Madagascar Egypte Djibouti
Guinee Equatoriale Gambie, Ghana Malawi, Ile Maurice Lybie Erythree
RCA, RDC, Congo Guinee, Guinee Bissau Mozambique Maroc Ethiopie
Rwanda Liberia, Mali Namibie, Seychelles Tunisie Kenya
Sao-Tome et Principe Mauritanie, Niger Seychelles, Swaziland Somalie, Soudan
Tchad Nigeria, Senegal Afrique du Sud Tanzanie
Sierra-Leone, Togo Zambie, Zimbabwe Ouganda
Source: Les auteurs.

© 2017 The Authors. African Development Review © 2017 African Development Bank
Analyse des effets de la qualite des institutions 679

Tableau 3: Statistiques descriptives


Variables Observations Moyennes Ecart-type Minimum Maximum

Volatilite des IDE (ecart-type) 269 4,113 9,469 –48,605 95,983


Formation Brute du Capital Fixe 978 5,028 9,118 –62,076 149,973
Capital Humain 562 42,703 25,666 5,165 121,612
Infrastructures 1001 29,837 38,915 0 214,750
Aide Publique au Developpement 958 9,795 12,136 –0,260 181,103
Ressources naturelles 209 782,15 763,78 0 2534,75
Polity 2 967 1,006 5,291 –9 10
Taille du marche 1006 9,164 8,574 –2,629 40,821
Corruption 1007 –0,606 0,606 –2,057 1,250
Etat de droit 1007 –0,711 0,675 –2,669 1,057
Qualite de la regulation 1007 –0,700 0,648 –2,665 1,001
Qualite du service public 1007 –0,720 0,635 –2,480 1,131
Stabilite politique 1007 –0,571 0,948 –3,324 1,192
Liberte d’expression 1007 –0,686 0,734 –2,234 1,025
Source: Calcul des auteurs.

5. Estimations et discussion des resultats

5.1 Qualite des institutions et volatilite des entrees des IDE en Afrique
Le tableau 5 contient les resultats de l’estimation par les GMM en systeme. La volatilite des IDE en Afrique est prise comme
variable dependante dans les deux modeles. Dans le modele 1, la volatilite des IDE est estimee par l’ecart-type, alors que dans le
modele 2, elle est estimee par le filtre de Prescott.
Les resultats du modele 1 montrent que les coefficients associes aux variables institutionnelles regime politique, etat de droit,
stabilite politique et liberte d’expression ont des effets positifs et statistiquement significatifs sur la volatilite des IDE en Afrique.
Autrement dit, a l’etat actuel, ces variables accroissent la volatilite des entrees d’IDE en Afrique respectivement de 0,11; 0,20;
1,72 et de 0,23. Ce resultat confirme celui obtenu par Daude and Stein (2007). Sur l’ensemble des pays de l’echantillon, seuls le
Botswana, la Mauritanie, le Senegal, le Burkina-Faso, le Nigeria et le Kenya ont fait des avancees significatives du point de vue
democratique. En depit du fait que l’Afrique ait connu un processus de liberalisation politique substantielle marquee par la
democratie, on note peu d’avancees significatives en matiere de droits civil et politique qui sont pour la plupart bafoues. Par
consequent, les structures erigees et les ideologies qui emergent en Afrique incarnent pour la plupart des inefficiences rendant
volatiles l’attractivite des IDE.
La corruption et la qualite de la regulation ont des effets negatifs et statistiquement significatifs sur la volatilite des IDE en
Afrique. La qualite du service public par contre, a un effet negatif et non significatif sur la volatilite des IDE en Afrique.
Au rang des variables de contr^ole, la Formation Brute du Capital Fixe et les infrastructures concourent aussi a la volatilite des
IDE qui s’explique par l’incapacite a reduire les co^uts de transaction, et a impulser une dynamique de productivite aux industries.
Les resultats sur les variables institutionnelles et les variables de contr^ole sont conformes a ceux de Buchanan et al. (2012).
Les resultats du modele 2 suggerent deux analyses. D’une part, la volatilite accrue des IDE en Afrique est davantage du ressort
de la FBCF, du regime politique et de la taille du marche. Les autres variables mises a contribution de cette volatilite des IDE bien
de maniere marginale sont la qualite du service publique et la stabilite politique. D’autre part, la volatilite des IDE est davantage
attenuee par la liberte d’expression et la qualite de la regulation.

5.2 Qualite des institutions et volatilite des entrees d’IDE dans les sous regions de l’Afrique
Le tableau 6 donne l’impact de la qualite des institutions sur la volatilite des IDE dans les differentes sous-regions de l’Afrique
par la technique des doubles moindres carres.
© 2017 The Authors. African Development Review © 2017 African Development Bank
680

Tableau 4: Matrice de correlation


V.IDE FBCF Cap_Hum Infrastr APD Res_Nat Polity2 Taille_mar Corrup Etat_droit Qual_Reg Qual_Ser Stab_Pol Lib_Exp

V.IDE 1
FBCF 0,117 1
Cap_Hum –0,18 –0,13 1
Infrastr –0,02 –0,15 0,315 1
APD 0,142 –0,06 –0,352 –0,402 1
Res_Nat 0,102 –0,09 0,089 0,448 –0,40 1
Polity2 0,244 –0,33 –0,231 0,030 0,702 0,079 1
Taille_mar 0,260 –0,12 –0,431 0,243 –0,14 0,818 0,280 1
Corrup 0,100 –0,37 0,238 –0,120 0,127 –0,281 0,034 –0,329 1
Etat_droit 0,018 –0,25 0,606 –0,108 –0,11 –0,15 –0,17 –0,467 0,792 1
Qual_Reg 0,061 –0,35 0,467 0,040 –0,13 –0,096 0,064 –0,330 0,678 0,824 1
Qual_Ser 0,085 –0,32 0,285 0,046 0,065 –0,058 –0,02 –0,199 0,791 0,736 0,728 1
Stab_Pol –0,27 0,118 0,707 0,163 –0,37 0,186 –0,39 –0,319 0,334 0,711 0,508 0,414 1
Lib_Exp 0,154 –0,40 0,456 –0,050 –0,04 0,018 0,081 –0,188 0,763 0,862 0,904 0,860 0,493 1
Source: Calcul des auteurs.

