Policy Research Working Paper 10624
Stuck in a Conflict Trap
The Case of the Central African Republic Civil War
Pierre Mandon
Vincent Nossek
Diderot Sandjong Tomi
Macroeconomics, Trade and Investment A verified reproducibility package for this paper is
Global Practice available at http://reproducibility.worldbank.org,
December 2023 click here for direct access.
Policy Research Working Paper 10624
Abstract
This paper uses the synthetic control method to assess the civil war led to a significant drop in gross domestic product
impact of the civil war in the Central African Republic per capita (41.6 percent), nighttime light intensity (33.8
on the main socioeconomic indicators. Based on a donor percent), industrial production (34.1 percent), manufac-
pool of low-income countries, the paper builds a synthetic turing value added (33.7 percent), and the human asset
counterfactual to evaluate the magnitude of the socioeco- index (20.2 percent), from 2013, which is considered as
nomic impacts of the civil war. The results indicate that the the starting point of the ongoing political and civil crisis.
This paper is a product of the Macroeconomics, Trade and Investment Global Practice. It is part of a larger effort by the
World Bank to provide open access to its research and make a contribution to development policy discussions around
the world. Policy Research Working Papers are also posted on the Web at http://www.worldbank.org/prwp. The authors
may be contacted at pmandon@worldbank.org. A verified reproducibility package for this paper is available at http://
reproducibility.worldbank.org, click here for direct access.
The Policy Research Working Paper Series disseminates the findings of work in progress to encourage the exchange of ideas about development
issues. An objective of the series is to get the findings out quickly, even if the presentations are less than fully polished. The papers carry the
names of the authors and should be cited accordingly. The findings, interpretations, and conclusions expressed in this paper are entirely those
of the authors. They do not necessarily represent the views of the International Bank for Reconstruction and Development/World Bank and
its affiliated organizations, or those of the Executive Directors of the World Bank or the governments they represent.
Produced by the Research Support Team
Stuck in a Conflict Trap: The Case of the Central African Republic Civil War
Pierre Mandona, Vincent Nosseka,b, Diderot Sandjong Tomia,c
a: World Bank Group
b: CERDI, University Clermont Auvergne (UCA)
c: University of Ottawa
The findings, interpretations, and conclusions expressed in this paper are entirely those of the authors.
They do not necessarily represent the views of the International Bank for Reconstruction and
Development/World Bank and its affiliated organizations, or those of the Executive Directors of the World
Bank or the governments they represent.
Keywords: C31; D74; O11
JEL codes: Central African Republic; Civil war; Development trap; Synthetic control method
Introduction
To what extent have repeated cycles of conflict and violence derailed socio-economic
development of the Central African Republic (CAR)? The cycle of conflict in this paper refers to
the CAR civil war that officially began in December 2012 and which is still ongoing, mainly in the
northern and western borders as well as in the eastern part of the country. Following the
signature of the peace accord (the Political Agreement for Peace and Stability, or APPR in French
acronym) in February 2019 between the government and leaders of the 14 armed groups, the
prospect for a gradual return to peace and stabilization surfaced but was quickly thwarted in
early 2021 following the presidential elections of December 27, 2020. In contrast to previous
coups and the first civil war officially called the “CAR Bush war’’ of 2004-07, CAR’s civil war is
exceptional in many ways. First, its duration is remarkable since it lasted 10 years and is still
ongoing. Second, the conflict has had a significant extension. For instance, between mid-
December 2012 and January 2014, the capital city of Bangui was captured by rebel groups, and
transport along the country’s two main international transit corridors, namely the Ubangi River
and Douala-Bangui road axis, was severely disrupted during the initial phase of the war. Third,
the parties to the conflict display a high degree of fragmentation as evidenced by the involvement
of numerous armed groups and the “de facto” partition of the country into a network of fiefdoms.
Finally, the intensity of the conflict is alarming. It has been marked by widespread ethnic
cleansing, including extrajudicial killings, rape, and massive forced internal and international
displacement. As identified by the latest World Bank Country Economic Memorandum (1) these
factors have had massive impacts on the overall fragility of the country. The appendix to this
paper provides more background and context of the CAR’s civil war.
The contemporary academic literature regarding the economic costs of conflicts originated with
the work of Organski and Kugler (1977) (2) who examined the impact of World Wars I and II (WW
I and WW II) on European countries. Since then, an extensive literature has emerged to quantify
the impact of conflicts (in a broad sense) on economic outcomes across different countries and
periods (3). Different techniques have been adopted to capture the economic costs of conflicts,
including the cost accounting method (4), cross-country regression methods (5), econometrics
on time-series data (6), and econometrics on panel data (7). In the 2000s, a new methodology,
proposed by Abadie and Gardeazabal (2003) (8) called the synthetic control method (SCM
hereafter), emerged, originally to quantify the GDP loss due to local conflict in Spain. More
precisely, they analyzed the economic costs of the conflict, with the armed Basque nationalist
and far left separatist organization Euskadi Ta Askatasuna (ETA), in the Basque Country from 1975
to 1997 by constructing a `synthetic’ region composed of two other Spanish regions (Catalonia
and Madrid) as a comparator or “counterfactual”. Based on this counterfactual, they concluded
that per capita GDP in the Basque region fell by about 10 percent compared to what it would
have been if the conflict had not occurred. More recently, Matta et al. (2017) (9) found a loss of
about 5.7 percent of per capita GDP due to a drop in gross capital formation in Tunisia in the
three years following the Arab Spring; Kešeljević and Spruk (2023) (10) found a loss of about 14
percent of per capita GDP in the Syrian Arab Republic in the nine years following the civil war.
2
Beyond these three papers, the SCM has been used to assess the economic impacts of several
revolts and conflicts, including the Arab Spring in the Arab Republic of Egypt (11) and Libya (12),
the Donbass war in Ukraine (13), the Kurdish separatism in Türkiye (14,15), the Second Intifada
for the Israeli economy (16), and the Yugoslav civil war (17). The SCM is a powerful statistical
technique that has gained prominence in the field of program evaluation and causal inference. It
has numerous advantages and is well adapted to our case of study for the following reasons
developed in Abadie et al. (2010, 2015) (18,19):
• The SCM is suitable for undertaking causal inference. One of the primary advantages
of SCM is its ability to provide causal estimates. SCM enables researchers to assess
the causal impact of a policy, event or treatment by constructing a synthetic control
group that closely resembles the treated unit before the intervention. By comparing
the post-intervention outcomes of the treated unit with the synthetic control,
researchers can estimate the causal effect of the policy or treatment.
• Estimates derived from the SCM are robust even in a small sample setting. The SCM
is particularly valuable in settings where the number of treated and control units are
small, which is typically an inadequate situation for the use of traditional matching or
difference-in-differences (DiD) approaches.
• The SCM considers the pre-intervention dynamics of the treated unit. This implies that
it does not rely on parallel pre-implementation trends like DiD for instance; unlike
other methods that rely on post-intervention data only, SCM constructs a synthetic
control group that captures the pre-intervention trends and characteristics of the
treated unit. This makes the method robust to potential confounding factors and
allows for more accurate causal inferences.
• The SCM enables researchers to estimate the counterfactual scenario by constructing
a synthetic control that represents what would have happened to the treated unit in
the absence of the policy or treatment. This counterfactual estimation is crucial for
evaluating the effectiveness of interventions and guiding policy decisions, or in our
case, to evaluate the impact of the CAR civil war.
