Journal Issue
Share
Article

Algeria: Selected Issues

Author(s):
International Monetary Fund
Published Date:
February 2005
Share
  • ShareShare
Show Summary Details

II. Relationship Between Government Expenditure and GDP34

Government expenditure in Algeria has moved with hydrocarbon production. This chapter examines whether government expenditure has induced growth during the period 1967–03. It finds that movements in real government capital expenditure have an impact on real nonhydrocarbon GDP (NHGDP) and, as one would expect, that real government current expenditure follows real hydrocarbon GDP (HGDP).

A. Introduction

1. Since independence in 1962, Algeria’s government policy has largely focused on government expenditure as a means to develop the economy and generate employment for its burgeoning population.35 Government expenditure averaged about 30 percent of GDP during 1967-2003, and was financed in large part by hydrocarbon revenue and, when these declined, external borrowing. Government capital expenditure, in the 1970s and early 1980s, was largely concentrated on large scale public investment projects and was channeled towards heavy industry, with the underlying belief that strong forward and backward linkages would stimulate growth in other sectors. Though the investment drive existed in earlier years, the oil price hike of 1979 provided an opportunity to raise government investment considerably. Government capital expenditure increased from about 5 percent of GDP in 1967 to its peak of about 18 percent in 1983 (Figure 1). In the late 1980s, in the face of civil strife, government expenditure was increasingly reoriented towards current expenditure and government capital expenditure declined to about 6 percent of GDP by 1991. Since then, it has gradually recovered to about 11 percent of GDP in 2003. Government current expenditure also fluctuated during the period moving from about 21 percent of GDP in 1967 to about 23 percent in 2003.

Figure 1.Algeria: Nominal Government Expenditure as Share of Nominal GDP

2. Government expenditure depends strongly on hydrocarbon budgetary revenue. Movements in nominal government expenditure and previous year’s nominal HGDP growth display a strong correlation, of about 0.75, between 1967 and 2003 (Figure 2). Real government expenditure as a share of real NHGDP has increased from 16 percent in 1967 to over 50 percent in 2003. Real NHGDP growth has averaged about 4 percent per year between 1967 and 2003. This chapter examines the government expenditure/growth nexus in Algeria.

Figure 2.Algeria: Change in Nominal Government Expenditure and Hydrocarbon GDP

(In percent)

3. It is unclear a priori whether increases in government expenditure induce higher output, and therefore reduce unemployment, or whether government expenditure has responded to the level of output. The literature remains divided over the relationship between government expenditure and changes in real GDP.36 One view is that government expenditure is a policy instrument to be used to increase output. Others consider that public expenditure is endogenous and follows economic growth. This latter proposition, known as Wagner’s law, is taken to mean that government expenditure increases endogenously in order to meet the protective, administrative and educational functions of the state.37 Empirical studies on the relationship between government expenditure and growth in hydrocarbon-rich countries are limited. Ghali and Al-Shamsi (1997) find evidence in favor of the first view, while Al-Faris (2002) and Ghali (1997) find evidence in support of Wagner’s law. Fasano and Wang (2001) find no conclusive evidence of causality in GCC countries.

4. This study finds that movements in real government capital expenditure induce real NHGDP growth while movements in real government current expenditure do not. This observation is consistent with recent empirical studies that find a strong and fairly robust impact of government capital expenditure on growth.38 With regards to a causal relationship from NHGDP to government expenditure, the results are inconclusive. This result is not surprising given that high oil revenues and external borrowing have insulated government expenditure from developments in the nonhydrocarbon sector.

5. The findings support the idea that government capital expenditure can be used as an instrument to stimulate growth and reduce unemployment. Increases in government capital expenditure would also entail some increase in current expenditure, in particular wages and other recurrent spending needed to make the public investment productive.

6. The paper is organized as follows:Section B describes the trends in government expenditure and GDP in Algeria. Section C presents the empirical findings. Section D concludes. The appendices discuss the econometric methodology utilized and present details of the empirical results.