© 2017 The Authors. African Development Review © 2017 African Development Bank
B. E. Ongo Nkoa et J. S. Song
Analyse des effets de la qualite des institutions 681

Tableau 5: Impact de la qualite des institutions sur la volatilite des IDE en Afrique
Variable dependante: Volatilite des IDE
Methode d’estimation: GMM-Systeme
Modele 1: Avec ecart-type Modele 2: Avec filtre de Prescott
Coefficients P< t Coefficients P< t

Formation Brute du Capital Fixe 0,4052 0,008 1,0668 0,075


Capital Humain 0,4174 0,261 3,6114 0,289
Infrastructures 0,5130 0,083 –2,1038 0,116
Aide Publique au Developpement –0,1798 0,772 1,0417 0,257
Taille du marche –0,9438 0,363 0,7878 0,076
Polity 2 0,1176 0,002 0,8616 0,000
Corruption –0,0152 0,062 –0,075 0,025
Etat de droit 0,0208 0,064 –0,747 0,484
Qualite de la regulation –0,0018 0,000 –0,2421 0,086
Qualite du service public –1,6505 0,570 0,0635 0,000
Stabilite politique 1,7261 0,079 0,0559 0,093
Liberte d’expression 0,2346 0,051 –0,9791 0,08
Constante –32,099 0,795 –28,131 0,716
Observations 145 145
Nombre de pays 53 53
Test de Sargan 104,41 109,84
P-value de Sargan 0,58 0,618
Test AR1 0,000 0,000
Test AR2 0,547 0,485
Notes:  ,  , et  indiquent respectivement les significativites a 10%, 5%, et 1%.
Source: Auteurs.

Tableau 6: Qualite des institutions et volatilite des IDE avec ecart-type: analyse sous regionale
Variable dependante: Volatilite des IDE avec ecart-type Methode d’estimation: DMC
Afrique Centrale Afrique de l’Ouest Afrique de l’Est Afrique Australe Afrique du Nord
Coefficients P< t Coefficients P< t Coefficients P< t Coefficients P< t Coefficients P< t

Formation Brute du Capital 1,4228 0,613 0, 2174 0,116 0,1307 0,029 0,1416 0,088 0,082 0,035
Fixe
Capital Humain 0,118 0,608 0,1058 0,044 0,0503 0,108 –0,003 0,001 –0,019 0,042
Infrastructures –0,0128 0,922 0,0708 0,009 0,0003 0,983 0,0138 0,020 –0,0009 0,002
Aide Publique au –0,343 0,084 0,7163 0,000 0,1511 0,028 0,078 0,315 0,1648 0,677
Developpement
Taille du marche 0,1912 0,745 0,1837 0,017 0,0325 0,805 0,107 0,027 –0,0736 0,036
Ressources naturelles 0,4352 0,093 0,0385 0,000 6,6273 0,000 –0,0154 0,885 –1,2296 0,264
Polity 2 0,4256 0,007 –0,3705 0,070 0,069 0,009 –0,0154 0,727 –0,1102 0,002
Corruption –0,2711 0,848 –1,256 0,687 2,5581 0,040 2,534 0,069 –0,4487 0,04
Etat de droit 0,1444 0,038 –1,4307 0,657 3,8403 0,013 1,8684 0,385 –0,7137 0,485
Qualite de la regulation –1,0976 0,078 –3,7434 0,296 0,1557 0,070 0,8217 0,029 –0,1618 0,097
Qualite du service public 5,5577 0,403 –0,9016 0,755 1,7545 0,022 –1,4933 0,059 0,918 0,001
Stabilite politique –1,8421 0,004 –0,8332 0,614 1,6117 0,004 –1,3904 0,007 0,367 0,053
Liberte d’expression 5,5213 0,722 2,3844 0,406 1,9611 0,123 –3,9097 0,019 –2,805 0,03
Constante 7,256 0,609 –12,775 0,000 4,2704 0,078 –1,0473 0,648 1,2547 0,403
Observations 96 163 64 146 63
Prob>F 0,0001 0,000 0,000 0,0008 0,007
R2 0,4003 0,3455 0,692 0,4663 0,4125
Notes:  ,  , et  indiquent respectivement les significativites a 10%, 5%, et 1%.
Source: Auteurs.

