African Development Review, Vol. 29, No. 3, 2017, 471–484 Threshold Effects of Debt on Economic Growth in Africa Arcade Ndoricimpa Abstract: This study aims at examining the debt threshold effects on economic growth in Africa. Non-dynamic and dynamic panel threshold regression approaches are used. The findings indicate that the estimated debt threshold is sensitive to the estimation technique used and to growth control variables included in the estimation. Existence of nonlinearities in the debt-growth nexus cannot be denied. The findings show that while low debt is neutral or growth-enhancing, high debt is consistently detrimental to growth for all the cases considered. This study shows that caution is needed when suggesting a debt threshold since this can be sensitive to modelling choice and to growth control variables. Nonlinearities in the debt-growth nexus are established but further analysis is needed to suggest a policy. 1. Introduction Africa, especially sub-Saharan African countries, have accumulated large stocks of debt over the years due to external (oil shocks, change in commodity prices, interest rates, etc.) and domestic shocks (deficits, wars, poor governance, etc.) (Presbitero, 2010). High public debt in sub-Saharan Africa has been a problem since the 1980s leading to repayment difficulties and slow growth due to high debt servicing (Pattillo et al., 2011). It is believed that at low levels, debt is growth-enhancing through capital accumulation and productivity growth (Pattillo et al., 2004) but high levels of debt harm long-run growth via higher long-term interest rates, higher future distortionary taxation, inflation, and greater uncertainty about prospects and policies, all of which discourage investments (Kumar and Woo, 2010). A question arising here is then, when does the debt level go from good to bad? Beyond which level does debt become a drag to growth? Because of the debt problems in developing countries since the 1980s, a number of studies sought to determine a debt overhang threshold, that is, a threshold at which the marginal effect of debt turns negative. Early studies on debt overhang threshold in developing countries are found in Kaminsky and Pereira (1996), Deshpande (1997), and Elbadawi et al. (1997). Just to mention a few, Elbadawi et al. (1997) on a sample of 99 developing countries, find a debt overhang threshold at around 97 per cent of GDP, while Pattillo et al. (2002), using various nonlinear specifications for a sample of 100 developing countries, find that the impact of debt on growth becomes negative at about 35–40 per cent of GDP. Similarly, Cordella et al. (2005) for 79 developing countries use different estimation procedures and find a debt overhang threshold between 10 and 35 per cent of GDP. Pattillo et al. (2011), on a panel of 93 developing countries, suggest that the impact of debt is negative when it is about 35–40 per cent of GDP. While in the 1980s and 1990s, debt problems were only for developing countries, with the recent global financial crisis, the sovereign debt has become a problem to a broad range of countries, even to industrial countries (Chancellor, 2010). Consequently, examining the threshold effects of debt on growth in developed countries has gained pre-eminence in the last few years (see, for example, Reinhart and Rogoff, 2010; Cecchetti et al., 2011; and Pescatori et al., 2014). Reinhart and Rogoff (2010) suggest a debt threshold of 90 per cent, Cecchetti et al. (2011) find that debt becomes a drag to growth when it is beyond 85 per cent of GDP, while Pescatori et al. (2014) find no evidence of any particular debt threshold above which medium-term growth prospects are dramatically compromised. The observation from the above discussion is that existing empirical literature gives mixed evidence on the level of debt threshold, and most of the recent empirical studies on the debt threshold effects on growth seem to focus on developed countries, OECD and Euro area countries and less on developing countries such as African countries that have suffered recurrent debt problems since the 1980s. It should be noted that although the World Bank and IMF provided a debt relief to a number of poor  University of Burundi, Faculty of Economics and Management, PO Box 1280 Bujumbura, Burundi; e-mail: arcade_ndoricimpa@yahoo.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. 471 472 A. Ndoricimpa and heavily indebted African countries under the HIPC Initiative since 1996, debt relief should not be considered as a medium or long-term solution since debts could escalate again due to the moral hazard effect (Arnone et al., 2008). In addition, although debt levels have in general declined in Africa during the past decade due to debt relief programmes, some African countries such as Eritrea, Ghana, Sudan, Cabo Verde, Mauritania, Sao Tom
e and Pr
ıncipe, the Seychelles, and so on, still have a high debt-to-GDP ratio. And as Ncube and Brixiov
a (2015) point out, the global financial crisis has left some African countries with fiscal challenges deteriorating debt sustainability. Also, the IMF1 recently argued that if African countries are not careful in their choice and financing of infrastructure projects, they could fall in the excessive debt trap again. Against this backdrop, there is a need to re-examine nonlinearities in the debt-growth nexus in Africa to suggest a debt threshold which would serve as a basis for fiscal and debt management. Besides, studies which exist on developing countries (see for instance, Imbs and Ranciere, 2005; Pattillo et al., 2011) combine in their samples, countries from Africa, Latin America and Asia. According to Sepehri and Moshiri (2004), not considering heterogeneity among country groups can lead to biased estimate of a threshold level. Moreover, it is inappropriate to set a single policy target applicable to all developing countries. Furthermore, while previous studies determine debt thresholds by estimating a quadratic relationship (see for instance, Elbadawi et al., 1997; Pattillo et al., 2002; Cordella