B. Developments in Government Expenditure and GDP

7. Despite the decline in hydrocarbon prices that started in 1981, government expenditure remained high, at about 30 percent of GDP, and was increasingly financed by external borrowing.39 The emergence of large fiscal imbalances and the associated buildup of external debt prompted the government to undertake a fiscal adjustment in the context of two Fund-supported programs, in 1989 and 1991, and capital expenditure reached its lowest level since 1967 in 1991, of about 8 percent of NHGDP (Figure 3). However in 1992, in the face of political uncertainties and civil strife, the government adopted an expansionary fiscal policy aimed at stimulating economic growth. Fiscal deficits mounted and reserve losses accelerated, partly reflecting the reluctance to adjust the exchange rate, in the face of declining hydrocarbon prices (Figure 4).40

Figure 3.Algeria: Nominal Government Expenditure as Share of Nominal NHGDP

(In percent)

Figure 4.Algeria: Nominal Government Expenditure and Overall Balance (in percent of GDP) and Oil Prices (US$ per barrel)

8. The country embarked on two Fund-supported program in 1994 and 1995 during which both demand management and structural reforms were implemented.41 Real wages were reduced and layoffs occurred in the context of public enterprise restructuring. Nevertheless, government expenditure remained high, also relative to other hydrocarbon countries, averaging over 32 percent of GDP between 2000 and 2003. Figure 5 presents government expenditure and net lending in Algeria and six other hydrocarbon exporters.

Figure 5.Algeria: Government Expenditure and Net Lending, as Share of GDP

9. Despite high government expenditure during most of the period, the total factor productivity (TFP) of the economy and the efficiency of government investment have been low, with TFP growth even turning negative during periods characterized by large increases in government investment. Figure 6 charts TFP for the nonhydrocarbon economy. To quantify TFP, a Cobb Douglas production function is used with physical and human capital, and labor as factors of production.42

Figure 6.Algeria: Evolution of TFP of NHGDP (3 - year moving average, 1970 = 100) and Government Capital Expenditure

(In percent of NHGDP)

10. Algeria had a high government ICOR in the 1980s, averaging over 7; however it declined to an average of 2.4 between 2000 and 2003. Nevertheless, the government ICOR remains high relative to other oil producing countries, suggesting that a higher level of government investment is needed relative to other countries to obtain a similar growth rate (Figure 7). A high ICOR is associated with low efficiency because it implies that for a given level of investment the resulting change in GDP growth is low. The ICORs are calculated by summing government investment over a 9-year period and scaling this by the change in GDP during the same period. However, for the period after 1999, a four-year period is utilized.

Figure 7.Algeria: Trends in Government Investment ICOR

C. Causality Testing of Government Expenditure and GDP

11. In light of the apparent low productivity of government capital expenditure and the large share of government expenditure in GDP, it is important to determine whether government expenditure induces movements in NHGDP. In view of the possibility that the sub-components of government expenditure may have differential impact on economic growth, the paper also breaks down government expenditure into current (Gcons) and capital expenditure (Gcap). Annual data for the period 1967-2003 are from the Algerian authorities, IMF and IFS. All variables are expressed in logarithms. (The econometric methodology employed in estimating the model and testing for Granger causality is explained in see Appendix III).

12. various technical approaches have been developed to examine the direction of causality. Due to the sensitivity of Granger causality tests to misspecification of the model and the robustness of the rank-transformed model of Holmes and Hutton (1988) to misspecification, the paper restricts discussion of findings to those from the rank-transformed model.43 This is particularly important given the significant structural changes in the Algerian economy during the period 1967–03. However, in order to illustrate how a simple monotonic transformation, such as replacing variables with their ranks, can affect the power of Granger-causality tests the results from standard approaches, Ordinary Least Squares (OLS) and Vector Error Correction Model (VECM), are also presented. (See Appendix II for discussion of the models).

13. The results indicate that there is a one-way causal relationship from real government capital expenditure to real NHGDP (Model 2a in Table 1). The data do not support the existence of a causal relationship from government current expenditure to NHGDP.

Table 1.Results from Causality Tests on Government Expenditure and NHGDP1/
ModelDetermining direction of granger causality betweenRankOLSVECM
government expenditure and NHGDP(a)(b)(c)
1Does current expenditure Granger-cause NHGDP?NoNoNo
2Does capital expenditure Granger-cause NHGDP?Yes*NoNo
3Does expenditure Granger-cause NHGDP?Yes**NoNo
4Does NHGDP Granger-cause current expenditure?NoNoYes**
5Does NHGDP Granger-cause capital expenditure?NoNoNo
6Does NHGDP Granger-cause expenditure?NoNoNo

* and ** denotes rejection of the null of no causality at 5 percent and 1 percent, respectively. The values in parenthesis indicate probabilities of Wald tests.

* and ** denotes rejection of the null of no causality at 5 percent and 1 percent, respectively. The values in parenthesis indicate probabilities of Wald tests.

