(2) EXTERNAL DEBT, DEBT BURDEN AND ECONOMIC GROWTH NEXUS: EMPIRICAL EVIDENCE AND POLICY LESSONS FROM SELECTED WEST AFRICAN STATES

Đưa bài này lên theo đề nghị của học viên cao học 

Trường Đại học Kinh tế thành phố Hồ Chí Minh:

(2) EXTERNAL DEBT, DEBT BURDEN AND ECONOMIC GROWTH NEXUS: EMPIRICAL EVIDENCE AND POLICY LESSONS FROM SELECTED WEST AFRICAN STATES

 BY

CHINEDU SAMUEL OKONKWO

SCHOOL OF ECONOMICS, UNIVERSITY OF NOTTINGHAM, UK

&

GBADEBO OLUSEGUN ODULARU

REGIONAL POLICIES AND MARKETS ANALYST,

FORUM FOR AGRICULTURAL RESEARCH IN AFRICA (FARA), ACCRA, GHANA.

 

4.0     EMPIRICAL RESULTS

Introduction

The aim of our specified model is to investigate the relationship between external debt and growth. To do this, we follow a full cointegration technique. We start by running a unit root test to identify the order of integration of our variables. Next, we test for cointegration among variables integrated of the same order. Lastly, an Error Correction Mechanism (ECM) is set up to capture short run dynamics as well as speed of adjustment of disequilibrium in the long run growth path. 

Unit Root Test

For the unit root test, we first plotted each series to examine its behaviour. Where a trend-like behaviour was discovered, we assume presence of a trend[1] and intercept but where there were no signs of a trend, we assume only an intercept for the unit root test. Real GDP per capita for Ghana and Cote d’Ivoire were observed to be upward trending.

The result in table 4.1 below is a summary of unit root test for all variables at levels and first difference. The result revels presence of unit root for all variables at levels. The variable become stationary at first difference since it appears to be integrated of order zero – I(0). For example, both ADF and PP unit root test of LRYPC for Cote d’Ivoire are less than the 5% critical values in absolute terms. Thus, we cannot reject the null hypothesis that the series is non stationary. After a test on the first difference, the ADF and PP t statistic values (-4.072 and -4.092 respectively) were greater than the 5% critical value( -2.946) in absolute terms and hence, we conclude that it is stationary.

For the ADF based unit root test on first differences, only that of variable LEXTY for Cote d’Ivoire seemed to show signs of unit root but this was strongly rejected at 1% level by the PP test. With this exception, all variables at first difference were found to be stationary at either 1% or 5% levels for both ADF and PP tests. For Senegal, a conflict of result was noticed between the unit root tests at levels for FDIY. However, given ADF support for non stationarity and a confirmation of stationarity at first difference, we assume the variable to be integrated of order one - I(1).

Table 4.1: Summary of ADF and PP Tests for Unit Root

	LEVELS					FIRST DIFFERENCE
Variables	ADF Test		PP Test			ADF Test		PP Test
	T- stat	5% CV	T- stat	5% CV	I(d)	T- stat	5% CV	T- stat	5% CV	I(d)
	COTE D'IVOIRE
LRYPC	-2.359 (1)	-3.540	-2.397	-3.537	I(1)	-4.072* (0)	-2.946	-4.092*	-2.946	I(0)
FDIY	-2.635 (0)	-2.943	-2.663	-2.943	I(1)	-8.340 * (0)	-2.946	-8.340*	-2.946	I(0)
LEXTY	-2.043 (2)	-2.948	-2.112	-2.943	I(1)	-1.990 (1)	-2.948	-4.440 *	-2.946	I(0)
LGDIY	-1.213 (0)	-2.949	-1.312	-2.943	I(1)	-4.933* (0)	-2.946	4.946*	-2.946	I(0)
LTDSE	-1.258 (1)	-2.946	-0.967	-2.943	I(1)	-3.87* (0)	-2.946	3.889*	-2.946	I(0)
	GAMBIA
LRYPC	-1.604 (0)	-2.943	-1.721	-2.943	I(1)	-5.709* (0)	-2.946	-5.720*	-2.946	I(0)
FDIY	-0.139 (1)	-2.946	-1.387	-2.943	I(1)	-10.329* (0)	-2.946	-11.462*	-2.946	I(0)
LEXTY	-2.479 (0)	-2.943	-2.287	-2.943	I(1)	-4.690* (0)	-2.946	-4.882*	-2.946	I(0)
LGDIY	-2.039 (0)	-2.943	-2.075	-2.943	I(1)	-5.748* (0)	-2.954	-5.799*	-2.946	I(0)
LTDSE	-1.915 (2)	-2.948	-2.667	-2.943	I(1)	-7.195* (1)	-2.948	-8.363*	-2.946	I(0)
	GHANA
LRYPC	-0.821 (0)	-3.537	-0.573	-3.537	I(1)	-4.218* (0)	-2.946	-4.218*	-2.946	I(0)
FDIY	-2.054 (0)	-2.943	-2.054	-2.943	I(1)	-7.194* (0)	-2.946	-8.021*	-2.946	I(0)
LEXTY	-1.375 (0)	-2.943	-1.476	-2.943	I(1)	-5.466* (0)	-2.946	-5.470*	-2.946	I(0)
LGDIY	-0.933 (0)	-2.943	-0.933	-2.943	I(1)	-6.503* (0)	-2.946	-6.548*	-2.946	I(0)
LTDSE	-0.621 (0)	-2.943	-0.621	-2.943	I(1)	-6.098* (0)	-2.946	-6.099*	-2.946	I(0)
	SENEGAL
LRYPC	-1.981 (0)	-2.943	-1.871	-2.943	I(1)	-7.280* (0)	-2.946	-7.285*	-2.946	I(0)
FDIY	-2.329 (1)	-2.946	-4.729*	-2.943	I(1)	-10.993* (0)	-2.946	-16.893*	-2.946	I(0)
LEXTY	-1.517 (0)	-2.951	-1.654	-2.943	I(1)	-4.657* (0)	-2.946	-4.847*	-2.946	I(0)
LGDIY	-1.01 (0)	-2.943	-1.042	-2.943	I(1)	-6.724* (0)	-2.946	-6.734*	-2.946	I(0)
LTDSE	-2.269 (0)	-2.943	-2.204	-2.943	I(1)	-7.307* (0)	-2.946	-7.238*	-2.946	I(0)

Note: *, ** indicate statistical significance at 1% and 5% levels respectively. Numbers in parenthesis by the T-statistic are optimal lag lengths. I(d) is the order of integration.

