Content uploaded by Santo Milasi
Author content
All content in this area was uploaded by Santo Milasi on Oct 24, 2017
Content may be subject to copyright.
Top marginal taxation and economic growth
Santo Milasia * , Robert J. Waldmannb
a International Labour Organization, Research Department, 4, route des Morillons, 1211 Genève 22, Switzerland.
b University of Rome “Tor Vergata”, Faculty of Economics, Department of Economics, Law and Institutions, via
Columbia 2, 00133, Roma, Italy.
Email addresses: santo.milasi@gmail.com (S.Milasi), robert.waldmann@gmail.com (R.Waldmann).
Abstract
This article explores the relationship between top marginal tax rates on personal income and economic growth. Using a
new dataset for consistently measured top marginal tax rates for a panel of 18 OECD countries over the period 1965-
2009, this paper finds evidence in favor of a quadratic top tax-growth relationship. This represents the first reported
evidence of a non-monotonic significant relationship between top marginal income tax rates and economic growth. The
point estimates of the regressions suggest that the marginal effect of higher top tax rates becomes negative above a growth
maximizing tax rate on the order of 60 percent. As top marginal tax rates observed after 1980 are below the estimated
growth maximizing level in most of the countries considered, a positive linear relationship between top marginal tax rates
and GDP growth is found over the sub-period 1980-2009. Overall, results show that raising top marginal tax rates which
are below their growth maximizing has the largest positive impact on growth when the related additional revenues are
used to finance higher productive public expenditure, reduce budget deficits or lower some form of distortionary taxation.
Keywords: top marginal taxation, economic growth, nonlinearity, growth-maximizing tax rate
JEL: E62; H20; O40
* This is the unrefereed Working Paper version of the article published on Applied Economics (23, October 2017) and
available online at: http://www.tandfonline.com/doi/full/10.1080/00036846.2017.1392001. The content of this paper
does not reflect the official opinion of the International Labour Organization. Responsibility for the information and
views expressed in the paper lies entirely with the author.
1. Introduction
Taxation on top incomes and its relationship with economic growth have been at the center
of the recent policy and academic debate. After decades of decreasing taxation on top incomes,
some OECD countries, such as Greece, Iceland, Ireland, Portugal and Spain, have recently
increased top marginal tax rates on personal income as a part of broader fiscal consolidations plans.
However, tax rates on top incomes have remained rather stable in the majority of OECD countries
despite the frequent need to increase tax receipts and respond to growing public demand for greater
tax progressivity. Today, top marginal tax rates in major OECD countries cluster around 40% with
no country exceeding 60%, whereas they averaged well above 60% during the 1960s and 1970s.
The reasons for keeping tax rates on top incomes at relatively low levels are typically grounded in
traditional “supply-side economics” arguments according to which higher top tax rates may give
rise to a series of behavioral responses among top incomes which may ultimately hamper economic
growth (Lindsey 1987; Feldstein, 1995). In particular, to the extent that top earners are highly
responsive to tax change and are responsible for a large fraction of aggregate savings, higher top
marginal tax rates may particularly distort investment decisions, discourage entrepreneurial activity,
reduce work efforts and ultimately undermine economic growth (Lee and Gordon, 2005). At the
same time, many governments refrain from raising top tax rates because of the growing concern that
top earners may migrate abroad (Simula and Trannoy, 2010; Akcigit et al. 2015) or shift their
income to capital gains in order to minimize the tax burden so leading to lower overall tax receipts
(Goolsbee, 2000). On the other hand, a number of recent works suggest that top marginal tax rates
observed in most of OECD countries during the last three decades have been below their “growth-
maximizing level” and have uniquely benefitted top incomes without generating positive spillovers
on aggregate growth (Piketty at al., 2014). To the extent this is true, higher tax rates on top incomes
could therefore spur growth by providing additional tax revenues to finance higher growth-
enhancing public expenditure (e.g. education, R&D, infrastructure). Besides the positive effect
through the financing of “productive” public expenditure, a number of studies suggest that increases
in top tax rates may be indirectly beneficial for economic growth through various other channels.
For instance, to the extent that lower middle-income recipients have a higher propensity to consume
than top earners (Dynan et al., 2004), tax increases on top incomes to finance tax cuts on lower
middle income groups may generate higher output (Zidar, 2014). Similarly, as budget deficits are
found to negatively affect growth (Bleaney et al. 2001), rising suboptimal top tax rates to finance
fiscal consolidation may foster economic growth. In addition, to the extent that high levels of
income inequality are detrimental for economic growth (Ostry et al., 2014), increases in top
marginal tax rates may positively affect economic growth by reducing the concentration of income
at the top.
This work aims at contributing to the ongoing debate. Focusing on 18 OECD countries over
the period 1965-2009, the paper explores the long-run relationship between top marginal taxation
on personal income and economic growth. With respect to previous studies, a series of
improvements are introduced by this paper. First, in contrast with most of past contributions that
assume a linear top tax-growth association, this work explores the case for non-linearity as
suggested by traditional model of endogenous growth. Second, thanks to a newer dataset by Piketty
at al. (2014), longer top marginal tax rates series, starting from 1965, are used. Including data
before 1980 allows consideration of an important part of the variation in top tax rates which was
outside the scope of most of existing works. Third, top tax rates series used in this paper
consistently combine top income tax rate at the central and sub-central level of government. The
paper shows that neglecting top tax rates at the local level, as in the majority of previous studies,
may have important consequences on the estimated results.
This paper finds evidence in support of a significant inverted U-shaped relationship between
top marginal tax rates and economic growth over the period 1965-2009. In line with the theoretical
predictions of endogenous growth models, these results suggest that, for low top tax rates, an
increase in top income taxation used to finance either higher growth-enhancing public expenditure
or lower other types of taxation may be beneficial to growth. However, as the top tax rate reaches
its growth-maximizing level, the distortionary effect of taxation prevails to the point that further
increases in the top marginal tax rate negatively affect growth. The estimated growth maximizing
top tax rate implied by the regression results is always higher than 60%. As since the early 1980s
most of the OECD countries show top tax rates persistently lower than the estimated growth
maximizing level, we explore the top tax-growth relationship on a subsample of observations
between 1980 and 2009. As expected, the coefficient on the quadratic term is no longer statistically
significant and a roughly linear and positive top tax-growth relationship emerges over this period.
As such, these results support the argument advanced in number of recent studies suggesting that
top marginal tax rates observed in the last three decades may be below their growth-maximizing
level. Moreover, following Kneller et al. (1999), the exploration of different specification of the
government budget constraint show that the positive impact of higher top tax rates on growth is
stronger when related additional revenues are implicitly used to finance either higher productive
expenditure (e.g. for education and infrastructure), budget deficits reductions or lower certain
taxation (e.g. social security contributions, corporate taxation and taxation on good and services).
