Average tax rate cyclicality in OECD countries: a test of three fiscal policy theories.
Furceri, Davide ; Karras, Georgios
1. Introduction
Is fiscal policy procyclical? Should it be? The standard Keynesian
model implies that fiscal policy should be countercyclical, that is,
government spending should increase in economic contractions and fall in
expansions, whereas taxes should follow the opposite pattern. At the
same time, Tax Smoothing models inspired by Barro (1979) suggest that
the government should smooth the tax rate by borrowing in recessions and
repaying in booms. In other words, changes in GDP will be positively
correlated with tax revenue but should be uncorrelated with tax rates
and tax revenue as a percentage of GDP. (1)
The empirical literature has tried to test the main recommendations
of both of these normative theories. (2) Focusing on the cyclicality of
government expenditure, the basic finding has been that government
spending tends to be slightly procyclical in developed countries and
strongly procyclical in developing countries. (3)
These findings have been interpreted as suggesting that fiscal
policy behavior is not consistent with Keynesian recommendations and
only partly consistent (for some developed countries) with the Tax
Smoothing hypothesis.
In contrast, both findings appear to be consistent with the
political economy theory of fiscal policy recently proposed by
Battaglini and Coate (2008). The economic model underlying this theory
is a dynamic stochastic general equilibrium model in which a single good
is produced using labor, whose productivity depends on the business
cycle. The political economy component of the model assumes that policy
choices in each period are made by a legislature comprising
representatives elected by single-member, geographically defined
districts. (4)
The main prediction of this theory is that fiscal policy is
procyclical. In particular, the model demonstrates that government
spending increases in booms and decreases during recessions, whereas tax
rates decrease during booms and increase in recessions. In other words,
the theory predicts that government spending as share of GDP should be
neutral over the cycle and that tax revenue as percentage of GDP should
be negatively correlated with changes of GDP. As we have already said,
these predictions appear to be broadly consistent with the empirical
evidence on government spending, but are they also consistent with the
evidence on tax revenue?
The aim of the present article is to contribute to the literature
by investigating the relation between tax rates and GDP changes. In
particular, we will try to assess whether tax revenues as a share of GDP
are positively or negatively correlated with the cyclical component of
real GDP. We will test the predictions of Battaglini and Coate's
(2008) positive theory and compare them with the normative frameworks of
the standard Keynesian model and the Tax Smoothing hypothesis.
To this purpose we compute the correlations between average
effective tax rates and cyclical output using a set of 26 OECD countries
for which we have data from 1965 to 2003. The results show that the
correlations for the great majority of countries are not statistically
significantly different from zero. Moreover, to differentiate between
different phases of the business cycle, we compute these correlations
over expansions (defined as periods with positive cyclical components)
and downturns (defined as periods with negative cyclical components).
Again, the results uncover almost no statistically significant
correlations.
All our findings are robust to the different types of taxes we
tried, which include (i) total taxes; (ii) taxes on income, profits, and
capital gains; (iii) social security contributions; (iv) taxes on
payroll and workforce; (v) taxes on property; (vi) taxes on goods and
services; and (vii) other taxes. Moreover, the results are also robust
to the use of a panel framework in assessing the relationship between
tax rates and changes in the cyclical component of GDP.
Thus, the empirical evidence of our article is largely consistent
with the Tax Smoothing hypothesis, whereas it seems to be inconsistent
with both the standard Keynesian recommendations and the predictions of
the theory proposed by Battaglini and Coate (2008). Taken together with
the literature's findings on the cyclical behavior of government
spending, our results call for a new theory of fiscal policy that is
consistent with the empirical evidence on the behavior of both
government spending and revenue (as well as debt) over the cycle.
The rest of the article is organized as follows. Section 2
describes the empirical methodology and the data we use to assess the
pattern of tax revenue over the cycle. Section 3 presents and discusses
the empirical results, and section 4 concludes.
2. Data and Empirical Methodology
All tax data are from the Revenue Statistics of OECD Member
Countries database, and measure various taxes as a percentage of GDP. In
particular, we consider the following average effective tax rates: (i)
total taxes; (ii) taxes on income, profits, and capital gains; (iii)
social security contributions; (iv) taxes on payroll and workforce; (v)
taxes on property; (vi) taxes on goods and services; and (vii) other
taxes.
Table 1 provides a list of these 26 OECD economies together with
country averages over 1965-2003 for the seven tax series. (5) The
average total tax rate has varied from 17.2% in Mexico to 46.7% in
Sweden. Although these OECD countries have relied very differently on
the various forms of taxes, income taxation has been the largest revenue
generator for most of them. (6) Taxes on income, profits, and capital
gains have ranged from 4.6% of GDP in Mexico to 25.1% of GDP in Denmark.
In nine of the countries, most of the revenues were raised by taxes on
goods and services. (7) Only in three countries has the largest share
been generated by social security taxes, (8) and in none of the
countries by property taxes, which are generally the smallest.
As shown in Figure I, total tax rates in these OECD countries have
generally increased over time, but with important differences among
countries in terms of starting and ending values.