Cointegration Test

As a necessary condition to proceed with the test for co-integration, all variables must be non stationary and integrated of the same order. Having obtained non-stationarity with integration of order 1 for all variables at levels, we immediately employ the Johansen’s technique to cointegration. This method does better when a multivariate system is considered. This method makes use of the trace statistic (λ_trace) and the maximum statistic (λ_max) to determine the rank of the cointegrating vectors in the VAR system.

Table 4.2: Johansen’s Test for Cointegration

	Null	Alt.	T Stat (λTrace)	5% CV for  λTrace	T Stat (λMax)	5% CV for  λMax
	r = 0	r ≥ 1	120.348*	88.8038	50.15189*	38.33101
COTE D'IVOIRE	r ≤ 1	r ≥ 2	70.19612*	63.8761	28.74268	32.11832
	r ≤ 2	r ≥ 3	41.45344	42.91525	19.57552	25.82321
	r ≤ 3	r ≥ 4	21.87792	25.87211	12.2037	19.38704
	r ≤ 4	r = 5	9.674223	12.51798	9.674223	12.51798
	r = 0	r ≥ 1	105.0939*	69.81889	57.65219*	33.87687
	r ≤ 1	r ≥ 2	47.44176	47.85613	23.75784	27.58434
GAMBIA	r ≤ 2	r ≥ 3	23.68392	29.79707	16.30871	21.13162
	r ≤ 3	r ≥ 4	7.375211	15.49471	5.937756	14.2646
	r ≤ 4	r = 5	1.437455	3.841466	1.437455	3.841466
	r = 0	r ≥ 1	93.07128*	88.8038	36.27307	38.33101
	r ≤ 1	r ≥ 2	56.79821	63.8761	27.99585	32.11832
GHANA	r ≤ 2	r ≥ 3	28.80236	42.91525	13.19479	25.82321
	r ≤ 3	r ≥ 4	15.60757	25.87211	9.377884	19.38704
	r ≤ 4	r = 5	6.22969	12.51798	6.22969	12.51798
	r = 0	r ≥ 1	95.81501*	88.8038	37.10218	38.33101
	r ≤ 1	r ≥ 2	58.71283	63.8761	29.09788	32.11832
SENEGAL	r ≤ 2	r ≥ 3	29.61495	42.91525	15.51528	25.82321
	r ≤ 3	r ≥ 4	14.09968	25.87211	8.537959	19.38704
	r ≤ 4	r = 5	5.561716	12.51798	5.561716	12.51798

Note: * denotes rejection of the null hypothesis at 5% significance level. ‘r’ is the number of cointegrating vectors.

Where the calculated test statistic exceeds the critical values from the Johansen’s tables, then we reject the null hypothesis that there are r cointegrating vectors and affirm the alternative which suggests that there are r +1 cointegrating vectors for the trace or more than r for the maximum eigenvalues. Selecting the optimal lag length is essential in testing for cointegration because the Johansen test is lag sensitive. As a result, we proceed by using the ‘lag length criteria’[2] test which suggests an optimal lag length to use for the model. We note that using many lags can ‘overwork’ the VAR and also considering that we employ annual time series data,  we employ the f+1 rule of thumb[3] by estimating the model starting with 2 lags and re-estimate again but each time dropping a lag. We select the optimal lag length as that which minimizes the AIC and SC criteria. With usual conflicting choices of AIC and SC, we scrutinize the result based on each lag and eventually select the lag that appears to better fit the data. Interestingly, we find the lag length to accept up 1 lags in most cases, i.e. a VAR(1) model. Table 5.3 below is a summary of results for the Johansen’s test for cointegration.

As summarized in the table 4.3 above, the result for the trace and the maximum value tests suggests rejection of the hypothesis of no cointegration for all countries. As a result, this implies that there is cointegration among variables in our specified model and hence, there exist a long run equilibrium relationship among the variables.

To sight an example using Gambia, we observe that the calculated trace statistic (105.09) exceeds the 5% critical values (69.81) and hence, we reject the null hypothesis that there are no cointegrating vectors. Again, we compare the trace value with the critical value at 5% level of significance for a test for more than one cointegrating vectors. It was not possible to reject the null that there are at least 1 cointegrating equations (r ≤ 1) since the trace value (47.44) is less than the 5% critical value (47.86). The same method employed in trace to identify the number of cointegrating vectors can also be applied in the max test to do the same. There was a conflict between the Trace and Max statistics for number of cointegrating vectors (CE) in Cote d’Ivoire, Ghana and Senegal.

For Cote d’Ivoire we find at least 1 CE hence, and test for the possibility of 2 CEs. For the case of Ghana and Senegal, while the trace statistic suggests the model has only 1 cointegrating equation, the max value suggested absence of any cointegrating equations but only just missing the 5% level of significance. However, if a 10% critical level was allowed for, we would be unable to reject that there is at least 1 cointegrating equation. Therefore, all comparison, based on a quick VECM supports the rejection of the null of no cointegration among our variables for all countries as the ECT become significant.

Since our major interest is in the growth equation, we initially utilize all our 4 other variables in explaining the output relation in the long run. To explain the long run relationship, we turn to the normalized cointegrating equation. However, after a critical examination of the normalized cointegration equation, we find it may not be to be truly revealing of our data. Hence, we proceed to check the short run error correction model. 