© 2017 The Authors. African Development Review © 2017 African Development Bank
682 B. E. Ongo Nkoa et J. S. Song

En Afrique Centrale, le regime politique et l’etat de droit accroissent la volatilite des IDE. Ceci se justifie par les faibles
niveaux de democratie, caracterises par le manque d’alternance au pouvoir dans les pays membres. La qualite de la regulation et
la stabilite politique reduisent la volatilite des IDE. Les legislations erigees en matiere de paix, de securite et d’amelioration du
climat des investissements motivent les investisseurs etrangers. Les ressources naturelles accroissent de maniere significative la
volatilite des IDE, alors que l’aide publique au developpement contribue a reduire bien que de maniere marginale, la volatilite
des IDE en Afrique Centrale.
En Afrique de l’Ouest, la volatilite accrue des IDE est expliquee par les variables economiques. Seul le regime politique reduit
de maniere significative la volatilite des IDE en raison des avancees significatives en matiere de democratie qui impulsent une
credibilite en matiere de droit de propriete et de stabilite sociopolitique.
En Afrique de l’Est, le regime politique, la corruption, l’etat de droit, la qualite de la regulation, la qualite du service public et
la stabilite politique engendrent une volatilite accrue des IDE. Ceci, s’explique par des niveaux moyens de democratie qui
favorisent des legislations et des revendications couplees a des destabilisations. Les ressources naturelles, l’aide publique au
developpement et la FBCF accroissent significativement la volatilite des IDE.
En Afrique Australe, la corruption et la qualite de la reglementation accroissent la volatilite accrue des IDE. La qualite du
service public, la stabilite politique et la liberte d’expression reduisent la volatilite des IDE. L’effet positif des variables
institutionnelles sur la volatilite des IDE s’explique par les institutions qui favorisent l’application inefficace et a co^uts eleves des
contrats sur les marches qu’induit a priori la corruption qui engendre des co^uts additionnels. La FBCF et l’aide publique au
developpement contribuent a accro^ıtre significativement la volatilite des IDE. A contrario, le capital humain contribue a reduire
la volatilite des IDE en Afrique Australe du fait d’une scolarisation de la jeunesse qui constitue une meilleure absorption et
stabilise les entrees d’IDE.
En Afrique du Nord, la qualite du service public et la stabilite politique accroissent la volatilite des IDE. En effet, les conflits
armes ont un impact negatif sur l’attractivite des IDE (Ezeoha et Ugwu, 2015). Dans une situation de conflit, il est fort probable
d’enregistrer des desinvestissements. Par contre, le regime politique, la corruption, l’etat de droit, la qualite de la reglementation
et la liberte d’expression reduisent la volatilite des IDE. Ces resultats s’expliquent par la vision optimiste portee sur l’alternance

Tableau 7: Qualite des institutions et volatilite des IDE avec filtre de Prescott: analyse sous regionale
Variable dependante: Volatilite des IDE avec filtre de Prescott Methode d’estimation: DMC
Afrique Centrale Afrique de l’Ouest Afrique de l’Est Afrique Australe Afrique du Nord
Coefficients P< t Coefficients P< t Coefficients P< t Coefficients P< t Coefficients P< t

Formation Brute du Capital 0.0497 0.809 0.462 0.000 –0.0008 0.987 0.1865 0.245 0.081 0.038
Fixe
Capital Humain –0.195 0.001 0.0446 0.376 0.0467 0.077 –0.094 0.006 –0.0295 0.356
Infrastructures 0.0482 0.009 –0.0159 0.485 –0.0229 0.045 0.054 0.001 0.004 0.064
Aide Publique au –0.0813 0.489 –0.393 0.006 0.093 0.000 –0.0648 0.538 0.2035 0.599
Developpement
Taille du marche 0.0720 0.528 –0.1532 0.420 0.421 0.000 –0.001 0.088 –0.103 0.023
Ressources naturelles 0.2154 0.045 0.253 0.009 4.892 0.000 0.019 0.095 –0.4356 0.674
Polity 2 0.052 0.045 –0.078 0.079 –0.0599 0.380 –0.103 0.000 –0.087 0.018
Corruption –1.8713 0.510 1.925 0.040 –1.907 0.062 0.0171 0.801 –0.675 0.006
Etat de droit 1.1232 0.735 1.023 0.006 –0.786 0.026 0.0060 0.780 0.992 0.005
Qualite de la regulation 3.444 0.090 –0.647 0.092 –0.546 0.006 –0.029 0.065 –0.613 0.613
Qualite du service public 1.2817 0.010 –4.8485 0.142 –0.528 0.058 0.821 0.029 1.0587 0.332
Stabilite politique 3.7263 0.078 0.6951 0.723 0.6813 0.149 0.0037 0.714 0.3677 0.453
Liberte d’expression –0.0408 0.049 –0.6824 0.828 –0.1628 0.871 –0.008 0.040 –2.181 0.002
Constante 14.52 0.000 1.0183 0.750 –3.628 0.033 9.332 0.000 1.6764 0.248
Observations 96 163 64 150 64
Prob>F 0.0008 0.0029 0.0001 0.0005 0.0053
R2 0.3206 0.5104 0.5001 0.1397 0.3964
Notes:  ,  , et  indiquent respectivement les significativites a 10%, 5%, et 1%.
Source: Auteurs.

© 2017 The Authors. African Development Review © 2017 African Development Bank
Analyse des effets de la qualite des institutions 683

Graphique 2: Filtres de Prescott des IDE en Afrique centrale

Notes: 1 ¼ Angola, 2 ¼ Burundi, 3 ¼ Cameroun, 4 ¼ DR Congo, 5 ¼ Gabon, 6 ¼ Guinee Equatoriale, 7 ¼ Republique


centrafricaine,8 ¼ Congo, 9 ¼ Rwanda, 10 ¼ Sao Tome et Principe, 11 ¼ Tchad.
Source: les auteurs.

au pouvoir qui vise a mettre fin aux differentes inefficiences de la gestion du pouvoir. Cette vision optimiste induit un
environnement institutionnel caracterise par une reglementation peu excessive qui facilite les transactions des investisseurs
etrangers et accelere leurs procedures dans un climat d’opacite administrative. Ainsi, l’environnement institutionnel instaure un
cadre legal efficace en reduisant les pots-de-vin qui incitent les investisseurs etrangers a acquerir des actifs.

