_{1}

^{*}

This study examines the impact of government spending on household consumption for the Economic Community of West African States (ECOWAS). As a modelling strategy, we use the Common Correlated Effect Mean Group (CCEMG) estimator that accounts for both parameter heterogeneity and cross-sectional dependence. The study provides various pieces of evidence through whole-panel and country-level analyses. The panel estimates indicate that government consumption has, on average, a negative effect on private consumption, implying that government and private consumption are substitutes. Country-level results reveal, however, considerable heterogeneity in the degree of substitutability across countries. They show crowding out effects in six countries, crowding in effects in one country and no significant effect in five countries. Therefore, government consumption is not a good instrument to stimulate aggregate demand and economic growth in ECOWAS countries.

The impact of government spending on private consumption and economic growth is one of the controversial issues in both theory and empirics. On the theoretical ground, there are different schools of thoughts on the issue. The Keynesian theory of absolute income hypothesis suggests that household current consumption is a function of their current disposable income. Thus, arise in government spending leads to increased output, employment and income, which further increases private consumption and generates thus crowding in effect. On the contrary, the standard real business cycle model, under the framework of neoclassical theory, predicts that a rise in government spending reduces private consumption [

On the empirical ground, there is no clear evidence on the impact of government spending on private consumption. A number of empirical studies found a positive impact of government spending on private consumption [

Literature mentioned above points towards the importance of the relationship between government spending and private consumption, however, this area of research is relatively understudied for African countries. Therefore, this study tries to contribute to the empirical literature by investigating the impact of government spending on private consumption for the member countries of the Economic Community of West African States (ECOWAS). We present empirical evidence on the relationship between government spending and private consumption for each of the ECOWAS member countries and for the panel as a whole. By doing this, we provide additional evidence on whether government spending generates “crowding-out” or “crowding-in” effects on the private sector and shed light on the importance of government spending in consumer welfare. To the best of our knowledge, this study is first of its kind to examine the long and short run effects of government spending on private consumption in ECOWAS countries. Secondly, from the methodological perspective, this study uses a more efficient and less restrictive econometric approach which improves the findings on the subject not only for ECOWAS countries but for overall empirical literature on the nexus between government and private consumption. More precisely, we use the Common Correlated Effect Mean Group (CCEMG) estimator introduced by [

The remainder of the paper is organized as follows. Section 2 outlines the econometric approach that will be used to carry out the empirical analysis. Section 3 reports and discusses the empirical findings of the study. Section 4 concludes the study and provides some policy recommendations.

The study examines the effect of government expenditure on private consumption using heterogeneous panel estimation techniques 1) to control for omitted variables and endogeneity bias and 2) to detect differences in the effect of government spending on private consumption across countries. In this section, we present the empirical model and discuss some econometric issues.

To test the effect of government consumption on private consumption, the following econometric model was estimated:

log C i t = α i + γ i log Y i t + β i log G i t + μ i t (1)

where C_{it} stand for household consumption, Y_{it} is income, G_{it} is government consumption, and μ_{it} stands for stochastic disturbance term assumed to follow a normal distribution. The coefficient on income is expected to be positive and lower than one. That on government consumption is ambiguous. The positive (negative) represents complementary (substitution) relationship between government spending and private consumption.

An important feature of our econometric model is that we do not impose a common β coefficient on government consumption. Accordingly, the parameter β is allowed to vary across countries. We are interested in the average value of β_{i}, reflecting the long-run effect of government consumption on private consumption.

The study uses annual time series data for 12 member countries of the Economic Community of West African States (ECOWAS), for the period from 1970 to 2016. The countries under study include: Benin, Burkina Faso, Cote d’Ivoire, Gambia, Ghana, Guinea-Bissau, Mali, Niger, Nigeria, Senegal, Sierra Leone and Togo. The coverage of countries and time period are dictated by the data availability for at least T = 30 observations. It is an unbalanced macro panel data analysis. The variables under investigation are household final consumption expenditure, gross domestic product (GDP) as a proxy for income and government final consumption expenditure.

An important issue in testing econometric relationship between variables is the definition and measurement of the variables. One may choose to work with variables expressed in levels, ratios or per capita terms. Previous studies have either used variables in real or real per capita terms. As robustness check, we estimate two models in this study. The first uses real household consumption, real government consumption and real gross domestic product (GDP) in constant 2010 US dollar. Real data on the variables were obtained from their respective shares in GDP. The second model considers the variables in real per capita terms using population.The dataset comes from the 2018 World Development Indicators of the World Bank. All the variables are transformed into natural logarithm to derive the direct estimation of elasticities.