et al., 2005), recently a number of estimation techniques have been developed which can help better capture nonlinearities in the relationships between variables such as debt and growth (see Hansen, 1999; Gonzalez et al., 2005; Kremer et al., 2013). As Egert (2015) points out, examining nonlinearities in the debt-growth nexus can be sensitive to modelling choices. The contribution of this study is three-fold. Firstly, while some previous studies exogenously determine debt threshold (see for instance, Reinhart and Rogoff, 2010; Kumar and Woo, 2010; and Pescatori et al., 2014), this study endogenously determines the debt threshold. Secondly, this study examines nonlinearities in the debt-growth nexus by focusing specifically on Africa. Thirdly, for robustness checks, this study applies two panel threshold estimation approaches, non-dynamic and dynamic panel threshold regression methods respectively, initiated by Hansen (1999) and Kremer et al. (2013). The rest of the paper is organized as follows. Section 2 highlights the stylized facts. Section 3 presents the methodology and data used. Section 4 presents and discusses the empirical results. Section 5 gives concluding remarks. 2. Public debt and Economic Growth in Africa: Stylized Facts We discuss in this section the correlation between public debt and economic growth by debt regimes. Following Reinhart and Rogoff (2010), we consider four arbitrary debt regimes, namely low public debt (below 30 per cent in ratio of GDP), medium-low public debt (between 30 per cent and 60 per cent), medium-high public debt (between 60 per cent and 90 per cent), and high public debt (above 90 per cent). Table 1 indicates that economic growth is higher in the low debt-regime and reduces as the level of debt increases. For countries with a debt-to-GDP ratio less than 30 per cent, the average economic growth rate is 2.78 per cent, those with a debt-to-GDP ratio between 30 per cent and 60 per cent experienced a growth rate of 1.57 per cent, and those with a debt ratio between 60 per cent and 90 per cent experienced a growth rate of 0.87 per cent, while countries that had a debt-to-GDP ratio above 90 per cent grew on average at 0.84 per cent. From the low debt-regime to medium-low debt-regime, and from medium-low debt to medium-high debt-regime, average economic growth reduces significantly, but it does not change significantly from medium-high public debt to high public debt-regime. Figure 1 shows the correlation between debt and growth in the four debt regimes. It suggests that low debt, medium-low debt and medium-high debt are positively associated with economic growth although not strongly. However, when debt-to-GDP ratio is above 90 per cent, there seems to be a negative correlation between debt and economic growth. Although exogenously Table 1: Debt-to-GDP ratio and economic growth by debt regimes Debt regimes d < 30% 30% < d < 60% 60% < d < 90% d > 90% Debt-to-GDP ratio Economic growth 20.87 45.28 75.98 134.51 2.78 1.58 0.88 0.84 Note: d stands for debt-to-GDP ratio. © 2017 The Author. African Development Review © 2017 African Development Bank Threshold Effects of Debt on Economic Growth 473 Figure 1: Correlation between debt and economic growth in Africa by debt regimes determined, this seems to indicate that the debt threshold we seek to determine should be in the high debt-regime, beyond 90 per cent of GDP. We endogenously determine the debt threshold level in Section 4 using threshold modelling techniques. 3. Methodology and Data This study applies non-dynamic panel threshold regression initiated by Hansen (1999) and for robustness checks employs dynamic panel threshold regression of Kremer et al. (2013). Hansen (1999) suggests a panel threshold modelling technique for balanced panels with individual specific effects whose specification is as follows, assuming the presence of one threshold in the relationship: yit ¼ mi þ b01 xit Iðqit  gÞ þ b02 xit Iðqit > gÞ þ eit ; © 2017 The Author. African Development Review © 2017 African Development Bank ð1Þ 474 A. Ndoricimpa where i ¼ 1; :::; N; t ¼ 1; :::; T; mi are individual specific effects; yit is the dependent variable; qit is the threshold variable which divides the observations into two regimes; g is the common threshold value; I(.) is the indicator function; and xit is a vector of the control variables. The elements of xit and the threshold variable qit are not time invariant. In case of one threshold in the relationship, the two regimes are distinguished by different regression slope coefficients b1 and b2 where b1 shows the marginal impact of the threshold variable on the dependent variable when the threshold variable is less or equal to the threshold value ðqit  gÞ and b2 captures the marginal impact of the threshold variable when the threshold variable is greater than the threshold value ðqit > gÞ. The error term eit is identically and independently distributed, that is eit  iid ð0; s 2 Þ. The individual effects mi are eliminated by removing individual specific means. After eliminating the individual specific effects, the optimal threshold value gb is obtained by minimizing the concentrated residual sum of squares, that is, b g ¼ arg min S 1 ðgÞ where S 1 ðgÞ is the residual sum g of squares for a given threshold value g. The search of g can be made for all values of the threshold variable qit or be limited to specific quantiles. To test whether the threshold effect is significant, Hansen (1999) suggests the following null hypothesis of linear relationship, that is, no threshold effect in the relationship. Rejecting this hypothesis means that there exists at least one threshold in the relationship.2 H0: b1 ¼ b2 If the linearity hypothesis is rejected, the remaining task is to know the number of thresholds governing the relationship, therefore another test further discriminates between one and two thresholds. According to Bick (2010), estimating the threshold model without including the regime intercept if it is present in the data generating process can lead to a bias proportional to d1 since