Fighting African Capital Flight: Trajectories, Dynamics and Tendencies

An April 2015 World Bank report on attainment of the Millennium Development Goal (MDG) extreme poverty target has revealed that extreme poverty has been decreasing in all regions of the world with the exception of sub-Saharan Africa (SSA), in spite of the sub-region enjoying more than two decades of growth resurgence. This study builds on a critique of Piketty’s ‘capital in the 21st century’ and recent methodological innovations on reverse Solow-Swan to review empirics on the adoption of common policy initiatives against a cause of extreme poverty in SSA: capital flight. The richness of the dataset enables the derivation of 14 fundamental characteristics of African capital flight based on income-levels, legal origins, natural resources, political stability, regional proximity and religious domination. The main finding reveals that regardless of fundamental characteristic, from a projection date of 2010, a genuine timeframe for harmonizing policies is between 2016 and 2023. In other words, the beginning of the post-2015 agenda on sustainable development goals coincides with the timeframe for common capital flight policies.


Introduction
There are at least four reasons for reviewing Asongu (2014a) on 'Fighting African Capital Flight: Empirics on Benchmarking Policy Harmonization': (i) recent disturbing extreme poverty trends in Sub-Saharan Africa (SSA); (ii) a critique of Piketty's 'capital in the 21 st century' that builds on capital flight to elucidate the sub-region's extreme poverty tragedy; (iii) a recent methodological innovation for common policy initiatives based on negative macroeconomic and institutional signals (reverse Solow-Swan) and (iv) the imperative to account for more fundamental characteristics of the sub-region's development in order to avail room for robustness and more policy implications.
First, an April 2015 World Bank report on attainment of the Millennium Development Goal (MDG) extreme poverty target has revealed that extreme poverty has been decreasing in all regions of the world, with the exception of Africa, where 45% of countries in SSA are substantially off-track from achieving the MDG extreme poverty target (World Bank, 2015). As documented in recent literature (Efobi et al., 2018;Asongu & Kodila-Tedika, 2018;Tchamyou, 2019aTchamyou, , 2019bTchamyou et al., 2019;Asongu & le Roux, 2017, whereas extreme poverty has been declining in all regions of the world, it has unfortunately been increasing in SSA. This is despite over two decades of growth resurgence that began in the mid 1990s. Second, building on the increasing poverty levels in SSA, Asongu and Nwachukwu (2016a) has presented a critique of Piketty's (2013) 'capital in the 21 st century'. Building on: (i) responses from Kenneth Rogoff and Joseph Stiglitz; (ii) post Washington Consensus paradigms and (iii) underpinnings from Boyce-Fofack-Ndikumana and Solow-Swan, Asongu and Nwachukwu (2016a) conclude that extreme poverty in SSA would increase as long as the return on political economy (or illicit capital flight) is higher than the growth rate in the sub-region.
Third, a recent stream of literature is building on theoretical underpinnings of neoclassical growth models to propose the need for common policies based on negative macroeconomic and institutional signals. In essence, whereas the theoretical underpinnings of income convergence have exclusively been limited to catch-up in positive signals, a new stream of literature is evolving on catch-up in negative signals. According to this stream, it is more relevant to initiate common policies based on negative signals because these are policy syndromes by conception and definition. The three studies in this stream of literature are to the best of our knowledge: (i) Asongu (2013a) on harmonizing policies against software piracy; (ii) Electronic copy available at: https://ssrn.com/abstract=3509773 Asongu and Nwachukwu (2016b) who have predicted the 2011 Spring using negative signals in institutional and macroeconomic variables and (iii) Asongu (2014a) on benchmarking policy harmonization against capital flight in SSA.
Fourth, Asongu (2014a) has used two fundamental characteristics to project horizons for common policies against capital flight in SSA. We extend the underlying study by accounting for income levels, legal origins, regional proximity and religious domination. In essence, accounting for more fundamental characteristics of the sub-region's development is essential in order to avail room for robustness and more policy implications. Accordingly, upholding blanket policies in the battle against capital fight may not be effective unless they are contingent on fundamental characteristics and prevailing trajectories of capital flight in SSA. Hence, policy makers are most likely to ask the following three questions before considering the harmonization of policies on capital flight. (1) Is capital flight converging within SSA? (2) If so, what is the degree and timing of the convergence process? (3) For which relevant fundamental characteristics of capital flight do answers to the first and second questions apply? While an answer to the first question will guide on the feasibility of harmonizing blanket policies, the answer to the second will determine an optimal timeframe for the blanket policies. But ultimately, the answer to the third (given that the first and second questions are already answered), will determine the feasibility-of, timeframe-for and exclusiveness (or non arbitrariness) of the common policies. This third question is most relevant because it underlines the need for common policies to be contingent on the prevailing speeds of and time for full (100%) convergence within each identified fundamental characteristic of capital flight.
The positioning of the research also departs from contemporary literature on capital flight which has been oriented towards, inter alia: the connection between fiscal policy and capital flight (Muchai & Muchai, 2016); lessons on causes and effects of capital flight from Africa (Ndikumana, 2016); the connection between capital flight and public social expenses in Madagascar (Ramiandrisoa & Rakotomanana, 2016) and Congo-Brazzaville (Moulemvo, 2016); insights into relationships between misinvoicing in trade and the flight of capital from Zimbabwe (Kwaramba et al., 2016); the nexus between natural resources and capital flight in Cameroon (Mpenya et al., 2016); how capital flight is related to tax income in Burkina Faso (Ndiaye & Siri, 2016); linkages between terrorism, capital flight and military expenditure (Efobi & Asongu, 2016;Asongu & Amankwah-Amoah, 2018); the institutional environment on the Electronic copy available at: https://ssrn.com/abstract=3509773 nexus between capital flows and capital flight (Gankou et al., 2016); the bundling and unbundling of institutions in the fight against capital flight (Asongu & Nwachukwu, 2017) and how terrorism sustains the addiction to capital flight .
The rest of the paper is organized in the following manner. Section 2 presents the data and methodology. The empirical analysis and discussion of results are covered in Section 3 while Section 4 concludes.