Real GDP data are obtained from the OECD's Economic Outlook
database. We consider the same set of 26 countries, for most of which we
have data on taxes from 1965 to 2003. (9)
We have tried several methods of obtaining the cyclical components
of GDP. First, we use three different filtering techniques to detrend
the (log) output series of each country and estimate its cyclical
component: (i) the Hodrick-Prescott (HP) filter with a smoothness
parameter equal to 100, (ii) the HP filter with a smoothness parameter
equal to 6.25, and (iii) a bandpass (BP) filter. These are among the
most commonly used detrending methods in the business cycle literature.
As an additional measure of cyclical GDP, we also considered data of the
output gap from the OECD Economic Outlook (2009). (10) Because our main
findings proved quite robust to these different cyclical measures, we
will only report results on the basis of the HP-6.25 filter in the next
section to preserve space. (11)
In Figure 2, we plot the business cycle component of real GDP
obtained using the HP filter with a smoothness parameter equal to 6.25,
our benchmark method. Casual inspection of Figure 2 reveals that
countries differ substantially in terms of cyclical behavior. This
variability, together with that of the tax rates, should facilitate the
empirical identification of the relationship between the tax rates and
the business cycle.
[FIGURE 1 OMITTED]
To assess the behavior of a tax rate over the cycle, we then
compute the correlation between the tax rate and the cyclical component
of the corresponding country's GDP,
corr([tax.sup.j.sub.i,t], [gdg.sup.c.sub.i,t]), (1)
where [gdp.sup.c.sub.i,t] is the cyclical component of (log) output
for country i at time t, and [tax.sup.j.sub.i,t] is +the tax rate j for
country i at time t.
Moreover, to differentiate between different phases of the cycle,
we compute these correlations over expansions (defined as times of
positive cyclical components) and downturns (defined as times of
negative cyclical components).
As a final test, we try to assess how the tax rates respond to
cyclical changes in GDP when all countries are considered together. To
this purpose, we estimate a simple bivariate relationship between tax
rates (tax revenue as percentage of GDP) and cyclical (log) GDP:
[(Tax / GDP).SUB.i.t] = [alpha] + [beta] x [gdg.sup.c.sub.i,t] +
[[epsilon].sub.i,t]. (2)
[FIGURE 2 OMITTED]
In particular, we estimate Equation 2 with the use of ordinary
least squares, country fixed effects (FE), country and time FE, and
Instrumental Variables (we instrument cyclical GDP with lagged values of
our dependent and independent variables).
3. Results
For a preliminary assessment of the behavior of tax revenue over
the cycle, we first compute the average total tax rates over the period
1965-2003 in periods of upturns (positive cyclical components of real
GDP) and downturns (negative cyclical components of real GDP) and the
value of total tax rates at peaks (the highest value of the cyclical
component of real GDP) and troughs (the lowest value of the cyclical
component of real GDP). The results are reported in Table 2. Looking at
the table, it is possible to see that the behavior of the total tax rate
over the phases of the cycle differs among countries. In particular,
focusing on the first two columns of the table, it emerges that, whereas
for 13 of the 26 countries the tax rate is bigger during downturns than
during upturns, for the other half of the sample, the tax rate is bigger
during upturns. The differences, however, are never sizable. This
balanced behavior during cyclical upturns and downturns is confirmed by
averaging the tax rate over all countries in the sample: These averages
are virtually identical in the two phases of the cycle. (12)
Similarly, focusing on the last two columns of the table, it seems
that Battaglini and Coate's (2008) prediction that tax rates
decrease during booms and increase in recessions cannot be confirmed. If
that prediction were true, we would expect tax rates at peaks to be
lower than tax rates at troughs. In fact, for most of the countries (15
of 26) we find the opposite result, and once again those tax rates
averaged over countries are virtually the same at peaks and troughs.
Next, as described in the previous section, we move to a more
formal assessment of tax cyclicality by computing the correlations
between tax rates and the cyclical components of (log) real GDP for each
of the 26 OECD countries over the period 1965-2003. The results are
reported in Table 3. The table also reports the correlation between tax
rates and GDP in cyclical downturns and upturns (columns 2 and 3).
Looking at the first column of Table 3, one can immediately see
that, with the exception of the United Kingdom (for which there is a
sizable and weakly statistically significant negative correlation), the
correlation between tax rates and GDP is slight and not statistically
significant. The result is broadly confirmed when we look at the
correlation during cyclical downturns and upturns. However, for some
countries (such as Greece and Italy) the absence of correlation over the
entire cycle is explained by the negative correlation of tax revenue to
cyclical upturns and positive correlations over downturns.
These results are robust to the different filtering methods (HP-100
and BP) used to retrieve the cyclical components of real GDP and to the
use of the OECD's output gap measure. In fact, looking at Table 4,
one can see that the absence of a strong or statistically significant
correlation, or both, is confirmed for the great majority of cases. We
note that this finding that the tax rate is essentially uncorrelated
with cyclical GDP is consistent with the Tax Smoothing hypothesis but
inconsistent with both the standard Keynesian recommendation and
Battaglini and Coate's (2008) prediction.