Error Correction Model

According to the Granger Representation Theorem, we can formulate an error correction model in order to capture the short run relationship. Hence, we formulate a Vector Error Correction Model (VECM) which includes our Error Correction Term (ECT). The ECT captures the speed of adjustment to a disturbance in the long run equilibrium in respective vectors of our model. Since all the terms in the VECM are stationary, we can apply the conventional t statistic in evaluating the estimates of the model. We concentrate on the growth relation of our model. Surprisingly, results of the full VECM give too many insignificant estimates (see table A1 in the appendix). Consequently, further checks were carried out to ensure that the model does not suffer from estimation bias.

We test for weak exogeneity of variables in the VAR system to determine whether the variables are weakly exogenous to the system. Results of the weak exogeneity test suggests that the variables for all our selected countries fits into the VAR system. Following this, we re-estimate the VAR system including many lags but find the model to suffer from degree of freedom problem as most diagnostics test could no longer be calculated. Although the estimates of the VECM became more significant, we find it to have multiple CEs and hence, fear that the model may have been ‘over worked’ to produce inconsistent results.

What the VAR model does not do is to provide estimates of contemporaneous terms. Understanding this, we proceed with Engel-Granger (EG) 2 step technique to cointegration in order to explore the error correction estimates. We find this to be more revealing since some estimates at contemporaneous terms seemed to matter more for our model. Sen et al (2007) finds the contemporaneous terms of a similar model to be very significant for explaining the debt-growth equation. For this reason, we estimate a short run model with dynamics based on EG 2 step technique. This method is similar to the VECM but instead of looking at a vector of equations, the ECM estimates individual equation, hence, allowing for a flexible dynamic specification[4].

To do this, we first employ the ADF residual based test as suggested by Engle & Granger (1987) to test for cointegration. The ‘cointegrating’ or ‘static regression’, which captures the long run relationship of variables at levels, is estimated and the generated residuals are tested for unit root using the ADF and PP tests. The null hypothesis for this test states that there is no cointegration and this can only be rejected where the residual is found to be stationary at levels. Table 4.3 below is a table of the cointegrating regression.

Table 4.3: Cointegrating Regression

VARIABLE	COTE D'IVOIRE	GAMBIA	GHANA	SENEGAL
Constant	6.989 [30.24]	5.535 [82.41]	5.947 [61.82]	6.282 [82.74]
FDIY	-0.045 [-4.21]	-0.003 [-1.59]	0.01 [0.88]	-0.005 [-0.75]
LEXDY	-0.302 [-7.33]	-0.008 [-0.34]	-0.049 [-1.40]	-0.067 [-4.45]
LGDIY	0.203 [4.70]	0.114 [3.91]	0.031 [0.76]	0.066 [3.09]
LTDSE	0.158 [6.49]	-0.007 [-0.38]	-0.12 [-5.35]	-0.007 [-0.359]
Adj. R-squared	0.936	0.356	0.757	0.70
S.E. of Reg.	0.058	0.045	0.067	0.035
F-statistic	136.21 (0.00)	6.11 (0.00)	29.89 (0.00)	22.53 (0.00)

*Note. We cannot test for statistical significance since the results of t-ratio in a co-integrating regression are not t-distributed.

The residual based test for unit root is summarized in table 4.4 below. The residuals based test of the cointegrating regression shows that the residuals for all countries (except Gambia) are void of unit root and hence we reject the null hypothesis that there is no cointegration. This was accepted by both the ADF and PP residual based tests. For Gambia, the calculated t statistics is less than the 10% critical value in absolute terms though it marginally missed the 10% significance level. However, taking inference from the Johansen’s test for cointegration, our conjecture is that there may cointegration, hence we test for it in the ECM.

Table 4.4: Residual Based Test for Cointegration (Growth Equation)

Country	Variables	ADF Test		PP Test		I(d)
		T- stat	10% CV	T- stat	10% CV
COTE D'IVORE	Resid	-4.189* (1)	-2.612	-3.682*	-2.610	I(0)
GAMBIA	Resid	-2.529 (1)	-2.612	-2.293	-2.610	I(1)
GHANA	Resid	-3.929* (3)	-2.612	-3.203**	-2.610	I(0)
SENEGAL	Resid	-3.922* (0)	-2.610	-3.949*	-2.610	I(0)

Note: *, **,*** indicate statistical significance at 1%, 5%, and 10%  levels respectively. Resid is the Residual. Numbers in parenthesis under t-statistics are optimal lag lengths. I(d) is the order of integration.

Specifying an error correction model in the single line equation enables us to see short term effects on the dependent variable. Since our focus is to investigate a debt-growth relationship of external debt, our model is specified as follows:

                                       (5.1)                                                                                       

Where all variables remain as discussed earlier and ECT_t-1 is the Error Correction Term.

Since we employ annual data, we initially start with a model with two lags and test down to obtain our parsimonious error correction model. Diagnostic tests on the model showed, to a great extent, that our models are rightly specified. However, we observe that the model for Cote d’Ivore seemed not to be cointegrated since the ECT for Cote d’Ivoire appeared negative. Adding more lags to the model could change this but could also yield inconsistent estimates though more lags would be suitable if the data were quarterly. Aside this, all the models pass diagnostic test and hence, table 4.5 below is used in interpreting our short run model above.

In the growth model for Ghana and Senegal, we observe that a substantial amount of debt relief was received in 2006. Considering this, we initially control for this by adding dummies to capture this effect. The dummies for Ghana seemed not to matter in the model where as that for Senegal seemed significant at 10% level. In that case, our conjecture is that the debt relief may have impacted more positively on Senegal hence, this is included in our parsimonious error correction model while that of Ghana was left out. We also estimate the growth relationship prior to 2006 and find the result to be insignificantly different for all countries[5].