orthogonality of the regressors is not preserved any more. Consequently, Bick (2010) extends Hansen’s (1999) model to account for regime intercepts. yit ¼ mi þ b01 xit Iðqit  gÞ þ d1 Iðqit  gÞ þ b02 xit Iðqit > gÞ þ eit ; ð2Þ where d1 is the regime intercept common to all cross-sections. Applying Bick’s (2010) extension of Hansen’s (1999) approach to the analysis of debt threshold effects on economic growth gives the following threshold model: grgdpit ¼ mi þ b1 d it Iðd it  gÞ þ d1 Iðd it  gÞ þ b2 d it Iðd it > gÞ þ azit þ eit ; ð3Þ where mi are country individual effects, grgdpit (growth rate of real GDP per capita) is the dependent variable, d it (debt-toGDP ratio) is the threshold variable and regime-dependent regressor, zit is a vector of control variables. b1 captures the marginal impact of debt on economic growth in the low-debt regime while b2 captures the marginal impact of debt in the high-debt regime. For robustness checks, this study also employs the dynamic panel threshold regression approach advanced by Kremer et al. (2013), to avoid any potential endogeneity bias in the model. The specification of the dynamic panel threshold regression model is the same as in Equation (3), the only difference is that in the vector zit of control variables, endogenous variable z2it is differentiated from exogenous variables z1it . In the dynamic model, the individual fixed effects are eliminated using the forward orthogonal deviations transformation suggested by Arellano and Bover (1995). Since the regression slope coefficients are obtained using GMM estimation, as in Arellano and Bover (1995), in this study, the lags of initial income (the endogenous variable), pcgdpit 2 ; pcgdpit 3 ; :::; pcgdpit p are used as instruments. In analysing the debt threshold effects on economic growth, following previous studies on growth modelling (see for instance, Azam et al., 2002; Kremer et al., 2013; Seleteng et al., 2013; Anyanwu, 2014; Akobeng, 2016), a number of control variables are considered, namely initial income, population growth rate, investment ratio (per cent of GDP), openness to trade, the ratio of foreign direct investment (per cent of GDP), the ratio of government spending (per cent of GDP), as well as an institutional variable to capture the level of democracy. Two more control variables are used for robustness checks, namely an HIPC completion point dummy to capture the effect of debt relief, and the ratio of total natural resource rent3 (per cent GDP). The list, definition, description and sources of data for the variables used are in Table 2. This study is based on a balanced panel of 38 African countries for the period 1980–2010 due to debt data availability. The sample of countries considered is in Table 3. Following Kremer et al. (2013), and Ibarra and Trupkin (2016), five-year averages of the data are used (i.e. 1980–1984, 1985–1989, 1990–1994, 1995–1999, 2000–2004, 2005–2009). In effect, using series of averages helps reducing business-cycle effects and measurement error (Jeanty, 2010). © 2017 The Author. African Development Review © 2017 African Development Bank 475 Threshold Effects of Debt on Economic Growth Table 2: Definition, description and descriptive statistics of the variables Variables grgdp d popgr inv fdi natres open gspend hipccp inst initial Definition, description and source Std. Dev. Min Max 1.3 4.9 –11.9 51.7 228 79.4 48.4 2.8 286.1 228 2.5 228 21.0 0.9 17.8 –4.6 6.3 –2.2 179.9 Obs. Mean Growth rate of real GDP per capita [Source: World Development Indicators, WDI (2015) and UNCTAD]. Debt-to-GDP ratio [Source: Historical Public Debt Database created by Abbas et al. (2010)] Growth rate of population [Source: WDI (2015)]. Investment ratio (% of GDP) captured by the GDP ratio of gross fixed capital formation [Source: WDI (2015) and Penn World Tables, PWT 8.1]. Foreign Direct Investment (FDI) ratio (% GDP) [Source: Data on FDI are from UNCTAD]. Total Natural Resource Rent (% GDP) [Source: WDI (2015)]. Log of openness, where openness is measured by the GDP ratio of the sum of exports and imports [Source: WDI (2015); UNCTAD]. The ratio of government spending (% GDP) [Source: Penn World Tables (PWT 8.1) and WDI, 2015]. HIPC completion point dummy taking the value of 1 from completion period onwards and 0 otherwise [Source: Own computation using IMF archive on HIPC completion periods]. An institutional variable proxied by Polity2, a political regime index that captures the level of democracy. The score of the Polity2 index ranges from þ10 (strongly democratic) to 10 (strongly autocratic). [Source: polity2 is from Polity IV Project database]. Log of one period-lagged real GDP per capita in 2005 constant prices [Source: Data for real GDP per capita is from WDI (2015) and UNCTAD, online database]. 228 228 2.5 4.2 –4.2 44.4 228 12.9 228 4.1 14.7 0.5 0.0 2.4 83.7 6.1 228 18.7 13.5 4.6 87.0 228 0.09 0.3 0 1 228 –1.6 5.7 –10 10 228 1.0 6.5 4.8 9.5 Note: The summary statistics is for five-year averages of data. Table 3: Mean debt-to-GDP ratio and mean economic growth in Africa (1980–2010) Country Algeria Benin Botswana Burkina Faso Burundi Cape Verde Cameroon Central African Rep. Chad Congo, Dem. Rep. Congo Republic C^ote d’Ivoire Egypt Equatorial Guinea Ethiopia Gabon Gambia Ghana Kenya Lesotho Madagascar Malawi Mali Mean d Mean growth Country Mean d 53.8 56.5 18.3 36.7 86.5 49.9 66.3 66.6 42.8 129.5 146.9 90.8 97.5 125.4 78.7 54.9 105.2 49.5 50.7 69.2 93.4 96.7 76.9 0.6 0.9 4.3 2.2 0.6 5.8 0.1 0.0 2.6 2.3 1.3 1.6 2.9 13.1 1.9 0.3 0.1 1.8 0.5 2.2 1.2 0.4 0.7 Mauritania Mauritius Morocco Niger Nigeria Rwanda Senegal Seychelles Sierra Leone South Africa Swaziland Tanzania Togo Tunisia Uganda Zambia 146.4 52.2 77.7 60.3 75.2 52.6 60.8 96.6 125.2 35.9 22.5 77.6 90.8 51.5 61.6 139.8 Mean growth 0.2 3.8 2.5 1.0 0.9 2.0 0.2 0.1 1.1 0.4 2.5 1.6 0.3 2.5 2.3 0.8 Note: d stands for the debt-to-GDP ratio. Source: Own computations using data from UNCTAD for real per capita GDP growth and Historical Public Debt Database created by Abbas et al. (2010)] for debt-to-GDP ratio. © 2017 The