Taxation aggregate activity and economic growth; further cross-country evidence on some supply-side hypotheses.
Garrison, Charles B. ; Feng-Yao Lee
TAXATION, AGGREGATE ACTIVITY AND ECONOMIC GROWTH: FURTHER
CROSS-COUNTRY EVIDENCE ON SOME SUPPLY-SIDE HYPOTHESES
This paper investigates the effect of marginal tax rates on the
level of economic
activity. Data from sixty-three countries for the period 1970-84
provide support for
Koester and Kormendi's method of estimating marginal tax
rates for individual
countries. However, their conclusion that increases in marginal
tax rates have negative
effects on the level of economic activity is not robust when we
extend the time period
from 1970-79 to 1970-84. Further, even for Koester and
Kormendi's own data set,
the negative relation does not hold when the sample is
disaggregated into industrial
countries and low-income countries.
I. INTRODUCTION
In a recent issue of this journal, Koester and Kormendi [1989] use
cross-country data to examine the effects of average and marginal tax
rates on economic activity. Koester and Kormendi provide a method for
estimating average and marginal tax rates from published tax revenue and
income data for a given country. They then use this method to derive tax
rates for sixty-three countries for the period 1970-79, and this enables
them to test the supply-side proposition that a reduction in a
country's marginal tax rate leads to an increase in the rate of
growth of economic activity and/or the level of economic activity. Based
on their results, they conclude that an increase in marginal tax rates
significantly reduces the level of per capita GDP. At the same time,
however, they find that once they control for the positive relation
between per capita GDP and the size of the government sector (as
reflected in a country's average tax rate), the negative relation
between marginal tax rates and the rate of economic growth disappears.
This paper employs Koester and Kormendi's method of estimating
tax rates and uses new income data made available by Summers and Heston
[1988] to extend the study period for the same sixty-three countries to
1985. The later data provide additional support for the reliability of
average and marginal tax rates estimated by the Koester and Kormendi
method. However, Koester and Kormendi's finding of a negative
relation between marginal tax rates and the level of economic activity
is not robust to extension of the time period to 1985. Further, we find
that, even for Koester and Kormendi's own data set, the negative
relation does not hold when the sample countries are grouped into
eighteen industrial countries and forty-five low-income countries. In
addition, the negative relation disappears if the full data set is
reduced from sixty-three to sixty-one by excluding two countries which
are "outliers" in the calculation of marginal tax rates. These
latter two findings are of interest because they indicate that Koester
and Kormendi's result of a significant negative relation between
marginal tax rates and per capita income is not robust but sensitive to
the selection of countries included in their study.
With respect to the influence of tax rates on the rate of economic
growth, our results for the extended time period (not shown here) tend
to confirm Koester and Kormendi's finding that neither average nor
marginal rates affect growth. In particular, once the level of per
capita GDP is taken into account, the apparently negative effect of tax
rates on growth disappears. Further, we find no evidence that an
increase in marginal tax rates reduces capital accumulation or labor
force growth. Combining these results with our findings regarding the
level of economic activity, we conclude that cross-country evidence
provides little or no support for the supply-side hypothesis that
increases in tax rates adversely affect economic activity.
II. THE EFFECT OF TAXES ON THE LEVEL OF ECONOMIC ACTIVITY
We follow Koester and Kormendi in estimating for each country the
following regression: (1) [TAXREV.sub.t] = [a.sub.0] + [a.sub.1]
[GDP.sub.t] + [u.sub.t [prime]] where TAXREV = total tax revenues,
[a.sub.0] and [a.sub.1] are parameters to be estimated, and u is the
disturbance term. The coefficient [a.sub.1] is the country's
marginal tax rate (MARTAX). We estimate [a.sub.1] for both 1970-79 and
the longer period 1970-84. The average tax rate (AVGTAX) for each
country is the mean of the ratio of tax collections to GDP for the years
in the sample period for which data are available. The results appear
reasonable; for example, estimations of equation (1) separately for
sixty-three countries for the period 1970-84 yield an average [R.sup.2]
of 0.77 and an average t-statistic for [a.sub.1] of 11.13. Further, the
average t-statistic for [a.sub.0] is 3.65; this supports the hypothesis
that average and marginal rates differ over the set of countries
studied.