The remainder of the paper is organized as follows. Section 2 presents a review of the
economic literature strictly related to the top tax rates-growth nexus. Section 3 describes data on top
marginal tax rates providing a detailed comparison between those used in this work and the ones
used by previous studies. Section 4 presents a first visual inspection of the nonlinear top tax-growth
link. Section 5 introduces the growth model and the estimation techniques used. Section 6 presents
and discusses the results. Section 7 concludes.
2. Literature review
While the discussion on the efficiency costs associated with top income taxation remains a
living matter in the empirical micro-level literature (Saez et al., 2012), very few studies have
explored the growth-effect of top income taxation at the macro-level. Existent contributions
typically share the same theoretical background based on models of endogenous growth, thereby
increases in distortionary taxation, such as the one on top incomes, can either enhance or hamper
economic growth depending on the initial level of taxation as well as on how tax revenues are spent
(Barro 1990; Barro and Sala-i-Martin, 1992). Despite endogenous growth model explicitly allow for
a non-linear effects of taxation on economic growth, existing empirical studies typically assume a
linear top tax-growth relationship, leading to quite mixed evidence.
To the best of our knowledge, Lee and Gordon (2005) is the first paper to take into account
top marginal tax rate on personal income in growth regressions based on endogenous growth model.
The authors argue that, to the extent that top marginal tax rates may well approximate tax rates
imposed on potential entrepreneurs, high top rates may discourage growth by reducing risk-taking
and entrepreneurial activity and ultimately growth.
1
Yet, in a sample of 70 mostly developing
countries over the period 1970-1997, the authors find that top marginal tax rates on personal income
do not have a significant impact on economic growth. Angelopoulos et al. (2007) find similar
results in a panel of 23 OECD countries over the period 1970-2000. Gemmell et al. (2011b), using a
Pooled Mean Group estimator on an annual panel of 15 OECD countries starting from the late
1970s, are the first to report a negative and significant impact of top marginal tax rates on economic
growth in the long-run. In particular, the author find that such negative effect on aggregate growth
is mainly due to the effect of top marginal tax rates on multi-factor productivity. Yet, the magnitude
of the estimated effect is quite small: a one percentage point fall in top tax rates increases growth by
around 0.05 percentage point. On the other hand, a number of recent studies suggest that a positive
relationship between taxation on top incomes and growth may be observed in developed countries
since 1980 when top marginal tax rates potentially are below their growth-maximizing level. For
instance, Rothschild and Scheuer (2011) and Piketty et al. (2014) point out that wages for high
1
This prediction is supported by Arnold (2011) which shows that top marginal personal income tax rates have a more
negative effect on TFP in industries characterized by high firm entry rates. Instead, Kneller and McGovan (2012) find
that increases in the marginal tax rate imposed on low income levels reduce the level of entrepreneurial activity, while
higher personal income marginal tax rate imposed on top incomes are associated with an increase in firms’ entry rate and
a decrease in the exit rates.
income earners may not be equal to their marginal product as they may partially be the result of
unproductive rent-seeking economic activity. As such, higher top tax rates may lead to higher
economic growth by discouraging rent-seeking behavior and simultaneously increasing public
revenues to finance growth-enhancing public expenditure (e.g. education, public R&D,
infrastructure). Piketty et al. (2014), using annual data for 18 OECD countries over the period 1960-
2010, partially confirm this argument by finding that top tax rates on personal income are positively
associated with economic growth, though this relationship is not always robust. Consistently, Zidar
(2014), focusing on U.S tax reforms in the last thirty years, finds that a one percent of GDP tax cut
for the bottom 90% leads to a 2.7% of GDP growth over a two year period, while the corresponding
tax cut applied to the top 10% results in a statistically insignificant 0.13% growth effect. Moreover,
the author shows that cuts to taxation imposed on the bottom 90% also have a larger effect on the
growth of employment, consumption, and investment than the ones imposed on the top 10%.
Finally, Dackehag and Hansson (2012) is the only study that investigate a non-linear relationship
between top marginal tax rates on personal income and economic growth.
2
Using a set of 25 OECD
countries over the period 1975-2010, the authors find limited evidence of a quadratic relationship
between top marginal tax rate and growth.
As it will be explained in greater detail, this work differs in various aspects from the studies
reviewed above. First, this paper estimates both a linear and a nonlinear relationship over a longer
time series starting from 1965. This allows consideration of a large part of variation in top tax rates
which was outside the scope of previous studies. Second, this work uses a different data source for
top marginal tax rates which, unlike most of existing studies, combine central and sub-central
government top income tax rates consistently measured both across and within-countries. Third, this
2
Bania et al. (2007) also analyze nonlinearities in relationship between income taxation and growth in the United
States. The authors find a significant quadratic relationship between average taxation and economic growth at state
level.
paper estimates the growth-maximizing top tax rate derived by the nonlinear regression model and
provides an analysis of the related implications.
3. Data on top marginal tax rates: some important remarks
This work uses a new data set on personal income top marginal tax rates recently compiled by
Piketty, Saez and Stancheva (2014) (PSS). The data series cover 18 OECD countries over the
period 1960-2010. The main sources are the OECD annual “Taxing Wages” publications for the
period 1984-2010 and the OECD (1986) “Personal income tax system for the period 1975-1983”
publication. Additional sources have been used by the authors to collect data for specific countries
and to extend top tax rates series back to 1960.
3
According to the OECD definition the top marginal
tax rate is “the combined central government and sub-central government marginal personal income
tax rate at the earnings threshold where the top statutory personal income tax rate first applies. It is
calculated as the additional central and sub-central government personal income tax resulting from
a unit increase in gross wage earnings”. The earning threshold where the top tax rate first applies is
on average around 4 times the average income, though it may significantly vary across countries.
Yet, this should not bias our analysis insofar FE estimations are used and the within-country top
tax-growth relationship is analyzed.
It is worth noting that all the empirical studies reviewed in Section 2 use data obtained from the
World Tax Database (WTD) from the Office of Tax Policy Research at the University of Michigan.
The WTD usually provides top tax rates series for developed countries starting from 1975.
Data on top tax rates used in this paper have several advantages with respect to ones from the
WTD used by previous works. First, tax series start in 1960; this allows taking into account a large
part of variation in top tax rates not considered by previous studies. Second, the top tax rates series
used in this work consistently encompass both central and local level of taxation for all countries in
the sample. By contrast, the underlying level of government for top tax rate series in the WTD is
3
See Piketty, Saez and Stantcheva (2014) for a detailed description of data sources.
often inconsistently measured both across countries and within countries over time. In some cases
top tax rates from WTD refer either only to the central or only to the local level of government,
while in others they include both national and local taxation.
Whenever income taxation at the local level is small or non-existent, differences between the two
datasets are negligible. Nonetheless, local income taxation is actually considerable in many
countries, and discrepancies between the two datasets are quantitatively important.
Focusing on a set of common observations for 18 OECD countries over the period 1975-2000,
simple summary statistics for top marginal tax rates series provided in the two datasets are first
analyzed. Table 1 shows that both the mean, the minimum and the maximum value differ across the
two dataset. The simple correlation is equal to 0.76, so the two variables are far from being
perfectly correlated.