So far, our analysis has focused on the total tax rate. We now
continue by looking at various types of taxes and their correlation with
cyclical GDP. This is more than a robustness exercise: In fact, many of
the relevant theories (including Battaglini and Coate 2008) are
formulated specifically in terms of income taxes. Thus, it is important
to differentiate between different tax rates. In Table 5, we report the
correlation between cyclical GDP and (i) taxes on income, profits, and
capital gains; (ii) social security contributions; (iii) taxes on
payroll and workforce; (iv) taxes on property; (v) taxes on goods and
services; and (vi) other taxes.
Starting with the results for income taxes (first column of Table
5), one readily observes that the correlations are not significantly
different from zero for all OECD countries in the sample (including the
United Kingdom). This result suggests that income tax rates (i.e.,
income taxes as a percentage of GDP) are neutral over the cycle,
contradicting Battaglini and Coate's predictions of a negative
correlation. (13) Similar implications can be drawn for all the
different tax rates, with the exception of the payroll tax rate for
Canada, Denmark, and Ireland and the tax rate on goods and services for
Mexico, which are the only cases to show a statistically significant
negative correlation with cyclical GDp. (14)
So far, we have been assuming that the correlations between
cyclical GDP and tax rates are time-invariant. We now relax this
assumption and allow these correlations to vary with time by estimating
them for rolling 20-year windows for each country. Figure 3 shows how
these time-varying correlations of the total tax rate with cyclical GDP
have evolved over time. Two basic results clearly stand out. First, as
expected, the results differ substantially among countries, and second,
the assumption of time-invariant correlations appears questionable. For
some countries (e.g., Australia, Denmark, Finland, Japan, Norway, Spain,
and Sweden) the correlation is increasing over time; for other countries
(e.g., Austria, Canada, France, Italy, Korea, Turkey, and United States)
it fluctuates around values close to zero (or slightly negative),
whereas for the rest of the sample (Belgium, Iceland, Ireland, and
Luxembourg) it decreases toward sizeable (and statistically significant)
negative values.
Again, however, the results suggest that Battaglini and
Coate's prediction of a consistently negative correlation between
GDP and the tax rate cannot be confirmed by the empirical evidence in
most of the countries analyzed.
As a next test, we try to assess how tax rates respond to a
cyclical change of GDP when all countries are considered together. For
this purpose, we estimate a simple bivariate relationship between tax
rates (again defined as tax revenue as a percentage of GDP) and cyclical
GDP (Eqn. 2). The results are reported in Table 6, where in each column
we present the results obtained with a different estimation method. (15)
Analyzing the results, one can observe that although the point estimate
of the coefficient measuring the degree of cyclicality of the tax rate
is negative, it is never statistically significant in any specification.
[FIGURE 3 OMITTED]
Next, we allow for a richer lag structure to permit the tax rates
to respond to cyclical output changes with some delay. To this end, we
re-estimate the simple regression of tax rates with one, two, and three
lags of cyclical output. (16) The results, reported in Table 7, still
point to the absence of a significant relation.
As a final test, we re-estimate tax rates as functions of cyclical
GDP and several control variables and interaction terms. In this way, we
try to control for possible omission bias, and we assess whether the
smoothness of tax rates might depend systematically on structural
country characteristics. In particular, the variables included in the
analysis are (i) trade openness (measured by the sum of total exports
and imports as a fraction of GDP); (ii) country size (measured by the
log of total population); (iii) the Gini coefficient for income
inequality; and (iv) a political dummy, left, that takes the value of 1
for "left-leaning" governments and equals zero otherwise. (17)
These variables are generally thought to be associated with tax revenue
as a share of GDP. For example, smaller and more open economies are
generally characterized by a larger government size (Alesina and
Wacziarg 1998). Left-leaning governments could also be associated with a
larger tax-to-GDP share, and countries with higher inequality might have
an incentive to increase tax rates for redistribution purposes. The
results in Table 8 show that indeed these variables are associated with
the share of total tax revenue in GDP; however, they do not affect the
response of tax rates to cyclical GDP. In particular, the estimated
coefficients suggest that smaller and more open economies tend to have
higher tax rates, and that income inequality also raises tax rates. In
contrast, the effect of a left-leaning government on taxes is not
statistically significant. The inclusion of these variables, both as
control and as interaction terms, however, does not affect the estimated
response of tax rates to cyclical GDP, so that the absence of a
correlation between the tax rates and cyclical GDP is still confirmed.
Our results are also robust to using different tax rates in the
panel regressions. Table 9 reports the results of estimating regression
(2) separately for each of the six types of taxes we considered above.
As the results indicate, the overwhelming majority of these estimates
are statistically insignificant. In particular, the coefficients for the
income tax, as well as taxes on goods and services, payroll, and
property taxes, are never statistically significantly different from
zero. In contrast (at least, when we include country fixed effects)
social security contributions appear to be negatively related to
cyclical changes, whereas "other taxes" are positively
associated with them. Thus, this method also seems to confirm a
generally neutral behavior of tax rates over the cycle.
4. Conclusions
This article investigated the cyclicality of the average effective
tax rate to investigate the validity of three theories of fiscal policy.