Table 4.5: Error Correction Model (Output Growth Equation)

Variables	COTE D'IVOIRE	GAMBIA	GHANA	SENEGAL
DFDIY	-0.005 [-0.52]	-0.005 [-1.81]***	-0.004 [-0.66]	0.002 [0.41]
DLEXDY	-0.125 [-1.93]***	-0.051 [-1.77]***	-0.075 [-1.75]***	-0.098 [-2.96]*
DLGDIY	0.082 [1.80]***	0.07 [2.06]**	0.02 [0.80]	-0.001 [-0.05]
DLTDSE	0.025 [0.79]	0.001 [0.04]	-0.031 [-1.74]***	-0.03 [-2.17]**
D(LRYPC(-1))	--	0.279 [1.42]	0.35 [2.37]**	--
D(FDIY(-1))	--	0.001[0.38]	--	--
D(FDIY(-2))	--	0.004[1.97]***	--	--
Dummy06	--	--	--	0.074 [1.85]***
ECTt-1	--	-0.287 [-1.89]***	-0.40 [-3.66]*	-0.47 [-3.24]*
Constant	-0.008 [-1.13]	0.005 [0.87]	-8.49E [-0.001]	-0.071 [-1.82]***
R-squared	0.18	0.35	0.467	0.548
Adjusted R2	0.078	0.158	0.356	0.458
DW Stat	1.52	1.6	1.90	1.858
F-statistic	1.76 (0.16)	1.8 (0.12)	4.23 (0.003)*	6.06  (0.000)*

Note: *, **,*** indicate statistical significance at 1%, 5% and 10%  levels respectively. Numbers in square parenthesis are t-statistics while number in parenthesis by the F stats are probability values.

Some work which investigate the debt-growth relationship like Elbabawi (1996) find that accumulated debt indicators (measured by lagged value of debt indicators) seemed to influence growth negatively. For this, we estimate our short run equation using up to two lags for our debt variables as well as our other explanatory variables. Again, we find that the accumulated debt variables captured by lags have low explanatory power to our model. A full display of this result is presented in table A2 in the appendix. Hence, we re-estimate our short run model relevant variables and exempt some insignificant variables.

In table 4.5 above, the short run model suggests that contemporaneous terms seemed to matter greatly for the model. As is generally the case, lagged values of output seemed to exhibit a positive relationship with the output equation for all relevant countries. It is expected that growth in a country in a previous year could be leveraged upon to enhance growth in the next period. However, lagged period output growth seemed to exhibit a positive effect for Ghana and Gambia   alone though that of Gambia seemed insignificant[6]. The result indicates that a 1% increase in the first period lag of DLRYPC corresponds with about 0.35% increase in output growth for Ghana. 

The net foreign investment variable has mixed signs at current and lagged periods for relevant countries. Foreign direct investment (FDI) is expected to foster technological transformations which could accelerate growth. For some countries, the variable was significant even up to the second lag (for Gambia). This is somewhat plausible since some investments are long term and could take several periods to become beneficial. All countries except Senegal exhibited immediate short run negative relationship with economic growth. Though not statistically significant at the first lag, the impact of the second period lag appeared to be statistically significant at the 10% level. Since the signs changed from an initial negative to significant positive in the lagged period, it could be that Gambia has not been quick to absorb the pros of foreign direct investment.

The external debt as a percentage of GDP seemed to be negatively relevant for all countries. This seemed to be in contrast with findings of Adegite et al (2008) and Frimpong et al (2006) who both find it to be insignificantly positive for Nigeria and Ghana respectively. However, our result is in line with findings of Fosu (1996) who finds debt to have a direct negative effect on economic growth in SSA countries by affecting productivity. Were (2001) also finds the same thing for Kenya. What this may suggest is that increase in external debt stocks have some direct negative effect on output growth.

Ideally, we may expect that increases in external borrowing could make available funds for productive purposes. However, this may not have been the case as activities to which these funds are channelled could have been unproductive. Many of such projects may be ‘white elephant’ projects which are politically motivated with no positive net present value. Furthermore, it may be that most of these countries have gone beyond the point after which debt stimulates growth. (Elbadawi,1996). The result suggests that Cote d’Ivoire appeared to suffer the greatest debt impact amongst our selected countries. This is not surprising for a nation whose external debt was valued at about $13.8 billion in 2007 which is more than 3 times that of Ghana and yet having a debt service to export ratio of 0.05. If we use this variable to make an inference on debt overhang as done by Cordella et al (2005) and Sen et al (2007) then the result show that our selected countries are still ‘overhung by debt’. However, assuming that the accumulation of debt (now captured by lagged debt ratios) reflects debt overhang as proposed by Elbadawi (1996) and Were (2001) then our observed result (which were insignificant) does not support the debt overhang.

On the whole, the short run estimates showed that gross domestic investment as a percentage of GDP seemed to have an immediate positive impact on output except for Senegal where it appeared negative, seemed statistically insignificant and had an elasticity near zero. Though not statistically significant, the negative sign for Senegal may go a long way to suggest that external debt may have provided a disincentive for domestic capital formation such that the output growth has an immediate negative effect on output growth. Some studies like Frimpong et al (2006) found first period lagged domestic investment to GDP ratio to be negative and statistically significant for the period of 1970 -1999 and claims that it may be an effect of the debt overhang effect. Sen et al (2007) paper finds differently on this subject. They find gross domestic investment to GDP ratio to have a significant effect on output growth for Latin American and Asian economies. For low income countries[7], Clements (2003) finds gross domestic investment to GDP to matter positively to output growth. By results, we cannot infer prevalence of debt overhang effect through the channel of gross domestic investment to GDP ratio on output growth.

Debt service tends to rise with growth in external debt accumulation. The rise in debt service is necessitated by repayment obligations. The debt service ratio is observed to be very important in explaining output growth. All countries but Cote d’Ivoire and Gambia appeared to have their output growth to be negatively influenced by total debt servicing ratio. However, associated with the positive signs for the two countries are statistical insignificant coefficients. The total debt servicing relation appeared to be statistically significant for Ghana and Senegal. The negative sign is seemingly plausible and theoretically intuitive since we expect some form of loss in potential output growth which could have been gained if the amount for servicing debt were used for productive economic activities. Our finding is generally consistent with studies like Savvides (1992), Metwally et al (1994), and Karagol (2002) among others who also find a negative relationship. Hence, for Ghana and Senegal, it appears that 1% increase in debt servicing led to approximately 0.03% fall in output growth in both countries. The positivity sign associated with Cote d’Ivoire may be due unsustainable debt management policies. This may also be reflective in the debt service ratio which fell from 24.2% in 2000 to 3.7% in 2005 meanwhile real GDP per capita was also fallen over the same period. 