• The SCM is a transparent and replicable method as the construction of the synthetic
control group involves clear and systematic processes, allowing other researchers to
reproduce the results and assess the robustness of the findings. This transparency
enhances the credibility and reliability of the estimated causal effects.
In summary, the SCM offers several advantages that make it an indispensable tool for causal
inference and impact evaluation. Specifically, there are three key issues that motivated our
3
preference for the SCM compared to Difference-in-Difference (DiD) approaches. First, a DiD
approach could conflate differences in outcomes associated with the war with differences due
to pre-treatment characteristics associated with war if the control group was not similar enough
to the CAR. The SCM is guaranteed to be at least as similar to CAR as a simple weighted mean of
control states or any one other control country (19). Second, traditional DiD methods assume
unobserved confounders are time constant, whereas the effects of pre-war confounders do not
have to be time constant in the synthetic control method (20). Third, traditional DiD methods
provide asymptotic large-sample inference that is inappropriate given the comparative case
study design, whereas the synthetic control method uses exact inference (21).
Compared to the previous studies, the contributions of this paper are threefold. First, we assess
the costs of conflict in the specific context of the CAR civil war from 2013, mainly through a
descriptive analysis. Second, we take advantage of recent progress in satellite nighttime lights,
or NTL measurements (22)) to bypass the lack of data inherent to low-income countries, or LICs
(like the CAR), and to control for potentially artificially inflated GDP figures due to political
economy reasons (23). Finally, we expand the analysis beyond the potential loss of GDP (and
lights intensity) and explore the potential detrimental impact of the civil war on the sectoral
composition of the economy, and key social indicators (acknowledging the quantitative literature
on the social costs of conflicts is much scarcer (10,24,25)) to establish the diagnostic of a possible
conflict trap.
We recognize however some limitations in the analysis. The choice to select all other LICs for the
potential donor pool to build our counterfactual is explained both for practical and transparency
reasons (see the Methodology section for more details) and for structural reasons (although each
country is unique and has a specific national context, all LICs face the challenges of structural
vulnerability and development trap (26)). These states, however, are also more likely to
experience large-scale social unrest and localized armed conflicts (27). If some of the LICs in the
donor pool are selected ex-post in the synthetic control, while they also suffered consequences
of conflicts during 2013-17, the estimates provided here can be considered conservative.
We did not find any other paper assessing the socio-economic impacts of the CAR civil war using
the SCM, so far. Although SCM has been previously used in the literature to assess the economic
impact of local conflicts and revolts (8–10) and the social and institutional impacts of civil war
(10), we expand the analysis on a variety of key macro-socio indicators, including the innovative
satellite NTL extended time series (22), the sectoral composition of the economy, and a reduced
version of the human asset index (28) to establish the formal diagnostic of a conflict trap (29), or
fragility trap (1). Over the period 2012-17, we find that the civil war led to a significant drop in
GDP per capita (41.6percent), NTL intensity (33.9 percent), industrial production (34 percent),
manufacturing value added (33.7 percent), and human asset index (20.2 percent). These results
are consistent with Abadie and Gardeazabal (2003) and Matta et al. (2017) (8,9), although the
magnitude of the adverse socio-economic impacts of the CAR’s civil war seem significantly worse,
with cumulative GDP per capita loss found to be between 2.9 times and 7.3 times larger than in
4
these two studies. We also highlight an economic loss about 3.1 higher than found in Kešeljević
and Spruk (2023) (10) although the authors also identify dramatic human development and infant
mortality costs of the Syrian civil war. To the best of our knowledge, the estimated cumulative
loss of GDP per capita during the CAR civil war is among the highest in the academic literature,
just below the Libyan case between 2011-14 (12)– which was a mix of revolts following the Arab
Spring, a civil war and an international war– and comparable in magnitude with the local
cumulative loss of GDP found in the Donetsk and Luhansk Oblasts of Ukraine between 2013-16
(13).
It is important to note that we do not discuss here any potential solution for the CAR, including
the potential role of official development aid, structural reforms and financing programs (1,30–
34). The remainder of the paper is structured as follows: The first section presents the
methodology used to estimate the socio-economic impacts of the 2013 civil war in CAR as well
as our identification strategy. The second section elaborates on the key variables used and data.
The third section presents the empirical results. The last section provides some concluding
remarks.
Methodology and identification strategy
To assess the socio-economic impacts of the 2013 civil war in CAR, we consider country-year level
data, with a post-`treatment’ period starting in 2013. We distinguish two phases in the civil war:
the period 2013-14, when the Séléka rebel coalition took over Bangui and a subsequent period
(2015-17) when the conflict has fragmented into several localized skirmishes concentrated
mostly in regions (called locally Préfectures) close to the northern and western borders and in
the East of the country. We do not go beyond 2017 to avoid noisy effects possibly coming from
more recent events (e.g., Wagner group starting to operate in 2018, leading to the cut of budget
support from France and other traditional partners; the COVID-19 lockdowns, the crypto-asset
law of 2022). As the main macro-socio indicators, we focused on GDP per capita, industrial value
added and manufacturing value added (over GDP) from the World Development Indicators (35),
satellite NTL from the DMSP data (22), and a reduced version of the (harmonized) human asset
index.
We use a SCM to estimate the expected levels of our key outcomes in the absence of war and
compare theses to the observed actual series to quantify the war’s impact on CAR’s socio-
economic development during the period 2013-17. This method provides causal estimates and
incorporates pre-intervention dynamics, especially with small-N settings. It has previously been
used to estimate the effects of local conflicts and massive revolts (8,9) although these papers
focused only on the evolution (and demand decomposition) of GDP per capita series after the
events. More recently, the SCM has also been used to assess the social and institutional impacts
of the Syrian civil war (10). We used permutation (or placebo) tests to assess whether our results
could be due to chance.
5
We rely on the SCM initially proposed by Abadie and Gardeazabal (2003) (8), then developed
and proposed by Abadie et al. (2010, 2015) (18,19). The method uses a weighted combination of
unit of observations (i.e., countries) to create a `synthetic’ control CAR, which provides an
estimate of expected socio-economic outcomes of interest (i.e., GDP per capita, satellite NTL,
industrial and manufacturing valued added, human asset index) if the civil war (our `treatment’)
had not occurred. The countries that comprise CAR’s synthetic control are selected by using all
pre-2012 outcomes (36) among a donor pool of comprising all LICs according to the World Bank
classification to avoid pre-selection bias in the composition of the donor pool, with the exception
of Afghanistan, the Syrian Arab Republic and the Republic of Yemen which faced a mix of
interstate and civil armed conflicts, including in the perimeter of their capital cities during the
same period (see Table A1, for more details). Those that are best able to predict the pre-2012
outcome trends in CAR are chosen for the synthetic control group. The expected socio-economic
outcomes for CAR from 2012-17 in the absence of civil war are then compared against the
observed outcomes. A difference between the observed and expected values can therefore be
interpreted as the impact of the civil war on our outcomes of interest.
