EMMA GALLI [*]
This article proposes refined econometric estimates of effective marginal income tax rates for 23 OECD countries from 1951 to 1990. Panel regressions find such measures negatively correlated with economic growth. These results are consistent with endogenous growth theories and opposite to those of most empirical literature, which relies on measures of effective average tax rates. The negative correlation is also robust to consideration of other growth determinants. (JEL H21, O11)
In a 1989 issue of this journal, Koester and Kormendi analyzed the relationship between taxation and output levels and dynamics in a cross-section of 63 countries during the 1970s. They concluded that no significant partial correlation exists between effective tax rates and economic growth. Such a result was at odds with the then-emerging endogenous growth paradigm, which maintains that high marginal tax rates on the "engines of growth" of the economy, like physical and human capital accumulation, negatively affect the long-run output growth rate.  The refutation of theory's predictions has spurred a series of empirical analyses on the relationship between growth and taxes. So far, this line of research has provided ambiguous results. Engen and Skinner (1992) find a negative correlation between average tax rates and economic growth in a sample of 107 countries during the 1970-85 time interval. Easterly and Rebelo (1993) use a variety of fiscal indicators to explain output levels and dynamics. They fail to find significant correlations between tax rates and growth and attribute this outcome to the potential endogeneity of fiscal policy to the scale of the economy. Finally, Mendoza et al. (1997) notice that measures of effective average tax rates on labor income, capital income, and consumption are negatively correlated with investments but not with growth rates in a panel of 18 Organization of Economic Cooperation and Development (OECD) countries from 1965 to 1991. They propose these results as evidence in favor of Harberger's (1964a, 1964b) "superneutrality" hypothesis, namely, that consumption/saving and labor/leisure choices are invariant to changes in the tax structure.
We argue that this lack of support for the predictions of endogenous growth models may be due to a misspecification of the fiscal variables. Theory posits that marginal effective tax rates on factor incomes distort the efficient allocation of resources and therefore reduce output growth. Marginal effective tax rates are not observable, however, and there is no obvious estimate of them. Most studies resort to average effective tax rates as proxy. Ram (1986), Engen and Skinner (1992), Plosser (1993), and Easterly and Rebelo (1993) simply use the ratio between tax revenues and GDP. This specification, as Easterly and Rebelo (1993) recognize, poses problems of collinearity with government expenditures. The coefficient on average tax pressure then becomes difficult to interpret, all the more so as theory suggests that government spending may be growth enhancing or retarding according to its end use. Mendoza et al. (1994, 1997) also argue that average fiscal pressure is too aggregated for a comprehensive test of t he theory. They construct a measure of effective average tax rates, distinguishing taxes on consumption and factor incomes.
Measures of effective average tax rates are characterized by a relative computational ease and enable one to construct yearly and therefore long time series. They also make it easier to simulate fiscal pressure on different tax bases. Yet they are not the marginal values that the optimization problems embedded in the theory yield. As OECD tax structures are generally progressive , average tax rates underestimate marginal tax rates. Studies that rely on average values, whatever their level of disaggregation, cannot therefore be considered proper tests of the predictions of growth theories about taxation. …