14. The findings imply that government capital expenditure can be utilized to increase NHGDP. Since capital expenditure tends to be channeled towards investment projects and current expenditure towards wages and consumer subsidies, the findings are consistent with observations that over the long-run, investment tends to have a higher impact on growth than consumption.44 Nevertheless, for increases in capital expenditure to be fully productive, current expenditure on maintenance and operations will also have to be increased. Therefore, the absence of a causal relationship running from current expenditure to NHGDP implies less a need to curtail current expenditure than a need for efficient resource allocation to ensure that current expenditure is in line with what is required to make government capital expenditure productive.

15. Real HGDP growth induces movements in real government current expenditure but not capital expenditure (Model 4a in Table 3). While capital expenditure had the long-run objective of diversifying the economy, the large share of public sector investment that has flowed to the hydrocarbon sector resulted in strong causal relationship from real capital expenditure to real HGDP (Model 2a in Table 3).

Table 2.Results from Causality Tests on Government Expenditure and HGDP 1/
ModelDetermining direction of granger causalityRankOLS
between government expenditure and NHGDP(a)(b)
1Does current expenditure Granger-cause HGDP?NoNo
2Does capital expenditure Granger-cause HGDP?Yes**No
3Does expenditure Granger-cause HGDP?Yes*No
4Does HGDP Granger-cause current expenditure?Yes*No
5Does HGDP Granger-cause capital expenditure?NoNo
6Does HGDP Granger-cause expenditure?NoNo

* and ** denotes rejection of the null of no causality at 5 percent and 1 percent, respectively. The values in parenthesis indicate probabilities of Wald tests.

* and ** denotes rejection of the null of no causality at 5 percent and 1 percent, respectively. The values in parenthesis indicate probabilities of Wald tests.

D. Conclusion

16. This study finds a one-way causal relationship from government capital expenditure to NHGDP. These results are obtained using the approach that is most robust to misspecification. There is no conclusive causal relationship from NHGDP to government expenditure, current and capital. This suggests that government capital expenditure can be used as a policy instrument to stimulate growth. Since government capital expenditure requires recurrent spending, current expenditure would be required in order to utilize and maintain the projects undertaken. This suggests that there is a need for an efficient allocation of government expenditure and hence underscores the importance of conducting public expenditure reviews.

17. However, given that capital expenditure is already high, further increases in capital expenditure may not necessarily translate into higher growth. The models provide no guidance on the extent to which increasing government capital above the currently high level would improve growth.

APPENDIX I Definition of Variables

HGDP = the log of nonhydrocarbon GDP in real terms. The real hydrocarbon GDP series is generated by deflating nominal hydrocarbon GDP with the hydrocarbon GDP deflator. For earlier years where the hydrocarbon GDP deflator is unavailable, the GDP deflator is utilized.

NHGDP = the log of nonhydrocarbon GDP in real terms. The real nonhydrocarbon GDP series is generated by deflating the nominal nonhydrocarbon GDP with the nonhydrocarbon GDP deflator. For earlier years where nonhydrocarbon GDP deflator is unavailable, the GDP deflator is utilized.

Gcons = the log of government current expenditure in real terms. The real government current expenditure series is generated by deflating the nominal government current expenditure, excluding net lending, by the consumer price index (CPI).

Gcap = the log of government capital expenditure in real terms. The real government capital expenditure series is generated by deflating the nominal government capital expenditure by the investment deflator. For earlier years where investment deflator is unavailable, a proxy using share of CPI and GDP deflator is utilized to deflate the series.

Gexp = the log of total government expenditure in real terms. The real government expenditure series is generated by taking the sum of Gcap and Gcons.

APPENDIX II

A. Rank Transformation

A critical failing of Vector Error Correction Models (VECM) and other standard models is that results from the parametric statistical techniques are sensitive to the functional form specification of the estimating equations, to the lag structure specified, and to filtering techniques used to achieve stationary variables (Bessler and Kling, 1984; Nelson and Kang, 1984; Holmes and Hutton, 1988).

A rank order transformation of the variables prior to Granger testing provides causality results which are robust to functional form specification and to possible heteroscedasticity and non – normality of the error structure in the estimation equation (Holmes and Hutton, 1988).45 In testing for causality it is important to use a methodology that requires the fewest assumptions about the nature of a relationship which may not even exist.

The qualitative relationship between the variables before rank transformation is equivalent to that between the rank transformed variables.Holmes and Hutton (1988) argue that if Y is a function of X, any strictly monotonic transformation of some or all of the variables in (X,Y) will not eliminate this causal or functional relationship. This invariance to a monotonic transformation encompasses rank transformation. Therefore if X causes Y,

Since economic theory does not generally specify a particular functional form, this suggests that qualitative inferences (that is, zero and sign restrictions) can and should be pursued using a rank transformed model similar to equation 3.