According to the standard Keynesian theory, tax policy should be
countercyclical; that is, tax rates should decrease in economic
contractions and increase in expansions. The Tax Smoothing hypothesis
instead implies that changes in GDP should be uncorrelated with tax
rates. In contrast to these normative theories, a recent positive theory
proposed by Battaglini and Coate (2008) predicts that the tax rate will
be negatively correlated with GDP. (18)
This article relies on these very different theoretical
implications to test the three theories. We compute the correlations
between tax revenue as a fraction of GDP and cyclical output for each of
26 OECD countries for which we have data from 1965-2003. The results
show that the correlations are generally quite small and statistically
indistinguishable from zero.
This finding is quite robust. It is obtained when we look at the
relationship between tax rates and cyclical GDP over upturns and
downturns or at business cycle peaks and troughs. It is also valid when
we look at different tax rates, in addition to total taxes, such as
income taxes, social security contributions, payroll taxes, property
taxes, taxes on goods and services, and "other" taxes. The
results are also robust to assessing the relationship between tax rates
and cyclical changes of GDP in a panel framework.
Overall, the empirical evidence of our article seems to be more
consistent with the Tax Smoothing hypothesis than either the
recommendations of the standard Keynesian model or the predictions of
Battaglini and Coate's (2008) theory. This result calls for a new
theory of fiscal policy that will be consistent with the empirical
evidence on both government spending and government revenue (as well as
debt) over the cycle.
Appendix: Filtering Methods
The first two methods use the Hodrick-Prescott (HP) filter proposed
by Hodrick and Prescott (1980). The filter decomposes the series into a
cyclical ([c.sub.i,t]) and a trend ([g.sub.i,t]) component by minimizing
with respect to [g.sub.i,t] for [lambda] > 0, the following quantity:
[T.summation over (t=1)] [([y.sub.i,t] - [g.sub.i-t]).sup.2]+
[lambda] [T-1.summation over (t=2)] [([g.sub.i,t] - [g.sub.i-t]).sup.2].
(A1)
The first method uses the value recommended by Hodrick and Prescott
for annual data for the smoothness parameter ([lambda]) equal to 100.
The second method sets the smoothness parameter ([lambda]) equal to
6.25. In this way, as pointed out by Ravn and Uhlig (2002), the
Hodrick-Prescon filter produces cyclical components comparable to those
obtained by the bandpass (BP) filter.
The third method makes use of the BP filter proposed by Baxter and
King (1995) and evaluated by Stock and Watson (1998) and Christiano and
Fitzgerald (1999), who also compare its properties to those of the HP
filter. The low-pass (LP) filter[alpha](L), which forms the basis for
the BP filter, selects a finite number of moving average weights
[[alpha].sub.h]), to minimize,
Q [[integral].sup.[pi].sub.-[pi]] [|[delta] ([omega]|.sup.2]
d[omega], (A2)
where
[alpha](L) = [[summation].sup.K.sub.h=-K] [[alpha].sub.h] [L.sup.h]
(A3)
and
[alpha](K)([omega]) = [[summation].sup.K.sub.h=-K] [[alpha].sub.h]
[e.sup.-I [omega]h] (A3) (A4)
The LP filter uses [[alpha].sub.k]([omega]) to approximate the
infinite MA filter [beta][omega]. Let's define [delta]([omega])
[equivalent to] [beta]([omega]) - a[omega]. Then, minimizing Q, we
minimize the discrepancy between the ideal LP filter [beta]([omega]) and
its finite representation [[alpha].sub.K][omega] at frequency [omega].
The main objective of the BP filter as implemented by Baxter and King
(1995) is to remove both the high-frequency and low- frequency component
of a series, leaving the business cycle frequencies. This is obtained by
subtracting the weights of two LP filters. We define[[omega].sub.L].
and[[omega].sub.H], the lower and upper frequencies of two LP filters as
8 and 2, respectively, for annual data. We therefore remove all
fluctuations shorter than two or longer than eight years. The frequency
representation of the BP weights becomes [[alpha].sub.K]
([[omega].sub.H]) - [[alpha].sub.K]([[omega].sub.H]) and forms the basis
of the Baxter-King filter which provides an alternative estimate of the
trend and the cyclical component.
References
Alesina, A., F. Campante, and G. Tabellini. 2008. Why is fiscal
policy often procyclical? Journal of the European Economics Association
6:1006-36.
Alesina, A., and R. Wacziarg. 1998. Openness, country size and
government. Journal of Public Economics 69:305-21.
Barro, R. J. 1979. On the determination of the public debt. Journal
of Political Economy 87:940-71.
Battaglini, M., and S. Coate. 2008. Fiscal policy over the real
business cycle: A positive theory. NBER Working Paper No. 14047.
Baxter, M., and R. King. 1995. Measuring business cycles:
Approximate band-pass filters for economic time series. NBER Working
Paper No. 5022.
Beffy, P. O., P. Ollivaud, P. Richardson, and F. Sedillot. 2006.
New OECD methods for supply-side and medium-term assessments: A capital
services approach. OECD Economics Department Working Papers No. 482.
Catao, L., and B. Sutton. 2002. Sovereign defaults: The role of
volatility. International Monetary Fund Working Paper 02/149.
Christiano, L., and T. Fitzgerald. 1999. The band pass filter. NBER
Working Paper No. 7257.