Author. African Development Review © 2017 African Development Bank 476 A. Ndoricimpa 4. Empirical Results and Discussion 4.1 Baseline Results The baseline estimation results based on the non-dynamic panel threshold regression of Hansen (1999) are reported in Table 4. The upper panel of Table 4 presents the sequential tests for the number of thresholds between public debt and economic growth. In the first step, the null hypothesis of no threshold (i.e. linear relationship) is tested against the hypothesis of at most one threshold (i.e., existence of two regimes in the debt–growth nexus), then the hypothesis of at most one threshold is tested against the hypothesis of at most two thresholds. The results show that the null hypothesis of no threshold is rejected at the 5 per cent significance level, indicating rejection of a linear relationship between public debt and economic growth in Africa. However, the null hypothesis of the presence of at most one threshold in the relationship between debt and economic growth cannot be rejected. This confirms the presence of nonlinearities in the debt–growth nexus in Africa. The results suggest a debt threshold of 92.78 per cent. However, as Kremer et al. (2013) point out, the inclusion of an endogenous variable in panel threshold modelling can create an endogenous bias leading to biased estimate of the threshold. We therefore exclude initial income among the control variables but find the same debt threshold of 92.78 per cent. The middle panel of Table 4 presents the results of the impact of debt on economic growth in the low- and high-debt regimes. For both models, including and excluding initial income among the control variables, the estimation results indicate that public debt has a positive but insignificant effect on economic growth in the low-debt regime (i.e. when debt-to-GDP ratio is below 92.78 per cent); but in the high-debt regime (i.e. when debt-to-GDP Table 4: Estimation results using non-dynamic panel threshold regression Test for the number of thresholds (Bootstrap p-value) H0: No threshold H0: At most one threshold 0.030 0.236 0.039 0.241 Estimated debt threshold g 95% confidence interval 92.78% [67.96%, 179.74%] 92.78% [67.96%, 182.02%] Impact of regime-dependent regressors Debt-GDP-ratio b1 b2 Estimated coefficients Estimated coefficients 0.011 (0.016) –0.052 (0.012) 0.013 (0.015) –0.048 (0.011) Impact of regime-independent regressors initialit popgrit invit openit gspendit f diit instit d1 Estimated coefficients Estimated coefficients –0.858 (1.059) 0.403 (0.242) 0.107 (0.036) 2.463 (1.021) –0.113 (0.032) 0.570 (0.090) 0.036 (0.054) –7.342 (1.849) – 0.427 (0.241) 0.111 (0.036) 2.416 (1.018) –0.107 (0.031) 0.560 (0.089) 0.034 (0.054) –7.164 (1.833) Notes: Standard errors are in parentheses.  ,  ,  indicate significance at 10%, 5% and 1%. Estimation results are from a GAUSS code written by Bick (2010). The sequential test procedure indicates that the number of thresholds is K ¼ 1. One thousand (1,000) bootstrap replications are used to obtain the p-values in testing the number of thresholds © 2017 The Author. African Development Review © 2017 African Development Bank Threshold Effects of Debt on Economic Growth 477 ratio is above 92.78 per cent), public debt is harmful to growth although the marginal impact seems to be small. The coefficient of debt in the high-debt regime is b2 ¼ 0:052 (when initial income is included), which means that a 1 percentage point increase in the debt-to-GDP ratio is associated with a slowdown in annual real per capita GDP growth of around 0.05 percentage points per year. The regime-dependent intercept d1 is also statistically significant at the 1 per cent level. The lower panel of Table 4 presents the estimation results for the impact of the regime-independent regressors on economic growth. Estimation results indicate that the coefficients associated with population growth, the ratio of investment, openness to trade, the ratio of government spending and the ratio of foreign direct investment are statistically significant. The coefficients of population growth, investment, openness to trade and foreign direct investment are positive while the coefficient of government spending is negative. This suggests that population growth,4 investment, openness to trade and foreign direct investment enhance economic growth while government spending harms growth in Africa, which is in accordance with most empirical literature on economic growth. Kremer et al. (2013), Mijiyawa (2013) and Thanh (2015) also find that investment enhances growth, Vinayagathasan (2013) finds that openness to trade spurs economic growth as well. Indeed, trade openness spurs growth by raising productivity and competitiveness, and by allowing technological imitation (Edwards, 1997). Foreign direct investment promotes host countries’ economic growth through its impact on their financial resources and investment, by enhancing their technological capabilities, by boosting their export competitiveness and by generating employment and strengthening their skills base (UNCTAD, 1999; Ndoricimpa, 2014). The negative effect of government spending found is in accordance with Barro and Sala-i-Martin (1997) who point out that ‘high level of public expenditures drains out the most efficient private investment and inhibits growth’. Seleteng et al. (2013) find also a negative impact of government spending on economic growth for SADC countries. However, the estimation results show that the coefficient of initial income is negative but insignificant, rejecting the hypothesis of conditional convergence. In addition, the level of democracy does not significantly affect growth in Africa although its coefficient is positive. Burkhart and Lewis-Beck (1994), Barro (1996) and Giavazzi and Tabellini (2005) reach the same conclusion. 