Coefficients of the error correction term (ECT) in our growth equation were observed to be statistically significant for Gambia, Ghana and Senegal. Surprisingly, that of Cote d’Ivoire seemed to be statistically insignificant implying that there may not be any cointegration among the variables. This was not supported by the Johansen’s test for cointegration which is based on a ‘systems’ framework. Since we are considering a residual based approach to the error correction model, the insignificance value of the ECT despite the right sign may be suggestive the there is no cointegration among our variables for Cote d’Ivoire.

All the ECT terms possessed the expected negative sign. The negative signs suggest that when output growth exceeds its long-run equilibrium value, then there exist a mechanism which corrects disequilibrium in the growth rate in the subsequent period. For example, the results show that a distortion in the long run equilibrium output growth in Gambia, Ghana and Senegal is corrected in the next period by 28.7%, 40% and 47% respectively. Additionally, we do not find any ECT to be overshooting[8].

4.5 A Debt-Investment Relation

‘Liquidity constraint hypothesis suggest that debt service obligations reduce availability of fund for productive investment (Hoffman et al, 1991). Therefore, it would be interesting to briefly investigate the effect of economic debt accumulation and debt servicing on investment since it is believed that external debt servicing could ‘crowd out’ investment. We note that we cannot examine fully the effect of crowding out since our investment variable does not distinguish between private and public investments. 

We employ the same EG residual approach to cointegration in order to examine the effect of external debt on investment in the short run. Again, all variables seemed to be cointegrated after testing their residual for unit roots[9]. The formulation of the model is similar to that of the growth equation only that gross investment becomes the subject of the equation. The summary of the short run equation can be seen in table 5.6 below.

Consistent with some theoretical hypothesis like that of debt overhang, It is expected that external debt would be deleterious to investment. Interestingly, it was observed that external debt had an immediate positive influence on investment for all countries. However, only that of Gambia appeared statistically significant. This finding tallies with Fosu (1996) who shows that investment to be higher in high debt countries than lower debt countries. Again, accumulated investment, captured by the first and second period lags, appeared statistically irrelevant in all countries except for Senegal where there was a significant negative impact based on the second lagged period. This may be suggestive that adverse effect of accumulated external debt on investment may take some time to impact in Senegal. 

On the other hand, debt servicing is expected to act as a negative effect on investment (especially private). From the result, it appears that investment in Ghana seemed to be affected negatively by the debt service ratio. Cote d’Ivoire also had the expected negative relationship but was statistically insignificant. For Gambia and Senegal, both the current and lagged terms appeared to have model positive effects, exhibiting no signs of ‘taxing or crowding out’ domestic investment. Were (2001) also discovered that lagged debt service ratio did not seem to affect private investment negatively in Kenya.

Table 4.6: Error Correction Model (Investment Equation)

Variables	COTE D'IVOIRE	GAMBIA	GHANA	SENEGAL
DFDIY	0.123 [4.03]*	0.019 [2.3]**	0.105 [3.28]*	-0.01 [-0.38]
DLEXDY	0.365 [1.54]	0.348 [2.39]**	0.17 [0.91]	0.232 [1.40]
DLRYPC	0.923 [1.54]	1.453 [1.89]***	1.883 [2.2]**	0.221 [0.23]
DLTDSE	-0.145 [-1.35]	0.101 [2.18]**	-0.031 [-1.83]***	0.03 [0.26]
D(LGDIY(-1))	0.416 [2.56]**	--	--	--
D(LEXDY(-1))	--	--	--	-0.097 [-0.60]
D(LEXDY(-2))	--	--	--	-0.568 [-2.38]**
D(LTDSE(-1))	--	2.469[2.47]**	--	0.156 [1.71]***
ECTt-1	-0.55 [-3.52]*	-0.374 [-3.16]*	-0.513 [-3.51]*	--
Constant	-0.013 [-0.57]	-0.01 [-0.38]	0.005 [0.15]	0.037 [1.23]
R-squared	0.475	0.496	0.431	0.275
Adjusted R2	0.367	0.392	0.339	0.09
DW Stat	2.03	1.41	1.91	1.90
F-statistic	4.37 (0.00]*	4.76 [0.00]*	4.69 (0.00)*	1.46 (0.22)

Note: *, **,*** indicate statistical significance at 1%, 5%, and 10%  levels respectively.

Though this debt- investment model is used to make only suggestive inferences, it is interesting to note that the immediate effect of external debt on investment appears positive but could become negative as experienced by Senegal alone. However, in the growth equation, it seemed to generally have a negative and significant immediate impact on growth. Thus, this prompts that there could be more factors or channels through which external debt accumulation affects economic growth.

5.0       CONCLUSION AND POLICY IMPLICATIONS

Conclusion

The study sets out with the aim to uncover the existent relationship between external debt, debt burden and economic growth for the period 1970 to 2007 in four selected African countries(Cote d’Ivoire, Gambia, Ghana, and Senegal). To do this, we employed time series econometric techniques of cointegration to evaluate the relationship.

We initially start off with a VAR method to examine the a systems effect of indebtedness but find accumulated debt, captured by lagged values, not to matter relatively much. The analysis becomes difficult to evaluate based on this. Hence, we proceed with the Engel Granger 2 step cointegration technique to examine the effect of indebtedness on output growth in an error correction model. Again, immediate past debt appeared to have insignificant effect, we find external debt to exert an immediate and significant negative influence on output growth in all our countries. This finding seems to be in consonance with the “direct effect of debt hypothesis” outlined by Fosu (1996). Furthermore, our result suggests that debt servicing impacts negatively and significantly on output growth for Ghana and Senegal while it was positive for the other two countries but did not seem to matter statistically.