Gavin. M., and R. Perotti. 1997. Fiscal policy in Latin America.
NBER Macroeconomics Annual 12:11-71.
Hodrick, R., and E. Prescott. 1980. Postwar U.S. business cycles:
An empirical investigation. Discussion paper 451, Carnegie Mellon
University.
Ilzetzki, E., and C. Vegh. 2008. Procyclical fiscal policy in
developing countries: Truth or fiction? NBER Working Paper No. 14191.
Kaminsky, G., C. Reinhart, and C. Vegh. 2004. When it rains, it
pours: Procyclical capital flows and macroeconomic policies. NBER
Working Paper No. 10780.
Lane, P. 2003. The cyclical behaviour of fiscal policy: Evidence
from the OECD. Journal of Public Economics 87:2261-675.
Organisation for Economic Co-Operation and Development (OECD).
Revenue statistics of OECD member countries database. OECD. Economic
outlook database.
Persson, T. 2001. Do political institutions shape economic
policy'? NBER Working Paper No. 8214.
Persson, T., and G. Tabellini. 2001. Political institutions and
policy outcomes: What are the stylized facts? CEPR Discussion Paper No.
2872.
Ravn, M. O., and H. Uhlig. 2002. On adjusting the Hodrick-Prescott
filter for the frequency of observations. The Review of Economics and
Statistics 84:371-5.
Romer, David. 2006. Advanced Macroeconomics. 3rd edition. New York:
McGraw-Hill.
Sorensen, B., L. Wu, and O. Yosha. 2001. Output fluctuations and
fiscal policy: U.S. state and local governments 1978- 1994. European
Economic Review 45:1271-310.
Stock, J., and M. Watson. 1998. Business cycle fluctuations in U.S.
macroeconomic time series. NBER Working Paper No. 6528.
Talvi, E., and C. Vegh. 2005. Tax base variability and procyclical
fiscal policy. Journal of Economic Development 78:156-90.
(1) Versions of the theory also predict that changes in GDP should
be uncorrelated with government spending but negatively correlated with
government spending as a proportion of GDP and that debt will be
increasing in booms and decreasing in the recessions (i.e., debt is
positively correlated with changes in GDP). For an example of the
"traditional" Keynesian theory, see Romer (2006, chapter 5).
For an exposition of the Tax Smoothing hypothesis in terms of the
average effective tax rate, see Romer (2006, chapter 11).
(2) See. for example, Gavin and Perotti (1997); Sorensen, Wu, and
Yosha (2001); Lane (2003); Talvi and Vegh (2005); Alesina, Campante, and
Tabellini (2008): and Ilzetzki and Vegh (2008).
(3) A possible explanation can be found in the literature of credit
supply. In fact, as argued by Catao and Sutton (2002) and Kaminsky,
Reinhart, and Vegh (2004), developing countries become credit
constrained during economic downturns or can borrow only at very high
interest rates. This precludes governments from effectively smoothing
economic fluctuations. As a result, government expenditure increases
during good times and decreases during bad times (i.e., fiscal policies
become procyclical). Another explanation to account for different
behavior of fiscal policy is political factors. Persson (2001), Persson
and Tabellini (2001), and Alesina, Campante, and Tabellini (2008) find
that political and institutional variables also matter for fiscal
responsiveness. In particular, Alesina, Campante, and Tabellini (2008)
show that most of the procyclicality of fiscal policy in developing
countries can be explained by high levels of corruption in those
countries. The authors build a model wherein a high degree of corruption
means that surpluses accumulated in good times when fiscal policy is
countercyclical are not optimal because the electorate knows that
corrupt political representation would misappropriate them. It is then
rational for the electorate to demand higher spending during expansions
(i.e., procyclical fiscal policy).
(4) The model assumes that the government can raise revenues in two
ways (via a proportional tax on labor income and by issuing one-period
risk-free bonds) and that public revenues are used to finance the
provision of public goods. The level of public debt and the persistent
level of productivity are the state variables, creating a dynamic
linkage across policymaking periods.
(5) Country selection is dictated by data availability only.
(6) To be specific, for 14 countries out of the 26: Australia,
Belgium, Canada, Denmark, Finland, Japan, Luxembourg, the Netherlands,
New Zealand, Norway, Sweden, Switzerland, the United Kingdom, and the
United States.
(7) The nine are Austria, Greece, Iceland, Ireland, Korea, Mexico,
Portugal, Spain, and Turkey.
(8) These three are France, Germany, and Italy.
(9) The exceptions are Iceland (1980-2003), Korea (1972-2003), and
Mexico (1980-2003).
(10) See Beffy et al. (2006) for details.
(11) We provide a more detailed discussion of these filtering
methods and their properties in the Appendix. We have also preferred the
HP-6.25 filter to simple differencing because the HP produces a cyclical
component that has a long-run average value of zero. This is a desirable
property for a cyclical component, but it is strongly violated by output
growth rates.
(12) Similar results are obtained for different tax rates. They are
available upon request.
(13) In fact, for the majority of countries, the point estimates of
these correlations of cyclical GDP with the income tax rate are
positive. It is tempting to observe that this positive sign (although
definitely the wrong sign for the Battaglini and Coate prediction) is
consistent with the standard Keynesian recommendation, whereas the lack
of statistical significance is consistent with the Tax Smoothing
hypothesis.