Graphique 3: Le filtre de Prescott des IDE en Afrique de l’Ouest

Notes: 1 ¼ Benin, 2 ¼ Burkina Faso, 3 ¼ Cape Vert, 4 ¼ C^ote d’Ivoire, 5 ¼ Gambie, 6 ¼ Ghana, 7 ¼ Guinee, 8 ¼ Guinee-Bissau,
9 ¼ Liberia, 10 ¼ Mali, 11 ¼ Mauritanie, 12 ¼ Niger, 13 ¼ Nigeria, 14 ¼ Senegal, 15 ¼ Sierra-Leone, 16 ¼ Togo.
Source: les auteurs.
© 2017 The Authors. African Development Review © 2017 African Development Bank
684 B. E. Ongo Nkoa et J. S. Song

Graphique 4: Le filtre de Prescott des IDE en Afrique de l’Est

Notes: 1 ¼ Comores, 2 ¼ Djibouti, 3 ¼ Erythree, 4 ¼ Ethiopie, 5 ¼ Kenya, 6 ¼ Somalie, 7 ¼ Soudan, 8 ¼ Tanzanie, 9 ¼ Ouganda.
Source: les auteurs.

Le tableau 7 presente les resultats de l’impact de la qualite des institutions sur la volatilite des IDE dans les differentes sous
regions. Il ressort des estimations que les institutions en Afrique centrale et en Afrique de l’Ouest, accroissent la volatilite des
IDE. Dans les autres trois regions, les institutions creent un climat favorable a l’investissement contribuant ainsi a une croissance
constante.

Graphique 5: Le filtre de Prescott des IDE Afrique Australe

Note: 1 ¼ Botswana, 2 ¼ Lesotho, 3 ¼ Madagascar, 4 ¼ Malawi, 5 ¼ Maurice, 6 ¼ Mozambique, 7 ¼ Namibie, 8 ¼ Seychelles,


9 ¼ Afrique du Sud, 10 ¼ Swaziland, 11 ¼ Zambie, 12 ¼ Zimbabwe
Source: les auteurs.
© 2017 The Authors. African Development Review © 2017 African Development Bank
Analyse des effets de la qualite des institutions 685

Graphique 6: Le filtre de Prescott des IDE en Afrique du Nord

Notes: 1 ¼ Algerie, 2 ¼ Egypte, 3 ¼ Lybie, 4 ¼ Maroc, 5 ¼ Tunisie.


Source: les auteurs.

Les graphiques 2, 3, 4, 5 et 6 donnent les filtres de Prescott de la volatilite des IDE dans chaque region et pour chaque pays.
L’observation importante ici est que les pays dotes de faibles institutions ont des entrees d’IDE tres volatiles.
En Afrique centrale, le filtre de Prescott indique que la volatilite des IDE est accentuee en Angola, en Guinee Equatoriale, au
Congo, a Sao Tome et Principe et au Tchad (Graphique 2).

6. Conclusion et recommandations de politique economique


Dans cet article nous avons analyse les effets de la qualite des institutions sur la volatilite des Investissements Directs Etrangers en
Afrique. Les investigations empiriques menees, nous ont conduit a faire usage des donnees de panel estimees par la Methode des
Moments Generalises et des Doubles Moindres Carres. Des differentes estimations, il ressort qu’en Afrique, la volatilite des IDE est
expliquee par la qualite des institutions, notamment, le regime politique, l’etat de droit, la stabilite politique, la liberte d’expression et
la qualite de la reglementation. Les autres variables mises a contribution de ce resultat sont la Formation Brute du Capital Fixe, le
capital humain, les infrastructures, l’aide publique au developpement, la taille du marche et les ressources naturelles. De plus, les
variables institutionnelles affectent de maniere heterogene la volatilite des IDE dans les differentes sous-regions. En Afrique Centrale,
la volatilite des IDE est expliquee par le regime politique, l’etat de droit, la qualite de la reglementation et la stabilite politique. En
Afrique de l’Ouest, cette volatilite est uniquement expliquee par le regime politique. En Afrique de l’Est, elle est justifiee par toutes les
variables institutionnelles. En Afrique Australe, la volatilite des IDE repose sur l’ampleur de la corruption, la qualite de la
reglementation, la stabilite politique, la qualite du service public et la liberte d’expression. En Afrique du Nord, la volatilite des IDE est
expliquee par toutes les variables institutionnelles excepte la democratie.
Trois principales recommandations sont formulees: (i) l’assainissement de l’environnement institutionnel a travers la mise en
place des regles economique, politique et legale qui delimitent et soutiennent l’activite transactionnelle; (ii) la promotion des politiques
de bonne gouvernance afin de reduire les inefficiences institutionnelles, de stimuler la competitivite dans les systemes de marche, et de
garantir la protection des droits de propriete; (iii) la reforme les institutions sur l’executif, visant a stabiliser les flux d’IDE.