4.2 Robustness Checks Three robustness checks are undertaken in this study. Firstly, as we know, a number of African countries that reached the completion period5 have received debt relief from the World Bank and IMF under the HIPC6 Initiative that started in 1996. In addition, in the last decade economic growth in a number of African countries has been driven by a natural resource boom. Thus, as a first robustness check, in examining the debt threshold effects on growth in Africa, two more control variables are added among the control variables, namely a HIPC completion point dummy to capture the effect of debt relief, and the ratio of total natural resource rent7 (per cent GDP). The estimation results are reported in Table 5a and indicate that linearity hypothesis is still rejected and that the presence of one threshold in the relationship between debt and growth cannot be rejected. The estimation results suggest that adding the HIPC completion point dummy does not change the debt threshold estimate since it remains 92.78 per cent. However, the inclusion of the ratio of total natural resource rent, which is statistically significant at the 5 per cent level, leads to a change in the estimate of the debt threshold. A debt threshold of 102.58 per cent is instead estimated. On the impact of debt on economic growth, the presence of nonlinearities is still confirmed but in the estimations where the ratio of total natural resource rent is included, the impact of debt on economic growth is significant in both debt regimes (below and above the threshold). On the impact of the control variables, the results are similar with the baseline results. Investment, openness to trade, foreign direct investment, and population growth are growth-enhancing while government spending harms economic growth. The ratio of total natural resource rent is also statistically significant with a positive impact on economic growth. The remaining control variables do not affect economic growth. As a second robustness check, we check whether the estimated debt threshold is not driven by the inclusion of some countries in the study sample. To achieve this, we exclude from the analysis some sets of countries and re-estimate the debt threshold. First, we exclude from our initial sample study, four8 largest economies in Africa, namely Nigeria, South Africa, Egypt and Algeria. Second, as it can be observed from Table 3, our study sample is a heterogeneous group of countries with some countries having a high mean debt-to-GDP ratio and others a relatively low mean debt-to-GDP ratio. Thus, we also separately exclude from the analysis five countries with the highest mean debt-to-GDP ratio (i.e. Mauritania, DRC, Zambia, Congo Republic and Equatorial Guinea), and five countries with the lowest mean debt-to-GDP (i.e. Botswana, Swaziland, South Africa, Burkina Faso and Chad). The estimation results are reported in Table 5b. They show that excluding the four largest economies or the five countries with the highest mean debt-to-GDP ratio gives the same debt threshold of 90 per cent. However, excluding the five countries with the lowest mean debt-to-GDP ratio gives a higher debt threshold of 102.58 per cent and the confidence interval widens. The © 2017 The Author. African Development Review © 2017 African Development Bank 478 A. Ndoricimpa Table 5a: Estimation results with non-dynamic panel threshold regression and added control variables Test for the number of thresholds: p-value H0: No threshold H0: At most one threshold 0.048 0.244 0.037 0.610 0.037 0.640 Estimated debt threshold 92.78% [67.96%, 182.02%] g 95% confidence interval 102.58% [66.53%, 182.02%] 102.58% [66.53%, 182.02%] Impact of regime-dependent regressors Debt-GDP ratio b1 b2 Estimated coeff. Estimated coeff. Estimated coeff. 0.009 (0.016) –0.052 (0.012) 0.027 (0.013) –0.043 (0.014) 0.026 (0.013) –0.042 (0.014) Impact of regime-independent regressors initialit popgrit invit openit gspendit f diit instit hipccpit natresit d1 Notes: Standard errors are in parentheses.  ,   , Estimated coeff. Estimated coeff. Estimated coeff. –0.845 (1.062) 0.413 (0.244) 0.106 (0.036) 2.509 (1.027) –0.114 (0.032) 0.574 (0.091) 0.043 (0.055) –0.794 (1.431) – –7.194 (1.872) –1.336 (1.075) 0.459 (0.240) 0.113 (0.036) 1.821 (1.033) –0.095 (0.033) 0.510 (0.094) 0.021 (0.054) – 0.087 (0.040) –5.948 (2.076) –1.315 (1.079) 0.466 (0.241) 0.113 (0.036) 1.866 (1.042) –0.096 (0.033) 0.514 (0.095) 0.026 (0.056) –0.597 (1.438) 0.085 (0.041) –5.818 (2.106) indicate significance at 10%, 5% and 1% level. 90 per cent debt threshold compares to 92.78 per cent obtained with the baseline results (see Table 4). Also, the 102.58 per cent threshold is similar to what was obtained in Table 5a with non-dynamic panel threshold regression with some added control variables. Otherwise, the conclusion on nonlinearity between debt and growth, the impact of debt on growth in both debt regimes (below and above the threshold), and the impact of the regime-independent control variables, is still the same as before. As a third robustness check, the study uses dynamic panel threshold regression initiated by Kremer et al. (2013) to account for the potential endogeneity problem which can arise from the inclusion of initial income among the regressors in the model. As Kremer et al. (2013) argue, given the importance of initial income for the convergence debate in the economic growth literature, excluding initial income or not accounting for the potential endogeneity bias created by its inclusion among the control variables can lead to biased estimate of the threshold. Four estimations are performed, excluding and including a HIPC completion point dummy and the ratio of total natural resource rent. The estimation results are reported in Table 5c. For all the four estimations, the results suggest a debt threshold of 90.34 per cent which is close to 92.78 per cent obtained with the non-dynamic panel threshold regression. However, including the HIPC completion point dummy and the ratio of total natural resource rent widens the interval confidence of the debt threshold estimate. For all the cases considered, nonlinearity in the debt–growth nexus still holds, and on the impact of debt on growth, the conclusion is the same with the baseline results in both debt