(14) Unlike the results for the income tax, the preponderance of
the point estimates of the correlations of cyclical GDP with social
security contributions and taxes on goods and services are negative.
This sign (although not the lack of statistical significance) is
consistent with the Battaglini and Coate (2008) prediction.
(15) We consider as instruments two lags of the cyclical GDP. The
Anderson statistics, the Cragg Donald Wald statistics, and the Sargan
test of overidentification restrictions suggest that the choice of the
instruments is appropriate. For simplicity, we report only the Sargan
test's p value in Table 6. Additional results are available from
the authors upon request.
(16) We are grateful to an anonymous referee for suggesting this
extension, as well as the one outlined in the next paragraph.
(17) Data for trade openness and populations are taken from the
OECD Economic Outlook (2009), data for Gini coefficients are from the
OECD Main Economic Indicators (2009), and data for left governments are
from the Database of Political Institutions political database.
(18) It might be helpful to remind the reader that our locus has
been on average effective tax rates (tax-to-GDP ratios) rather than
marginal tax rates. Although the latter is the relevant variable for
many important economic issues (e.g., the effects of taxes on labor
supply or long-run economic growth), our focus here has been on the
cyclical properties of aggregate tax revenue, for which the former is
appropriate.
Davide Furceri, OECD, 2 rue Andre Pascal. 75775 Paris Cedex 16:
E-mail davide.furceri@oecd.org: furceri.economia@ unipa.it.
Georgios Karras, University of Illinois at Chicago. Department of
Economics, 601 South Morgan Street, Chicago, IL 60607, USA: E-mail
gkarras@uic.edu; corresponding author.
This work has been carried out while Davide Furceri was working as
an economist at the Fiscal Policy division of the European Central Bank
(ECB). Tile opinions expressed herein are those of the authors and do
not necessarily reflect those of the ECB or the Eurosystem, the OECD, or
its member countries.
Received September 2009: accepted May 2010.
Table 1. Average Tax Rates (% of GDP, Averaged over the Period
1965-2003)
Income,
Total Profits, Social
Country Tax Capital Gains Security
Australia 27.1 15.0 --
Austria 26.4 10.6 12.3
Belgium 41.9 15.9 13.0
Canada 32.8 14.9 4.0
Denmark 44.4 25.2 1.4
Finland 40.0 16.2 8.8
France 40.4 7.2 16.3
Germany 36.1 11.8 12.8
Greece 27.7 5.1 8.7
Iceland 32.2 10.0 1.9
Ireland 31.6 11.0 4.1
Italy 34.2 10.6 11.7
Japan 24.6 10.6 7.5
Korea 17.9 5.1 1.3
Luxembourg 38.1 15.3 10.5
Mexico 17.3 4.6 2.5
Netherlands 41.0 12.5 15.8
New Zealand 31.6 19.7 --
Norway 40.3 15.4 8.9
Portugal 26.1 6.3 7.3
Spain 26.3 7.0 10.2
Sweden 46.7 20.4 11.4
Switzerland 26.4 11.4 6.9
Turkey 19.5 6.9 2.8
United Kingdom 35.3 13.7 6.0
United States 26.6 12.6 5.9
Average 32.7 12.2 8.5
Goods and
Country Payroll Property Services Others
Australia 1.5 2.5 8.2 --
Austria 2.7 1.0 12.8 0.4
Belgium -- 1.1 11.8 --
Canada 0.7 3.4 9.8 0.3
Denmark 0.3 2.1 15.6 0.1
Finland 0.9 1.0 13.5 0.1
France 0.9 2.4 12.3 1.3
Germany 0.2 1.3 10.1 --
Greece 0.3 1.8 12.0 --
Iceland 0.9 2.4 17.0 0.7
Ireland 0.5 2.2 14.0 --
Italy 0.2 1.5 10.0 1.9
Japan -- 2.4 4.3 0.1