Though our study does not distinguish between private and public investments, gross domestic  investment seemed to generally exhibit positive effect on GDP per capita. This contradicts findings on debt overhang effect through the effect of gross domestic investment to GDP ratio on output growth. Furthermore, a preliminary investigation of the debt-growth relationship showed that external debt does not impact negatively on investment growth in all countries. However, that of Gambia seemed to be more positive and significant which infers that growth in external debt seemed to tally with higher investment growth for smaller countries where investment levels are minimal. Our findings appeared mixed on the effect of debt servicing on investment. It appeared there is a trade-off of potential investment for debt servicing in two countries while the other two countries exhibited a positive relationship.

Though our estimates appears to reveal some plausible results for external debt- growth relationship, the study also encounter some major drawbacks. First, we employ relatively short time series data spanning the period 1970 - 2007. Though, our time horizon is larger than some similar earlier studies, results of the model would be more revealing should there be more observations. Secondly, we are unable to obtain some key variables such as private investment in order to make calculated inferences on other hypothesised debt effects such as the crowding out effect of private investment. Furthermore, additional variables could be also added to augment the model. Lastly, we do not consider in detail the debt relief and forgiveness granted to these nations over the study period. It could be interesting to investigate the effects over time.

Conclusively, we find that external debt and debt burden has country specific effects due to structural differences and appears to have significant immediate impacts. Though few studies have tried to investigate the external debt-growth relationship using a non-linear model, some have found the relationship to be insignificant (like Adegbite et al (2008), for Nigeria and Schclarek (2004) for both developing and industrial countries). However, exploring the possibility of a non-linear relationship and ascertaining the turning point could prove to be interesting in a further study for individual countries. Furthermore, we note that external debt accumulation could be induced by real borrowings and/or fluctuations in interest rate or terms of agreement on loans with different lending institutions. This could provide greater insight into the real effect of borrowings on the economy.

Policy Implications

Generally, econometric result of our study lends, to a large extent, credence to some underlying theoretical propositions like the ‘direct effect of debt hypothesis.’ Changes in external debt has been found to be an important factor influencing economic growth. Steaming from this result, it would be erroneous to support extravagant foreign borrowing. Debt could also have a knock on effect by restricting returns to domestic capital formation which then affects growth negatively in some countries. From our findings, we concur with popular recommendation that external debt should only be sourced for ‘high priority projects’ which are well appraised and having direct positive net present value.

Additionally, external lending from the international community like the World Bank, and IMF in the form of real money should be discouraged for developing countries especially HIPCs unless there is certification that it is used for productive economic activities. This is because there could be a strong adverse selection problem. Hence, there may be also need for provision of technical assistance and possible monitoring of projects to ensure completion and standardization of developmental projects.

Following this, the debt relief initiative for HIPCs may prove abortive if policies in receiving countries remain undesirable and unchanged. There must be operational policies which are development oriented and which lends support for sustainable growth before debt relief can be provided. As a contention, “a once and for all program is greatly superior to a gradual program of increasing relief” (Easterly, 2002). This should free up some capital which would have gone for debt service. Notwithstanding, for ‘once and for all’ program to be effective it must be the case that there is evidence of sound policies in receiving countries and that, in the future, relief will  never be provided again. 

Finally, improving the debt management policies and encouraging transparency may prove to be effective in tackling the debt serving burdens. Our analysis may also suggest that establishing and strengthening policies on foreign direct investment and domestic capital formation should be  fostered in order to benefit fully from their potentially positive influence. It would also reduce dependence on foreign loans, avoid use for non real economic sector and politically motivated selfish aims.

BIBLIOGRAPHY

Adegbite, E. O., Ayadi, F. S. and Ayadi, O. F. (2008). ‘The Impact of Nigeria’s External Debt on Economic Development’. International Journal of Emerging Markets, Vol. 3, 3, 2008, pp. 285-301.

Afxentiou, P. C. and Serletis, A. (1996^a). ‘Foreign Indebtedness in Low and Middle Income Developing Countries’. Social and Economic Studies, Vol. 45, 1, pp. 133-159.

Afxentiou, P. C. and Serletis, A. (1996). ‘Growth and Foreign Indebtedness in Developing Countries: An Empirical Study Using Long-term Cross-Country Data.’ The Journal of Developing Areas, 31, Fall, pp. 25-40.

Amoateng, K. and Amoako, A. B. (1996). ‘Economic Growth, Export and External Debt Causality: The Case of African Countries’. Applied Economics, Vol. 28, 1, pp. 21-27.

Asteriou, D. and Hall, S.G. (2007). ‘Applied Econometrics, A Modern Approach’. Palgrave Macmillan, Basingstoke.

Durbarry, R., Padachi, K. and Seetanah, B. (2008). ‘An Empirical Assessment of External Debt on Economic Growth: The Case of Africa in a Panel VAR’. Journal of International Finance and Economics. Retrieved on 12 August, 2010 from http://findarticles.com/p/articles/mi_6765/is_1_8/ai_n28552128

Boyce, J.K. and Ndikumana, L. (2002). ‘Africa’s Debt: who owes whom?’. University of Massachusetts, Department of Economics and Political Economy Research Institute, Working Paper 32.

Brooks, C. (2002). ‘Introductory Econometrics for Finance’. Cambridge University Press, Cambridge.

Bullow, J. and Rogoff, K. (1990). ‘Cleaning up third world debt without getting taken to the cleaners’.  Journal of Economic Perspectives, Vol. 4, 1, pp. 31-42.

Butts, H. C. 2009. ‘Short Term External Debt and Economic Growth—Granger Causality: Evidence from Latin America and the Caribbean’. The Review of Black political economy, Vol. 36, pp. 93–111.

CIA (2010). ‘The World FactBook’. Retrieved on 15 July, 2010 from https://www.cia.gov/library/publications/the-world-factbook/index.html

Chenery, H.B. and Strout, A. (1966). ‘Foreign Assistance and Economic Development’. American Economic Review, Vol. 56, pp. 679-733.