Formally, the synthetic outcomes are computed as follows:
������������
�������������������������������������������������ℎ;������������ = � ������������������������
������������ ∗
������������������������,������������ ,
������������=1
where ������������������������ and ������������������������������������������������������������ℎ stand for respectively for the outcome of interest in country ������������ and synthetic
∗
CAR, at year ������������, ������������ stands for the vector of countries in the donor pool, and ������������������������ stands to the optimal
weight attached to country ������������. The impact of civil war on CAR’s development outcomes can
therefore be calculated as follows:
�������������������������������������������������ℎ;������������,
������������������������ = ������������������������������������������������;������������ − ������������
for years ≥ 2013, where ������������������������ stands for the post-treatment impact and ������������������������������������������������ stands for the
outcome of interest in CAR.
While traditional statistical inference techniques do not work with this approach, given the small
sample size, a permutation test (also called placebo test) can be used to assess how unusual such
an effect would be if it were due to chance and thus control for the effect size. This permutation
test involves implementing the synthetic control technique for each country in our donor pool,
as though it were the one that had experienced civil war during 2012-17. The estimated effect
for CAR can then be compared to the size of these other effect estimates. Typically, the
permutation test results are compared for states in which pre-war outcome trends are well
predicted by the synthetic control. Results of the test for all countries, in addition to the results
for those states with up to 5 times the mean squared prediction error (MSPE) observed for CAR
are presented. Countries with a poorly matched synthetic control might indeed appear to have
more extreme differences as an artifact of poor prediction.
6
By examining the effects of the CAR civil war on GDP per capita, NTL intensity, and various
economic indicators such as industrial and manufacturing value added, as well as social indicators
represented by the human asset index, it becomes possible to determine whether the observed
shock to GDP is genuine or merely a statistical artifact. Additionally, this analysis helps evaluate
the negative consequences of the civil conflict on crucial macroeconomic factors, ultimately
affecting long-term development.
Data and variables
As a proxy of economic development, we consider the GDP per capita in purchasing power parity
(in constant 2017 international US dollars) from the World Development Indicators (35) from
1990 (the first year available for CAR) up to 2017, as in Matta et al. (2019) (9).
When accounting for satellite NTL intensity, we use the Defense Meteorological Satellite Program
(DMSP) data from the Earth Observation Group (EIG) at Colorado Schools of Mine’s (Ghosh et al.,
2021) (22), as they provide original DMSP-OLS data from 1992 to 2013 but also an extended
version of the DMSP NTL from 2013 to 2021. The regular DMSP NTL series has been discontinued
after 2013 for two mains reasons. First the launch of a new generation of satellites equipped with
more capable sensors, named Visible Infrared Imaging Radiometer Suite (VIIRS), providing more
accurate NTL data. Second the orbital shifts from day/night to dawn/dusk of previously used
DMSP satellites. The shift to NTL data collected by VIIRS prevented the use of NTL data for long
temporal analysis and pushed researchers to produce harmonized data series combining DMSP
and VIIRS data. Ghosh et al. (2021) provide consistent time series for the NTL from 1992 to 2021
by exploiting the constant orbital shift of satellites, which brought back older DMSP satellites
(F15 and F16) into a usable day/night cycle to continue data collection with DMSP satellites. To
have harmonized series for CAR, we select the period 2005-2017 as 2005 saw major structural
reforms for Enerca (the main energy utility company of the CAR) and a policy shift towards
liberalization of the electricity sector. We extract the sum of the NTL within a buffer zone of a 10
km radius around the capital location of each country. The sum of the NTL gives an aggregate
that is a good proxy for electric power output (37) as well as economic activity (38) both at the
national and local level. We use the capital as a reference city for a country rather than the all
country to improve the signal-to-noise ratio of the NTL time series, as LICs tend to have poor
electricity infrastructure, and the NTL data would measure large areas of noisy data. Capitals are
usually the most infrastructure rich and economically dynamic locations (especially in LICs) and
are therefore indicative proxies in the case of economic shocks.
Data on the share of industrial (including construction) and manufacturing value added over GDP
are derived from the World Development Indicators from 2009 (the first year available for CAR)
up to 2017. Due to lack of data the pre-treatment period for these two outcomes is limited, but
the use of these indicators enables us to check consistency in our results. Due to large differences
in sectoral composition among our donor pool of LICs and CAR, we convert the indicators in index
base 100 in 2012 (i.e., just before the first clashes) which enables us to capture sectoral dynamics
after the start of the civil war.
7
Finally, we use a reduced version of the human asset index taken from Feindouno and Goujon
(2019) (28). Their HAI is a simple composite index from 0 (low score) to 100 (best score) which
includes four subcomponents: i) un undernourishment index, ii) the under five years old mortality
index, iii) the adult literacy rate index, and iv) the secondary school enrollment index. We
recompute the HAI by excluding the adult literacy rate index as it is difficult to interpret this
indicator in post-war situation for CAR due to the significant number people registered as
international refugees (744, 000 or 13.3 percent of the population at end of 2022), mostly in
neighboring countries, comprising an unknown proportion of literate and illiterate adults. The
reduced version of HAI is available from 1990 (the first year available for CAR) up to 2014, which
enables us to only capture the short-term impact of the civil war during the rule of Michel
Djotodia as the President of the Republic after the Séléka rebel coalition took over Bangui.
Table A1 lists the countries included in the synthetic controls, along with their weights. It also
provides pre-war and post-war outcome values in CAR for the synthetic controls. Additionally,
this table displays CAR MSPE in the pre-war period for each type of outcome, offering a
comprehensive overview of the synthetic controls' performance.
Empirical results
CAR has faced a persistent lack of economic growth since gaining independence in 1960. The
country’s history of political instability is characterized by a series of coups, which led to a fragile
institutional framework and a decline in overall productivity. This decline is noticeable in the
absence of structural transformation, limited technological progress and innovation, and poor
management of both public and private enterprises. Moreover, the country’s GDP per capita in
2012 was equivalent to the level recorded in 1990, highlighting the stagnation that had plagued
the nation for over two decades (see Figure 1).
In addition, the subsequent civil war in 2013 triggered a devastating divergence in CAR's actual
GDP per capita compared to its synthetic control. This divergence remained massive in 2017 (see
Figure 2). A permutation test reveals that this impact is not only significant but also the most
substantial when compared to other countries, surpassing even the six states (out of 19 in the
donor pool) with MSPE less than or equal to five times that of CAR (see Figure 3). The disparities
in GDP per capita between CAR and its synthetic control from 2012 to 2014 indicate a drastic
collapse in income levels during the fall of Bangui, amounting to approximately USD 490 per
capita. Furthermore, the extended period from 2012 to 2017 witnessed a staggering decline of
approximately USD 503 per capita. These figures represent normalized gaps of 40.6 percent and
41.6 percent, respectively, when compared to the actual GDP per capita in 2012. Consequently,
it is not surprising that CAR is unlikely to reclaim its pre-war economic development level until
the late 2030s (1). In the analysis that follows, we take a closer look at the factors contributing to
this gloomy outlook, which reinforces the plausibility of this scenario.
Figure 1: Income level in CAR, 1990-2017
8
1300
GDP per capita (PPP, constant 17 int. USD)
1200
1100
1000
900
800
700
1990 1992 1994 1996 1998 2000 2002 2004 2006 2008 2010 2012 2014 2016
Year
Source: GDP per capita series are from the World Development Indicators (35), from 1990-2017. Note: The grey bar
indicates the beginning of the civil war up to the departure of Séléka rebel coalition from Bangui (from December
2012 to January 2014). The subsequent period translates into localized conflicts areas in Northwestern and Eastern
parts of the country, outside of major cities.