[Table 1 around here]
The differences in summary statistics shown in Table 1 may seem moderate, but, when all
country-year data points are analyzed in greater detail, differences in reported top tax rates between
PSS and WTD are indeed substantial. As shown in Table 2, for some countries, like Canada,
Denmark, Switzerland, Norway, Sweden, top tax rates from PSS are more than 20 percentage points
higher than the ones reported by the WTD for many years of the sample period. Differences are
somehow smaller for other countries, such as France and Portugal, but still quantitatively relevant.
As already noted, the explanation behind such huge discrepancies between the series reported in the
two dataset is that data in PSS, unlike WTD, also consider top tax rates applying at the local level of
government.
Although top tax rates data from the WTD have been probably used by previous papers because
of lack of alternative data sources, none of the studies discusses the implications of using such data.
In fact, neglecting top tax rates at the local level may have serious implications for the estimation of
the effect of top income taxation on growth. First, when estimating the relationship between
economic growth and top tax rates, there are no theoretical reasons to prefer top marginal tax rates
which only refer to the central level of government. The consolidated level of top taxation, central
plus local government taxation, is indeed expected to be the one influencing top incomes’ decisions
and in turn economic growth. Second, for any given growth rate, when lower-than-actual top tax
rates from WTD are assigned to certain countries, like the Nordic ones, which instead have overall
top marginal tax rates well above OECD countries’ average, the estimated effect of top marginal tax
rates on growth may be significantly biased downward.
Moreover, top tax rates series in the WTD are inconsistently defined, not only across country,
but also within countries over time. As Table 2 shows, implausible low top tax rates for a certain
country are often observed only during a certain time interval, suggesting that before and after that
interval top tax rates from WTD are equal to the ones from PSS.
4
Since most of the studies on the
top tax rates-growth nexus look at the within country relationship, such within country
inconsistency in top tax rates definitions further exacerbates the problems of using the WTD in the
context under consideration.
[Table 2 around here]
In sum, what is shown above casts some doubts on the robustness of the findings in Lee and
Gordon (2005), Angelopoulos et al. (2007), Gemmel et al. (2011b) and Dackehag and Hansson
(2012) which all use data from World Tax Database. Indeed, both the sign and the statistical
significance of their results could be partly driven by the use of inconsistently defined data top tax
rates. While showing how the top tax rate-growth relationship changes when PSS is used in place of
WTD is outside the scope of this work, using PSS top tax rate series, may provide more accurate
insights on the top tax rate-growth nexus.
4
For example, while large differences between the two dataset are observed in data for Denmark before 1987, only
small discrepancies are reported for this country after 1993.
4. Top marginal taxation and growth: preliminary evidence
Before turning to a more elaborate empirical analysis, a visual inspection of the top tax-growth
nexus is presented. Such type of analysis is useful to investigate the existence of nonlinearities in
the relationship between top marginal taxation and growth. Using data for 18 OECD countries over
the whole data period available (1960-2009), Panel A of Figure 1 shows the relationship between 5-
year average annual growth rates (Growth) and 5-year average top marginal tax rates (TMTR).
Along with the linear regression line (dotted line) a quadratic regression line (dashed line) and a
nonparametric Lowess fit are plotted (solid line).
5
The Lowess procedure estimates a local weighted
bivariate regression for each data point in the scatterplot.
6
While a quadratic model restricts the
relationship between top tax rates and growth to a (possibly inverted) U shape, the nonparametric
Lowess fit allows the data to suggest the appropriate model specification
7
. Panel A of Figure 1
shows that the fitted curve from the Lowess regression closely tracks the quadratic regression curve.
This result suggests that imposing a quadratic fit does not force the top tax-growth relationship to a
particular functional form far from the true relationship in the data. Therefore, prime evidence
suggests that a quadratic model is well suited to detect the existence of a nonlinear relationship
between top tax rates and growth.
[Fig. 1 around here]
Moreover, Panel A of Figure 1 suggests that a quadratic model, of the type Growthit=α +
β1TMTRit+ β2 (0.01*TMTRsqit) + εit, fits the data better than a linear regression model. Note that
TMTRsq is rescaled by multiplying it by 0.01. Such transformation is useful to provide an easier
5
Lowess stands for locally weighted scatterplot smoother (Cleveland, 1979)
6
For an extensive discussion on nonparametric fitting procedures see, among others, Dinardo and Tobias (2001) and
Jacoby (2000)
7
The Lowess fit presented in Panel A of Figure 1 is obtained by using “lowess yvar xvar” STATA command. The
bandwidth value chosen is equal to 0.8. Using smaller values (i.e 0.7 and 0.6) returns a functional form similar to the
one presented in Panel A of Figure 1. Standard tricube weights are used.
representation of the associated coefficient and standard errors. Simple OLS estimation suggests
that both β1 and β2 are statistically significant; the coefficient on the linear term is equal to 0.126,
while the one on the rescaled squared term is -0.082.
8
As a result, the growth maximizing tax rate
implied by a simple quadratic specification is roughly 76% (dashed vertical line in Panel A).
9
In
sum, the quadratic regression line suggests that increases in top tax rates are positively associated
with economic growth but after a certain threshold (around 76%) further increases in top tax rates
become detrimental for economic growth.
Nevertheless, as noted by Haque, Pesaran and Sharma (1999) nonlinearities in cross-country
studies may emerge because of slope heterogeneity and not because a genuine nonlinear
relationship exists. In other words, if the relationship between top marginal tax rate and growth is
positive in some countries and negative in others, an artificial nonlinear relationship may emerge
from estimating a pooled model. In order to check for this possibility, in Panel B of Figure 1,
country-specific quadratic regression line are plotted along with the quadratic fit for the pooled
model (dashed line - shaded area is its 95% confidence interval). The graph supports the idea that
the nonlinear relationship between top tax rates and economic growth shown for the entire sample is
not an artifact arising from pooling countries. Indeed, although there is some level of heterogeneity,
most of the countries show an inverted-U shape relationship similar to the one obtained for the
entire sample. This is consistent with the idea that 18 quite homogeneous OECD countries, which
experienced comparable pattern in top tax rates in last forty years, may show similar top tax-growth
relationship over time. However, the growth-maximizing top tax rate for some of the countries in
our sample is lower than the one obtained for the pooled model (dashed vertical line in Panel B).
This is probably due to the presence of some countries which, showing a nearly linear relationship
8
β1 and β2 are statistically significant even after netting out country-specific effects.
9
The growth-maximizing tax rate (τ*) is found after differentiating Growth with respect to TMTR, and setting the result
to zero (i.e. dGrowth/dTMTR=0). Therefore, τ*= [-β1/(2β2*0.01)]
between top tax rates and growth, increase the estimated growth-maximizing top tax rate for the
pooled model.