Chowdhury, K. (1994). ‘A structural analysis of external debt and economic growth: some evidence from selected countries in Asia and the Pacific’. Applied Economics, 26, pp. 1121-1131.

Cordella, T., Ricci, A. and Arranz R. (2005). ‘‘Debt Overhang or Debt Irrelevance? Revisiting the Debt-Growth Link’. WP/05/223, International Monetary Fund, Washington DC.

Clements, B., Bhattarcharya, R. and Nguyen, T.Q. (2003). ‘External Debt, Public Investment, and Growth in Low-Income Countries’. IMF Working Paper No. 03/249, IMF, Washington, DC.

Cohen, D. (1997). ‘Growth and external debt: A new perspective on the African and Latin American tragedies’. Centre for Economic Policy Research Discussion Paper, No. 1753.

Easterly, W. (2002). ‘How Did Heavily Indebted Poor Countries get Heavily Indebted? Reviewing Two Decades of Debt Relief’. World Development, Vol. 30, 10, pp. 1677-1696.

Elbadawi, I.A., Ndulu, B.J. and Ndungu, N. (1999), in Iqbal, Z. and Ravi, K. (Eds), ‘Debt Overhang and Economic Growth in Sub-Saharan Africa’. External Finance for Low-income Countries, IMF Institute, Washington, DC, Ch. 5.

Ellig, J. (Ed.) (2001). ‘Dynamic Competition and Public Policy: Technology, Innovation and Antitrust Issues’, Cambridge University Press, Cambridge.

Engle, R. F. Granger, C. W. J. ( 1987). ‘Co-Integration and Error Correction: Representation, Estimation, and Testing.’ Econometrica, Vol. 55 No. 2 (March 1987) pp. 251-276.

Fosu, A. K. (1996). ‘The Impact of External Debt on Economic Growth in Sub-Saharan Africa’. Journal of Economic Development, Vol. 21, 1, pp. 93-118.

Fosu, A. K. (1999). ‘The External Debt Burden and Economic Growth in the 1980s: Evidence from Sub-Saharan Africa’. Canadian Journal of Development Studies, Vol. 20, 2, pp. 307-318.

Frimpong, J. M. and Oteng-Abayie, E.F. (2006). ‘The Impact of External Debt on Economic Growth in Ghana: A Cointegration Analysis’. Journal of Science and Technology, Vol.26, 3. Retrieved on 1 May 2010 from http://dspace.knust.edu.gh/dspace

Geiger, L. T. (1990). ‘Debt and Economic Development in Latin America’. The Journal of Developing Areas, 24, pp. 181-194.

Hofman, B. and Reisen, H. (1991) ‘Some Evidence on Debt-Related Determinants of Investment and Consumption in Heavily Indebted Countries’. Weltwirtschafliches Archiv, Vol. 127, 2, pp. 280-297.

International Development Association (IDA) and International Monetary Fund (IMF). (2007). ‘Heavily Indebted Poor Countries (HIPC) Initiative and Multilateral Debt Relief Initiative (MDRI) − Status of Implementation’. A report prepared on September 28, 2007 for the October 21, 2007 Development Committee Meeting.

International Monetary Fund (IMF). (1989). World Economic Outlook, Supplementary Note 1, Washington, USA. Retrieved on 20 July, 2010 from http://www.allbusiness.com/finance/110797-1.html

International Monetary Fund (IMF). (2010). ‘Debt Relief under the Heavily Indebted Poor Countries (HIPC) Initiative’. A fact sheet, July. IMF, Washington, D. C

Iyoha, M. A. (1996). ‘External Debt and Economic Growth in Sub-Saharan African Countries: An Econometric Study’. Paper presented at AERC workshop, Nairobi

Johansen, S. (1988). ‘Statistical Analysis of Cointegrating Vectors’. Journal of Economic Dynamics and Control, Vol. 12, pp. 231-254.

Johansen, S. and Juselius, K. (1990). ‘Maximum Likelihood Estimation and Inference on Cointegration – with Application to the Demand for Money’. Oxford Bulletin of Economics and Statistics, Vol. 52, 2, pp. 169-211.

Karagol, E. (2002). ‘The Causality Analysis of External Debt Service and GNP: The Case of Turkey’. Central Bank Review, Vol. 2, 1, pp. 39-64.

Krugman, P. (1988). ‘Financing Versus Forgiving a Debt Overhang’. Journal of Development Economics, Vol. 29, 3, pp. 253–68.

Krumm, K. L. (1985). ‘The External Debt of Sub-Saharan Africa’. World Bank Staff Working Paper No. 741 (June).

Mallik G. (2008). ‘Foreign Aid and Economic Growth: A Cointegration Analysis of the Six Poorest African Countries’. Economic Analysis and Policy, Vol. 38, pp. 251-260.

Metwally, M. M. and Tamaschke, R. (1994). ‘The interaction among foreign debt, capital flows and growth; case studies’. Journal of Policy Modelling, Vol. 16, 6, pp. 597-608.

Morisset, J. (1991). 'Can debt reduction restore economic growth in highly indebted countries?’. Revie D’economique Politique, Vol. 101, 4, pp. 639-666.

Nakatani, P and Herrera, R. (2007). ‘The south has already repaid its external debt to the north, but the north refuses to recognize its debts to the south’. CES Working Paper, 2007.14.

Patillo, C., Poirson, H., and Ricci, L. (2002). ‘External debt and growth’. IMF Working Paper 02/69, IMF, Washington, DC.

Patillo, C., Poirson, H., and Ricci, L. (2004). ‘What are the channels through which external debt affects growth?’. IMF Working Paper 04/15, IMF, Washington, DC.

Robinson, J. A. and Torvik, R. (2005). ‘White Elephants.’ Journal of Public Economics, Vol. 89, pp. 197– 210.

Savvides, A. (1992). ‘Investment Slowdown in Developing Countries During the 1980s: Debt Overhang Or Foreign Capital Inflows’. Kyklos, Vol. 45, 3, pp. 363-378.

Schclarek, A. (2004). ‘Debt and Economic Growth in Developing and Industrial Countries’. Working Papers 2005:34, Lund University, Department of Economics.