References
Adams, S. et O. Opoku (2015), ‘Foreign Direct Investment, Regulations and Growth in Sub-Saharan Africa’, Economic Analysis
and Policy, Vol. 47, pp. 48–56.
© 2017 The Authors. African Development Review © 2017 African Development Bank
686 B. E. Ongo Nkoa et J. S. Song

Ajide, K. B. et O. E. Osode (2017), ‘Does FDI Dampen or Magnify Output Growth Volatility in the ECOWAS Region?’, African
Development Review, Vol. 29, No. 2, pp. 211–22.
Ali, F., N. Fiess et R. Macdonald (2010), ‘Do Institutions Matter for Foreign Direct Investment?’, Open Economies Review,
Vol. 21, No. 2, pp. 201–19.
Alley, I. et I. Poloamina (2015), ‘Private Capital Flow Shocks and Sub-Saharan African Macroeconomic Performance’, Journal
of International Economic Studies, No. 29, pp. 61–84.
Anyanwu, J. C. (2012), ‘Why Does Foreign Direct Investment Go Where It Goes? New Evidence From African Countries’,
Annals of Economics and Finance, Vol. 13, No. 2, pp. 433–70.
Anyanwu, J. C. et N. D. Yameogo (2015a), ‘What Drives Foreign Direct Investments into West Africa? An Empirical
Investigation’, African Development Review, Vol. 27, No. 3, pp. 199–215.
Anyanwu, J. C. et N. D. Yameogo (2015b), ‘Regional Comparison of Foreign Direct Investment to Africa: Empirical Analysis’,
African Development Review, Vol. 27, No. 4, pp. 345–63.
Arellano, M. et S. Bond (1991), ‘Some Tests of Specification for Panel Data: Monte Carlo Evidence and an Application to
Employment Equations’, Review of Economics Studies, Vol. 58, No. 2, pp. 277–97.
Arellano, M. et O. Bover (1995), ‘Another Look at the Instrumental Variable Estimation of Error-Components Models’, Journal
of Econometrics, Vol. 68, No. 1, pp. 29–51.
Asamoah, M. E., C. K. D. Adjasi and A. L. Alhassan (2016) ‘Macroeconomic Uncertainty, Foreign Direct Investment and
Institutional Quality: Evidence from Sub-Saharan Africa’, Economic Systems, available at: https://doi.org/10.1016/j.
ecosys.2016.02.010
Asiedu, E. et D. Lien (2011), ‘Democracy, Foreign Direct Investments and Natural Resources’, Journal of International
Economics, Vol. 84, pp. 99–111.
Avom, D. et N. B. E. Ongo (2013), ‘Foreign Direct Investment in Central Africa: What Are the Relevant Determinants? An
Empirical Investigation’, African Integration and Development Review, Vol. 6, No. 2, pp. 29–47.
Barrios S., H. Gorg et E. Strobl (2005), ‘Foreign Direct Investment, Competition and Industrial Development in the Host
Country’, European Economic Review, Vol. 49, pp. 1761–84.
Benassy-Quere, A., M. Coupet et T. Mayer (2007), ‘Institutional Determinants of Foreign Direct Investment’, The World
Economy, pp. 764–82.
Blundell, R. et S. Bond (1998), ‘Initial Conditions and Moment Restrictions in Dynamic Panel Data Models’, Journal of
Econometrics, Vol. 87, pp. 115–43.
Bokpin, A. G. (2017), ‘Foreign Direct Investment and Environmental Sustainability in Africa: The Role of Institutions and
Governance’, Research in International Business and Finance, available at: https://doi.org/10.1016/j.ribaf.2016.07.038.
Broto, C., J. Dıaz-Cassou et A. Erce (2011), ‘Measuring and Explaining the Volatility of Capital Flows to Emerging Countries’,
Journal of Banking and Finance, Vol. 35, pp. 1941–53.
Buchanan, B., V. Quan et R. Meenakshi (2012), ‘Foreign Direct Investment and Institutional Quality: Some Empirical
Evidence’, International Review of Financial Analysis, Vol. 21, pp. 81–89.
Cleeve, E., Y. Debrah et Z. Yiheyis (2015), ‘Human Capital and FDI Inflow: An Assessment of the African Case’, World
Development, Vol. 74, pp. 1–14.
CNUCED (2016), Base de donnees de la CNUCED, available at: www.unctad.org/fdistatistics.
Daude, C. et E. Stein (2007), ‘The Quality of Institutions and Foreign Direct Investment’, Economics and Politics, Vol. 19,
No. 3, pp. 317–44.
Dunne, J. P. et N. Masiyandima (2017), ‘Bilateral FDI from South Africa and Income Convergence in SADC’, African
Development Review, Vol. 29, No. 3, pp. 403–15.
© 2017 The Authors. African Development Review © 2017 African Development Bank
Analyse des effets de la qualite des institutions 687