regimes. Low debt has a positive but insignificant effect on growth while high debt harms economic growth. Similarly, the conclusion on the impact of the control variables on growth is the same with the baseline results. Investment, openness to trade, and foreign direct investment enhance economic growth while government spending harms it. However, initial income, population growth, and the level of democracy do not affect growth although their coefficients are well signed. The newly included control variables, namely, HIPC completion © 2017 The Author. African Development Review © 2017 African Development Bank 479 Threshold Effects of Debt on Economic Growth Table 5b: Estimation results with non-dynamic panel threshold regression after excluding some countries (1) (2) (3) Test for the number of thresholds: p-value H0: No threshold H0: At most one threshold 0.044 0.219 0.051 0.834 0.043 0.808 Estimated debt threshold g 95% confidence interval 90.00% [70.32%, 183.98%] 90.00% [70.32%, 184.74%] 102.58 [41.16%, 183.98%] Impact of regime-dependent regressors Debt-GDP-Ratio b1 b2 Estimated coeff. Estimated coeff. Estimated coeff. 0.012 (0.017) –0.048 (0.013) 0.017 (0.019) –0.049 (0.013) 0.028 (0.013) –0.035 (0.013) Impact of regime-independent regressors initialit popgrit invit openit gspendit f diit instit hipccpit natresit d1 Estimated coeff. Estimated coeff. Estimated coeff. –0.858 (1.115) 0.494 (0.245) 0.138 (0.038) 1.300 (1.091) –0.085 (0.035) 0.492 (0.098) 0.073 (0.045) –0.211 (1.450) 0.019 (0.060) –6.947 (1.891) –1.001 (1.162) 0.518 (0.255) 0.139 (0.039) 1.154 (1.180) –0.093 (0.036) 0.482 (0.102) 0.083 (0.047) –0.252 (1.523) 0.023 (0.063) –7.471 (2.002) –1.370 (1.122) 0.598 (0.244) 0.104 (0.037) 2.546 (1.048) –0.114 (0.033) 0.682 (0.112) 0.059 (0.041) –1.465 (1.476) 0.001 (0.055) –5.339 (2.060) Notes: Standard errors are in parentheses.  ,  ,  indicate significance at 10%, 5% and 1% level. In column (1) are the estimation results when excluding the four largest economies. In columns (2) are the estimation results after excluding five countries with the highest mean debt-to-GDP ratio. In column (3) are the estimation results after excluding five countries with the lowest mean debt-to-GDP ratio. point dummy and the ratio of total natural resource rent, do not affect economic growth either although the sign of the ratio of total natural resource rent is positive. It should be noted that the debt threshold estimated in this study, in the range of 90 per cent – 102.6 per cent is comparable to what was found by Elbadawi et al. (1997), and Kumar and Woo (2010) respectively of 97 per cent and 90 per cent for samples of developing countries. However, it is higher than 35–40 per cent found by Pattillo et al. (2002), 10–35 per cent obtained by Cordella et al. (2005), 60 per cent found by Imbs and Ranciere (2005), and 35–40 per cent found by Pattillo et al. (2011). The difference in the debt threshold estimates can be explained by different estimation techniques used and different samples of countries considered in the analysis. 4.3 Debt Sustainability Testing We delineate countries using the 90 per cent debt threshold9 obtained in this study and form two groups of countries, and then test for debt sustainability for the two groups. The first group contains countries with mean debt-to-GDP ratio less than the 90 per cent threshold and the second group is formed by countries with mean debt-to-GDP ratio greater than that threshold. © 2017 The Author. African Development Review © 2017 African Development Bank 480 A. Ndoricimpa Table 5c: Estimation results with dynamic panel threshold regression Estimated debt threshold g 95% confidence interval 90.34% [64.06%, 102.86%] 90.34% [26.04%, 102.58%] 90.34% [26.04%, 102.58%] 90.34% [25.26%, 101.44%] Estimated coefficients Estimated coefficients Impact of regime-dependent regressors Estimated coefficients Debt-GDP ratio b1 b2 Estimated coefficients 0.004 (0.016) –0.039 (0.022) 0.004 (0.017) –0.038 (0.021) 0.008 (0.014) –0.038 (0.021) 0.007 (0.016) –0.038 (0.021) Impact of regime-independent regressors initialit popgrit invit openit gspendit f diit instit hipccpit natresit d1 Estimated coefficients Estimated coefficients Estimated coefficients Estimated coefficients –4.580 (3.202) 0.176 (0.486) 0.104 (0.047) 2.622 (0.823) –0.153 (0.038) 0.579 (0.097) 0.035 (0.045) – – –5.244 (2.013) –4.389 (2.964) 0.189 (0.483) 0.105 (0.046) 2.634 (0.828) –0.150 (0.035) 0.576 (0.097) 0.038 (0.048) –0.129 (0.785) – –5.211 (2.103) –3.661 (2.979) 0.236 (0.455) 0.115 (0.046) 2.388 (0.862) –0.126 (0.034) 0.531 (0.104) 0.033 (0.045) – 0.041 (0.039) –5.468 (2.061) –3.557 (2.839) 0.242 (0.457) 0.117 (0.045) 2.401 (0.880) –0.125 (0.032) 0.530 (0.105) 0.034 (0.047) –0.060 (0.804) 0.039 (0.040) –5.418 (2.128) Observations N Low-debt regime High-debt regime 156 38 72 38 Notes: Standard errors are in parentheses.  ,  ,  indicate significance at 10%, 5% and 1%. Estimation results are from a Matlab code written by Kremer et al. (2013). Following Hamilton and Flavin (1986), this study uses stationarity tests to test for debt sustainability in the two groups of countries. Panel unit root tests advanced by Breitung (2000), Im et al. (2003; hereafter IPS test), Levin et al. (2002; hereafter LCC test) and Pesaran (2007) are applied in this study. Panel unit root tests results on the debt-to-GDP ratio for the two groups of countries are reported in Table 6. For the first group of countries (denoted as Group 1 in Table 6), the results indicate that when Table 6: Panel unit root tests on the debt-to-GDP ratio Group 1 Group 2 Panel unit root tests Constant Constant and Trend Constant Constant and Trend Breitung LLC IPS PESCADF 0.004  0.085 0.057 0.078 0.470 0.002 0.561 0.138 0.595 0.963 0.598 0.929 0.423 0.888 0.944 0.254 Note: LLC stands for Levin, Lin and Chu (2002). IPS stands for Im, Pesaran and Shin (2003). PESCADF is the Pesaran (2007) panel unit root test in the presence of cross-sectional dependence. Values reported in the table are p- values.  