9
Figure 2: Income level in CAR and its synthetic control, 1990-2017
1400
GDP per capita (PPP, constant 17 int. USD)
1300
1200
1100
1000
900
800
700
1990 1992 1994 1996 1998 2000 2002 2004 2006 2008 2010 2012 2014 2016
Year
Seleka occupation CAR Synthetic CAR
Source: GDP per capita series are from the World Development Indicators (35), from 1990-2017. Note: In upper chart,
the blue (lighter in greyscale) line indicates actual GDP series, and the orange (darker in greyscale) line indicates
synthetic GDP series. The grey bar indicates the beginning of the civil war up to the departure of Séléka rebel coalition
from Bangui (from December 2012 to January 2014). In the lower chart, the dark continuous line indicates the gap
between actual and synthetic GDP per capita series (normalized to 0 in 2012), the dark dash horizontal line indicates
a null gap, and the two red vertical lines indicates the period of Séléka occupation in Bangui (2012, excluded, up to
2014, excluded). The subsequent period translates into localized conflicts areas in Northwestern and Eastern parts of
the country, outside of major cities.
10
Figure 3: Differences between income level for CAR and control countries and their respective
synthetic controls, 1990-2017
Source: GDP per capita series are from the World Development Indicators (35), from 1990-17. Note: The dark dash
line indicates the gap between actual and synthetic GDP per capita series for CAR and the gray lines indicates the
gap between actual and synthetic GDP per capita series for control countries (normalized to 0 in 2012). The dark dash
horizontal line indicates a null gap, and the two red vertical lines indicates the period of Séléka occupation in Bangui
(2012, excluded, up to 2014, excluded). The subsequent period translates into localized conflicts areas in
Northwestern and Eastern parts of the country, outside of major cities.
11
Martínez (2022) (23) finds that authoritarian regimes are prone to overstate yearly self-reported
GDP growth and using satellite NTL intensity helps to assess the reality of the economic activity.
Although GDP series are checked by international institutions, it is possible that the relatively
good economic performance observed under former president François Bozizé was
overestimated (GDP per capita grew by 25.5 percent in 2012, from USD 963 per capita in 2003 to
USD 1209 per capita in 2012) while the country was under scrutiny for traversing a trajectory
toward authoritarianism, as suggested by its Polity2 score (i.e., a common autocratic-democratic
score produced by Ted Robert Gurr and Monty G. Marshall and released by the Center of
Systemic Peace) of -1 at the time, corresponding to a closed anocracy (39). Consequently, the
subsequent drastic correction of GDP from 2013 could be partly artificial. Accordingly, we check
the dynamics of NTL intensity in the perimeter of Bangui following the first clashes, for robustness
check. Although we observe a divergence of NTL intensity over the period 2011-12 (before the
start of the civil war) between actual CAR series and its synthetic control, the divergence stems
from the dynamism of synthetic CAR, while the intensity of the real lights stagnated for CAR. For
subsequent years the growing divergence also comes from the crash of actual luminosity in the
area of Bangui, as attested by the artificial gap closure in 2012 to check for post-war divergence
(Figure 4). The permutation test shows that this divergence is larger than both the average and
median of all other donor countries. The divergence is also larger than the average and median
of the eight states (out of 24 in the donor pool) with MSPE less than or equal to five times than
that of CAR (see Figure 5). Only Asmara (Eritrea), and Bissau (Guinea-Bissau) are doing worse
over the whole period in terms of NTL dynamics. The differences in luminosity between CAR and
its synthetic control indicate a drop in light intensity of approximately 22.5 percent from 2012-
14 and a drop of 33.9 percent for the extended period 2012-17. The growing gap over time
suggests that power production and grid transmission capacity in Bangui were durably impacted,
indicating a medium-to-long-term deterioration of essential network infrastructure for shared
prosperity and poverty reduction (40).
12
Figure 4: NTL intensity in CAR and its synthetic control, 2005-17
5500
5000
Nighttime lights intensity 4500
4000
3500
3000
2500
2000
2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017
Year
Seleka occupation CAR Synthetic CAR
5500
5000
Nighttime lights intensity
4500
4000
3500
3000
2500
2000
2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017
Year
Seleka occupation CAR Synthetic CAR
13
Source: Satellite NTL series are from Gosh et al. (2021) (22) and authors’ estimate. Note: In upper and middle charts,
the blue (lighter in greyscale) line indicates actual light series, and the orange (darker in greyscale) line indicates
synthetic light series. The grey bar indicates the beginning of the civil war up to the departure of Séléka rebel coalition
from Bangui (from December 2012 to January 2014). In middle and lower charts, we normalize the gap between CAR
and its synthetic control for 2012 and subsequent years, for consistency. In the lower chart, the dark continuous line
indicates the gap between actual and synthetic NTL intensity (normalized to 0 in 2012), the dark dash horizontal line
indicates a null gap, and the two red vertical lines indicates the period of Séléka occupation in Bangui (2012, excluded,
up to 2014, excluded). The subsequent period translates into localized conflicts areas in Northwestern and Eastern
parts of the country, outside of major cities.
14
Figure 5: Differences between NTL intensity for CAR and control countries and their
respective synthetic controls, 2005-17
No pre-NTL's MSPE five times higher than CAR's
4000
Gap Nighttime light intensity
2000
0
-2000
-4000
2005 06 07 08 09 10 11 12 13 14 15 16 17 18
Seleka occupation
Year
Source: Satellite NTL series are derived from Gosh et al. (2021) (22), and cover the period 2005-17. Note: The dark
dash line indicates the gap between actual and synthetic light series for CAR and the gray lines indicates the gap
between actual and synthetic light series for control countries (normalized to 0 in 2012). The dark dash horizontal
line indicates a null gap, and the two red vertical lines indicates the period of Séléka occupation in Bangui (2012,
excluded, up to 2014, excluded). The subsequent period translates into localized conflicts areas in Northwestern and
Eastern parts of the country, outside of major cities.
15
Despite a pre-treatment period starting only in 2009, this paper shows the existence of a massive
and sharp divergence of the share of actual industrial and manufacturing value added over GDP
from synthetic controls between 2013 throughout 2017 (see Figures 5 and 6). This result
underscores the fact that structural transformation remains a major challenge for the economic
development of African countries (41) and especially for resource-rich economies (42),
preventing them from escaping the natural resource curse (43). Moreover, results from the
permutation tests indicate the existence of a significant drop in both the industrial (-23.9 percent)
and manufacturing (-35.5 percent) value added over GDP compared to other countries during
the Séléka rebels’ occupation of Bangui in 2013-14. Even when keeping the closest comparator
countries with CAR in terms of MSPE, only Sierra Leone (due to the Ebola crisis between 2014-
16) and Rwanda (due to the boom of ecotourism and more generally a service-oriented economy
on labor intensive activities (44)) are doing worse in terms of manufacturing dynamics if we
extend the analysis up to 2017 (see Figures 7 and 8). The differences between CAR and its
synthetic controls suggest that the shares of industrial and manufacturing value over GDP
declined by approximately 34.1 percent and 33.7 percent respectively between 2012 and 2017,
probably due to the destruction and flight of capital (45). Although closely interconnected, these
two results do not induce the same macro issues:
i) The collapse of the industrial sector as a whole (including construction activity) is in
line with the previous results on light intensity and indicates the degradation of the
ability to rebuilt and maintain essential public and network infrastructure to bridge
the infrastructure gap (46).
ii) The relative decline of manufacturing in the sectoral composition of the CAR’s
economy indicates that the sector failed to advance towards more sophisticated
products (47).
iii) Altogether, these results reassess the vicious cycle of the lack of state building, civil
conflicts over control of political power and natural resource wealth, and the poverty
trap (48).
