3.2  Fiscal-Growth Estimates for the 1990s.

Table 6 demonstrates that the net effect on growth of the changes to the fiscal budget in EMU countries is generally quite small. The exceptions to this are Spain and Finland, where the long-run growth rate of the economy decreased by over one percentage point per annum. As long-run effects, these figures are probably too large to be credible and serve to highlight several limitations in an application such as this using currently available empirical estimates and fiscal data which, for those countries, may not completely eliminate cyclical effects. Perhaps the most important limitation in this regard is the possibility of heterogeneity in the effect of fiscal policy changes across countries because of differences in institutional characteristics, as noted in Section III. Caution over the size of these estimated effects leads us to remove these countries from the sample from this point onwards.[11]

Among the EMU countries, the average growth rate is expected to have increased due to fiscal policy in France and the Netherlands, and decreased in Austria and Germany. It is clear from Table 6 that whether fiscal effects on average growth rates are positive or negative depends both on whether fiscal deficits increase or decrease and on the mix of taxes and expenditure.

Note: Overall growth effects include effects from other revenues and expenditures. rdis = distortionary tax revenues; eprd = productive expenditure; surp/def budget surplus/deficit.

Using the variance-covariance matrix to create confidence intervals around these estimates suggests that there is a 95% probability that the net effect of the changes made to fiscal policy in the 1990s is positive in France and the Netherlands and negative in Germany and Austria. These growth effects are reasonably precisely estimated and, though rather modest, are composed of much larger effects from individual changes to policy (see Table 6). For example, the dominant factor explaining the 0.2 percentage point per annum net increase in average growth in France is the increase in productive forms of expenditure (row 4, column 2). By itself this added over half a percentage point to the long-term growth rate. In the Netherlands the net positive effect appears to arise almost entirely because of a reduction in the deficit, given that the large positive growth effect from decreasing distortionary taxes (1.2 percentage points) was completely offset by the decrease in productive expenditures.

The predicted decline in average growth rates in Germany in the table is a consequence of increased revenues from distortionary taxation (0.3 percentage points) and decreased expenditure on productive goods and services (0.2 percentage points). Proposed EMU membership and the re-unification probably explain much of this combination of policy changes in Germany. Finally, in Austria an increase in revenues from distortionary taxation again appear to offer much of the explanation for the net decline in growth (1.1 percentage points), although this effect was to some degree offset by the effect of increasing productive expenditure (0.8 percentage points).

The remaining EU countries used their increased deficit to fund increased productive or unproductive expenditures, with some compensating or additional movement in tax revenues. In Denmark, Sweden and the UK, unproductive expenditures rose 1% of GDP, but whereas no substantive changes were made in the remainder of the budget in Denmark, Sweden experienced declines in distortionary taxation and the UK saw increases in non-distortionary tax revenues. As already noted above, the use of deficit financing meant that public spending in the UK and Sweden grew by around 2 percentage points of GDP from the late 1980s, whereas in Denmark the public sector actually decreased. The reduction in revenues from distortionary taxation is the principal explanation of the long-run growth increase in Sweden compared to Denmark and the UK. This change alone added 1 percentage point per annum to growth. Again the confidence interval placed round these estimates suggest a 95 per cent chance that the net effect of the changes made to fiscal policy in the UK and Denmark had a negative effect on growth and a positive effect in Sweden.

3.3  Disaggregated Expenditures

By disaggregating expenditure categories, greater detail can be added to the government budget and to estimated fiscal-growth effects. The final column of Table 6 reports growth effects using regressions that disaggregate productive expenditures into education, health and ‘other productive’. The alternative parameter estimates have some impact on the estimated growth effects. Perhaps the most obvious change is that for Sweden where the net effect alters from +0.20 to –0.11 percentage points per annum. The principal explanation for this switch in sign is the negative growth effect from reducing expenditures on education over the 1990s that was not fully captured using the aggregated data. Unlike the forecasts for the other countries this new forecast is outside the 95 per cent confidence interval made using results from aggregated data.

The confidence intervals using disaggregated expenditure data are larger, suggesting some cost to using this greater level of detail in the data. Nevertheless, in only two of the ten forecasts does the confidence interval cross zero (Sweden, France), indicating that while we must be cautious about the precise magnitude of net growth effects, we can remain reasonably confident regarding whether the net effect of fiscal policy was positive or negative.