Although the above visual inspection of the association between top tax rates and economic
growth provides some suggestive evidence, results reported above may be due to omitted variable
bias. Therefore, a more elaborated econometric analysis is necessary in order to deepen the
understanding of the top-tax growth relationship.
5. Model and methodology
The empirical framework of this research draws on models typically used in aggregate country-
level analysis of the effect of taxation on growth. The model is estimated for 18 OECD countries
and nine non-overlapping 5-year periods over the time span 1965-2009. A shorter time series going
from 1980 to 2009 is also considered to allow a better comparison between this paper’s results and
those of previous studies. The specification of the growth regression is the following:
Growthit=α + β1TMTRit+ β2 (0.01*TMTRsqit) + γXit + δZit +μi + Өt + εit (1)
The dependent variable, Growth, is the 5-year average annual growth rate of real GDP per
capita. In line with other studies, all the independent variables are averaged over the 5-year growth
period (see Kneller et al. 1999; Fölster and Henrekson 2001; Angelopoulos et al. 2007). μi are the
country-specific fixed effects, Өt is a set of time period dummies and εit is the unobservable error
term. Including country fixed effects helps to control for country-specific growth-related
institutional, political and social factors which are relatively stable over time. Period dummies are
included to control for the effect of macroeconomic shocks which influence the growth rate and are
common to all countries.
TMTR is the 5-year average personal income top marginal tax rates. Following the preliminary
analysis presented in Section 5, its squared term (TMTRsq ) is also included to control for the
existence of a quadratic top tax-growth relationship. As previously highlighted, TMTRsq has been
rescaled by multiplying it by 0.01.
10
Rescaling helps the reader to observe the magnitude of the
coefficient and of the standard errors related to the quadratic term in the tables hereunder presented.
Indeed, if TMTRsq were not rescaled, the associated coefficients and standard errors would be equal
to too small numbers (e.g. 0.0001) to be efficiently reported and appropriately informative.
Using top marginal tax rates has advantages. First, as noted by Myles (2007) and Gemmel et al.
(2011b), having a marginal dimension, they may well capture distortions to individual’s decisions.
Second, unlike other measure of taxation, which are often expressed as public revenues as
percentage of GDP, top marginal tax rates, being calculated on the base of statutory top tax rate, are
direct discretionary policy instruments and hence less subject to endogeneity issues induced by
business-cycle co-variation. However, there could still be a problem of reverse causality over the 5-
year period. Indeed, countries could cut top tax rates when there is or is expected to be a slowdown
in GDP growth. Nonetheless, as shown by Galli (2002) and Milasi (2016) for a number of OECD
over the last few decades, GDP growth is not a statistically significant determinant of either levels
or changes in top marginal tax rates on personal income. Therefore, despite the fact that some
countries may have cut top tax rates as a response to slow growth remains possible, there is little
evidence that reverse causality could be an important driver of this paper’s results. However, top
marginal tax rates at the beginning of the 5-year period are also considered to rule out reverse
causality issues in this paper’s result.
Xit is a set of standard growth determinants; including: the natural logarithm of GDP per capita
measured at the beginning of the 5-year growth period (InitGDP) to capture the process of
countries’ convergence in income levels; the gross investments share in GDP (Inv) and the average
years of schooling (School) as a proxy for the level of physical and human capital investment
respectively; the growth rate of population (∆Pop); and the level of inflation (Infl). Moreover, we
10
In the remainder of this paper, whenever the text refers to TMTRsqit , it is implicitly referring to the rescaled squared
top marginal tax rate. In order to ease the written exposition, the rescaling factor (i.e. 0.01) is omitted from the text.
control for total tax revenues as a percentage of GDP (Total tax rev) since the level of top marginal
tax rates is likely to be associated with the overall tax burden.
Following Kneller et al (1999), Zit is set of traditional fiscal policy measures commonly included
simultaneously in recent growth regression models.
11
Kneller et al. (1999) indeed note that the
estimated effect on growth of a specific tax instruments may be biased when only one side of the
government budget is considered. According to the authors, this is because the government budget
constraint is a “closed system”: changes in one budget component must correspond to equal
changes in at least one other component. As such, according to which components of the
government budget are included in, or excluded from, the regression model, the magnitude and the
significance of the coefficients on TMTR and TMTRsq may change and will reflect the effect on
growth of a unitary change in top marginal tax rates financed by a change in the omitted budget
components.
In order to handle the implications of the government budget constraint correctly and allow the
effect of top marginal taxation on growth to be isolated and properly interpreted, we control for
several dimensions of both the revenue and the expenditure sides of the government budget. These
include: the average personal income tax rate (Average inc. tax), social security contributions (SSC),
corporate taxation (Corp tax), taxation on property (Property tax) and taxes on good and services
(Good&Services). All these variables are defined as a percentage of GDP. In addition, we also
control for the expenditure side of the government budget. Unfortunately, because of the lack of
detailed data on government spending back to 1965, the level of general government consumption
as a share of GDP
12
(GovCons) is the only component of the expenditure side of the public budget
considered in regressions over the period 1965-2009. Yet, GovCons include most of the portion of
government spending which is considered as “unproductive” (Kneller et al. 1999), so still allowing
11
See inter alia Bleaney at al. (2001); Arnold (2008); Benos (2009); Gemmel at al. (2008); Muinelo-Gallo and Roca-
Sagales (2011); Gemmel et al. (2011b).
us to explore the growth effect of top marginal tax rates when “productive” expenditure is among
the implicit financing sources. Moreover, thanks to the availability of additional data, a more
complete specification of the government budget is provided for estimations over the period 1980-
2009. Over such period total government spending is disaggregated into “productive”,
“unproductive” and “other expenditures”. A measure for the budget deficit as a percentage of GDP
is also included in regressions over this period. Finally, we also control for the level income tax
progressivity (Progressivity) since top marginal tax rates may simply reflect the degree of
progressivity of the tax systems. Detailed definitions and sources for all the variables described
above are provided in Table A1 of the Appendix.
Finally, equation (1) is estimated using Fixed Effect (FE). Empirical studies on the tax-growth
nexus usually relies on this methodology. The advantage of using this technique is that it controls
for persistent country characteristic which may be correlated both with the growth rate and the level
of top tax rates (e.g. persistent social factors, institutional setting, etc…). As in other studies, a
Hausman test suggests the fixed effect model is the appropriate one to estimate equation (1).
However, fixed effect models only rely on within-country variation and, enhancing the medium-
term co-variation in the data, are more likely to capture business cycle-related correlation among
the variables. Thus, Pooled OLS estimation is also considered to check how results change when
cross-country variation in the data is taken into account.
6. Results
In Table 3 a quadratic relationship between top tax rates and growth over the period 1965-2009
is explored.