16
Figure 6: Industrial value added (over GDP) in CAR and its synthetic control, 2009-17
130
Industry value added (base 100 =2012)
120
110
100
90
80
70
60
2009 2010 2011 2012 2013 2014 2015 2016 2017
Year
Seleka occupation CAR Synthetic CAR
20
Gap Industry value added (index 100 =2012)
10
0
-10
-20
-30
-40
-50
2009 10 11 12 13 14 15 16 17
Seleka occupation
Year
Source: Industrial value-added series are from the World Development Indicators (35), from 2009-17. Note: In upper
chart, the blue (lighter in greyscale) line indicates actual industrial value-added series, and the orange (darker in
greyscale) line indicates industrial value-added synthetic series. The grey bar indicates the beginning of the civil war
up to the departure of Séléka rebel coalition from Bangui (from December 2012 to January 2014). In the lower chart,
the dark continuous line indicates the gap between actual and synthetic industrial value-added series (normalized to
0 in 2012), the dark dash horizontal line indicates a null gap, and the two red vertical lines indicates the period of
17
Séléka occupation in Bangui (2012, excluded, up to 2014, excluded). The subsequent period translates into localized
conflicts areas in Northwestern and Eastern parts of the country, outside of major cities.
Figure 7: Manufacturing value added (percent of GDP) in CAR and its synthetic control, 2009-
17
130
Manufacturing value added (base 100
120
110
100
=2012)
90
80
70
60
2009 2010 2011 2012 2013 2014 2015 2016 2017
Year
Seleka occupation CAR Synthetic CAR
20
Gap Manufacturing value added (index 100 =2012)
10
0
-10
-20
-30
-40
-50
2009 10 11 12 13 14 15 16 17
Seleka occupation
Year
Source: Manufacturing value-added series are from the World Development Indicators (35), from 2009-17. Note: In
upper chart, the blue (lighter in greyscale) line indicates actual manufacturing value-added series, and the orange
(darker in greyscale) line indicates manufacturing value-added synthetic series. The grey bar indicates the beginning
of the civil war up to the departure of Séléka rebel coalition from Bangui (from December 2012 to January 2014). In
18
the lower chart, the dark continuous line indicates the gap between actual and synthetic industrial value-added series
(normalized to 0 in 2012), the dark dash horizontal line indicates a null gap, and the two red vertical lines indicates
the period of Séléka occupation in Bangui (2012, excluded, up to 2014, excluded). The subsequent period translates
into localized conflicts areas in Northwestern and Eastern parts of the country, outside of major cities.
Figure 8: Differences between industrial value added (percent of GDP) for CAR and control
countries and their respective synthetic controls, 2009-17
We keep all placebo countries
150
Gap Industry value added (index 100 =2012)
100
50
0
-50
-100
2009 10 11 12 13 14 15 16 17
Seleka occupation
Year
No pre-Industrial's MSPE five times higher than CAR's
150
Gap Industry value added (index 100 =2012)
100
50
0
-50
2009 10 11 12 13 14 15 16 17
Seleka occupation
Year
Source: Industrial value-added series are from the World Development Indicators (35), from 2009-17. Note: The dark
dash line indicates the gap between actual and synthetic light series for CAR and the gray lines indicates the gap
19
between actual and synthetic light series for control countries (normalized to 0 in 2012). The dark dash horizontal
line indicates a null gap, and the two red vertical lines indicates the period of Séléka occupation in Bangui (2012,
excluded, up to 2014, excluded). The subsequent period translates into localized conflicts areas in Northwestern and
Eastern parts of the country, outside of major cities.
Figure 9: Differences between manufacturing value added (percent of GDP) for CAR and
control countries and their respective synthetic controls, 2009-17
We keep all placebo countries
Gap Manufacturing value added (index 100 =2012)
100
50
0
-50
-100
2009 10 11 12 13 14 15 16 17
Seleka occupation
Year
No pre-Manufacturing's MSPE five times higher than CAR's
Gap Manufacturing value added (index 100 =2012)
100
50
0
-50
-100
2009 10 11 12 13 14 15 16 17
Seleka occupation
Year
Source: Manufacturing value-added series are from the World Development Indicators (35) , from 2009-17. Note:
The dark dash line indicates the gap between actual and synthetic light series for CAR and the gray lines indicates
20
the gap between actual and synthetic light series for control countries (normalized to 0 in 2012). The dark dash
horizontal line indicates a null gap, and the two red vertical lines indicates the period of Séléka occupation in Bangui
(2012, excluded, up to 2014, excluded). The subsequent period translates into localized conflicts areas in
Northwestern and Eastern parts of the country, outside of major cities.
With regard to the conflict-development nexus, the direction of causality is unclear (49), as we
observe a drop in CAR’s human asset index (HAI, hereafter) between 2011 and 2012, before the
beginning of the civil war (see Figure 9). After artificially closing the gap between CAR's actual
HAI and its synthetic control for 2012, the permutation tests confirm the significant drop in 2013-
14; this drop holds true even for the eight states out of the total 23 in the donor pool, where the
MSPE was less than or equal to five times that of CAR (see Figure 10). The time span for the
reduced HAI is limited to 1990-2014, so we only observe a short-term (but very sharp) drop of
20.2 percentage points compared to 2012 in terms of human development, during the first phase
of the war. In other words, even if the causality is uncertain, the CAR civil war is correlated with
a sharp aggravation in two subcomponents of the reduced HAI, namely the general
undernourishment of the population and the decline of secondary school enrollment. The latter
finding is in line with the recent literature on Cameroon’s Anglophone conflict (50). This is
extremely concerning when considering that the decline in general living conditions, specifically
in terms of food access (first dimension of the reduced HAI), may indicate the possibility of state
collapse in the near future (51). Additionally, the decrease in access to secondary schooling
(second dimension of the reduced HAI) not only hampers the country's economic recovery, but
also perpetuates informal, subsistence-level activities and provides a potential pool for
recruitment by armed groups (1), further exacerbated by the additional challenge of early
pregnancies (52).