Dunning, J. (1980), ‘Toward an Eclectic Theory of International Production Some Empirical Tests’, Journal of International
Business Studies, Vol. 11, No. 1, pp. 9–31.
Dupasquier, C. et P. N. Osakwe (2006), ‘Foreign Direct Investment in Africa: Performance, Challenges, and Responsibilities’,
Journal of Asian Economics, Vol. 17, pp. 241–60.
Eregha, P. (2015), ‘Foreign Direct Investment Inflow, Volatility, and Domestic Investment in West Africa’, The Journal of
Developing Areas, Vol. 49, No. 2, pp. 273–94.
Estrin, S., J. Korosteleva et T. Mickiewicz (2013), ‘Which Institutions Encourage Entrepreneurial Growth Aspirations?’,
Journal of Business Venturing, Vol. 28, pp. 564–80.
Ezeoha, A. et O. Ugwu (2015), ‘Interactive Impact of Armed Conflicts on Foreign Direct Investments in Africa’, African
Development Review, Vol. 27, No. 4, pp. 456–68.
Gankou, J-M., M. Bendoma et M. N. Sow (2016), ‘The Institutional Environment and the Link between Capital Flows and
Capital Flight in Cameroon’, African Development Review, Vol. 28, No. 1, pp. 65–87.
Insee (2016), Bulletin statistique de l’Institut national de la statistique et des etudes economiques 2016, pp. 24.
Kaufmann, D., A. Kraay et M. Mastruzzi (2010), ‘The Worldwide Governance Indicators: A Summary of Methodology, Data
and Analytical Issues’, World Bank Policy Research, Working Paper No. 5430.
Komlan, F. (2016), ‘Quality of Institutions, Natural Resources and Foreign Direct Investment (FDI) in Sub-Saharan Africa:
Dynamic Approach’, Journal of Economics and Development Studies, Vol. 4, No. 2, pp. 28–55.
Krugman, P. (1991), ‘Increasing Returns and Economic Geography’, Journal of Political Economy, Vol. 99, No. 3,
pp. 483–99.
Lensink, R. et O. Morrissey (2006), ‘Foreign Direct Investment: Flows, Volatility, and the Impact on Growth’, Review of
International Economics, Vol. 14, No. 3, pp. 478–93.
Malikane, Ch. et P. Chitambara (2017), ‘Foreign Direct Investment, Democracy and Economic Growth in Southern Africa’,
African Development Review, Vol. 29, No. 1, pp. 92–102.
Mijiyawa, A. G. (2015), ‘What Drives Foreign Direct Investment in Africa? An Empirical Investigation with Panel Data’,
African Development Review, Vol. 27, No. 4, pp. 392–402.
North, D. (1990), Institutions, Institutional Change, and Economic Performance, Cambridge University Press, Cambridge.
Okafor, G. (2015), ‘Locational Determinants of US Outward FDI into Sub-Saharan Africa’, The Journal of Developing Areas,
Vol. 49, No. 1, pp. 187–205.
Pournarakis, M., et N. C. Varsakelis (2002), ‘Foreign Direct Investment in Central and Eastern European Countries: Do
Institutions Matter?’, EIBA Annual conference, December, Athens, Greece.
Roodman, D. (2009), ‘How to Do Xtabond2: An Introduction to Difference and System GMM in Stata’, The Stata Journal,
Vol. 9, pp. 86–136.
Sane, M. (2016), ‘Determinants of Foreign Direct Investment Inflows to ECOWAS Member Countries: Panel Data Modelling
and Estimation’, Modern Economy, Vol. 7, pp. 1517–42.
Sevestre, P. (2002), Econometrie des donnees de panel, Dunod, Paris.
Vadlamannati, K., A. Tamazian et L. Irala (2009), ‘Determinants of Foreign Direct Investment and Volatility in South East
Asian Economies’, Journal of the Asia Pacific Economy, Vol. 14, No. 3, pp. 246–61.
Waqas, Y., H. H. Shujahat et I. N. Muhammad (2015), ‘Macroeconomic Factors and Foreign Portfolio Investment Volatility: A
Case of South Asian Countries’, Future Business Journal, available at: https://doi.org/10.1016/j.fbj.2015.11.002
WDI (2016), World Development Indicators, base de donnees de la Banque mondiale.
Wei, S. J. (2000) ‘How Taxing Is Corruption on International Investors?’, Review of Economics and Statistics, Vol. 82,
pp. 1–11.

© 2017 The Authors. African Development Review © 2017 African Development Bank
688 B. E. Ongo Nkoa et J. S. Song

WGI (2016), Worldwide Governance Indicators, base de donnees des indicateurs de gouvernance de la Banque Mondiale.
Wisniewski, T. et S. Pathan (2014), ‘Political Environment and Foreign Direct Investment: Evidence from OECD Countries’,
European Journal of Political Economy, Vol. 36, pp. 13–23.
Yasin, M. (2005), ‘Official Development Assistance and Foreign Direct Investment Flows to Sub-Saharan Africa’, African
Development Review, Vol. 17, No. 1, pp. 23–40.

© 2017 The Authors. African Development Review © 2017 African Development Bank
African Development Review, Vol. 29, No. 4, 2017, 689–690

Appreciation to Article Reviewers


Through Volume 29, No. 4 of the African Development Review, we are taking the opportunity to express our sincere appreciation
to the scholars and experts listed below who have taken from their time to serve as peer reviewers for the various articles
submitted to the African Development Review over the past few months. The contributions of the referees are greatly valued and
will continue to be appreciated. We look forward to their continued support.

Adebowale, Olabimtan
Aderemi, Taiwo
Akalpler, Ergin
Aniche, Ernest
Arthur, Eric
Ayanwale, Adeolu
Ayisi, Richard Kwabi
Azad, Md. Abul Kalam
Bakehe, Novice Patrick
Belbute, Jose
Bobbo, Amadou
Bolarinwa, Segun
Bouabid, Anis
Coothoopermal, Sachin
Egboro, Edwin
Fakere, Alexander
Felix, Winston
Gandjon Fankem, Gislain Stephane
Gbahabo, Paul
Gyimah, Adjei Gyamfi
Houngbonon, Georges
Iwaloye, Ojo
Izueke, Edwin
Labintan, Constant
Lakuma, Corti
Laurila, Hannu
Lemma, Alemseged
Malanou, Florentin
Mballa, Urbain Nerry Cyrille
Michael, Bryane
Mintah, Samuel
Nkoumou Ngoa, Gaston Brice
Noubbigh, Hedi
Novignon, Jacob
Ogbuabor, Jonathan
Ojo, Mathew
Omisakin, Olusegun
Oncioiu, Ionica

© 2017 The Authors. African Development Review © 2017 African Development Bank. Published by Blackwell Publishing Ltd,
9600 Garsington Road, Oxford OX4 2DQ, UK and 350 Main Street, Malden, MA 02148, USA. 689
690

Pilo, Mikemina
Selim, Hoda
Sene, Omar
Sene, Seydina
Tanle, Augustine

© 2017 The Authors. African Development Review © 2017 African Development Bank
African Development Review, Vol. 29, No. 4, 2017, 691

Corrigendum

In Nwosu and Orji 2017, an error has been identified in the published version.

