We now explore the well-known prediction (implied e.g. by Barro, 1990) that it is not the size of government per se that matters to growth, but the mix between size and efficiency/quality. As said in Section 2 above, behind this there are two key issues: first, that the effect of fiscal policy on growth is non-linear; and second, this non-linearity is driven by efficiency or quality in the public sector.
6. 1 A measure of public sector quality and econometric specification
We need a proxy for the quality of public sector in Greece, which also has an annual variation. In Angelopoulos and Philippopoulos (2005), we have constructed an index of relative public sector efficiency for a group of 64 developing and industrialized countries over four 5-year periods, 1980-1985, 1985-1990, 1990-1995 and 1995-2000, by following the methodology of Afonso et al. (2003) for the OECD countries. However, most of the data used for the construction of that index are not available on a time-series basis. As far as we know, only one of those indexes is available: the variable Electric Power Transmission and Distribution Losses, which is available from WDI on an annual basis for the whole period 1960-1999. Tanzi and Davoodi (1998) have also used this variable as a proxy for the quality of public infrastructure.19 Obviously, this index can only provide a crude approximation of the quality of public sector; however, it is fortunate that there exists such a variable which also fulfills our time-series requirements. All related studies, we are aware of, are cross-section studies (see Angelopoulos and Philippopoulos, 2005, for a review of the literature).
To get a proxy for the quality/efficiency of public sector, we first take the log of the inverse of Electric Power Transmission and Distribution Losses and denote this as govqual. In turn, by first-differencing, we get the growth rate of govqual, denoted as Δgovqual. Table 1 reports that Δgovqual is an I(0) variable. As a preliminary step, we just include Δgovqual in our basic regression (the one of column 3 of Table 2.) As can be seen in column 1 of Table 4, by simply adding efficiency into the growth regression does not alter anything.
To test our main idea, we work in two steps. In the first step, we use our proxy for government efficiency, Δgovqual, to divide years into efficient and inefficient. In the second step, we examine whether the effect of government size differs depending on whether we are in an efficient or an inefficient year.
Consider the first step. When Δgovqual is positive, which implies an improvement in the quality of public infrastructure relative to the previous year, we classify the government in that year as being “efficient”. Conversely, when Δgovqual is negative, we classify the government in that year as being “inefficient”. This simple classification rule implies that the Greek public sector is classified as “efficient” in 22 years and as “inefficient” in 17 years, over the period 1960-1999.20
6. 2 Estimation and tests
Given this classification, we move on to the second step by regressing growth on our principal measure of the overall size of government (namely, the growth rate of the share of government in GDP, Δgovshare) by allowing the size effect to differ between the efficient and the inefficient sub-sample (denoted respectively as Δgovshareeff and Δgovshareineff). The results, reported in column 2 of Table 4, reveal a non-linear relationship between fiscal spending and economic growth. As can be seen, Δgovshareineff is significantly negative at the 1% level, while Δgovshareff is quantitatively small (close to zero) and insignificant. The estimated quantitative effect of Δgovshare in the inefficient regime is important, as this is more than double than the average effect reported in Table 2. Note that these results are net of any direct effect of Δgovqual on growth, because we have controlled for Δgovqual in the regression (the latter is positive but not significant).
Notice that the effects of dummy variables are also affected relative to section 4 (compare new results with those in column 3 of Table 2). That is, the d(1960-1973) dummy is now insignificant although still positive, while the d(1980-1993) period remains negative and significant. It is interesting that the d(1994-2000) dummy now turns to be significantly negative. An obvious explanation is that all recent years have been classified as efficient, so that the positive effect on growth has been already controlled for (via the size-efficiency mix).
It is also important to note that the R2 in this model jumps to about 80%. This is a rather impressive fit. It implies that the growth in the share of government in GDP, when allowing for a non-linear Laffer curve-type effect, can explain - along with some simple political dummies - around 80% of the variation of the growth rate in the Greek economy over the last forty years. This again highlights the importance of fiscal policy (now both its size and quality) for macroeconomic outcomes.
Finally, notice that the specification in column 2 of Table 4 passes the serial correlation tests. Also, the RESET test for non-linear functional form cannot reject the null of a correct specification. Since the RESET tests of the regressions in columns 1-3 of Table 2 and column 1 of Table 4 reject the null, this adds to our confidence that there is a Laffer curve pattern from fiscal policy to growth, and this pattern is captured by our model specification in column 2 of Table 4.