In the multiple rank F test, the (set of) Y(s) and X(s) in a parametric model

(where Y = y - ỹ and X = x - x)

are replaced by their ranks where R() is the rank of a variable.

R(.) is defined as a vector of the deviations of ranks of a variable from the means of the ranks, r(Yi), of a variable Yi over N observations (i = 1,......,N). The rank of Y measured in deviations is (r(Y1),.....r(YN)) = R(Y) and the ranks of each of the (k-1) variables in the matrix X = [X1,....., Xk-1] are R(X) = [R(X1), .....,R(Xk-1)]. This results in the multiple rank equation

C(β) = (C11), .....C k-1k-1)) is such that, under the alternative hypothesis, if each Xi is ranked from lowest (being 1) to highest (being N) then βiC(βi) ≥ 0 and under the null hypothesis βiC(βi) = 0 if and only if βi = 0.

Multiple rank F test involves OLS estimation of equation 5 followed by performance of tests based on the residual variances from these rank regressions. Conover and Iman (1982) report that in small samples with stochastic X, the multiple rank F-test is robust against non normality of errors, while for the normal distribution, the loss in power associated with ranks versus raw data is slight. Furthermore, the power advantages of the multiple rank F-test increase the more extreme the departures from the assumptions of normality and homoscedasticity and as the relationship between the variables weakens. (Holmes and Hutton, 1990) Figures 8 illustrates the impact of rank transformation using the government expenditure series.

Figure 10.Algeria: Real Government Expenditure with and without Rank Transformation, in First Differences of the Logarithms

B. Granger Causality

The paper utilizes the tests for Granger (1980) prima facie causal relationship, as opposed to tests for a Granger (1969) causal relationship, between output, and aggregate government current and capital expenditures. The distinction between prima facie and direct causal relationship is important in a stochastic setting. A (set of) variable(s) X is a prima facie cause of another (set of) variable(s) Y if X precedes Y in time and if P(Y | X) ≠ P(Y). It is not possible to establish that a causal relationship exists as long as it is possible to conceive of other potential prima facie causes which have not been included in the model being analyzed.

APPENDIX III

A. Methodology for Testing for Granger Causality

NHGDP is Granger-caused by government expenditure if increases in government expenditure precede an increase in NHGDP. Granger-causality is investigated by estimating the following equations

Equations 1a and b must be tested to determine the direction of causality between the variables and establish whether Granger-causality, if it exists, is one-way or bi-directional, where Gexp is log of real government expenditure, α1 is the long-run constant elasticity of real NHGDP with respect to real government expenditure and α2 is the converse. All the variables are in differences.

If movements in government expenditure are induced by HGDP developments then HGDP Granger-causes government expenditure. To test for Granger causality, equations 2a and b are estimated where α3 captures the elasticity of real government expenditure with respect to real HGDP and α4 the converse.

Table 3 illustrates the specification of the models under the different approaches: rank-transformed model, VECM, and OLS.

Table 3.Causality Testing: Model Specification1/
Specification for Testing
Approach
1. Holmes and Hutton (1988)NHGDPtGconst,(1a),NHGDPtGcapt(1b)
dR(Gconst)=Σni=1bidR(Gconsti)+Σmi=1aidR(NHGDPti)+ɛt(1a)
dR(Gcapt)=Σni=1bidR(Gcapti)+Σmi=1aidR(NHGDPti)+ɛt(1b)
GconstNHGDPtandGcaptNHGDPt
dR(NHGDPt)=Σpi=1ajdR(NHGDPti)+Σqi=1bidR(Gconsti)+Σsi=1cidR(Gcapti)+wt(1c)
2. Granger (1969) OLSNHGDPtGconst(2a),NHGDPtGcapt(2b)
dGconst=Σmj=1bjdGconstj+Σnj=1ajdNHGDPtj+vt(2a)
dGcapt=Σni=1bidGcapti+Σmi=1ajdNHGDPti+ɛt(2b)
GconstNHGDPtandGcaptNHGDPt
dNHGDPt=Σpi=1ai(2c)dNHGDPti+Σqi=1bidRGconsti)+Σsi=1cidGcapti+wt
NHGDPtGconst,NHGDPtGcapt,GconstNHGDPt,andGcaptNHGDPt
3. Granger (1986) VECMdGconst=Σni=1a1idGconsgt1+Σni=1b1idGcapgt1+Σmj=1c1idNHGDPtj+dECTt1+μ2t

dGapt=Σi=1na2idGconsgt1+Σi=1nb2idGcapgt1+Σj=1mc2idNHGDPtj+dECTt1+μ2t

dNHGDPt=Σni=1a3idGconsgt1+Σni=1b3idGcapgt1+Σmj=1c3idNHGDPtj+dECTt1 +μ2t(3)

R(.) represents a rank order transformation. ECT refers to the error correction term, G reflects government expenditure implying that both government current expenditure (Gcons) and government capital (Gcap) expenditure are estimated. All the models can be written with total government expenditure (Gexp).