Figure 10: Reduced HAI in CAR and its synthetic control, 2005-17
35
Reduced Human Asset Index (0-100)
30
25
20
15
10
5
1990
1991
1992
1993
1994
1995
1996
1997
1998
1999
2000
2001
2002
2003
2004
2005
2006
2007
2008
2009
2010
2011
2012
2013
2014
Year
Seleka occupation CAR Synthetic CAR
21
35
Reduced Human Asset Index (0-100)
30
25
20
15
10
5
1990
1991
1992
1993
1994
1995
1996
1997
1998
1999
2000
2001
2002
2003
2004
2005
2006
2007
2008
2009
2010
2011
2012
2013
2014
Year
Seleka occupation CAR Synthetic CAR
10
Gap Reduced Human Asset Index
5
0
-5
-10
1990 92 94 96 98 2000 02 04 06 08 10 12 14
Seleka Occupation
Year
Source: The reduced HAI series are derived from Feindouno and Goujon (2019) (28), from 1990-2014. Note: In upper
and middle charts, the blue (lighter in greyscale) line indicates actual light series, and the orange (darker in greyscale)
line indicates synthetic light series. The grey bar indicates the beginning of the civil war up to the departure of Séléka
rebel coalition from Bangui (from December 2012 to January 2014). In middle and lower charts, we normalize the
gap between CAR and its synthetic control for 2012 and subsequent years, for consistency. In the lower chart, the
dark continuous line indicates the gap between actual and synthetic GDP per capita series (normalized to 0 in 2012),
the dark dash horizontal line indicates a null gap, and the two red vertical lines indicates the period of Séléka
occupation in Bangui (2012, excluded, up to 2014, excluded).
22
Figure 11: Differences between HAI for CAR and control countries and their respective
synthetic controls., 2009-17
We keep all placebo countries
20
15
Gap Reduced human asset index
10
5
0
-5
-10
-15
-20
1990 92 94 96 98 2000 02 04 06 08 10 12 14
Seleka Occupation
Year
No pre-HAI's MSPE five times higher than CAR's
10
Gap Reduced human asset index
5
0
-5
-10
1990 92 94 96 98 2000 02 04 06 08 10 12 14
Seleka Occupation
Year
Source: The reduced HAI series are derived from Feindouno and Goujon (2019) (28), from 1990-2014. Note: The dark
dash line indicates the gap between actual and synthetic light series for CAR and the gray lines indicates the gap
between actual and synthetic light series for control countries (normalized to 0 in 2012). The dark dash horizontal
23
line indicates a null gap, and the two red vertical lines indicates the period of Séléka occupation in Bangui (2012,
excluded, up to 2014, excluded).
Conclusion
We found overall evidence that the civil war in the CAR led to a massive drop in living standards
proxied by the GDP per capita from 2013 to 2017, and the results are confirmed with NTL
dynamics in the area of Bangui. Maybe even more worrying, our results on NTL also suggest a
long-lasting inability to ensure electricity provision, an essential public service, which is also
necessary for the emergence of the formal private sector. We also find a large drop in
industrial/manufacturing activities in the post-war sectoral composition of the economy and a
significant degradation of human capital, especially in the dimensions of undernourishment and
secondary school enrollment. The persistence of human capital losses is confirmed as the CAR
ranks at the bottom of the human capital and development indices (it was 188th out of 191
countries in 2022) (53).
Quantitatively, we found that the civil war led to a significant drop in GDP per capita (41.6
percent), NTL intensity (33.9 percent), industrial production (34.1 percent), manufacturing value
added (33.7 percent), and the human asset index (20.2 percent) over the period 2012-17. These
sharp declines in the four key socio-economic indicators are quite similar in magnitude, ranging
from 34 to 44 percent. The cumulative loss of GDP per capita is found to be about 3 times larger
than the effect of the civil war in the Syrian Arab Republic highlighted by Kešeljević and Spruk
(2023) (10), 4.2 times larger than the effect of ETA terrorism and direct actions in the Basque
Country highlighted by Abadie and Gardeazabal (2003) (8), and 7.3 times larger than the effect
of the Arab Spring in Tunisia highlighted by Matta et al. (2017) (9). The lower decline of NTL
compared with that of per capita GDP may also suggest overestimated GDP series under the
presidency of François Bozizé (23). Furthermore, this paper emphasizes that there are no signs
suggesting a positive trajectory for the CAR to regain its pre-war living standards by 2030. The
ongoing conflicts persisting as of early 2023, combined with the lack of any indications of
economic and social recovery from 2017, further reinforce this lack of progress (1). Overall, we
identify a vicious conflict-development cycle in the CAR.
We recognize however some limitations in the analysis. The choice to select all other LICs for the
potential donor pool is explained both for practical and transparency reasons (we use the World
Bank classification) and for structural reasons (although each country is unique and has a specific
national context, all LICs face the challenges of structural vulnerability and development trap
(26)). These states, however, are also likely to experience large social unrest and localized armed
conflicts(27). If some of the LICs in the donor pool are selected ex-post in the synthetic control,
while they also suffered consequences of conflicts during 2013-17, the estimates provided here
can be considered conservative. That is the reason we automatically discarded Afghanistan, the
Syrian Arab Republic, and the Republic of Yemen from the donor pool, as they also faced
generalized armed conflicts, including within their capital cities, during the period of interest. In
24
addition, LICs also suffer from a general lack of regular data collection, explaining the constraints
on timespan and data availability for our key macro-indicators.
Future studies might further explore i) the complex interaction between income shocks and civil
conflicts (54), ii) a more global overview of the socio-economic impact of war and conflicts, and
iii) potential solutions for collapsed states and their population.
25
References
1. World Bank. From Fragility to Accelerated and Inclusive Growth: A Country Economic Memorandum
for the Central African Republic. The World Bank Group; 2022.
2. Organski AFK, Kugler J. The Costs of Major Wars: The Phoenix Factor. Am Polit Sci Rev.
1977;71(4):1347–66.
3. Gardeazabal J. Methods for Measuring Aggregate Costs Of Conflict, in Garfinkel, M. R. and Skaperdas,
S. (eds). In: The Oxford Handbook of the Economics of Peace and Conflict. Oxford University Press;
2012.
4. Arunatilake N, Jayasuriya S, Kelegama S. The Economic Cost of the War in Sri Lanka. World Dev.
2001;29(9):1483–500.
5. Alesina A, Perotti R. Income Distribution, Political Instability, and Investment. Eur Econ Rev.
1996;40(6):1203–28.
6. Enders W, Sandler T. Causality between Transnational Terrorism and Tourism: The Case of Spain.
Terrorism. 1991;14(1):49–58.
7. Blomberg B, Hess G, Orphanides A. The Macroeconomic Consequences of Terrorism. J Monet Econ.
2004;51(5):1007–32.
8. Abadie A, Gardeazabal J. The Economic Costs of Conflict: A Case Study of the Basque Country. Am
Econ Rev. 2003;93(1):113–32.
9. Matta S, Appleton S, Bleaney M. The Impact of the Arab Spring on the Tunisian Economy. World Bank
Econ Rev. 2019;33(1):231–58.
10. Kešeljević A, Spruk R. Estimating the Effects of Syrian Civil War. Empir Econ. 2023;forthcoming.
11. Echevarría C A, García-Enríquez J. The Economic Cost of the Arab Spring: the Case of the Egyptian
Revolution. Empir Econ. 2020;59(3):1453–77.
12. Echevarría C A, García-Enríquez J. The Economic Consequences of the Libyan Spring: A Synthetic
Control Analysis. Def Peace Econ. 2019;30(5):592–608.
13. Bluszcz J, Valente M. The Economic Costs of Hybrid Wars: The Case of Ukraine. Def Peace Econ.
2022;33(1):1–25.
14. Bilgel F, Karahasan BC. The Economic Costs of Separatist Terrorism in Turkey. J Confl Resolut.
2017;61(2):457–79.