Korea 0.1 2.0 9.0 0.5
Luxembourg 0.3 2.7 9.3 0.1
Mexico 0.2 0.3 9.5 0.2
Netherlands -- 1.5 11.0 0.1
New Zealand 0.4 2.3 9.3 -
Norway -- 1.0 15.0 0.2
Portugal 0.4 0.8 11.1 0.4
Spain 0.2 1.4 7.5 0.2
Sweden 1.6 1.1 12.2 0.1
Switzerland -- 2.2 5.9 --
Turkey -- 1.0 7.3 2.4
United Kingdom -- 4.3 10.8 1.0
United States -- 3.3 4.9 --
Average 0.8 1.9 10.6 0.5
Table 2. Total Taxes over the Cycle (%, Averaged over the Period 1965-
2003)
Country Downturns Upturns Peaks Troughs
Australia 26.4 27.8 29.4 27.2
Austria 27.0 25.7 30.5 27.0
Belgium 42.0 41.8 37.9 40.6
Canada 33.2 32.4 34.8 33.1
Denmark 43.7 45.8 41.1 40.0
Finland 41.4 38.5 43.0 44.9
France 41.7 39.3 43.0 35.9
Germany 36.5 35.7 36.8 32.2
Greece 28.1 27.4 20.3 21.3
Iceland 31.7 32.7 29.7 28.3
Ireland 31.9 31.3 33.5 26.9
Italy 34.5 33.9 25.7 26.1
Japan 23.5 26.0 22.3 18.2
Korea 17.9 17.9 16.6 21.1
Luxembourg 37.9 38.4 31.4 42.4
Mexico 17.4 17.2 15.7 16.7
Netherlands 41.3 40.9 41.2 41.3
New Zealand 31.1 32.0 30.4 23.2
Norway 40.0 40.7 42.5 43.5
Portugal 25.1 27.0 18.4 20.8
Spain 25.6 27.3 17.5 16.5
Sweden 46.1 47.4 38.5 46.9
Switzerland 26.8 25.9 25.0 28.5
Turkey 20.3 18.7 28.4 35.1
United Kingdom 35.6 34.9 31.4 34.4
United States 26.7 26.6 25.5 27
Average 32.0 32.0 30.4 30.8
Table 3. Correlations between Total Tax Rates and Cyclical Output
(1965-2003)
Country Overall Downturns Upturns
Australia 0.10 -0.04 -0.19
Austria -0.01 0.41 0.30
Belgium -0.09 0.00 -0.31
Canada -0.11 -0.19 0.16
Denmark 0.09 0.10 -0.31
Finland -0.05 0.23 0.27
France -0.06 0.27 0.32
Germany 0.06 0.66 *** 0.16
Greece -0.10 0.45 ** -0.57 ***
Iceland 0.10 0.36 -0.38
Ireland -0.14 -0.20 0.00
Italy 0.01 0.48 ** -0.47 **
Japan 0.17 0.10 -0.44 **
Korea 0.06 0.03 0.22
Luxembourg -0.13 -0.02 -0.57 ***
Mexico -0.13 0.32 -0.40
Netherlands -0.08 -0.06 -0.06
New Zealand 0.03 0.30 -0.30
Norway 0.02 -0.27 0.10
Portugal 0.02 0.01 -0.33
Spain 0.08 0.09 -0.14
Sweden 0.09 0.07 -0.13
Switzerland -0.16 0.07 -0.19
Turkey -0.08 -0.39 * 0.51 **
United Kingdom -0.27 * 0.02 -0.35
United States 0.04 0.37 -0.10
Average -0.01 0.09 -0.08
*, **, and *** indicate statistical significance at 10%, 5% and 1%,
respectively.
Table 4. Correlations between Total Tax Rates and Output Business
Cycle (1965-2003)
Country HP6.25 HP100 BP (a) Output GAP
Australia 0.10 0.09 -0.03 -0.04
Austria -0.01 -0.05 -0.08 -0.31
Belgium -0.09 -0.13 -0.25 -0.48 ***
Canada -0.11 -0.05 -0.18 -0.20
Denmark 0.09 0.08 -0.06 0.18
Finland -0.05 -0.06 -0.14 -0.09
France -0.06 -0.07 -0.26 -0.31
Germany 0.06 0.08 0.06 0.00
Greece -0.10 -0.09 -0.27 -0.35 *
Iceland 0.10 0.26 0.14 0.07
Ireland -0.14 -0.20 -0.46 *** -0.73 ***
Italy 0.01 -0.05 -0.22 -0.46 **
Japan 0.17 0.05 -0.17 0.24
Korea 0.06 -0.05 -0.20 -0.10
Luxembourg -0.13 -0.16 -0.29 -0.61 ***
Mexico -0.13 -0.12 -0.29 0.31
Netherlands -0.08 -0.05 -0.33 * -0.34 **
New Zealand 0.03 0.09 0.16 -0.41 **
Norway 0.02 0.09 0.26 0.50
Portugal 0.02 0.03 -0.10 0.30
Spain 0.08 0.02 -0.20 0.63 ***
Sweden 0.09 0.08 -0.07 0.05
Switzerland -0.16 -0.18 -0.44 *** -0.26
Turkey -0.08 -0.01 0.09 -0.16
United Kingdom -0.27 * -0.25 -0.40 ** -0.27
United States 0.04 0.09 0.00 0.06
Average -0.01 0.09 -0.08 -0.11
*, **, and *** indicate statistical significance at 10%,
5% and 1%, respectively.
(a) BP correlations are based on fewer observations (6).