R(.) represents a rank order transformation. ECT refers to the error correction term, G reflects government expenditure implying that both government current expenditure (Gcons) and government capital (Gcap) expenditure are estimated. All the models can be written with total government expenditure (Gexp).

B. Summary of Econometric Details and Empirical Results

I. Unit Root tests

Since causality tests are based on the assumption of the existence of stationary stochastic processes, the empirical investigation begins with an analysis of the time series properties of the variables. The Augmented Dickey Fuller (ADF) test is used to determine the order of integration. All the variables are found to be integrated of order one (Table 4). Given these findings, both the rank transformed model and the standard OLS model are estimated using differenced stationary series.

Table 4.Augmented Dickey – Fuller Test Results for Unit Roots 1/
HGDPNHGDPGexpGconsGcap
Levels-2.23 c2-2.04 c2-1.86 c1-0.66 c0-1.91 c1
First difference4.69 1***-1.94*2-3.05***-3.87***0-3.66***0

*, **, and *** denotes significance at 10 percent, 5 percent, and 1 percent, respectively. The Schwartz Information Criteria (SIC) is used to select the lag structure. The values in superscript indicate number of lags, presence or absence of constant (c).

*, **, and *** denotes significance at 10 percent, 5 percent, and 1 percent, respectively. The Schwartz Information Criteria (SIC) is used to select the lag structure. The values in superscript indicate number of lags, presence or absence of constant (c).

II. Rank Transformed Model

Current and lagged values of the differences series on HGDP, NHGDP, Gcap, Gcons and Gexp in each model are treated as separate variables when calculating their ranks, for example, R(Gconstt) and R(Gconstt-1).

In specifying the model, the lagged differences are included to ensure that residuals are white noise and to capture proxy for other explanatory variables not included in themodel.46 The Akaike Information Criteria and Schwartz Information Criteria are used to select an “optimal” univariate lag length.

Does Government Expenditure induce NHGDP?

F-tests on the model strongly reject the null of no causality between government capital expenditure and NHGDP. However, inconclusive results are obtained from NHGDP to government expenditure, current or capital. The findings imply that there is uni-directional causality from government capital expenditure to NHGDP (Table 5).

Table 5.Causality Tests on OLS Estimates Employing Ranks 1/
dNHGDPtdGconstdGcaptdGexpt
Wald test: R(dGconst-1) → dNHGDP
Wald test: R(Σ2t-1dGcap) → dNHGDP6.65*
(0.01)
Wald test: R(Σ2t-1dGexp) → dNHGDP5.94**
Wald test: R(Σ5t-1dNHGDP) → dGcons(0.00)2.68
(0.07)
Wald test: R(Σ7t-1dNHGDP) → dGcap11.29
(0.13)
Wald test: R(Σ7t-1dNHGDP) → dGexp1.41
(0.27)

* and ** denotes rejection of the null of no causality at 5 percent and 1 percent, respectively. The values in brackets indicate probabilities of Wald tests.

* and ** denotes rejection of the null of no causality at 5 percent and 1 percent, respectively. The values in brackets indicate probabilities of Wald tests.

Does HGDP induce government expenditure?

F-tests indicate presence of one-way causality from HGDP to government current expenditure and one-way causality from capital expenditure to HGDP (Table 6).

Table 6.Causality Tests on OLS Estimates Employing Ranks 1/
dHGDPtdGconstdGcaptdGexpt
Wald test: R(Σ2t-1dGcons) → dHGDP
Wald test: R(Σ2t-1dGcap) → dHGDP4.61**
(0.00)
Wald test: R(Σ2t-1dGexp) → dHGDP4.24*
(0.01)
Wald test: R(Σ5t-1dHGDP) → dGcons3.44*
(0.02)
Wald test: R(Σ7t-1dHGDP) → dGcap1.02
(0.45)
Wald test: R(Σ7t-1dHGDP) → dGexp1.40
(0.29)

* and ** denotes rejection of the null of no causality at 5 percent and 1 percent, respectively. The values in brackets indicate probabilities of Wald tests.

* and ** denotes rejection of the null of no causality at 5 percent and 1 percent, respectively. The values in brackets indicate probabilities of Wald tests.

