Cointegration and Causal Relationship between Taxes and Spending for Kenya, Nigeria and South Africa

Abstract

Causal relationships between taxes and spending are examined for three African countries using the GDP as a control variable, and dummy variables to address structural changes in Nigeria and South Africa. There is one cointegration equation between nominal fiscal variables in all three countries, one cointegration equation for Kenya and two cointegration equations for Nigeria and South Africa for the real fiscal variables and their respective dummy variables. Short-term results of the nominal variables show fiscal independence for all three countries. In real terms, taxes cause spending for Kenya and Nigeria and a weak fiscal synchronization for South Africa. There is long run fiscal synchronization in nominal terms for all three countries, and in real terms for both Nigeria and South Africa, while real taxes cause spending in Kenya. Long-run estimates show a unit increase in nominal (real) taxes translating into a less than proportionate increase in nominal (real) spending for Kenya and South Africa, and a more than proportionate increase in nominal (real) spending for Nigeria. Fiscal imbalance is not a threat in the budgetary process in Kenya and South Africa, but an issue of concern in Nigeria, where oil revenues are a major source of support for budget short falls.

Keywords:

• Autoregressive distributed lag error-correction model
• taxes
• spending
• causality
• cointegration

JEL Classifications:

• C3
• E6
• H6

Acknowledgements

The author wishes to acknowledge with gratitude helpful comments of the editor and referee of this journal, and absolves them from any remaining errors and/or omissions.

Notes

¹Nigeria has the unique position of being the only African country in the current study that has paid off its debt in 2007. See also Figures 1, 2 and 3. Note that we have extended Narayan and Narayan's (2006) data sample of South Africa from 1966–2000 to 1960–2007.

²The unit roots tests have been skipped because they are routine equations.

³ Equation (3) is estimated by excluding dummy variables (ND and SAD) for Kenya, and including dummy variables ND for Nigeria and SAD for South Africa.

⁴See also the fully modified (FM) VAR model proposed by Phillips (1995) and Phillips and Hansen (1990).

⁵Dickey et al. (1986) recommend that the unit root test is more efficient when trends are excluded from deterministic variables employed in such tests.

⁶Exact maximum likelihood estimates of real government spending as a function of real taxes for the long run using a first order autoregressive, AR(1), process to correct for serial correlation, with p-values reported in square brackets are as follows:

Kenya: g _t =0.969 [0.00]τ _t , DW=2.224, , F(1, 35)=8874.40 [0.00].

Nigeria: g _t =1.045 [0.00]τ _t −0.576 [0.01]ND _t , DW=1.790, , F (2, 39)=852.03 [0.00].

South Africa: g _t =0.965 [0.00]τ _t −0.106 [0.03]SAD _t , DW=1.852, , F (2, 45)=9098.60 [0.00].