Country | T | C_{t } | GDP_{t } | G_{t} | ρ | |||
---|---|---|---|---|---|---|---|---|

Mean | Std. | Mean | Std. | Mean | Std. | |||

Benin | 47 | 21.798 | 0.434 | 22.013 | 0.524 | 19.981^{ } | 0.668^{ } | 0.927*^{ } |

Burkina | 47 | 21.768 | 0.544 | 22.080 | 0.637 | 20.293^{ } | 0.977^{ } | 0.941*^{ } |

Cote d’Ivoire | 47 | 23.233 | 0.326 | 23.676 | 0.278 | 21.745 | 0.233^{ } | 0.846*^{ } |

Gambia | 40 | 19.921 | 0.523 | 20.165 | 0.390 | 18.207^{ } | 0.310 | −0.355* |

Ghana | 47 | 23.247 | 0.484 | 23.482 | 0.538 | 21.331^{ } | 0.730 | 0.919*^{ } |

Guinea-Bissau | 47 | 20.070 | 0.355 | 20.213 | 0.318 | 18.295 | 0.335 | 0.147 |

Mali | 47 | 22.147 | 0.481 | 22.338 | 0.552 | 20.353^{ } | 0.685^{ } | 0.923*^{ } |

Niger | 47 | 21.761 | 0.274 | 22.014 | 0.336 | 19.970^{ } | 0.476 | 0.884*^{ } |

Nigeria | 36 | 25.592 | 0.552 | 25.970 | 0.507 | 23.525^{ } | 0.546 | 0.717*^{ } |

Senegal | 47 | 22.449 | 0.452 | 22.731 | 0.408 | 20.902^{ } | 0.345^{ } | 0.910*^{ } |

Sierra Leone | 47 | 21.151 | 0.366 | 21.307 | 0.299 | 18.922^{ } | 0.354 | 0.882* |

Togo | 47 | 21.183 | 0.510 | 21.497 | 0.326 | 19.535 | 0.266 | 0.411*^{ } |

Panel | 546 | 21.982 | 1.450 | 22.244 | 1.498 | 20.215 | 1.474 | 0.955*^{ } |

Note: ρ is the correlation coefficient between private and government consumption. *indicates significance at the 5% level.

In order to choose the suitable estimator to be used in coefficient estimations, we need to address two econometric key issues. The first issue is to control for the possible cross-sectional dependence across the panel units which results from unobserved common factors. In the earlier studies, it was assumed that errors were cross-sectionally independent. It has been demonstrated that ignoring cross-sectional dependence by employing standard panel estimation methods is likely to produce inconsistent and biased estimates [

The econometric literature provides various tests analyzing cross-sectional dependency in panel data [

The second important issue to test is whether or not the slope coefficients are homogeneous among panel members. In standard panel data estimation methods it is assumed that slope coefficients are identical across countries. If this assumption does not hold, these methods will provide inconsistent and misleading results [

Variables | Breusch-Pagan LM | Pesaran scaled LM | Pesaran CD |
---|---|---|---|

C_{t } | 2405.049 (0.000) | 203.588 (0.000) | 48.850 (0.000) |

GDP_{t} | 2452.981 (0.000) | 207.760 (0.000) | 49.283 (0.000) |

G_{t} | 1236.996 (0.000) | 101.922 (0.000) | 30.875 (0.000) |

Residuals | 255.643 (0.000) | 16.506 (0.000) | 1.760 (0.078) |

Note: * and ** indicate rejection of the null hypothesis of no cross-sectional dependence at the 5% and 10% significance levels, respectively.