Koester and Kormendi base their conclusion regarding the adverse
effects of high marginal tax rates on per capita GDP on the following
regression: (2) [YPC80.sub.j] = -0.77 + 2.48 [AVGTAX.sub.j]
(-1.43) (6.55)
-0.38 [MARTAX.sub.j] + [e.sub.j].
(-2.62)
[R.sup.2] = 0.48 Adjusted [R.sup.2] = 0.47
In this equation, [YPC80.sub.j] is per capita GDP in 1980 in
country j, and t-statistics are in parentheses. Koester and Kormendi
contend not only that the negative relation between per capita GDP and
the marginal tax rate is "...evidence in support of the supply-side
hypothesis that high marginal tax rates adversely affect the level of
economic activity" but also that "...marginal rates have
distinct effects from average rates of taxation as per supply-side
theory" [1989, 380]. The interpretation of the significantly
positive coefficient of AVGTAX is not that a high average tax rate
causes a high per capita income. Rather, Koester and Kormendi contend
that equation (2) provides a method of controlling for the known
positive relation between average tax rates and per capita income, so as
to isolate the effect of marginal tax rates on per capita GDP. Koester
and Kormendi conclude that the coefficient on MARTAX measures the effect
of marginal tax rates on per capita GDP, holding constant average tax
rates.
We now wish to determine whether Koester and Kormendi's
results are sensitive to (1) alternative grouping of countries and (2)
testing with alternative time periods. Koester and Kormendi consider the
possibility that high-income countries and low-income countries are
affected differently by tax rates, and accordingly stratify their sample
into two subsets: LDCs (less developed countries) and non-LDCs. Deriving
t-statistics of -1.9 and -1.8 on MARTAX from the two subsets, Koester
and Kormendi conclude that the benefits of reductions in marginal tax
rates are "...economically important (and statistically
significant)..." for both groups [1989, 381].
Koester and Kormendi do not specify the particular countries
included in their non-LDC and LDC subsets. They do, however, report the
average per capita GDP for the two groups. Based on these averages, we
surmised that Koester and Kormendi's two subsets include thirty-one
non-LDCs and thirty-two LDCs.(1) But this grouping seems arbitrary; the
danger is that the non-LDC subset, which has a very wide range of 1980
per capita income ($2,152 to $8,089) does not provide any additional
information over that conveyed by the entire sample set. Consequently,
we have estimated their equation (2) for a smaller number of high-income
countries: the eighteen countries identified in World Development Report
[1988] as "industrial market economies." This definition has
the advantage of reducing the range of per capita income within the
"high-income" group and raising the average per capita income
substantially: the range for these eighteen countries is $3,467 to
$8,089 and the mean is $6,051.(2) An implication of reducing the number
of countries in the high-income subset is that the number of countries
in the low-income subset is expanded to forty-five. A net effect of the
alternative grouping is that the disparity in average per capita income
between the high-income countries and the low-income countries is
greater than in Koester and Kormendi's grouping. Thus it might be
argued that the alternative grouping provides more information than does
Koester and Kormendi's grouping.
The result of this alternative grouping is that the coefficient on
MARTAX for the eighteen industrial countries is less than half that for
Koester and Kormendi's non-LDC subset. Further, the t-statistic is
only -0.68, so that marginal tax rates become statistically not
significant in explaining the level of economic activity in industrial
market economies. The result is much the same for the low-income subset,
where the coefficient on MARTAX is only one-fourth as large as in
Koester and Kormendi's LDC subset and the t-statistic is only
-0.37. Thus marginal tax rates are not statistically significant in
explaining the variation in YPC80 among the forty-five low-income
countries.
The sensitivity of Koester and Kormendi's results to
alternative grouping leads us to a second alteration. Koester and
Kormendi note that their data set includes two "outliers":
Israel and Zaire. Given their average tax rates, these two countries
have exceptionally high marginal tax rates. Further, Israel has a very
high marginal tax rate (1.42 for Koester and Kormendi's data set)
and a very high average tax rate (0.42) given its per capita income
level. Koester and Kormendi exclude these two countries in estimating
almost all their equations; nevertheless, they do not exclude them in
estimating equation (2).