III. OLS Model

Having established that the series is a unit root process, differenced stationary series are utilized where both the AIC and SIC information are utilized to select model with the “optimal” lag length.

Does government expenditure induce NHGDP?

F -tests do not reject the null of no causality therefore the OLS model does not find causality from government expenditure, current or capital, to NHGDP. Within the OLS context pair-wise granger causality tests are performed on the variables (Table 7).

Table 7.Causality Tests on OLS Model 1/
F- statistic
dNHGDP→dGcons0.48
dGcons→ dNHGDP(0.70)
dNHGDP→dGcap0.40
dGcap→ dNHGDP(0.67)
dNHGDP→dGexp0.94
dGexp→ dNHGDP(0.46)

Probability in brackets. * indicates significance at the 5 percent level. The null hypothesis is that x does not prima facie Granger-cause y in the first regression and the y does not Granger-cause x in the second regression.

Probability in brackets. * indicates significance at the 5 percent level. The null hypothesis is that x does not prima facie Granger-cause y in the first regression and the y does not Granger-cause x in the second regression.

Does HGDP induce government expenditure?

Similarly, the results do not reject the presence of causality from HGDP to any other government expenditure variables. (Table 8).

Table 8.Causality Tests on OLS Model 1/
F- statistic
dHGDPΣdGcons2.18
dGconsΣdHGDP(0.10)
dHGDPΣdGcap0.23
dGcapΣdHGDP(0.88)
dHGDPΣdGexp1.42
dGexpΣdHGDP(0.26)

Probability in brackets. * indicates significance at the 5 percent level. The null hypothesis is that x does not prima facie Granger-cause y in the first regression and the y does not Granger-cause x in the second regression.

Probability in brackets. * indicates significance at the 5 percent level. The null hypothesis is that x does not prima facie Granger-cause y in the first regression and the y does not Granger-cause x in the second regression.

IV. VECM Model

The Johansen trace and maximal eigenvalue statistics confirm the existence of a long run relation between NHGDP, government current and capital expenditure.

Does government expenditure Granger-cause NHGDP?

The findings indicate suggest that government expenditure is positively related to NHGDP (Table 9 and Table 10). The sign on the error correction term, a, confirms the existence of a long _ run relation between government expenditure and NHGDP providing support to the Granger representation theorem.47 Moving to the estimation of the cointegrating relationship, Tables 9 and 10 present the normalized cointegrating vector (β) and the corresponding adjustment coefficients, the error correction term, (α).

Table 9.Normalized Cointegration Vector and Weak Exogeneity tests 1/
NHGDPGovernment expenditure
CurrentCapital
β1-1.12-0.18
[-7.52][-4.43]
α-0.110.900.27
(-0.79)[4.27][0.75]

Model has lag length of 3. Equation includes an unreported constant. t.statistics are in parenthesis.

Model has lag length of 3. Equation includes an unreported constant. t.statistics are in parenthesis.

Table 10.Normalized Cointegration Vector and Weak Exogeneity Tests 1/
NHGDPGexp
β1-0.81
[-10.11]
α-0.130.59
(-1.20)[3.56]

Model has lag length of 3. Equation includes an unreported constant. t.statistics are in parenthesis.

Model has lag length of 3. Equation includes an unreported constant. t.statistics are in parenthesis.

Government capital expenditure is weakly exogenous consistent with observations that government expenditure decisions in oil rich economies tend to be influenced by hydrocarbon revenues, therefore hydrocarbon GDP.48 The t-statistics indicate that both NHGDP and government capital expenditure are weakly exogenous (Table 9 and Table 10).

Causality tests indicate that NHGDP has a causal effect on government current expenditure. There is no evidence of causality between the other variables. Therefore, on the basis of the VECM, it can be concluded that there is unidirectional causality from NHGDP to government current expenditure (Table 11 and Table 12).

Table 11.VECM Pairwise Granger Causality Tests 1/
D(NHGDP)D(GCONS)D(GCAP)
D(GCONS)1.674.26
(0.64)(0.23)
D(GCAP)1.844.10
(0.61)(0.25)
D(NHGDP)13.12**1.74
(0.00)(0.63)
All4.0014.28*6.67
(0.68)(0.03)(0.35)

χ2 presented. P-values in brackets. Column headings are the dependent variable of the equation under consideration. The null is exclusion of lags of the variable specified. Rejection of the null implies rejection of Granger-causality. ‘All’ refers to exclusion of all the endogenous variables from the VECM, other than the lags of the dependent variable. * and ** denotes significance at 5 percent, and 1 percent, respectively.