After issue publication, the author wants to add an acknowledgement to recognize the funding agency of his study.

This research work was carried out with financial and scientific support from the Partnership for Economic Policy (PEP)
(www.pep-net.org) with funding from the Department for International Development (DFID) of the United Kingdom (or UK
Aid), and the Government of Canada through the International Development Research Center (IDRC). The authors are also
grateful to Luca Tiberti, Jean-Yves Duclos and Guy Lacroix for technical support and guidance, as well as to John Cockburn
and Abdelkrim Araar for valuable comments and suggestions.

We apologize for this error.

Reference
Nwosu, E.O. and A. Orji (2017). ‘Addressing Poverty and Gender Inequality through Access to Formal Credit and Enhanced
Enterprise Performance in Nigeria: An Empirical Investigation’, African Development Review, Vol. 29, No. S1, pp. 56–72.

© 2017 The Authors. African Development Review © 2017 African Development Bank. Published by Blackwell Publishing Ltd,
9600 Garsington Road, Oxford OX4 2DQ, UK and 350 Main Street, Malden, MA 02148, USA. 691
African Development Review
AIMS, SCOPE AND AUTHOR GUIDELINES
Aims and Scope
African Development Review is a professional journal devoted to the study and analysis of development policy in Africa. Published four times a year
for the African Development Bank, the Review emphasizes policy relevance of research findings, rather than purely theoretical and quantitative issues.
The African Development Bank is a regional multilateral development finance institution, the members of which are all of the 53 countries in Africa
and 25 countries from Asia, Europe, North and South America. The purpose of the Bank is to further the economic development and social progress of African
countries, individually and collectively. To this end, the Bank promotes the investment of public and private capital for development, primarily by providing
loans and grants for projects and programs that contribute to poverty reduction and broadly based development in Africa.
The review contains:
• High quality articles analyzing recent critical issues relating to Africa’s economic and social development;
• Empirical analyses and case studies either of single countries, or of comparative nature, which shed light on development policy choices; and
• Book reviews, conference reports and comments on Review articles.
It covers, amongst others, the following subjects:
• Macroeconomic Policies (Fiscal, Monetary and Exchange Rate Policy);
• Economic and Structural Reforms including Issues of Financial Sector Reforms;
• Sectoral Issues on Agriculture, Energy, Mining and Industry;
• Issues of Building Infrastructure and Human Resource Capacity;
• Private Sector Development;
• Regional and International Concerns such as Debt, Trade, Capital Flows, Regional Integration, South-South Cooperation and Globalization; and
• Socioeconomic Issues, including Income Distribution and Poverty Alleviation.

Author Guidelines
Information for Contributors
The African Development Review is interested in publishing papers, research notes and book reviews on development issues in Africa. The journal
focuses on quality policy-oriented papers. Articles submitted should not be under consideration elsewhere. Manuscripts will be sent anonymously to
at least two referees.

Preparation and submission of manuscripts


1. African Development Review has now adopted ScholarOne Manuscripts, for online manuscript submission and peer review. The new system
brings with it a whole host of benefits including:
• Quick and easy submission
• Administration centralized and reduced
• Significant decrease in peer review times
Manuscripts should be submitted, double spaced throughout and typed clearly on one side of the paper only at http://mc.manuscriptcentral.com/
adrev. Authors will be able to track the status of their papers online. Full instructions and support are available on the site and a user ID and
password can be obtained on the first visit. If you require assistance then click the Get Help Now link which appears at the top right of every
ScholarOne Manuscripts page.
2. Papers should be no more than 7,000 words. There are no submission fees.
3. The paper must be preceded by an abstract of no more than 200 words which should be submitted in English or French, and should state, in
condensed form, the contribution made by the paper.
4. No more than 3 levels of headings should be used. Footnotes are to be avoided and should take the form of endnotes listed prior to the references.
The Harvard style of referencing is used (author’s name and date of publication bracketed in the text) and all works cited should be listed
alphabetically by the author after the main body of the text, style as follows:
Bosrup, E. (1981), “Populations and Technological Change”, University of Chicago Press, Chicago, USA.
Cohen, J. (1980), “Land Tenure and Rural Development in Africa”, in Agricultural Development in Africa, R. Bates and M. Lofchie (eds), Praeger,
New York, USA.
Nissanke, M. and Thorbecke, E. (2008), “Introduction: Globalization-Poverty Channels and Case Studies from Sub-Saharan Africa”, African
Development Review, Vol. 20, No.1, April, 1–19.
Venables, A. J. (2009), “Rethinking Growth in a Globalizing World: an Economic Geography Lens”, African Development Review, Vol. 21,
No.2, September, 331–351.
5. All editorial correspondence, outside manuscript submission and status of papers, should be sent to: the Editor, African Development Review, African
Development Bank, CCIA Building, Avenue Jean-Paul II, 01 BP 1387, Abidjan 01, Côte d’Ivoire; E-mail address: j.anyanwu@afdb.org or
Tel.: +225 2026 2225.

Tables and artwork


All tables and figures should be placed at their appropriate places within manuscript. Figures should be supplied as high quality, original artwork and
any lettering or line work should be able to sustain reduction to the final size of reproduction. Tints and complex shading should be avoided.

Copyright
The contribution of the author(s) should be completely original, it should in no way violate any existing copyright, and it should contain nothing of a libelous
or scandalous nature.