15. Bilgel F, Karahasan BC. Thirty Years of Conflict and Economic Growth in Turkey: A Synthetic Control
Approach. Def Peace Econ. 2019;30(5):609–31.
16. Horiuchi Y, Mayerson A. The Opportunity Cost of Conflict: Statistically Comparing Israel and Synthetic
Israel. Polit Sci Res Methods. 2015;3(3):609–18.
26
17. Kešeljević A, Spruk R. Long-Term Effects of Yugoslav War. Def Peace Econ. 2023;34(4):410–36.
18. Abadie A, Diamond A, Hainmueller J. Synthetic Control Methods for Comparative Case Studies:
Estimating the Effect of California’s Tobacco Control Program. J Am Stat Assoc. 2010;105(490):493–
505.
19. Abadie A, Diamond A, Hainmueller J. Comparative Politics and the Synthetic Control Method. Am J
Polit Sci. 2015;59(2):495–510.
20. Kreif N, Grieve R, Hangartner D, Turner AJ, Nikolova S, Sutton M. Examination of the Synthetic Control
Method for Evaluating Health Policies with Multiple Treated Units. Health Econ. 2016;25(12):1514–
28.
21. Bertrand M, Duflo E, Mullainathan S. How Much Should We Trust Differences-In-Differences
Estimates? Q J Econ. 2004;119(1):249–75.
22. Gosh T, Baugh K, Elvidge C, Zhizhin M, Poyda A, Hsu FC. Extending the DMSP Nighttime Lights Time
Series beyond 2013. Remote Sens. 2021;13(24):5004.
23. Martínez L. How Much Should We Trust the Dictator’s GDP Growth Estimates? J Polit Econ.
2022;130(2):2731–69.
24. Hoeffler A, Reynal-Querol M. Measuring the Costs of Conflict. Cent Study Afr Econ Work Pap. 2003;
25. Islam A, Ouch C, Smyth R, Wang LC. The Long-Term Effects of Civil Conflicts on Education, Earnings,
and Fertility: Evidence from Cambodia. J Comp Econ. 2016;44(3):800–20.
26. Guillaumont P. Caught in a Trap, Identifying the Least Developed Countries. Economica; 2009.
27. Laville C. Les Causes des Conflits Civils : Avancées Récentes à l’Aide de Données Désagrégées. Rev
Francaise Econ. 2019;34(3):123–65.
28. Feindouno S, Goujon M. Human Assets Index: Insights from a Retrospective Series Analysis. Soc Indic
Res. 2019;141(3):959–84.
29. Collier P, Elliott VL, Hegre H, Reynal-Querol M, Sambanis N. Breaking the Conflict Trap. Civil War and
Development Policy. The World Bank Group and Oxford University Press; 2003.
30. Moyo D. Dead Aid. Why Aid Is Not Working and How There Is a Better Way for Africa. Farrar, Straus
and Giroux; 2010.
31. Doucouliagos H, Paldam M. The Robust Result in Meta-analysis of Aid Effectiveness: A Response to
Mekasha and Tarp. J Dev Stud. 2013;49(4):584–7.
32. Mekasha TJ, Tarp F. A Meta-Analysis of Aid Effectiveness: Revisiting the Evidence. Polit Gov.
2019;7(2):5–28.
33. Balima H, Sokolova A. IMF Programs and Economic Growth: A Meta-Analysis. J Dev Econ.
2021;153(November):102741.
27
34. Mandon P, Woldemichael MT. Has Chinese Aid Benefited Recipient Countries? Evidence from a Meta-
Regression Analysis. World Dev. 2023;166(June):106211.
35. World Bank. World Development Indicators [Last updated 12/22/2022]. 2022; Available from:
https://databank.worldbank.org/source/world-development-indicators
36. Kol A, Klößner S, Pfeifer G, Schieler M. Standard Synthetic Control Methods: The Case of Using All
Preintervention Outcomes Together With Covariates. J Bus Econ Stat. 2022;40(3):1362–76.
37. Min B, Gaba KM, Fall OF, Agalassou A. Detection of Rural Electrification in Africa using DMSP-OLS
Night Lights Imagery. Int J Remote Sens. 2013;34(22):8118–41.
38. Hu Y, Yao J. Illuminating Economic Growth. J Econom. 2022;228(2):359–78.
39. Center for Systematic Peace. Polity5 Annual Time-Series, 1946-2018. 2022; Available from:
https://www.systemicpeace.org/csprandd.html
40. Afeikhena J. Infrastructure, Economic Growth and Poverty Reduction in Africa. J Infrastruct Dev.
2011;3(2):127–51.
41. Mc Millan M, Rodrik D, Verdusco-Gallo I. Globalization, Structural Change, and Productivity Growth,
with an Update on Africa. World Dev. 2014;63:11–32.
42. Venables A. Using Natural Resources for Development: Why Has It Proven So Difficult? J Econ
Perspect. 2016;30(1):161–84.
43. Cust J, Mandon P. Nonrenewable Natural Capital and Human Capital Distortions: Impact on
Accumulation, Gender, and the Public Sector. In: The Changing Wealth of Nations 2021: Managing
Assets for the Future. World Bank; 2021.
44. Nielsen H, Spenceley A. The Success of Tourism in Rwanda - Gorillas and More. World Dev Rep Backgr
Pap. 2011;
45. Collier P. On the Economic Consequences of Civil War. Oxf Econ Pap. 1999;51(1):168–83.
46. Calderón C, Cantu C, Chuhan-Pole P. Infrastructure Development in Sub-Saharan Africa: A Scorecard.
World Bank Policy Res Work Pap No 8425. 2018;
47. Hidalgo C, Klinger B, Barabasi AL, Hausmann R. The Product Space Conditions the Development of
Nations. Science. 2007;317(5837):482–7.
48. Besley T, Persson T. State Capacity, Conflict, and Development. Econometrica. 2010;78(1):1–34.
49. Ray D, Esteban J. Conflict and Development. Annu Rev Econ. 2017;9(1):263–93.
50. Galindo-Silva H, Tchuente F. Armed Conflict and Early Human Capital Accumulation: Evidence from
Cameroon’s Anglophone Conflict. Cornell Univ ArXiv. 2023;
51. Redl C, Hlatshwayo S. Forecasting Social Unrest: A Machine Learning Approach. IMF Work Pap No
2021263. 2021;
28
52. Duflo E, Dupas P, Kremer M. Education, HIV, and Early Fertility: Experimental Evidence from Kenya.
Am Econ Rev. 2015;105(9):2757–2597.
53. UN Development Programme. Human Development Report 2021-2022. Uncertain Times, Unsettled
Lives: Shaping our Future in a Transforming World. United Nations; 2022.
54. Laville C, Mandon P. Internal Conflicts and Shocks. A Narrative Meta-Analysis. World Bank Policy Res
Work Pap No 10315. 2023;
55. Sıradağ A. Explaining the Conflict in Central African Republic: Causes and Dynamics. Epiphany J
Transdiscipl Stud. 2016;9(3):86–103.
56. Stanyard J, Vircoulon T, Rademeyer J. The Grey Zone: Russia’s Military, Mercenary and Criminal
Engagement in Africa. Global Initiative Against Transnational Organized Crime; 2023.
57. Stapleton T. Central African Republic: Coups, Mutinies, and Civil War. In: Oxford Research
Encyclopedia of Politics. Oxford University Press; 2019.