Hausman-type comparison of fixed effects and mean group estimates. The results of these tests are presented in

To deal with both cross-section dependence and slope heterogeneity, we use the Common Correlated Effects Mean Group (CCEMG) estimator designed by [

X i t = α 1 i + ϕ i f t + γ i g t + η i t (2)

μ i t = α 2 i + ω i f t + e i t (3)

where f_{t} and g_{t} are unobservable time variant common factors with country-specific factor loadings ϕ_{i} and γ_{i}; and η_{it} and e_{it} are individual country-specific idiosyncratic errors assumed to be distributed independently of the common factors and across panel units. The error term, μ_{it}, is allowed to be correlated with the regressors X_{it}, through the presence of the factors f_{t} and g_{t}. This implies that if the factor loadings ϕ_{i} and ω_{i} are non-zero, estimating Equation (1) without accounting for this correlation will produce biased and inconsistent estimates of long run effects. The CCEMG estimator solves the issue of cross-section dependence by augmenting the regression equation with the cross-sectional averages of the dependent variable as well as the observed regressors:

log C i t = α i + γ i log Y i t + β i log G i t + d 1 i log C ¯ t + d 2 i log Y ¯ t + d 3 i log G ¯ t + e i t (4)

Equation (4) is estimated by OLS for each cross-section. The consistent mean group estimator is derived as the simple average of the group-specific estimates.

To test whether there is a long-run relationship between private consumption, income and government expenditure, we perform the residual-based panel

Test | Statistic | Prob. |
---|---|---|

Hausman test | 142.062*^{ } | 0.000 |

Swamy test | 563.80*^{ } | 0.000 |

Delta | 136.074*^{ } | 0.000 |

Delta adjusted | 182.307*^{ } | 0.000 |

Note: The Hausman test compares Fixed Effects model with Mean Group estimator. * indicates rejection of the null hypothesis at 5% significance level.

cointegration test for the CCEMG model. We apply the Cross-sectionally Augmented Dickey-Fuller (CADF) unit root test proposed by [

Δ log C i t = a i + θ i Δ log Y i t + φ i Δ log G i t + λ i e c t i t − 1 + v i t (5)

where Δ is the first difference operator and e c t i t − 1 is the lagged error correction term computed from the long-run cointegrating relationship of Equation (1), in which e c t i t − 1 = log C i t − 1 − α i − γ i log Y i t − 1 − β i log G i t − 1 . Equation (5) is estimated using CCEMG estimator.

Before carrying out the empirical analysis, we test for the order of integration of the variables by means of unit root tests. This step is necessary to make sure that we do not run spurious regressions. We first apply the well-known IPS test developed by [

The existence of cross-sectional dependency and slope heterogeneity among countries make the CCEMG estimator suitable for estimating the long and short run relationships between government and private consumption. To enable comparison of the results, we also run the Mean Group (MG) estimator which assumes independent errors. The results for the whole panel are depicted in

Level | First difference | |||
---|---|---|---|---|

IPS test | CADF test | IPS test | CADF test | |

Cons | 6.012 (1.000) | 0.367 (0.643) | −24.235* (0.000) | −6.534* (0.000) |

G | 1.587 (0.934) | −2.043* (0.021) | −22.517* (0.000) | −5.941* (0.000) |

GDP | 9.077 (1.000) | 1.193 (0.884) | −19.006* (0.000) | −6.046* (0.000) |

Notes: Tests are conducted for model with intercept. p-values are in parentheses. Optimal lag length was determined using AIC with a maximum of 5. * denotes rejection of the null hypothesis of unit root at the 5% significant level.

Dependent variable: log of private consumption | ||||
---|---|---|---|---|

Model 1 | Model 2 | |||

CCEMG | MG | CCEMG | MG | |

G | −0.096** (−1.87) | −0.110*(−2.13) | −0.123* (−2.73) | −0.125* (−2.49) |

GDP | 0.942* (11.74) | 1.123* (19.2) | 0.733* (7.91) | 0.782* (9.20) |

Intercept | 0.030 (0.04) | −0.787 (−0.87) | 0.503 (0.43) | 1.695* (3.15) |

RMSE | 0.060 | 0.083 | 0.061 | 0.083 |

CD test | −1.024 [0.305] | 3.390* [0.001] | −1.213 [0.225] | 3.36* [0.001] |

IPS | −12.492* [0.000] | −7.220* [0.000] | −12.227* [0.000] | −7.585* [0.000] |

CADF | −5.300* [0.000] | −3.583* [0.000] | −6.168* [0.000] | −3.392* [0.000] |

Note: In model 1 the variables are in real terms while in model 2 they are in real per capita terms. CCEMG is the Common Correlated Effects Mean Group estimator and MG refers to the Mean Group estimator. Figures in parentheses are t-statistics and those in brackets are p-values. The CD test statistics are Pesaran [

degree of cross-section dependence for the MG estimates, but no evidence of cross-section dependence for the CCEMG estimates. In addition the CCEMG estimator shows a lower root mean square error. For this reason, we rely on the CCEMG estimates for inference. From the CCEMG results, it can be seen that private consumption responds negatively to changes in government consumption. A one percent growth in government consumption expenditure leads to a 0.10 percent decline in private consumption. This is an indication of crowing out effect in the long run. This crowding out of private consumption could be explained by the existence of a negative wealth effect induced by increased government expenditure. Thus short term negative wealth effect offsets substitution effect. The negative relationship between government and private consumption is consistent with [