and  indicate rejection of null hypothesis of a unit root at 10% and 1% level. Group 1 (Group 2) is the group of countries for which the mean debt-to-GDP ratio is less (greater) than the debt threshold of 90%. © 2017 The Author. African Development Review © 2017 African Development Bank 481 Threshold Effects of Debt on Economic Growth the deterministic part considered in the test is the intercept, the null hypothesis of a unit root in the debt-to-GDP ratio is rejected for all the panel unit root tests used, at the 1 per cent level with the Breitung test and at the 10 per cent level with the LLC test and IPS test. But when the trend is considered, the debt-to-GDP ratio is stationary only with LLC test (at 1 per cent level). However, for the second group of countries (denoted as Group 2 in Table 6), the results indicate that for all the panel unit root tests used, the null hypothesis of the presence of a unit root in the ratio of debt-to-GDP cannot be rejected, suggesting that the ratio of debt-to-GDP is not mean-reverting. This implies that the hypothesis of debt sustainability seems to hold for the first group of countries and not for the second group. However, this is to be taken with caution since panel unit root tests have different asymptotic assumptions and tend to perform differently depending on the number of panels and the number of time periods considered. Indeed the first group contains 26 countries while the second contains only 12 countries (see Table 3). In addition, nonlinearity and the presence of breaks in the data are not taken into account in this study and could affect panel unit root test results. 5. Concluding Remarks This study aimed at examining the debt threshold effects on economic growth in Africa. Both non-dynamic and dynamic panel threshold regression approaches respectively initiated by Hansen (1999) and Kremer et al. (2013) are applied. The estimation results show that with Hansen’s (1999) approach, the debt threshold estimate can be sensitive to control variables added in the growth equation. However, when applying dynamic panel threshold regression approach, the debt threshold estimate remains the same even after adding more control variables. The estimation results point to the existence of nonlinearities in the debt–growth nexus in Africa. While low debt is neutral or growth-enhancing, high debt is consistently detrimental to growth for all the cases considered. On the impact of control variables, the results show that investment, openness to trade, foreign direct investment, as well as natural resource rents enhance economic growth in Africa. However, government spending affects growth negatively while the level of democracy does not affect growth. This study shows that caution is needed when suggesting a debt threshold since this can be sensitive to modelling choice and to growth control variables included in the model. Nonlinearities in the debt–growth nexus in Africa are established but further analysis is needed to suggest a policy. Notes 1. http://www.agenceecofin.com/gestion-publique/1105-38001-le-fmi-met-en-garde-les-pays-africains-contre-l-endettementlie-au-financement-des-projets-dinfrastructures 2. Under the null hypothesis, classical tests have non-standard distributions. Bootstrapping is therefore used to simulate the asymptotic distribution of the likelihood ratio test. In this study, 1,000 replications are used. 3. Economic growth in a number of African countries has been driven by a natural resource boom. 4. The empirical literature gives mixed evidence on the impact of population growth on economic growth in developing countries. For instance, Kremer et al. (2013) and Eggoh and Muhammad (2014) find a negative impact of population growth on economic growth, Furuoka (2009) finds a positive impact while Vinayagathasan (2013) finds that population growth has no impact. 5. Countries had to qualify for debt relief by complying with certain conditions, that is, carrying out strong programmes of macroeconomic adjustment and structural reforms designed to promote growth and reduce poverty and only countries that reach the completion period qualified. 6. Heavily Indebted Poor Countries. 7. Economic growth in a number of African countries has been driven by a natural resource boom. 8. Excluding seven largest economies does not change significantly the results. These results are not reported but are available upon request. 9. We do not use the 102 per cent threshold which was obtained for some cases since countries with mean ratio of debt/GDP greater than this threshold would be few, which can bias panel unit root testing. © 2017 The Author. African Development Review © 2017 African Development Bank 482 A. Ndoricimpa References Abbas, S. A., N. Belhocine, A. ElGanainy and M. Horton (2010), ‘A Historical Public Debt Database’, IMF Working Paper, WP/10/245. Akobeng, E. (2016), ‘Growth and Institutions: A Potential Medicine for the Poor in Sub-Saharan Africa’, African Development Review, Vol. 28, No. 1, pp. 1–17. Anyanwu, J. C. (2014), ‘Factors Affecting Economic Growth in Africa: Are There any Lessons from China?’, African Development Review, Vol. 26, No. 3, pp. 468–93. Arellano, M. and O. Bover (1995), ‘Another Look at the Instrumental Variables Estimation of Error Components Models’, Journal of Econometrics, Vol. 68, pp. 29–51. Arnone, M., L. Bandiera and F. Presbitero (2008), ‘Debt Sustainability Framework in HIPCs: A Critical Assessment and Suggested Improvements’, Available at: https://ssrn.com/abstract=871171 or https://doi.org/10.2139/ssrn.871171 Azam, J. P., A. Fosu and N. S. Ndung’u (2002), ‘Explaining Slow Growth in Africa’, African Development Review, Vol. 14, No. 2, pp. 177–220. Barro, R. J. (1996), ‘Democracy and Growth’, Journal of Economic Growth, Vol. 1, No. 1, pp. 1–27. Barro, R. J. and X. Sala-i-Martin (1997), ‘Technological Diffusion, Convergence, and Growth’, Journal of Economic Growth, Vol. 2, No. 1, pp. 1–26. Bick, A. (2010), ‘Threshold Effects of Inflation on Economic Growth in Developing Countries’, Economics Letters, Vol. 108, No. 2, pp. 126–29. Breitung, J. (2000), ‘The