Sen, S., Kasibhatla, K.M., and Stewart D.B. (2007). ‘Debt Overhang and Economic Growth - The Asian and the Latin American Experience’. Economic Systems, Vol. 31, 1, pp. 3-11

Ubok-Udom, E.U. (1978). ‘Development through Debt: Rationalizing the Costs of External Borrowing’. The Nigerian Journal of Economic and Social Studies, Vol. 20, pp. 48-56.

Were, M. (2001). ‘The Impact of External Debt on Economic Growth in Kenya: An Empirical Assessment’. UNU/ WIDER Discussion Paper No. 2001/120, Helsinki.

World Bank (2010). ‘Heavy Indebted Poor Countries’. Retrieved on 23 August, 2010 from http://go.worldbank.org/4IMVXTQ090.

 

APPENDIX

Table A1: Summary of VECM for the Growth Equation

Variables	COTE D'IVOIRE	GAMBIA	GHANA	SENEGAL
CointEq1	0.093 [ 1.36]	-0.015 [-1.68]	-0.032 [-2.26]**	-0.437 [-3.06]*
D(LRYPC(-1))	0.163 [ 0.81]	0.088 [ 0.46]	0.155 [ 0.924]	0.077 [ 0.39]
D(FDIY(-1))	-0.012 [-0.98]	-0.0001 [-0.03]	0.008 [ 1.22]	0.008 [ 1.56]
D(LEXDY(-1))	0.009 [ 0.12]	0.074 [ 1.5]	0.084 [ 1.96]***	0.078 [ 1.59]
D(LGDIY(-1))	0.077 [ 1.52]	-0.018 [-0.49]	0.004 [ 0.13]	-0.013 [-0.33]
D(LTDSE(-1))	-0.007 [-0.19]	-0.018 [-1.16]	-0.029 [-1.44]	0.005 [ 0.28]
Constant	-0.012 [-1.57]	0.002 [ 0.3]	0.0001 [ 0.01]	-0.001 [-0.09]
R-squared	0.244847	0.09875	0.358663	0.37961
Adj. R-squared	0.088609	-0.087716	0.225973	0.251254
F-statistic	1.567137	0.529588	2.703012**	2.957471**

CointEq1 is the Error Correction Term generated by the VAR(1)

Table A2: The Over-Parameterized ECM (Growth Equation)

Variables	COTE D'IVOIRE	GAMBIA	GHANA	SENEGAL
DFDIY	-0.004 [-0.34]	-0.004 [-1.61]	-0.001 [-0.11]	0.006 [1.14]
DLEXDY	-0.118 [-1.44]	-0.056 [-1.62]	-0.066 [-1.17]	-0.106 [-2.87]*
DLGDIY	0.077 [1.52]	0.073 [1.91]***	0.006 [0.18]	-0.012 [-0.39]
DLTDSE	0.03 [0.64]	-0.008 [-0.6]	-0.04 [-1.85]***	-0.03 [-1.87]***
DLRYPC(-1)	0.336 [1.46]	0.274 [1.23]	0.351 [2.08]**	0.056 [0.318]
DFDIY(-1)	0.005 [0.47]	-0.001 [-0.32]	0.003 [0.45]	0.008 [1.53]
DLEXDY(-1)	0.027 [0.36]	0.021 [0.52]	0.028 [0.76]	0.019 [0.60]
DLGDIY(-1)	0.003 [0.05]	-0.013 [-0.32]	-0.026 [-0.86]	-0.012 [-0.31]
DLTDSE(-1)	-0.016 [-0.41]	-0.01 [-0.66]	-0.026 [-1.29]	0.011 [0.66]
Constant	-0.006 [-0.76]	0.006 [0.93]	0.0001 [0.01]	-0.072 [-1.54]
Dummy06	--	--	-0.016 [-0.20]	-0.075 [-1.57]
ECT(-1)	-0.215 [-0.93]	-0.305 [-1.81]***	-0.387 [-2.64]**	-0.461 [-2.28]**
R-squared	0.341	0.275	0.517	0.59
Adj R-squared	0.077	-0.014	0.296	0.403
D-W stat	1.94	1.92	2.09	1.91
F-statistic	1.29 (0.29)	0.95 (0.51)	2.34 (0.04)***	3.15 (0.009)**

Table A3: Unit Root Test for co-integration (Investment Equation)

Country	Variables	ADF Test		PP Test		I(d)
		T- stat	10% CV	T- stat	10% CV
COTE D'IVORE	Resid	-3.778* (1)	-2.612	-3.14*	-2.610	I(0)
GAMBIA	Resid	-2.564* (0)	-2.612	-2.784***	-2.610	I(0)
GHANA	Resid	-3.737* (0)	-2.612	-3.737*	-2.610	I(0)
SENEGAL	Resid	-3.593* (0)	-2.610	-3.576*	-2.610	I(0)

Note: *, **,*** indicate statistical significance at 1%, 5%, and 10%  levels respectively.

---

[1] Where we include a trend for the unit root test, we also confirm that it is statistically significant in the unit root test. No trend was included in the first difference unit root tests since differencing de-trends a series.

[2] This is accessible using Eviews 6.0

[3] Where f is the frequency, annual.

[4] In an ECM, it is possible to drop unwanted or insignificant lags of specific variables as well as include contemporaneous terms meanwhile in a VECM is not possible to drop lags of specific variables such that each variable and equation in the  system is estimated with the same number of lags. Hence using the ECM gave us a better feel of our data.

[5] All tables are not shown to conserve space but they are readily available upon request.

[6] The first lag of output growth in the ECM for Gambia seemed to be relevant to the model though statistically insignificant. Taking it out weakens the model and causes the ECT term to change. We compare the model with a case where the lag is taken off but and find it better off. AIC. SC and diagnostic statistics were used to do this.

[7] All our countries fall into the low income countries bounds before 2008 according to the World Bank classification. 

[8] That is greater than 100%

[9] See table A3 in the Appendix.