13
Results in the first column show that the quadratic relationship between top tax rates
13
The case of a linear relationship over the 1965-2009 is also tested. Excluding TMTRsqit and estimating the same models
presented in columns 1-9 of Table 3, there is not statistically significant evidence of a linear relationship between top tax
rates and economic growth. Moreover, although data series for top tax rates are available since 1960, the econometric
analysis starts in 1965 as most of the other tax variables used as covariates in this paper’s regressions start being available
in this year.
and growth is statistically significant when a set of “core determinants” is included. As expected,
the coefficient associated with the linear term is positive while the quadratic term is negative,
suggesting a hump-shaped relationship between top marginal tax rates and GDP growth. The
bottom part of Table 3 reports the growth-maximizing top tax rate (τ*) implied by the quadratic
specification.
14
The estimated growth-maximizing top tax rate presented in column (1) is slightly
above 60%, suggesting that up to this level of top marginal taxation higher tax rates to finance an
increase in public expenditure or a decrease in budget deficits (see the implicit financing sources)
have a positive effect on growth. However, above that threshold, further increases are detrimental to
GDP growth.
Regarding the control variables, the coefficient on the initial GDP is always significant at the one
percent level in every regression of Table 3, confirming the existence of conditional convergence
across countries. Among the other “core variables”, coefficients on population growth and inflation
have the expected sign are they both statistically significant. The coefficient on investment is
surprisingly negative and often statistically significant.
15
Although unexpected, this result is not
unprecedented. Negative, though not statistically significant, coefficients on investment are indeed
found in some regressions by Kneller at al. (1999), Fölster and Henrekson (2001) and Angelopoulos
et al. (2007) using FE estimations over country samples comparable to the one used in this work.
This result is mainly the consequences of the demeaning process in FE effect estimation which
eliminates the cross-country variation in investment levels. Since, in our sample of 18 developed
countries, investment mainly varies across countries, country fixed effects are actually capturing a
large extent of the growth-relevant variation in investment level. Moreover, since this work use 5-
year intervals, the regression results on investments may be affected by the business cycle.
14
Referring to Eq. (1), the growth-maximizing tax rate is found after differentiating Growth with respect to TMTR, and
setting the result to zero (i.e. dGrowth/dTMTR=0). Therefore, τ*=[-β1/(2β2*0.01)]
15
Results on the other explanatory variable do not change markedly when only the linear term, TMTR, is included.
Investment is extremely low during recessions, and, in our sample except for the last two years,
recessions have been followed by recoveries with higher than average growth. Given the above
reasons, it is not surprising that the coefficient on Invit turns out to be negative. The demeaning
process in FE effect estimation is also likely to explain the not statistically significant, and
sometimes negative, coefficients on the average education level (School) which, similarly to the
investment level, mostly varies across OECD countries. Finally, the coefficient on the overall tax
burden (Total tax rev) is negative and statistically significant, confirming the well-known result that
more burdensome overall taxation is associated with slower growth.
16
As the coefficient on the
“core” control variables only slightly change across the various regressions of Table 3, what follows
will mostly focus on discussing how the coefficients on TMTRit and TMTRsqit change when
additional tax- and expenditure-related control variables are added to or excluded from the
estimation model.
First, as the top marginal tax rates might be correlated with the level of income tax progressivity,
the ratio between top marginal income tax rates and the average income tax rate is included in
column (2) as a proxy for the degree of income tax progressivity. As in Lee and Gordon (2005),
which use the same measure of income tax progressivity, the coefficient on Progressivity is
negative but not statistically significant. Moreover, its inclusion does not affect the significance of
the quadratic relationship, so suggesting that top marginal income taxation per se, rather than as a
proxy of progressivity, is relevant to economic growth.
17
Although the implicit financing sources
16
Results in column (1) remain almost the same of those reported when the total tax revenue as a share of GDP is
excluded. Results without Total tax rev. are not shown to save space, they are available upon request.
17
Even when TMTR is excluded from the regression, the coefficient on Progressivity remains not statistically
significant. A quadratic specification on Progressivity does not provide statistically significant results both when TMTR
is included in and excluded from the estimation. Moreover, Progressivity would remain not statistically significant if it
were included in any of the specification of Table 3; therefore starting from column (3) it is excluded in order to
remain unchanged with respect to column (1), the growth maximizing tax rate rises by almost 3
percentage points, probably reflecting the fact that part of the potentially negative growth-effect of
income tax progressivity was previously captured by the coefficients on top marginal tax rates.
In column (3), the overall tax burden is disaggregated into a wider set of tax measures including:
average personal income tax rate (Average inc. tax), social security contributions (SSC), corporate
taxation (Corp tax), taxation on property (Property tax) and taxes on good and services
(Good&Services). Such decomposition allows us to control if either a specific tax instruments or a
certain tax structure was driving the results shown in columns (1) and (2). The inclusion of this
larger set of variables leaves the coefficients on top tax rates basically unchanged, though the
growth-maximizing tax rate marginally falls to around 61%. In line with existing empirical
evidence, social security contributions and property taxes result to be the most harmful for growth,
whereas less distortionary taxations on average income and on good and services do not seem to
have significant impact on growth. The coefficient on the corporate taxation is instead found to be
positive, though rarely statistically significant. Such a positive coefficient may be due to the fact
that international tax competition may have led to sub-optimal corporate tax rates (Angelopoulos et
al., 2007).
18
In column (4), the expenditure side of the government budget constraint is also considered by
including the share of government consumption in GDP, which enters in the regression with a
negative and statistically significant coefficient. Note that the reported τ* is now at 66% and higher
than the ones found in columns 1-3. This may be explained by the inclusion of the government
consumption in the growth model. Indeed, when in columns 1-3, government consumption was
among the implicit financing sources, the coefficients on top tax rates also captured its negative
minimize the risk of collinearity with other tax variables. All these results are not shown to save space, they are
available upon request.
18
A quadratic specification for each of the government budget components has been also tested. Results suggest that this
specification is never significant for any of components; a linear specification is therefore preferred hereafter.
effect on growth, so lowering the estimated growth-maximizing top tax rates. Conversely, once the
negative effect of government consumption on growth is explicitly taken into account in column
(4), the positive effect of top tax rates is better isolated and the estimated growth-maximizing tax
rate rises. As the budget deficit and the portion of public expenditure not encompassed in our
measure of government consumption (i.e. the share of productive expenditure) remain the only
budget components which are not explicitly modeled in column (4), results in this column suggest
that increases in top marginal tax rates up to 66% to finance either higher productive expenditure or
lower budget deficit may be beneficial for growth.