29
Appendix
Background of the CAR civil war
The Central African Republic (CAR) civil war is a conflict that officially began in December 2012
and is still ongoing close to the northern and western borders and in the eastern part of the
country, despite the signature of a peace agreement (the Political Agreement for Peace and
Stability, or APPR in the French acronym) between the government and 14 armed groups in
February 2019. The conflict is characterized by violence between various armed groups, including
government forces, rebel groups, and ethnic militias. The roots of the conflict can be traced back
to the coup that brought President François Bozizé to power in 2003. Bozizé's presidency was
characterized by political instability, lack of achievements in terms of development, authoritarian
stance (ruled by decrees) and physical abuses on the opposition (55), and, in 2012, a coalition of
rebel groups known as Séléka launched an armed rebellion against the government.
The Séléka rebels, who were mainly from the northeast of the country, accused Bozizé's
government of neglecting their region and favoring his own ethnic group, the Gbaya. The rebels
quickly gained control of much of the country, and in March 2013 they overthrew Bozizé's
government. In response to the Séléka numerous exactions, a Christian militia known as the Anti-
Balaka was formed. The Anti-Balaka targeted Muslim civilians, leading to reprisal attacks by
Muslim militias and the displacement of hundreds of thousands of people. The conflict continued
to escalate, with various armed groups vying for control of territory and resources (forestry
concessions and extractive resources).
In 2014, the United Nations Security Council authorized the deployment of a peacekeeping
mission to the CAR, known as MINUSCA. The mission was tasked with protecting civilians,
supporting political dialogue, and helping to disarm armed groups. From 2018, the Russian
paramilitary group Wagner started to operate in the country (56) with the active support of the
newly elected president Faustin-Archange Touadéra and his government, and Rwanda also
deployed what it called “force protection troops’’ in 2020 under a bilateral agreement on
defense. Thanks to these external forces, the government regained control in Bangui region and
major cities, but despite the signature of a peace agreement in 2019, the conflict is still ongoing
at a regional level.
Contrary to previous coups and the Bush war of 2004-07, CAR’s civil war is exceptional in terms
of duration (10 years and still ongoing), extension (the capital city Bangui was overtaken between
2013-14 and the two main corridors, the Ubangi river and Douala-Bangui road axis were severely
disturbed in the first step of the war), fragmentation (regarding the number of armed groups
involved and the de facto partition of the country into a network of fiefdoms) and intensity
(widespread ethnical cleansing, including extrajudicial killings, rape, and massive forced internal
and international displacement) (57).
30
Table A1. Weights assigned to countries for CAR’s outcomes of interest
Potential donnor pool (SCM GDP per capita (PPP, constant 17 int. USD) Nighttime lights (NTL) intensity** Ind. value added (over GDP), index 100= 2012 Manuf. value added (over GDP), index 100= 2012 Tranformed human asset index (HAI)
weight in parenthesis) 1990-2017 2005-2017 2009-2017 2009-2017 1990-2014
Afghanistan No [automatically discarded] No [automatically discarded] No [automatically discarded] No [automatically discarded] No [automatically discarded]
Burkina Faso Yes (0%) Yes (0%) Yes (2.5%) Yes (0%) Yes (0%)
Burundi Yes (19.1%) Yes (10.5%) Yes (11.4%) No [unbalanced series] Yes (42.6%)
Chad Yes (0%) Yes (0%) Yes (3.4%) Yes (33.4%) Yes (0%)
Congo, Dem. Rep. Yes (5.7%) Yes (0%) Yes (2.4%) Yes (0%) Yes (0%)
Eritrea No [unbalanced series] Yes (0%) No [unbalanced series] No [unbalanced series] Yes (0%)
Ethiopia Yes (15.7%) Yes (7.6%) Yes (3.2%) Yes (0%) Yes (0%)
Gambia, The Yes (0%) Yes (2.1%) Yes (29.3%) Yes (0%) Yes (0%)
Guinea Yes (0%) Yes (0%) Yes (3.1%) Yes (0%) Yes (0%)
Guinea-Bissau Yes (0%) Yes (0%) Yes (2.4%) Yes (0%) Yes (0%)
Korea, Dem. People's Rep. No [unbalanced series] Yes (1.7%) No [unbalanced series] No [unbalanced series] No [unbalanced series]
Liberia No [unbalanced series] Yes (0%) Yes (5.1%) No [unbalanced series] Yes (0%)
Madagascar Yes (0%) Yes (0%) Yes (2.9%) Yes (0%) Yes (0%)
Malawi Yes (6.8%) Yes (2.1%) Yes (3.1%) Yes (0%) Yes (0%)
Mali Yes (0%) Yes (0%) Yes (2.3%) Yes (0%) Yes (0%)
Mozambique Yes (0%) Yes (0%) Yes (2.9%) Yes (0%) Yes (0%)
Niger Yes (36%) Yes (0%) Yes (2.5%) Yes (66.6%) Yes (16.6%)
Rwanda Yes (0%) Yes (0%) Yes (5.8%) Yes (0%) Yes (2.9%)
Sierra Leone Yes (0%) Yes (0%) Yes (1.6%) Yes (0%) Yes (0%)
Somalia No [unbalanced series] Yes (76%) No [unbalanced series] No [unbalanced series] Yes (30.2%)
South Sudan No [unbalanced series] Yes (0%) No [unbalanced series] No [unbalanced series] No [unbalanced series]
Sudan Yes (0%) Yes (0%) Yes (2.8%) No [unbalanced series] Yes (0%)
Syrian Arab Republic No [automatically discarded] No [automatically discarded] No [automatically discarded] No [automatically discarded] No [automatically discarded]
Togo Yes (16.7%) Yes (0%) Yes (5.2%) Yes (0%) Yes (0%)
Uganda Yes (0%) Yes (0%) Yes (3.3%) Yes (0%) Yes (0%)
Yemen, Rep. No [automatically discarded] No [automatically discarded] No [automatically discarded] No [automatically discarded] No [automatically discarded]
Zambia Yes (0%) Yes (0%) Yes (4.9%) Yes (0%) Yes (7.8%)
CAR MSPE 537.4 8,102.6 4.71E-22 2.8 0.2
CAR actual pre-war* 1,061.9 3,539.4 93.8 86.0 14.8
CAR synth. pre-war value 1,059.2 3,539.9 93.8 86.6 15.0
CAR actual post-war 796.5 3,362.9 78.6 81.3 24.1
CAR synth. post-war 1,299.6 4,822.7 112.7 115.0 29.5
Source: Authors’ construction. Notes: The list of low-income countries (LICs) is taken from the World Bank website
(https://datahelpdesk.worldbank.org/knowledgebase/articles/906519-world-bank-country-and-lending-groups) for fiscal year 2023 (FY23) or calendar year 2021. GDP per capita,
industrial value added, and manufacturing value added (over GDP) are taken from the World Development Indicators (35), satellite nighttime lights (NTL) intensity is derived from
the DMSP data (22), and the reduced version of the (harmonized) human asset index is derived from Feindouno and Goujon (2019) (28). *: Adjusted SCM values for NTL and reduced
HAI. **: For NTL intensity we focus on a 10km radius around capital cities of each country (enabling us to include every country in the donor pool, even South Sudan before country’s
independence, through Juba’s light intensity).
31