Data
The research focuses on 37 countries in Africa building on data for the period 1980 to 2010 from a plethora of sources: Boyce and Ndikumana (2012a); the African Development Indicators (ADI) and the Financial Development and Structure Database (FDSD) of the World Bank. The geographical and temporal scopes of the research are contingent on the availability of data at the time of the study. The capital flight data come from Boyce and Ndikumana (2012a) and at the time of study only 37 countries are available for the corresponding periodicity.
Insights into the sampled countries and related categories are disclosed in Appendix 4. In what follows, some essential points surrounding the selection of data are clarified, notably: (i) the determination of fundamental features, (ii) how the capital flight measure is comparable and compatible and (iii) choice of control variables.

Determination of fundamental characteristics
Building on the attendant scholarship, it is not feasible to establish convergence when sampled countries exhibit much heterogeneity (Asongu, 2013a). It is in view of improving the homogenous characterization of the dataset that the dataset are classified based on some fundamental characteristics pertaining to capital flight. In the choice of these fundamental features, governance (inter alia, regulation quality, corruption-control and transparency) and macroeconomic features are have the shortcoming of being dynamic over time. Therefore, an adopted threshold may be inconsistent within the sampled periodicity, especially given the length of the sample (i.e. a 30 year span).
In the light of the above, the research builds on Weeks (2012) in the selection of the fundamental features, namely: petroleum-exporting and conflict-affected countries, inter alia. To Electronic copy available at: https://ssrn.com/abstract=3509773 these features, this study includes the following categorizations: religious domination, legal origins and income levels. Whereas the categorization approach employed by Weeks (2012) is exclusive, there is a consensus in the literature that "conflicts" and political strife as well as a sector that is petroleum-dominated influence the macroeconomic performance of African countries (Boyce & Ndikumana, 2012a, 2012b. Moreover, there are some apparent issues in the assignment of countries to the selected categories on an exclusive and non-arbitrary basis. In order to avoid repetition, more information on the adopted categories can be found in Asongu (2014a) which has built on a body of literature for the categorization of countries, notably: Weeks (2012), Ndikumana (2003, 2012a), La Porta et al., (1998Porta et al., ( , 1999, Asongu (2014b).