Exclusive License Form


Authors will be required to sign an Exclusive License Form (ELF) for all papers accepted for publication. Signature of the ELF is a condition of
publication and papers will not be passed to the publisher for production unless a signed form has been received. Please note that signature of the Exclusive
License Form does not affect ownership of copyright in the material. (Government employees need to complete the Author Warranty sections, although
copyright in such cases does not need to be assigned). After submission authors will retain the right to publish their paper in various media/
circumstances (please see the form for further details). To assist authors an appropriate form is online and authors are notified appropriately.
Online production tracking is now available for your article through Wiley-Blackwell’s Author Services
Author Services enables authors to track their articles—once accepted—through the production process to publication online and in print. Authors can check the
status of their articles online and choose to receive automated e-mails at key stages of production. The author will receive an e-mail with a unique link that enables
them to register and have their article automatically added to the system. Please ensure that a complete e-mail address is provided when submitting the manuscript.
Visit http://authorservices.wiley.com/bauthor/ for more details on online production tracking and for a wealth of resources including FAQs and tips on article
preparation, submission and more.
African Development Review
Revue africaine de développement
Revue Africaine de développement
OBJECTIFS, PORTÉE ET DIRECTIVES AUX AUTEURS
Editor Periodical ID Statement: AFRICAN DEVELOPMENT
John C. Anyanwu, African Development Bank REVIEW (ISSN: 1017-6772) is published quarterly in March, Objectifs et Portée
June, September and December. US mailing agent: Mercury La Revue Africaine de développement est une revue professionnelle consacrée à l’étude et à l’analyse des politiques de développement en Afrique. Publiée quatre
Media Processing, LLC 1850 Elizabeth Avenue, Suite #C, fois par an pour la Banque africaine de développement, la Revue met l’accent sur la pertinence des résultats de la recherche, plutôt que sur des questions purement
Managing Editor
Rahway, NJ 07065 USA. Periodical postage paid at Rahway, NJ. théoriques et quantitatives.
Abebe Shimeles, African Development Bank
Journal Customer Services: For ordering information, claims La Banque africaine de développement est une institution régionale multilatérale de financement du développement qui compte les 53 pays africains comme
Associate Editors and any enquiry concerning your journal subscription please go to membres ainsi que 25 pays d’Asie, Europe et d’Amérique du Nord et du Sud. La Banque a pour vocation de contribuer au développement économique et au
www.wileycustomerhelp.com/ask or contact your nearest office: progrès social des pays africains, individuellement et collectivement. À cette fin, la Banque encourage l’investissement de capitaux publics et privés dans des
Augustin Fosu, University of Ghana, Ghana opérations en faveur du développement, essentiellement par l’octroi de prêts et de dons à des projets et programmes qui contribuent à la réduction de la pauvreté
Akin Iwayemi, University of Ibadan, Nigeria Americas: Email: cs-journals@wiley.com; Tel: +1 781 388 8598
or +1 800 835 6770 (Toll free in the USA & Canada) et à un développement à base élargie en Afrique.
Abdul B. Kamara, African Development Bank
Stephen Karingi, Economic Commission for Africa Europe, Middle East and Africa: Email: cs-journals@wiley.com; La revue contient :
Laeticia Mukarasi, East African Development Bank Tel: +44 (0) 1865 778315 • Des articles de qualité qui analysent les grandes questions récentes relatives au développement économique et social de l’Afrique ;
Asia Pacific: Email: cs-journals@wiley.com; Tel: +65 6511 8000 • Des analyses empiriques et des études de cas, soit de pays individuels, soit à caractère comparatif, qui apportent des éclaircissements sur des choix de poli-
Germano Mwabu, University of Nairobi, Kenya
Japan: For Japanese speaking support, Email: cs-japan@wiley. tiques de développement ; et
Tchétché Nguessan, Centre Ivoirien de Recherches
com; Tel: +65 6511 8010 or Tel (toll-free): 005 316 50 480. • Des critiques de livres, des rapports de conférences et des commentaires sur des articles parus dans la Revue.
Economiques et Sociales, Cote d’Ivoire
Visit our Online Customer Get-Help available in 7 languages Elle couvre, notamment, les thèmes suivants :
Catherine Pattillo, International Monetary Fund at www.wileycustomerhelp.com
• Les politiques macroéconomiques (Politiques budgétaire, monétaire et de change) ;
Advisory Board This journal is available online at Wiley Online Library. Visit • Les réformes économiques et structurelles, y compris les questions touchant les réformes du secteur financier ;
Olu Ajakaiye, African Centre for Shared Development, wileyonlinelibrary.com to search the articles and register for table • Les questions sectorielles sur l’agriculture, l’énergie, les mines et l’industrie ;
Ibadan, Nigeria of contents e-mail alerts. • Les questions relatives au renforcement de l’infrastructure et des compétences des ressources humaines ;
Ernest Aryeetey, African Research Universities Alliance • Le développement du secteur privé ;
For submission instructions, subscription and all other information • Les questions régionales et internationales comme la dette, le commerce, les flux de capitaux, l’intégration régionale, la coopération Sud-Sud et la mondialisation ; et
Jean-Paul Azam, ARQADE Université de Toulouse, France
visit: http://wileyonlinelibrary.com/journal/AFDR • Les questions socioéconomiques, notamment, la répartition des revenus et la lutte contre la pauvreté.
Arne Bigsten, Goteborg University, Sweden
Paul Collier, Oxford University, UK Production Editor: Lazel Arcangel
Hakim Ben Hammouda, formerly African Development Bank (email: afdr@wiley.com)
Directive