In order to further explore how the magnitude and the significance of the coefficients on TMTRit
and TMTRsq as well as the associated τ* change when different financing assumptions are
considered, one budget component at a time is removed from regressions in columns 5-9. In column
(5), the set of implicit financing sources is therefore broadened by first omitting the average income
tax rate. Consistently with the fact that the average income tax rate did not result as an important
driver of growth (see columns 3 and 4), the significance and the magnitude of the coefficients on
TMTR and TMTRsq are not affected by its exclusion, though τ* falls to 63%. Similarly the
significance of the coefficients on our variables of interest does not change when the corporate tax
rate is omitted in column (6). In this case τ* rises again to above 68%, so reflecting the fact that
rising top tax rates which are below this threshold to finance lower corporate tax rate may be
beneficial to growth. Conversely, when property taxes are excluded from the model in column (7),
the negative coefficient on TMTRsqit is no more statistically significant, whereas the positive one on
the linear term remains significant at the 10-percent level. This is explained by the fact that property
taxes are found to be detrimental to growth (columns 3-6), so the positive effect of higher top
taxation prevails when related revenues are used to lower property taxes, giving rise to an almost
linear positive relationship between growth and top tax rate. The coefficients on TMTRit and
TMTRsqit are instead both strongly significant again when social security contributions and taxation
on good and services are omitted from the regressions in columns (8) and (9) respectively. In both
columns τ* is estimated to be at around 65%.
Note that the significance of the quadratic top tax-growth relationship in each of the
specifications in Table 3 is robust to the use of POLS in place of FE. Moreover, results are robust to
the use of top marginal tax rates at the beginning of the 5-year period in place of the averaged ones,
so supporting the argument that our estimates are not driven by reverse causality over the 5-year
period.
19
[Table 3 around here]
In sum, the quadratic relationship between top marginal tax rates on personal income and
growth is strongly significant across the different specifications in Table 3. The implied growth-
maximizing top tax rate is consistently above 60 per cent, ranging from 60.6% per cent in column
(1) to 68.0% per cent in column (6). Support to these results is provided by a number of recent
contributions estimating that the revenue-maximizing top tax rate in major OECD countries should
be around 70 per cent or higher.
20
For instance, IMF (2013) shows that the revenue-maximizing top
marginal tax rate for a number of developed countries range between 60% and 70%. A number of
other contributions find even higher revenue-maximizing top marginal tax rates. Romer and Romer
(2012), focusing on the interwar period in the United States, find that tax revenue would indeed be
maximized with a top tax rate around 84%. Bach et al. (2012) estimate that the revenue-maximizing
top tax rate in Germany should be around 75%. Finally, Piketty at al. (2014) show that the socially
optimal top tax rate would be roughly 83%.
Since many OECD countries currently show top marginal tax rates between 40 and 50 per cent,
Table 4 presents the marginal effect on growth of increasing top tax rates when they are whether at
19
Results are not shown to save space. They are available upon request.
20
To the extent that the growth-maximizing top tax rate found in this work is correct, the revenue-maximizing top tax
rate is expected to be higher than 60 percent. This is because the peak of the Laffer curve with proper dynamic scoring
would be at a higher top tax rate if the quadratic specification found in this paper is valid.
the 40% or at the 50% level. In order to interpret the coefficient on TMTRit as the marginal effect on
growth when top tax rates are at the X% level, a minor transformation on the squared term is
necessary (i.e. [0.01*(TMTRit – X)2]). Such transformation does not change the results previously
discussed; indeed all the coefficients remain exactly as shown in Table 3.
21
In order to save space
they are not reported. The only thing that changes is the magnitude and interpretation of the
coefficient on TMTRit which now represents the marginal effect of increasing top tax rates when
they are at the X% level. Table 4 shows that increasing top tax rates when they are at the 40%, has a
positive and significant marginal effect on growth. Consistently with theory, the positive marginal
effect of increasing top tax rates is decreasing in its level. Indeed, coefficients on TMTRit are lower
in magnitude and less significant when the marginal effect is evaluated at the 50% level of top
taxation.
[Table 4 around here]
6.1. Top tax rates and growth after 1980
The analysis performed in Table 3 is replicated in Table 5 over a shorter sample period starting
in 1980. First, this allows investigating whether the top tax-growth relationship changes when a
“low top tax rate” period is considered. Moreover, focusing on the post-1980 period allows to better
compare this paper’s results with the ones in previous studies, which, as already discussed, mostly
focus on sample starting around 1980.
22
The inverted U-shaped relationship between top tax rate and growth found over the period 1965-
2009 is not apparent in estimations after 1980 for which he coefficients on both TMTRit and
TMTRsqit are both individually and jointly statistically insignificant. This result is not surprising
insofar the growth-maximizing top tax rate of roughly 60% reported in Table 3 is correct. Indeed,
21
Note that the coefficients on the squared term are indeed equal to the ones reported in column 4-6 of Table 3.
22
Except Piketty et al (2014), the sample period used by previous studies on the top tax-growth nexus always starts around
1980.
there would be no reason to expect a statistically significant quadratic top tax-growth relationship
over the period 1980-2009 when the average top tax rate for the entire sample was at 51%, with
only three countries (i.e. Denmark, Netherlands and Sweden) showing average top tax rates higher
than 60%. Therefore, a linear top tax-growth relationship is instead investigated over the period
1980-2009. Results in Table 5 show that the linear coefficient on TMTRit is always positive and
statistically significant in most of the cases. This result does not conflict with the inverted U-shape
relationship reported for the 1965-2009 sample period in Table 3; but it further corroborates its
existence. As suggested by models of endogenous growth, the linear positive coefficient on TMTR
after 1980 may be indeed interpreted as the effect on growth of increasing top taxation around the
upward-sloping side of the “growth-hill. Results are indeed consistent with these models’
theoretical predictions: increases in top marginal taxation which are below their growth-maximizing
level are beneficial to growth when the related additional revenues are used to finance either higher
productive expenditure or reductions in budget deficits (columns 4 and 5). This positive impact of
higher top tax rates persist when corporate taxation, property taxes and SSC are added among the
implicit financing sources in column (7), (8) and (9) respectively. The positive coefficient on TMTR
becomes even stronger when taxation on good and services is added to the financing sources in the
last column. This may be consistent with the fact that higher top tax rates to finance reductions in
taxation on good and services may spur consumption, and in turn growth, especially among lower
middle income earners which typically have the highest marginal propensity to consume.
Conversely the coefficient on TMTR looses significance when the average income taxation is added
among the implicit financing sources in column (5). This may be consistent with the idea that top
marginal taxation is more relevant than average income tax rate in affecting individuals’ decisions
and in turn growth.
The coefficients on the other explanatory variables are similar to the ones found in Tables 3. The
only coefficient that changes is the one on property taxes which is no longer statistically significant,
so suggesting that the distorsive effect of higher property taxes may have been limited in the post-
1980 period. Conversely, the negative impact of higher SSC is now larger in magnitude and always
statistically significant.
The positive linear coefficient on TMTRit found in this paper stands in contrast to other existing
studies which, using a similar country-time series sample, find a negative, though only rarely
significant, effect of top marginal tax rates on growth. Such a discrepancy is likely due to the fact
that top marginal tax rates used in this paper also take into account top tax rates at the local level
taxation along the ones at the central level of government. However, differences may be also due to
how the expenditure side of the government budget constraint is specified in the growth regression.
In order to check for this possibility, a wider set of expenditure-related variables is considered in
line with previous papers (e.g. Angelopoulos et al., 2007 and Gemmel et al., 2011b), In particular,
following Kneller et al. (1999), total public outlays are classified into three groups: productive,
unproductive and other expenditure.