The results further show that there is a positive and significant relationship between consumption and current income. An increase of one percent in income causes household consumption expenditure to rise by 0.9 percent, by keeping other things constant. The coefficient on real total income is greater than that on government consumption, indicating that household consumption strongly depends on current income. This finding is consistent with the Keynesian Absolute Income Hypothesis and other empirical studies [

The mere fact that the results for the whole panel reveal a negative and significant relationship between government and private consumption does not necessarily imply that the crowing-out effect holds in each individual country. To make sure that this result is not driven by a few countries, we look at the country-level CCEMG estimates. The results are reported in

Dependent variable is log of private consumption | ||||
---|---|---|---|---|

Country | Model 1 | Model 2 | ||

G | GDP | G | GDP | |

Benin | −0.109* (−3.21) | 0.929* (10.59) | −0.182*(−4.69) | 0.581* (3.00) |

Burkina Faso | −0.197* (−5.46) | 1.250* (15.65) | −0.263* (−6.54) | 1.230* (8.59) |

Cote d’Ivoire | 0.234* (3.51) | 0.660* (7.32) | 0.200* (3.61) | 0.449* (5.83) |

Gambia | −0.389* (−6.02) | 1.556* (7.07) | −0.335* (−6.75) | 0.335 (0.65) |

Ghana | −0.238* (−5.65) | 1.222* (12.75) | −0.220* (−5.08) | 1.135* (8.84) |

Guinea−Bissau | −0.037 (−0.80) | 1.047* (8.76) | −0.085** (−1.96) | 0.947* (7.26) |

Mali | −0.288*(−4.00) | 0.818* (6.72) | −0.216* (−3.19) | 0.682* (4.98) |

Niger | −0.214* (−5.11) | 0.845* (12.38) | −0.226* (−3.98) | 1.174* (19.95) |

Nigeria | 0.028(0.59) | 0.621* (3.03) | 0.018 (0.37) | 0.711* (3.12) |

Senegal | 0.074 (1.07) | 0.703* (6.99) | −0.087 (−1.59) | 0.333* (2.54) |

Sierra Leone | 0.062 (0.78) | 0.818* (7.97) | 0.088 (1.16) | 0.733* (7.46) |

Togo | −0.082 (−0.81) | 0.836* (4.45) | −0.176 (−1.90) | 0.485** (1.92) |

Note: In model 1 the variables are in real terms while in model 2 they are in real per capita terms. Figures in parentheses are t-statistics. * and ** indicate significance at the 5% and 10% levels, respectively.

of a positive response of private consumption expenditure to government expenditure supports the Keynesian effects of fiscal policy on private consumption. The elasticity of consumption with respect to current income is positive and significant in all countries, confirming the Keynesian absolute income hypothesis. This means that increase in economic growth has significant impact on private consumption in ECOWAS member countries. This finding is in contrast to the work of [

Since the variables under study are cointegrated, we have estimated the error correction model to obtain short run dynamic relationship and results are reported in

Dependent variable is private consumption growth rate | ||||||
---|---|---|---|---|---|---|

Country | Model 1 | Model 2 | ||||

G | GDP | ECT | G | GDP | ECT | |

Benin | −0.106* (−2.34) | 0.907* (4.19) | −0.586* (−3.94) | −0.162* (−2.97) | 0.872* (3.87) | −0.438* (−3.20) |

Burkina Faso | −0.176* (−3.70) | 1.156* (4.64) | −0.341* (−2.57) | −0.196* (−3.97) | 1.166* (4.57) | −0.334* (−2.42) |

Cote d’Ivoire | 0.102** (1.81) | 0.775* (4.76) | −0.482* (−3.81) | 0.156* (2.76) | 0.663* (4.05) | −0.637* (−4.15) |