Control variables
In accordance with Asongu (2014a), 14 variables are adopted for the conditioning information set. These elements in the conditioning information set are used in two distinct specifications that account for both trade and financial globalization (i.e. trade openness, private capital flows and foreign direct investment), expenditure of the government (i.e. public investment and government spending), economic prosperity (i.e. GDP per capita growth and GDP growth), institutional quality (i.e. rule of law and regulation quality), the stability of prices (i.e. inflation), financial development (i.e. liquid liabilities and money supply) and development assistance (entailing total foreign aid and foreign aid from the DAC 2 countries). It is worthwhile to clarify that the choice of the variables is consistent with the theoretical insights into conditional convergence which maintain that if there are disparities between countries in institutional and macroeconomic features that are exogenous to capital flight, conditional convergence is likely to be apparent. According to Asongu (2015), globalization drives capital flight. Boyce and Ndikumana (2012b) maintain that one of the most critical mechanisms via which government funds are stolen is through public spending. Weeks (2012) posits that capital flight is associated with high dependence on foreign aid and low quality of institutions. It has been documented in the literature that investors prefer investing in economies that are less characterized by features of ambiguity (Kelsey & le Roux, 2017 such as very high inflation. In line with Boyce and Ndikumana (2003), high levels of economic growth that are not driven by petroleum exports are linked with lower levels of capital flight, in the light of higher anticipated returns from investment. Insights into the summary statistics, correlation matrix and definitions of variables are presented respectively, in Appendix 1, Appendix 2 and Appendix 3.

Methodology
This research uses the beta (β) convergence technique that is in line with the methodological motivations of the paper, consistent with the bulk of literature on the imperative of the adopted estimation technique to be consistent with data behavior and study objective (Chao et al., 2019;Zhang et al., 2019;Li et al., 2014Li et al., , 2016Kou et al., 2012Kou et al., , 2014Kou et al., , 2016Kou et al., , 2019aKou et al., , 2019b. This procedure of estimation is typically in line with the income catch-up scholarship that has been assessed building on models of neoclassical growth, notably: Baumol (1986); Sala-i-Martin (1992, 1995) and Mankiw et al. (1992). The attendant theoretical insights have been extended to other areas of development studies, inter alia: financial markets and financial intermediary developments (Narayan et al., 2011;Tchamyou & Asongu, 2017;Tchamyou et al., 2018;Efobi et al., 2019).
(2) below are the main specifications used to assess conditional convergence if t i W , is taken as strictly exogenous. It follows that the country-specific effect i  articulates other drivers of the steady state of the country that are not observed in In order to eliminate fixed effects that can cause endogeneity owing to the correlation between the lagged outcome variable and fixed effects, the difference of Eq. (2) it taken to produce Eq. (3). Eq.
(2) and Eq. (3) are then combined within a framework of a system Generalised Method of Moments (GMM) that ensures parallel conditions between the dependent variables and error terms by using lagged differences of the regressors as instruments in Eq.
(2) lagged Electronic copy available at: https://ssrn.com/abstract=3509773 levels of the regressors as instruments in Eq.
(3). The choice of the difference estimator of the GMM technique (Arellano & Bond, 1991) relative to the system estimator of the same technique (Arellano & Bover, 1995;Blundell & Bond, 1998) is motivated by the need to obtain more efficient estimates as documented by Bond et al. (2001). The specification is two-step in order to account for heteroscedasticity.
As maintained by Islam (1995, 14), it is not appropriate to assess convergence using yearly time spans because these are too short and consequently, short term disturbances may persist during such short time spans. Therefore, given a dataset spanning 31 years, the research follows Asongu (2013a) in employing two-year data averages in terms of non-overlapping intervals (NOI). In addition to the justification provided above, four more additional motivations are worth clarifying. (i) While NOI that are characterized by higher numerical values absorb more short term disturbances, there is also an associated shortcoming of having estimated models that are weakened in the light of the information criteria used to assess and validate the estimated models. Therefore, the selection of the two-year NOI over NOIs with higher numerical values is also motivated by the need to take on board as many time series properties as possible. (iv) From an exploratory visual analysis, it is apparent that evidence of persistence in short term or business cycle disturbances is not associated with capital flight. Hence, the coefficient of auto-regression is 2 (i.e. is set to 2) and the research computes the implied convergence rate by calculating /2. Accordingly, the estimated coefficient of the lagged difference outcome variable is divided by the number of NOI (i.e. 2) because it has been employed to absorb short term disturbances. In essence, the criterion for assessing convergence is that the absolute value of the estimated lagged coefficient should be between the interval of zero and one ( 1 0    given that the country is converging to a steady state (Asongu, 2013).