23
A measure for the budget deficit is also considered in order to
have a full specification of the government budget constraint. These variables were not included in
estimations since 1965 because of lack of data availability. With respect to Table 5, the advantage
of using a more detailed disaggregation as in Table 6 is that the growth effect of top marginal
taxation when productive expenditure is the only implicit financing source can be discerned from
the case when the financing sources only include unproductive spending. On the downside,
disaggregating government expenditure into productive and unproductive spending reduces the
number of observation because of shorter time series data availability and existence of missing
points in data series.
24
[Table 5 around here]
23
See Table A1 for detailed description of what each expenditure aggregate includes.
24
The number of observations indeed drops to 89. Nonetheless, evidence may be still relevant. In fact, existing studies
on the tax-growth nexus, like Angelopoulos et al. (2007) and Kneller et al. (1999) are often based on less than 100
observations.
Results in Table 6 are in line with those presented in Table 5: over the period 1980-2009, the
impact on growth of increasing top marginal tax rates, which are potentially below their growth-
maximizing level, is always positive regardless of the implied financing assumption considered.
This positive impact is again stronger when potential higher revenues from increased top marginal
taxation are used to either reduce budget deficit or lower taxation on good and services (columns 1
and 8). The positive effect of an increase in TMTR is somewhat less significant when additional
revenues are implicitly used to either increase productive expenditure (column 2) or lower taxation
on corporate income, SSC or property taxes (columns 5-7). Not surprisingly, and in line with
economic theory, the effect of an increase in TMTR is not significant when potentially higher
revenues are used to finance either higher nonproductive expenditure or other types of expenditure
(column 3).
[Table 6 around here]
7. Conclusions
Results presented in Section 6 suggest two main conclusions. First, evidence supports the
theoretical prediction of a non-monotonic relationship between top tax rates and GDP growth.
Indeed, a clear concave top tax-growth relationship emerges from the estimation of a quadratic
model over the period 1965-2009. Second, the point estimates of the regressions suggest that the
marginal effect of top marginal tax rates on growth becomes negative when top tax rates are higher
than roughly 60%. Consistently with this finding, the concave top tax-growth relationship loses
statistical significance over the period 1980-2009, when average top tax rates are on average
substantially below 60%, and a roughly linear positive top tax-growth relationship arises.
Although it is difficult to draw any definitive policy conclusions from these results, this study
adds to a growing amount of evidence suggesting that current top marginal tax rates are below their
growth-maximizing level. In particular, this paper shows that higher top marginal tax rates on
personal income may foster aggregate growth if the additional potential revenues were used to: i)
finance higher growth-enhancing public expenditure (e.g. R&D, infrastructure, education); ii)
reduce persistent budget deficit; iii) lower SSC; iv) decrease taxes on good and services. Moreover,
the fact that the positive relationship between top marginal taxation and economic growth holds
regardless of the implicit financing sources considered is consistent with the arguments in
Rothschild and Scheuer (2011) and Piketty et al. (2014) thereby higher top tax rates could
discourage top incomes’ unproductive rent-seeking economic activity and in turn fostering growth.
Nevertheless, increases in top tax rates may entail a series of negative growth effects which,
given the nature of this paper, cannot be easily taken into account. First, higher top tax rates may
encourage migration of top earners and, as shown by Simula and Trannoy (2010) in their study on
France, this possibility would significantly lower the optimal top tax rate. In this regard, it is
therefore important to understand which the degree of mobility exists among top incomes is. As
shown by Kleven et al. (2010) for European soccer players – a potentially highly mobile income
group – top incomes’ mobility may not be significant.
In addition, the possibility that top incomes facing higher top tax rates may resort to “accounting
manoeuvres” in order to minimize taxation remains a considerable disincentive for governments for
raising taxation at the top. Within this context, measures to broaden the tax base and limit tax
avoidance opportunities may be important prerequisite, if not substitute, for increasing taxation on
top incomes.
Yet, even though higher taxation at the top could have a direct negative effect on economic
growth, it could importantly attenuate the surge in income of income inequality observed in many
OECD countries, especially in the aftermath of the 2008 financial crisis. Besides being a desirable
outcome from a social standpoint, lower income inequality could be also important for economic
growth. Indeed, the extent that high levels of income inequality produce a number of highly
detrimental effects for economic growth (Ostry et al., 2013), increases in top marginal tax rates may
foster equity in income distribution and in turn have a net positive effect economic growth.
References
Akcigit, U., Baslandze, S., Stantcheva, S. 2015. Taxation and the International Mobility of
Inventors (No. w21024). National Bureau of Economic Research.
Angelopoulos, K., Economides, G., Kammas, P., 2007. Tax-spending policies and economic
growth: theoretical predictions and evidence from the OECD. European Journal of Political
Economy 23, 885-902.
Bach, S., Corneo, G., Steiner, V., 2012. Optimal top marginal tax rates under income splitting for
couples. European Economic Review
Bania, N., Gray, J.A., Stone, J.A., 2007. Growth, taxes, and government expenditures: growth hills
for US states. National Tax Journal 60.2:193.
Barro, R., 1990. Government Spending in a Simple Model of Endogenous Growth. Journal of
Political Economy 98 (5 pt 2)
Barro, R., Sala-i-Martin, X., 1992. Public Finance in Models of Economic Growth. Review of
Economic Studies 59, 645-661.
Bleaney, M., Gemmell, N., Kneller, R., 2001. Testing the Endogenous Growth Model: Public
Expenditure, Taxation, and Growth over the Long Run. Canadian Journal of Economics 34, 36-57.
Cleveland, W.S., 1979. Robust locally weighted regression and smoothing scatterplots. Journal of
American Statistical Association74, 829–836.
Dackehag, M., Hansson, Å., 2012. Taxation of income and economic growth: An empirical analysis
of 25 rich OECD countries. Working Paper no. 2012: 6, University of Lund.
DiNardo, J., Tobias, J.L., 2001. Nonparametric density and regression estimation. The Journal of
Economic Perspectives. 15, 11-28.
Dynan, K.E., Skinner, J., Zeldes, S.P.,2004. Do the Rich Save More? Journal of Political Economy
112.
Feldstein, M., 1995.The Effect of Marginal Tax Rates on Taxable Income: A Panel Study of the
1986 Tax Reform Act. Journal of Political Economy, vol. 103, no. 3, pages 551-572.
Fölster, S., Henrekson, M., 2001. Growth effects of government expenditure and taxation in rich
countries. European Economic Review 45, 1501- 1520.
Galli, E., 2002. Tax progressivity and economic and political determinants: an empirical analysis
for the OECD Countries. mimeo, Università di Roma “La Sapienza”.
Gemmell, N., Kneller, R., Sanz, I., 2008. Fiscal policy impacts on growth in the OECD: Are they
long-run? Working Paper, University of Nottingham, 2008.