Local Power of Some Unit Root Tests for Panel Data’, in B. H. Baltagi (ed.), Advances in Econometrics, Volume 15: Nonstationary Panels, Panel Cointegration, and Dynamic Panels, Amsterdam: JAY Press, pp. 161–78. Burkhart, R. E. and M. S. Lewis-Beck (1994), ‘Comparative Democracy: The Economic Development Thesis’, The American Political Science Review, Vol. 88, No. 4, pp. 903–10. Cecchetti, S. G., M. S. Mohanty and F. Zampolli (2011), ‘The Real Effects of Debt’, BIS Working Paper No. 352, September. Chancellor, E. (2010), ‘Reflections on the Sovereign Debt Crisis’, White Paper, July. Cordella, T., L. Ricci and M. Ruiz-Arranz (2005), ‘Debt Overhang or Debt Irrelevance? Revisiting the Debt-Growth Link’, IMF Working Paper WP/05/223, December. Deshpande, A. (1997), ‘The Debt Overhang and the Disincentive to Invest’, Journal of Development Economics, Vol. 52, pp. 64–111. Edwards, S. (1997), ‘Openness, Productivity and Growth: What Do We Really Know?’ NBER Working Paper No. 5978, March. Egert, B. (2015), ‘Public Debt, Economic Growth and Nonlinear Effects: Myth or Reality?’, Journal of Macroeconomics, Vol. 43, pp. 226–38. Eggoh, J. C. and K. Muhammad (2014), ‘On the Nonlinear Relationship between Inflation and Economic Growth’, Research in Economics, Vol. 68, No. 2, 133–43. Elbadawi, I., B. Ndulu and N. Ndung’u (1997), ‘Debt Overhang and Economic Growth in Sub-Saharan Africa’, in Z. Iqbal and R. Kanbur, External Finance for Low-income Countries, International Monetary Fund, Washington, DC. Furuoka, F. (2009), ‘Population Growth and Economic Development: New Empirical Evidence from Thailand’, Economics Bulletin, Vol. 29, No. 1, pp. 1–14. Giavazzi, F. and G. Tabellini (2005), ‘Economic and Political Liberalizations’, Journal of MonetaryEconomics, Vol. 52, No. 7, pp. 1297–330. © 2017 The Author. African Development Review © 2017 African Development Bank Threshold Effects of Debt on Economic Growth 483 Gonzalez, A., T. Terasvirta and D. van Dijk (2005), ‘Panel Smooth Transition Regression Models’, SSE/EFI Working Paper Series in Economics and Finance, No. 604. Hamilton, J. D. and M. Flavin (1986), ‘On the Limitations of Government Borrowing: A Framework for Empirical Testing’, American Economic Review, Vol. 76, No. 4, pp. 808–19. Hansen, B. E. (1999), ‘Threshold Effects in Non-Dynamic Panels: Estimation, Testing, and Inference’, Journal of Econometrics, Vol. 93, pp. 345–68. Ibarra, R. and D. Trupkin (2016), ‘Reexamining the Relationship between Inflation and Growth: Do Institutions Matter in Developing Countries?’ Economic Modelling, Vol. 52, pp. 332–51. Im, K. S., M. H. Pesaran and Y. Shin (2003), ‘Testing for Unit Roots in Heterogeneous Panels’, Journal of Econometrics, Vol. 115, pp. 53–74. Imbs, J. and R. Ranciere (2005), ‘The Overhang Hangover’, World Bank Policy Research Working Paper 3673, August. Jeanty, P. W. (2010), ‘Using the World Development Indicators Database for Statistical Analysis in STATA’, The STATA Journal, Vol. 10, No. 1, pp. 30–45. Kaminsky, G. L. and A. Pereira (1996), ‘The Debt Crisis: Lessons of the 1980s for the 1990s’, Journal of Development Economics, Vol. 50, No. 1, pp. 1–24. Kremer, S., A. Bick and D. Nautz (2013), ‘Inflation and Growth: New Evidence from a Dynamic Panel Threshold Analysis’, Empirical Economics, Vol. 44, No. 2, pp. 861–78. Kumar, S. M. and J. Woo (2010), ‘Public Debt and Growth’, IMF Working Paper 174, July. Levin, A., C.-F. Lin and C.-S.J. Chu (2002), ‘Unit Root Tests in Panel Data: Asymptotic and Finite-sample Properties’, Journal of Econometrics, Vol. 108, pp. 1–24. Mijiyawa, A. G. (2013), ‘Africa’s Recent Economic Growth: What Are the Contributing Factors?’, African Development Review, Vol. 25, No. 3, pp. 289–302. Ncube, M. and Z. Brixiov
a (2015), ‘Public Debt Sustainability in Africa: Building Resilience and Challenges Ahead’, Working Paper Series No. 227, African Development Bank, Abidjan. Ndoricimpa, A. (2014), ‘Heterogeneous Panel Causality between Exports and Growth in COMESA Countries’, Journal of Developing Areas, Vol. 48, No. 4, pp. 349–61. Pattillo, C., H. Poirson and L. Ricci (2002), ‘External Debt and Growth’, IMF Working Paper WP/02/69, April. Pattillo, C., H. Poirson and L. Ricci (2004), ‘What Are the Channels through which External Debt Affects Growth?’, IMF Working Paper WP/04/15, January. Pattillo, C., H. Poirson and L. A. Ricci (2011), ‘External Debt and Growth’, Review of Economics and Institutions, Vol. 2, No. 3, Article 2. doi: 10.5202/rei.v2i3.45. Pesaran, M. H. (2007), ‘A Simple Panel Unit Root Test in the Presence of Cross-section Dependence’, Journal of Applied Econometrics, Vol. 22, No. 2, pp. 265–312. Pescatori, A., D. Sandri and J. Simon (2014), ‘Debt and Growth: Is There a Magic Threshold?’, IMF Working Paper, WP/14/ 34. Presbitero, A. F. (2010), ‘Total Public Debt and Economic Growth in Developing Countries’, MoFiR Working Paper No. 44, October. Reinhart, C. M. and K. S. Rogoff (2010), ‘Growth in a Time of Debt’, American Economic Review, Vol. 100, No. 2, pp. 573–78. Seleteng, M., M. Bittencourt and R. van Eyden (2013), ‘Non-linearities in Inflation-Growth Nexus in the SADC Region: Panel Smooth Transition Regression Approach’, Economic Modelling, Vol. 30, pp. 149–56. © 2017 The Author. African Development Review © 2017 African Development Bank 484 A. Ndoricimpa Sepehri, A. and S. Moshiri (2004), ‘Inflation-Growth Profiles across Countries: Evidence from Developing and Developed Countries’, International Review of Applied Economics, Vol. 18, pp. 191–207. Thanh, S. D. (2015), ‘Threshold Effects of Inflation on Growth in the ASEAN-5 Countries: A Panel Smooth Transition Regression Approach’, Journal of Economics, Finance and Administrative Science, Vol. 20, pp. 41–48. UNCTAD (1999), ‘Foreign Direct Investment and the Challenge of Development’, World Investment Report, UNCTAD, Geneva. Vinayagathasan, T. (2013), ‘Inflation and Economic Growth: A Dynamic Panel Threshold Analysis for Asian Economies’, Journal of Asian Economics, Vol. 26, pp. 31–41. © 2017 The Author. African Development Review © 2017 African Development Bank