χ2 presented. P-values in brackets. Column headings are the dependent variable of the equation under consideration. The null is exclusion of lags of the variable specified. Rejection of the null implies rejection of Granger-causality. ‘All’ refers to exclusion of all the endogenous variables from the VECM, other than the lags of the dependent variable. * and ** denotes significance at 5 percent, and 1 percent, respectively.

Table 12.VECM Pairwise Granger Causality Tests 1/
D(NHGDP)D(Gexp)
D(Gexp)1.63
(0.65)
D(NHGDP)6.73
(0.08)

χ2 presented. P-values in brackets. Column headings are the dependent variable of the equation under consideration. The null is exclusion of lags of the variable specified. Rejection of the null implies rejection of Granger-causality. *, and ** denotes significance at 5 percent, and 1 percent, respectively.

χ2 presented. P-values in brackets. Column headings are the dependent variable of the equation under consideration. The null is exclusion of lags of the variable specified. Rejection of the null implies rejection of Granger-causality. *, and ** denotes significance at 5 percent, and 1 percent, respectively.

Does HGDP induce government expenditure?

Cointegration tests on both two variable model, Gexp and HGDP and three variable model, Gcap, Gcons and HGDP, do not reject null of no cointegration therefore a VECM is not modeled.

References

    AhsanS.M.1996Cointegration and Wagner’s Hypothesis: Time Series Evidence for Canada,Applied Economics Vol. 28 pp. 105558.

    Al-FarisA.F2002Public Expenditure and Economic Growth in the Gulf Cooperation Council Countries,Applied Economics Vol. 34 pp. 118793.

    • Crossref
    • Search Google Scholar
    • Export Citation

    AutyR.M and A. H.Gelb2001Political Economy of Resource-Abundant States,in Resource Abundance and Economic Developmented. byR.M.Auty;study prepared for the World Institute for Development Economics Research of the United Nations University (New York: Oxford University Press).

    • Search Google Scholar
    • Export Citation

    BarroR. and Jong-WhaLee2000International Data on Educational Attainment, Updates and Implications.NBER Working Paper 7911. (Cambridge,MA: National Bureau of Economic Research).

    • Search Google Scholar
    • Export Citation

    BaneerjeeA. D.F.Hendry and G.W.Smith1986Exploring Equilibrium Relationships in Economics through Static Models: Some Monte Carlo Evidence,Oxford Bulletin of Economics and Statistics Vol. 48 pp. 25377.

    • Crossref
    • Search Google Scholar
    • Export Citation

    BosworthB. and S.Collins2003The Empirics of Growth: An Update,” (unpublished; Washington: Brookings Institution and Georgetown University).

    • Search Google Scholar
    • Export Citation

    CalderónC. and L.Servén2003Infrastructure Compression and Public Sector Solvency in Latin Americain The Limits of Stabilization – Infrastructure Public Deficits and Growth in Latin Americaed. ByW.Easterly and L.Servén (Washington: World Bank).

    • Search Google Scholar
    • Export Citation

    CohenD. and M.Soto2001Growth and Human Capital: Good Data, Good Results,OECD Technical Papers 179 (Organization for Economic Cooperation and Development).

    • Search Google Scholar
    • Export Citation

    ConoverW.J. and R. L.Iman1982Analysis of Covariance Using the Rank TransformationBiometrics Vol. 38 pp. 71524.

    EasterlyW. and S.Rebelo1993Fiscal Policy and Economic Growth,Journal of Monetary Economics Vol. 32(December) pp. 41758.

    EkenS. T.Helbling and A.Mazarei1997Fiscal Policy and Growth in the Middle East and North Africa Region,IMF Working Paper 97/101 (Washington: International Monetary Fund).

    • Crossref
    • Search Google Scholar
    • Export Citation

    EngleR.F. and C.W.J.Granger1987Co-integration and Error Correction Representation, Estimation, and Testing,Econometrics Vol. 55(March) pp. 25176.

    • Crossref
    • Search Google Scholar
    • Export Citation

    FasanoUgo and Q.Wang2001Fiscal Expenditure Policy and Non-oil Economic Growth: Evidence from the GCC Countries,IMF Working Paper 01/195 (Washington: International Monetary Fund).

    • Search Google Scholar
    • Export Citation

    GelbA.1988Oil Windfalls: Blessing or curse? A World Bank Research Publication (New York: Oxford University Press).

    GhaliK.H.1997Government Spending and Economic Growth in Saudi Arabia”,Journal of Development Economics Vol. 22pp 16572

    GhaliK.H. and Al-ShamsiF1997Fiscal Policy and Economic Growth: A Study Relating to the United Arab Emirates”,Economia Internazional Vol. 50pp 51933

    • Search Google Scholar
    • Export Citation

    GrangerC.W.J.1969Investigating Causal Relations by Econometric Models and Cross-Spectral Methods,Econometrica Vol. 37pp 42438.

