4  Convergence and Divergence in EU Fiscal Policy

The above results appear to suggest that fiscal policy has had a significant, and fairly robust, effect on long-run growth in European countries, and more widely in the OECD. Evidence from, for example, Barro and Sala-i-Martin (1995) and de la Fuente (1998) also suggests a tendency towards per capita income convergence within and between European countries. This raises the obvious question: has fiscal policy contributed towards, or acted against, income convergence in Europe? There are, of course, several forces acting towards convergence of particular fiscal variables across the EU. For example, the ‘convergence criteria’ for monetary union include restrictions on budgetary deficits for participating countries, while tax harmonisation guidelines have operated more widely within the EU for some time, encouraging moves towards similar indirect tax rates in particular. Convergence of deficits and taxes would, of course, via the government budget constraint, imply some convergence of public expenditures. However, as we have seen, within tax and expenditure totals, EU countries continue to have considerable discretion over distortionary/non-distortionary and productive/unproductive components. Since it is these components which are important for growth, growth convergence need not necessarily be fostered by fiscal changes.

Figure 1 – Parameter Heterogeneity

We begin by considering whether the five fiscal/GDP ratios (2 x tax; 2 x expenditure; budget surplus) are converging over time for a sample of EU countries. When examining income convergence, the usual measure adopted is σ-convergence – measuring changes in the standard deviation of income levels. In our case however, since we wish to be able to compare across fiscal categories as well as over time, standard deviations are not very helpful due to the absence of any normalisation. An alternative, used by Sanz and Velazquez (2001), is to construct ‘similarity indices’ which measure the share of particular expenditure categories in total expenditure for country i relative to the average for all countries in the sample. A similar measure could be used for our GDP ratios. However, while this measure has a lower bound of zero (identical values across countries), it has no upper bound.

For present purposes, Gini coefficients provide a preferable alternative. Applied to the relevant fiscal category, these provide a measure of the degree of inequality (dissimilarity) in fiscal variables across countries. Since our public expenditure or tax/GDP ratios are simply expenditures (or tax revenues) measured in GDP units, we can consider the spatial distribution of expenditures (taxes) across EU countries analogously to the distribution of income across individuals.[14] Thus a Gini = 0 implies complete equality (identical values across countries), and Gini = 1 implies complete inequality (i.e. one country spends the EU total, all others spend zero). Table 8 below shows Gini coefficients for total revenue and expenditure, while Figure 2 shows Ginis for the two expenditure (eprd, enprd) and tax (rdis,rndis) components. Since it is important that the sample is unchanged across periods, these are calculated for five 5-year periods from 1970-1995, for a sample of 10 EU countries for which comparable data are available.[15] (Period averaging has been used to smooth short-term variations). Similar Gini coefficients for surplus/deficits are shown in Figure 3.[16]

Table 8 suggests a considerable degree of similarity across EU countries and that both total tax and expenditure ratios have become more similar (equal) since the mid-1970s.[17] However, it can be seen from Figure 2 that, while there is some evidence of unconditional convergence (declining Gini) for the tax components, there is not much evidence for expenditure components. Final period values for expenditures are not very different from those at the start of the period. Deficits, in Figure 3, show clear divergent tendencies (increasing Gini).[18]

To examine the strength of these effects we test for statistically significant changes in Ginis, and also use two tests for convergence based on changes in variances. These are the Variance Ratio (VR), and Likelihood Ratio (LR) tests proposed by Lichtenberg (1994) and Carree and Klomp (1995) respectively. The VR test is a simple ratio of initial and final variances while the LR test also uses the covariance. Table 9 presents the results of these tests, which indicate that for cases of potential convergence, the null hypothesis of no convergence cannot be rejected (with the possible exception of rndis). Note that the alternative hypothesis, H₁, is of divergence for unproductive expenditure and surpluses, since for these cases, the final period variance (or Gini) exceeds initial period values. The Ginis suggest significant divergence for the budget surplus.

Though there is little evidence of unconditional convergence across all of our EU sample, it may nevertheless be the case that (i) a subset of countries share a steady-state; and (ii) that countries converge (conditionally) to their own, or a shared, steady-state. We can test for this using the fixed-effects regression:

(6)    

where y is the relevant fiscal variable, the αs and β are parameters and ε is a classical error term. The parameter β captures convergence (β < 0) from short-run disequilibrium, towards a steady-state. The parameter α_t captures common time-varying shocks, while α_i captures country fixed effects, such that α_i ≠ 0 implies country i does not share a steady-state with country N – the ‘default’ country.

Results from this exercise are shown in Table 10. There is clear evidence of within-country β-convergence with, perhaps unsurprisingly, especially strong equilibrating tendencies for the budget surplus/deficit.[19] However, tests of the null that generally reject the hypothesis, suggesting that the steady-state values for the various fiscal ratios differ across our OECD sample countries (though for rndis, H₀ is rejected only at the 10% level). Similar rejection is evident for the EU10 sample.

Note: the null hypothesis is based on whether the ratio of final to initial variances (VR test) exceeds or falls short of unity. *(**) = exceeds 10% (5%) critical value.