Table 5. Correlations between Various Tax Rates and Output Business
Cycle (1965-2003)
Income, Profits,
Country Capital Gains Social Security Payroll
Australia 0.16 -- 0.02
Austria 0.05 -0.05 -0.05
Belgium -0.04 -0.07 --
Canada 0.00 -0.13 -0.64 ***
Denmark 0.02 0.06 -0.29 **
Finland 0.11 -0.10 -0.18
France 0.12 -0.08 -0.26
Germany 0.06 0.00 -0.22
Greece -0.07 -0.11 0.00
Iceland 0.04 -0.10 -0.02
Ireland -0.02 0.01 -0.45 **
Italy 0.03 -0.03 0.07
Japan 0.31 0.01 --
Korea 0.07 -0.03 --
Luxembourg -0.11 -0.24 -0.12
Mexico 0.37 0.10 0.29
Netherlands 0.02 -0.18 --
New Zealand 0.13 -- -0.27
Norway 0.05 -0.03 --
Portugal 0.09 0.01 -0.19
Spain 0.11 0.05 --
Sweden 0.30 0.00 0.03
Switzerland -0.15 -0.15 --
Turkey -0.03 -0.17 --
United
Kingdom -0.17 -0.07 --
United States 0.22 -0.12 --
Average 0.06 -0.06 -0.07
Goods and
Country Property Services Other
Australia -0.01 -0.02 --
Austria -0.03 -0.10 0.12
Belgium 0.06 -0.30 --
Canada -0.26 -0.07 -0.23
Denmark 0.09 0.34 --
Finland 0.04 -0.6 --
France -0.06 -0.11 -0.10
Germany -0.08 0.18 --
Greece -0.03 -0.12 --
Iceland 0.00 0.22 0.02
Ireland -0.11 -0.22 --
Italy -0.15 0.04 0.37
Japan -0.02 -0.04 --
Korea 0.08 0.15 -0.03
Luxembourg 0.13 -0.12 --
Mexico 0.36 -0.40 ** -0.20
Netherlands 0.16 0.06 0.02
New Zealand 0.08 -0.07 --
Norway -0.10 0.15 0.27
Portugal 0.14 -0.03 -0.06
Spain 0.20 0.06 -0.15
Sweden 0.10 -0.09 -0.06
Switzerland -0.03 -0.09 --
Turkey -0.16 -0.09 0.33
United
Kingdom 0.03 -0.17 -0.22
United States -0.02 -0.13 --
Average 0.02 -0.05 0.01
*, **, and *** indicate statistical significance at 10%, 5% and 1%,
respectively.
Table 6. Response of Tax Rates to Output Business Cycle (1965-2003)
Country and Time
OLS Country FE FE
Degree of cyclicality (R) -4.21 -4.51 -6.32
(-0.23) (-0.79) (-1.04)
Number of observations 976 976 976
[R.sup.2] 0.00 0.00 0.13
Sargan statistic p-value
IV
Degree of cyclicality (R) -26.15
(-0.42)
Number of observations 947
[R.sup.2] --
Sargan statistic p-value 0.84
t-statistics in parentheses. *, **, and *** indicate statistical
significance at 10%, 5%n and 1%, respectively.
Table 7. Response of Tax Rates to Output Business Cycle
(1965-2003)--Ordinary Least Squares
[gdp.sup.c.sub.i.t-1]
Degree of cyclicality ([beta]) 5.76
(0.32)
[gdp.sup.c.sub.i.t-2]
Degree of cyclicality ([beta]) 5.61
(0.31)
[gdp.sup.c.sub.i.t-3]
Degree of cyclicality ([beta]) 10.559
(0.52)
t-statistics in parentheses. *, **, and *** indicate statistical
significance at 10%, 5% and 1%, respectively.
Table 8. Response of Different Tax Rates to Output Business Cycle
(1965-2003)
I II
Degree of cyclicality ([beta]) -6.16 -6.01
(-0.39) (-0.35)
Openness 8.14 --
(19.36) *** --
Openness*Cyclical output -15.83 --
(-0.60) --
Population -- -1.67
-- (-9.37) ***
Population*Cyclical output -- 0.22
-- (0.02)
Gini -- --
-- --
Gini*Cyclical output -- --
-- --
Left government -- --
-- --
Left gov.*Cyclical output -- --
Number of observations 949 975
[R.sup.2] 0.28 0.08
III IV
Degree of cyclicality ([beta]) -1.02 -28.07
(-0.05) (-1.00)
Openness -- --
-- --
Openness*Cyclical output -- --
-- --
Population -- --
-- --
Population*Cyclical output -- --
-- --
Gini 31.56 --
(5.71) *** --
Gini*Cyclical output -97.29 --
(-0.27) --
Left government -- -0.167
(-0.24)
Left gov.*Cyclical output -- 86.21
(1.80) *
Number of observations 835 634
[R.sup.2] 0.04 0.01
t-statistics in parentheses. *, **, and *** indicate statistical
significance at 1O%, 5% and 1%, respectively.
Table 9. Response of Different Tax Rates to Output Business Cycle
(1965-2003)
OLS Country FE Country and Time FE IV
Income tax 5.63 5.22 1.75 0.97
(0.51) (1.13) (0.44) (0.03)
Social security -6.14 -7.09 -6.92 -24.48
(-0.41) (-1.66) * (-2.32) ** (-0.81)
Payroll -1.40 -1.09 -0.39 2.98
(-0.58) (-1.19) (-0.36) (0.29)
Property -0.65 0.10 0.42 -1.05
(-0.08) 0.09 0.36 (-0.14)
Goods and -3.69 -3.81 -2.79 -1.48
services (-0.55) (-1.36) (-0.98) (-0.06)
Others 0.69 2.30 2.72 1.77
(0.35) (2.11) ** (2.23) ** (0.24)
t-statistics in parentheses. *, **, and *** indicate statistical
significance at 10%, 5%, and 1%, respectively.