Presentation of results
This section looks at three principal concerns: (i) investigation of the presence of convergence; (ii) computation of the speed of convergence and (iii) determination of the time needed for full (100%) convergence. The summary of overall findings is presented in Table 1 in which the three concerns are addressed. Findings for absolute (unconditional) and conditional convergence are presented in Table 2 and Tables 3-4 respectively.
Absolute convergence is estimated with only the lagged difference of the endogenous variable as independent variable whereas conditional convergence is in the presence of the conditioning information set (control variables). Hence, unconditional convergence is estimated without t i W , : vector of determinants (government expenditure, trade, FDI, GDP growth, regulation quality, financial depth, development assistance and inflation) of capital flight 3 .
Accordingly, in order to assess the validity of the model and indeed the convergence hypothesis, we perform two tests, notably: (i) the Sargan test which assesses the over-identification restrictions and (ii) the Arellano and Bond test for autocorrelation which examines the null hypothesis of no autocorrelation. The Sargan-test investigates if the instruments are uncorrelated with the error term in the equation of interest. The null hypothesis is the stance that the instruments as a group are strictly exogenous (do not suffer from endogeneity), which is necessary for the validity of the GMM estimates. The p-values of estimated coefficients are disclosed in brackets in the line following the reported values of the estimated coefficients. We broadly observe that the null hypothesis of the Sargan test is not rejected in all the regressions.
Priority is given to the second order autocorrelation: AR(2) test in first difference because it is more relevant than AR(1) as it detects autocorrelation in difference. For almost every model, we are unable to reject the AR(2) null hypothesis for the absence of autocorrelation, especially for conditional convergence specifications. Therefore, there is robust evidence that most of the models are free from autocorrelation at the 1% significance level. Table 1 presents a summary of the findings from Tables 2-4. This entails results for Absolute Convergence (AC), Conditional Convergence (CC), the Speed of Absolute Convergence (SAC), the Speed of Conditional Convergence (SCC) and the rate required to achieve full (100%) convergence in both types of convergences.
From a general perspective, the following conclusions could be drawn. (i) Conditional convergence findings based on the second specification (Table 4) are substantially more significant than those based on the first specification (Table 3). Therefore, conditional convergence is based on the variables we observe and empirically test (or model), which may not reflect all determinants of capital flight that facilitate the convergence process. Hence, the discussion of findings will be based only on the second specification for conditional