3.4  Heterogeneous fiscal-growth effects

In estimating individual country fiscal-growth effects, we have assumed so far that the homogeneous parameters estimated over all countries applies to each. This sub-section addresses an alternative source of potential differences between countries: that these marginal effects differ across countries. That is, do regression parameters differ across the sample such that, ceteris paribus, some EU countries experience stronger fiscal effects on growth than others?

It is known that the results from a dynamic fixed effects (DFE) regression are likely to be biased if, as Pesaran and Smith (1995) suggest, the assumption of homogeneity of the short-run parameter estimates across countries cannot be accepted. They show that this may be a more serious problem than the bias generated by the inclusion of lagged dependent variables and can lead to inconsistent and misleading results even for large T and large N. To overcome this bias they suggest the use of either the pooled mean group (PMG) or mean group (MG) estimators (Pesaran, et al., 1999). A comparison of the results from these two has the additional advantage of allowing us to address formally the question of whether the long-run effect of fiscal policy on growth is identical across countries.[12] Acceptance of this restriction implies that the results from the PMG estimator are more efficient than those from the MG estimator (Pesaran, et al., 1999).

The estimated regression for the MG model is of the following ARDL form,

(5)    

where i indicates the country, t is time, g is the rate of growth, F is a matrix of fiscal variables, φ, β and γ are parameters to be estimated and ε_it a classical error term. The test for the long run effect of fiscal policy is made on the parameter β₁ (the long run fiscal policy parameter adjusted for lagged growth). Consistent with the general-to-specific approach, the lag structure of the regression is chosen on the basis of the Schwarz information criteria. The long run effect of fiscal policy across countries is taken as the (unweighted) average of the estimates from the N individual country regressions. The PMG model differs from these single country time series regressions by imposing homogeneity of the long-run parameters: φ_i and β_1i become φ and β₁respectively. A Hausman test can be used to test the statistical plausibility of this restriction.[13]

The disadvantage of the MG and PMG estimators is of course that unless the available time series is very long a degrees of freedom problem is soon reached. For this reason we restrict the right-hand-side variables to include the investment rate and three fiscal variables: the surplus, distortionary taxation and productive expenditure. These are chosen in light of the results from BKG (2001) and it is worth remembering that the coefficients on these terms must be interpreted as conditional, on those excluded fiscal variables, some of which the results from BGK suggest may be significant. We are also forced to restrict the regression equation to include a maximum of two lags of the dependent variable.

We begin by estimating equation (5) for the 16 OECD countries and then for the 10 EU countries. As reported in Table 7, we provide the individual test statistics (p-values) from the Hausman test of homogeneity of the long-run parameters as well as the test statistic from a joint test. For the OECD16 we find we can accept homogeneity both collectively and individually, although for distortionary taxation (rdis) acceptance is at the 13 per cent. Similarly for the EU10 we can again accept the restriction that the long-run parameters are identical, although for productive expenditures (eprd) acceptance is at the 11 per cent level. Having accepted homogeneity of the long-run parameters we choose to report the results from the PMG estimator in the table. (We omit the short-run parameters to conserve space).

Concentrating on the fiscal parameters, consistent with the results from BGK (2001) the surplus/deficit and productive expenditures are found to affect the growth rate positively whereas distortionary taxation is found to lower growth. All of these long-run parameter estimates are significant at standard confidence levels. The parameter estimates are broadly in line with those from BGK, the effect of distortionary taxation and the surplus are very close to the estimates found in that paper whereas the effect of productive expenditure is slightly lower. In order to test whether this is because of the use of a different set of countries, a slightly longer time period and the removal of several variables from the right hand side of the regression we re-estimate the DFE model of BGK with these restrictions. Regression 3 (Table 7) imposes the same length of lag structure as BGK. The DFE regression with a long lag structure in fact produces similar results to the PMG regressions. The coefficients are similar in value, although the standard errors are somewhat larger such that the coefficient on the budget surplus is no longer significant and that on productive expenditure is significant only at the 10 per cent level.

Bassanini & Scarpetta (2001) argue that in small country samples the estimated parameters may be sensitive to the inclusion or exclusion of any one country, even when the Hausman tests do not reject the assumption of homogeneity of the long-run parameters. Following their example, we re-estimate the PMG regression 1 excluding in turn one country from the sample. Figure 1 reports the coefficients for each of the fiscal variables when a single country is omitted. We also indicate the standard errors from the full sample results to provide 95% confidence intervals for the results. As can be seen the parameter estimates remain stable from such a test and never stray outside of the confidence bands. The parameter estimates from the EU10 sample in Table 7 also lie within these confidence bands.

Note: t-statistics in parentheses.