    GrangerC.W.J.1980Testing for Causality: A Personal Viewpoint,Journal of Economic Dynamics and Control Vol. 2 pp. 32952.

    GrangerC.W.J.1986Developments in the Study of Cointegrated Economic Variables,Oxford Bulletin of Economics and Statistics(August) pp. 21328

    • Search Google Scholar
    • Export Citation

    HolmesJ.M. and P.A.Hutton1988A Functional-Form, Distribution-Free Alternative to Parametric Analysis of Granger Causal Models,Advances in Econometrics Vol. 7 pp. 211225.

    • Search Google Scholar
    • Export Citation

    HolmesJ.M. and P.A.Hutton1990On the Causal Relationship between Government Expenditures and National Income,The Review of Economics and Statistics Vol. 72 pp. 8795.

    • Crossref
    • Search Google Scholar
    • Export Citation

    KhanM.S. and M.S.Kumar1997Public and Private Investment and the Growth Process in Developing Countries,Oxford Bulletin of Economics and Statistics Vol. 59 pp. 6988.

    • Crossref
    • Search Google Scholar
    • Export Citation

    KremersJ.J.M. N.R.Ericsson and J. J.Dolado1992The Power of Integration Tests,Oxford Bulletin of Economics and Statistics Vol. 54 pp. 325-48

    • Crossref
    • Search Google Scholar
    • Export Citation

    La FeearaE. and M.Marcellino2000TFP, Costs, and Public Infrastructure: An Equivocal Relationship,IGIER Working Paper No. 176 (Milan Italy: Innocenzo Gasparini Institute for Economic Research).

    • Search Google Scholar
    • Export Citation

    MilbourneR. G.Otto and G.Voss2003Public Investment and Economic Growth,Applied Economics Vol. 35pp 527540

    NashashibiK. P.Alonso-Gamo S.Bazzoni A.Féler N.Laframboise and A. P.Horvitz1998Algeria: Stabilization and Transition to the Market,IMF Occasional Paper 165 (Washington: International Monetary Fund).

    • Search Google Scholar
    • Export Citation

    NelsonC.R. and H.Kang1984Pitfalls in the Use of Time as an Explanatory Variable in Regression,Journal of Business and Economic Statistics Vol. 2pp 7382.

    • Search Google Scholar
    • Export Citation

    NehruV. and A.Dhareshwar1993A New Database on Physical Capital Stock: Sources, Methodology and Results.Revista Análisis de Económico Vol. 8pp 3759.

    • Search Google Scholar
    • Export Citation

    ParkW.K1996Wagner’s Law Versus Keynesian Paradigm: The Korean Experience,Public Finance Vol. 51 pp. 7191.

    PorterA.K and M.McSweeney1974Comparison of Three Common Experimental Designs to Improve Statistical Power when Data Violate Parametric Assumptions”,presented at the Meeting of the American Education Research Association.

    • Search Google Scholar
    • Export Citation

    SinhaD.1998Government Expenditure and Economic Growth in Malaysia,Journal of Economic Development Vol. 23(December) pp. 7180.

    • Search Google Scholar
    • Export Citation
34This paper was prepared by Nkunde Mwase.
37Wagner (1967)
40The government was unwilling to adjust the exchange rate because it wanted to contain the cost of external debt service, which in 1992–93 amounted to 80 percent of export proceeds.
41The 1995 program was a three year Fund-supported program
42Drawing on Bosworth and Collins (2003), the production function is based on the assumption of a factor share of 0.35 for physical capital and 0.65 for labor and human capital combined. Human capital is proxied by the average years of schooling for the population. Data for human capital is taken from Barro and Lee (2000 and Cohen and Soto (2001). Capital stock is generated using the perpetual inventory method with the initial capital data obtained from Nehru and Dhareshwar (1993). In light of the high share of government and public sector investment in total investment, total investment is utilized.
43Examination of alternative approaches to estimate causal relationships suggests that rank order transformation should be employed in order to address the sensitivity and weaknesses of standard approaches in the presence of functional form misspecification, heteroscedasticity and non-normality of errors.
44See Eken and others (1997) for a discussion of the theoretical and empirical literature.
45Holmes and Hutton (1988) present a small sample multiple rank test attributable to Porter and McSweeney (1974), Conover and Iman (1982). The results extend to the inferences of lagged dependent variable models common to econometric studies, for example, prima facie granger causality.

Other Resources Citing This Publication