Discussion of results
Before we dive into the discussing the results, it is important first and foremost to understand the economic intuition motivating absolute and conditional convergence of capital flight in the African continent. Absolute convergence in capital flight occurs when countries share the same fundamental characteristics with regard to bases governing capital flight such that only cross-country variations in initial levels of capital flight exist. Absolute convergence thus, results from factors such as, inter alia: significant export of petroleum; political instability due to conflicts; the emphasis legal origin places on property rights, enforcements of the rights and fight against corruption; the manner in which economic prosperity affects the propensity by which the extra-wealth is saved abroad. Absolute convergence also occurs because of adjustments common to fundamental characteristics (conflict-affected, high-income or English common-law countries for example). Hence, based on the above intuition we could expect capital flight to be higher in petroleum and conflict-affected countries. This is a necessary but not a sufficient condition for speedy convergence because of disparities in initial conditions of capital flight. These differences in initial conditions depend on: (i) time-dynamic evidence of significant petroleum exports, either because of recent discovery or substantial decline in productions; (ii) spontaneous reoccurrence of conflicts after relatively stable periods or arbitrary and unilateral violation of peace accords and (iii) the diffusion of legal cultures transmitted by colonial powers over time through regionalization and globalization such that the legal origin fundamental holds less ground.
On the other hand, conditional convergence is that which is contingent on cross-country disparities in structural and institutional characteristics that determine capital flight. In accordance with the economic growth literature (Barro & Sala-i-Martin, 1992, conditional convergence depicts the kind of convergence whereby, one's own long-term steady state (equilibrium) is contingent on structural characteristics and fundamentals of its institutions in particular and its economy in general. For example, non-petroleum exporting countries may differ significantly in the level of globalization, institutional quality, economic prosperity, financial development, price stability, foreign aid…etc To this end, our model for conditional convergence is contingent on institutional quality (rule of law and regulation quality), globalization (trade, FDI and private capital flows), financial development (at overall economic and financial system levels), economic prosperity (GDP growth at macro and micro levels), inflation and development assistance (total NODA and NODA from DAC countries) 4 . Due to constraints in degrees of freedom, some models have not been conditional on all the determinants of capital flight outlined above. This is not a major issue because some conditional specifications in mainstream literature are limited to two macroeconomic control variables (Bruno et al., 2012).
We have observed the following from the findings. (i) Based on continental results, findings on 'Petroleum exporting', 'σorth Africa' 'French civil-law', 'Middle-income' and 'Upper-middle-income' countries significantly affect the absolute convergence process. The corresponding lower (higher) rate (time) of (to full) convergence is the result of differences in initial conditions of capital flight. For instance, the difference in petroleum countries could be explained by significant variations in initial conditions of capital flight discussed above: timedynamic evidence of significant petroleum exports, either because of recent discovery or substantial decline in productions. (ii) Within the perspective of CC, but for the 'Conflictaffected' and 'Low-income' countries results, African findings are broadly consistent across other fundamental characteristics. 'Conflict-affected' and 'Low-income' countries significantly have a higher (lower) rate (time required) of (for full) conditional converge because of substantially lower cross-country differences in macroeconomic and institutional characteristics determining capital flight. Hence, cross-country differences in factors governing capital flight among "Conflict-affected" and "Low-income" countries are not very substantial. (iii) Regardless of fundamental characteristic, from a projection date of 2010, a genuine timeframe for harmonizing policies is between 2016 and 2023. This empirically indicates that (both in absolute and conditional terms) countries with lower rates of capital flight are catching-up their counterparts with higher rates. Consistent with the intuition motivating this analysis on policy harmonization, two inferences could be drawn: (i) on the one hand, convergence implies that, adopting common policies against the scourge is feasible and (ii) full (100%) convergence within the specified time horizon reflects the implementation (or harmonization) of the feasible policies without distinction of nationality or locality. IPRs: Intellectual Property Rights.

Declaration
Availability of supporting data: the data for this paper is available upon request.

Acknowledgement:
The authors are indebted to the editor and reviewers for constructive comments.

Compliance with Ethical Standards
The authors are self-funded and have received no funding for this manuscript. The authors also have no conflict of interest.