Gambia | −0.374* (−6.88) | 0.919* (2.13) | −0.430* (−2.36) | −0.318* (−5.78) | 0.755** (1.81) | −0.482* (−3.02) |

Ghana | −0.251* (−4.84) | 1.436* (6.32) | −0.884* (−5.18) | −0.250* (−4.82) | 1.435* (6.28) | −0.825* (−4.67) |

Guinea−Bissau | −0.007 (−0.13) | 0.841* (6.57) | −0.719* (−5.08) | −0.038 (−0.72) | 0.792* (6.09) | −0.707* (−4.88) |

Mali | −0.142* (−2.42) | 0.076 (0.33) | −0.433* (−2.86) | −0.130* (−2.45) | 0.088 (0.43) | −0.544* (−3.88) |

Niger | −0.221* (−4.33) | 0.763* (5.71) | −0.864* (−5.21) | −0.169* (−3.12) | 0.837* (5.63) | −0.565* (−4.16) |

Nigeria | 0.063 (1.02) | 0.494 (1.57) | −1.007* (−4.86) | 0.088 (1.38) | 0.465 (1.43) | −0.991* (−4.68) |

Senegal | 0.033 (0.48) | 0.589* (5.13) | −0.445* (−3.65) | 0.041 (0.57) | 0.514* (3.97) | −0.514* (−2.76) |

Sierra Leone | −0.035 (−0.59) | 1.039* (6.85) | −0.647* (−4.82) | −0.018 (−0.28) | 0.992* (5.90) | −0.669* (−4.62) |

Togo | 0.069 (0.80) | 0.942* (4.00) | −1.280* (−8.43) | 0.030 (0.31) | 0.859* (3.31) | −1.178* (−7.27) |

Panel | −0.087* (−2.02) | 0.828* (8.38) | −0.677* (−8.26) | −0.080** (−1.92) | 0.786* (7.92) | −0.645* (−8.86) |

Note: In model 1 the variables are in real terms while in model 2 they are in real per capita terms. Figures in parentheses are t-statistics. * and ** indicate significance at the 5% and 10% levels, respectively.

The country-level estimates show a crowding out effect of government consumption on private consumption in Benin, Burkina Faso, Gambia, Ghana, Mali and Niger, and no significant effect in Guinea-Bissau, Nigeria, Senegal, Sierra Leone and Togo. However, there is a crowding in effect of government consumption on private consumption in the case of Cote d’Ivoire. As can be seen, a one percent increase in government consumption leads to an increase in private consumption by 0.10 percent. Hence, Keynesian hypothesis of positive relationship between government and private consumption holds only in Cote d’Ivoire. Therefore, private consumption cannot be held responsible for any crowding-out effect that government spending might have on aggregate demand in Cote d’Ivoire. The estimated coefficient of the income variable is positive and significant in all countries except Mali and Nigeria.

In this study, we examined the effect of government consumption on household final consumption expenditure for 12ECOWAS countries over the period from 1970 to 2016. We employed a panel estimation technique that is specifically designed to deal with the key econometric problems plaguing previous studies, namely cross-section dependence and slope heterogeneity. Our empirical strategy deals with these issues using the Common Correlated Effects Mean Group estimator developed by [

The results of this study call for some plausible reasons behind the nature of the relationship between government and private consumption. Why countries show different patterns in this nexus? The study has used aggregate data and as we know individual components of government spending might have different relationships with household consumption. In fact, various components of government spending would be valued differently by households and would affect their consumption decision differently. For instance, if government spends more on education and health, households would then have to spend less in those items. On the contrary, if government spending is intending to improve public transport system, households would use public transport more frequently and spend more on transportation. Therefore, the composition of government spending may determine the differential and heterogeneous effects on household consumption across countries. Thus, there is scope for further research by disaggregating government consumption into different components. Another plausible explanation of our results could lie in the way used by governments to finance their spending. Finally, our estimation method has assumed the effect of government spending on private consumption to be the same regardless the level of household consumption. This effect could be different for households with different consumption levels. Thus, as another future avenue of research, we suggest analyzing the asymmetric effects of government spending along the distribution of private consumption. We leave all these challenging avenues for future researches.

The author declares no conflicts of interest regarding the publication of this paper.

Keho, Y. (2019) Impact of Government Spending on Private Consumption: Evidence from ECOWAS Countries. Modern Economy, 10, 600-614. https://doi.org/10.4236/me.2019.103041