We wish to determine whether Koester and Kormendi's results
are sensitive to the exclusion of Israel and Zaire from their data set;
the results are reported in Table I. For the entire sample, now
consisting of sixty-one countries, MARTAX no longer is statistically
significant in explaining YPC80. Further, for Koester and
Kormendi's two groups (non-LDCs, now consisting of thirty
countries, and LDCs, now consisting of thirty-one countries) MARTAX is
not statistically significant. Consequently, we are skeptical of their
test of the supply-side hypothesis that high marginal tax rates
adversely affect the level of economic activity, since their results
rely so heavily on the inclusion of one or two countries. We next
attempt to determine whether their results are sensitive to an extension
of the time period.
Summers and Heston [1988] have now provided a new set of
international estimates of real GDP at 1980 international prices for the
period 1950-85. Incorporating their estimates into our data set enables
us to now estimate equation (2) with YPC85 (real GDP per capita in 1985)
as the dependent variable. Tax rates for the YPC85 equation are for the
period 1970-84, and are based on updated estimates of tax revenues from
the IMF's Government Finance Statistics Yearbook. Our real GDP and
GDP deflator are from World Bank's most recent World Table [1989]
on magnetic tape.
We show in Table II estimates of equation (2) using the later data.
The most striking result is that, for the full sixty-three-country
sample, MARTAX is not statistically significant when tax rates are
calculated from 1970-84 data and the dependent variable is YPC85. Thus
Koester and Kormendi's results are not upheld when YPC85 is the
dependent variable and tax rates are calculated for the longer period.
Further, MARTAX is not statistically significant for any subset of
countries in the sample, except for the subset of thirty-two
less-developed countries. However, exclusion of two outliers (Ethiopia
and Zaire) from the LDC group reduces sharply the size of the
t-statistics for MARTAX, from -2.51 to -1.56.(3)
We note further that for several of the subsets the equation
performs very poorly in explaining per capita income (as reflected in
the values for the adjusted [R.sup.2]), suggesting that variables other
than the marginal tax rate are much more important in explaining the
level of economic activity.(4) [Tabular Data I and II Omitted]
(1)Our calculations of mean YPC are $4,703 and $1,108, respectively,
for the thirty-one non-LDC countries and thirty-two LDC countries.
Koester and Kormendi, however, report means of $4,698 and $1,113. The
sixty-three countries cannot be divided into two groups which have means
closer to those reported by Koester and Kormendi. (2)The eighten
industrial countries are Australia, Austria, Belgium, Canada, Denmark,
Finland, France, Germany, Ireland, Italy, Netherlands, New Zealand,
Norway, Spain, Sweden, Switzerland, United Kingdom and United States.
(3)Ethiopia was not an outlier in the Koester and Kormendi study.
However, Koester and Kormendi, who estimated a marginal tax rate of only
0.275 for Ethiopia, collected their tax data from the 1981 volume of
Goverment Finance Statistics Yearbook, which includes data from 1972 to
1977 for Ethiopia. We collected data from the 1988 volume, which
includes tax data from 1972 to 1979 for Ethiopia. (4)We tried a number
of alternative grouping schemes but could find none in which MARTAX was
statistically significant when the dependent variableis YPC85 and tax
rates are calculated for the longer period. For example, if Japan (for
which sufficient data are available to calculate tax rates for 1970-84
but not for 1970-79) is added to the list of industrial countries to
create a nineteen-country subset, the t-statistic on MARTAX is only -
1.62. If the small countries of Iceland and Luxembourg (both of which
have high incomes but populations of less than one million) as well as
Japan are added to create a subset of twenty-one countries, the
t-statistic falls to -0.95.
REFERENCES
Koester, Reinhard B., and Roger C. Kormendi. "Taxation,
Aggregate Activity and Economic Growth: Cross-Country Evidence of Some
Supply-side Hypothesis." Economic Inquiry, July 1989, 367-86.
Summers, Robert, and Alan Heston. "A New Set of International
Comparisons of Real Product and Price Levels Estimates for 130
Countries, 1950-85." Review of Income and Wealth, Series 34, March
1988, 1-25. International Monetary Fund. Government Finance Statistics
Yearbook. Washington, D.C.: International Monetary Fund, 1980, 1981,
1988. World Bank. World Development Report. Washington, D.C.: The World
Bank, 1982, 1988. __. World Table on magnetic tape. Washington, D.C.,
The World Bank, 1989.
CHARLES B. GARRISON and FENG-YAO LEE, The University of Tennessee,
Knoxville, Department of Economics. We wish to thank Roger Kormendi,
Frank C. Wykoff, and an anoymous referee for helpful suggestions. The
results in the paper, however, are solely the responsibility of the
authors.