Gemmell, N., Kneller, R. and Sanz, I., 2011b. The growth effects of corporate and personal income
taxes in the OECD. Globalisation and Development Centre, Bond University, Australia, Working
Paper No. 49.
Goldsbee, A. 2000. What Happens When You Tax the Rich? Evidence from Executive
Compensationl. Journal of Political Economy 108, 352.
Ul Haque, N., Pesaran, M.H., Sharma, S., 1999. Neglected heterogeneity and dynamics in cross-
country savings regressions.
International Monetary Fund. Fiscal Monitor, Taxing Time. October, 2013.
Jacoby, W. G., 2000. Loess: a nonparametric, graphical tool for depicting relationships between
variables. Electoral Studies 19, 577-613.
Jaimovich, N., Rebelo, S., 2012. Non-linear Effects of Taxation on Growth. No. w18473, National
Bureau of Economic Research.
Kleven, H., C. Landais, and E. Saez, 2010, “Taxation and International Migration of Superstars:
Evidence from the European Football Market,” NBER Working Paper No. 16545 (Cambridge,
Massachusetts: National Bureau of Economic Research).
Kneller, R., Bleaney, M., Gemmell., N., 1999. Fiscal Policy and Growth: Evidence from OECD
Countries. Journal of Public Economics 74, 171-90.
Kneller, R., McGowan, D., 2011. Tax Policy and Firm Entry and Exit Dynamics: Evidence from
OECD Countries. Discussion Paper no. 12/01, University of Nottingham. Available at SSRN
1857458.
Lee, Y., Gordon, R., 2005, Tax structure and economic growth, Journal of Public Economics 89,
1027-1043.
Lindsey, L., 1987. Individual taxpayer response to tax cuts: 1982-1984, with implications for the
revenue maximizing tax rate», Journal of Public Economics, vol.33, no. 2, pages 173-206.
Milasi, S., 2016. The determinants of taxation on top incomes: evidence from the last half century.
https://www.researchgate.net/publication/320281106_The_determinants_of_taxation_on_top_inco
mes_evidence_from_the_last_half_century
Myles, G.D., 2007. Economic growth and the role of taxation. Prepared for the OECD. Unpublished
report, University of Essex, IFS
Muinelo-Gallo, L., Roca-Sagales, O., 2011. Economic growth and inequality: the role of fiscal
policies. Australian Economic Papers 50, 74-97.
Organisation for Economic Co-operation and Development, National Accounts Database
Ostry, J., D., Berg, A., Tsangarides C., G. 2014 Redistribution, Inequality, and Growth.
International Monetary Fund, Research Department, April 2014.
Piketty, T., Saez, E., Stantcheva, S., 2014. Optimal Taxation of Top Labor Incomes: A Tale of
Three Elasticities. American Economic Journal: Economic Policy, 6(1): 230-71.
Roine, J., Vlachos, J., Waldenström, D., 2009.The long-run determinants of inequality: What can
we learn from top income data? Journal of Public Economics 93, 974-988.
Romer, C.D., Romer, D.H., 2012. The incentive effects of marginal tax rates: evidence from the
interwar era. No. w17860. National Bureau of Economic Research.
Rothschild, C., Scheuer, F., 2011. Optimal Taxation with Rent-Seeking. No. w17035.
National Bureau of Economic Research, 2011.
Saez, E., Slemrod, J., Giertz, S.H., 2012. The elasticity of taxable income with respect to marginal
tax rates: A critical review. Journal of Economic Literature 50, 3-50.
Simula L., Trannoy, A. 2010. Optimal income tax under the threat of migration by top-income
Earners. Journal of Public Economics, n. 94(1), pp. 163-173.
World Tax Database, Office of Tax Policy Research at the University of Michigan,
http://www.bus.umich.edu/otpr/otpr/default.asp
Zidar, Owen M. Tax Cuts for Whom? Heterogeneous Effects of Income Tax Changes on Growth &
Employment, December 2014
Appendix
[Table A1 here]
Tables:
Table 1: Summary Statistics for top marginal tax rates series from PSS and WTD, 1975-2000
Variable
Unit and Sources
Mean
S.D.
Min.
Max.
Obs.
Corr.
Top Individual Income Tax rate
%, PSS
55.31
12.15
28
88
400
0.76
Top Individual Income Tax rate
%, WTD
51.85
17.06
11.5
91
400
Table 2: differences in reported data between Piketty, Saez and Stantcheva (2011) and the World Tax Database
Country
Year
PSS (%)
WTD (%)
Difference
Country
Year
PSS (%)
WTD (%)
Difference
Canada
1975
61
47
14
Norway
1987
56
34
22
Canada
1984
50
34
16
Norway
1988
48
23
25
Canada
1985
52
34
18
Norway
1989
54
19
35
Canada
1986
55
34
21
Norway
1990
54
17
37
Canada
1987
53
34
19
Norway
1991
50
14
36
Canada
1988
46
29
17
Norway
1992
41
13
28
Canada
1989
47
29
18
Norway
1993
42
14
28
Canada
1990
48
29
19
Norway
1994
42
14
28
Canada
1991
48
29
19
Norway
1995
52
14
38
Canada
1992
49
29
20
Norway
1996
52
14
38
Canada
1993
51
29
22
Norway
1998
52
28
24
Canada
1994
51
29
22
Norway
1999
54
28
26
Canada
1995
52
29
23
Portugal
1988
55
50
5
Canada
1996
52
29
23
Spain
1992
56
53
3
Canada
1997
49
29
20
Spain
1998
56
48
8
Canada
1998
49
29
20
Spain
1999
48
40
8
Canada
1999
48
29
19
Sweden
1981
87
85
2
Denmark
1975
60
40
20
Sweden
1982
88
85
3
Denmark
1984
68
40
28
Sweden
1991
51
20
31
Denmark
1985
68
40
28
Sweden
1992
51
20
31
Denmark
1986
69
40
29
Sweden
1993
51
30
21
Denmark
1987
69
22
47
Sweden
1994
51
30
21
Denmark
1993
70
68
2
Sweden
1995
56
30
26
Denmark
1995
65
64
2
Sweden
1996
56
30
26
Denmark
1998
60
58
2
Sweden
1997
56
30
26
France
1980
75
60
15
Sweden
1998
56
31
25
France
1981
66
60
6
Sweden
1999
56
31
25
France
1982
70
60
10
Switzerland
1979
44
41
3
France
1987
57
53
4
Switzerland
1981
44
42
2
France
1988
57
52
5
Switzerland
1982
44
40
4
France
1989
57
52
5
Switzerland
1984
44
12
33
Norway
1984
64
41
23
Switzerland
1985
44
12
33
Norway
1985
63
40
23
Switzerland
1986
44
12
33
Norway
1986
62
40
23
Switzerland
1987-1994a
44
12
33
Note: countries for which differences in reported top tax series between the two dataset are minimal for the whole sample period (i.e. on the order
of 2/3 percentage point), are not reported in this table. a Differences between top tax rates reported in two datasets for Switzerland over the period
1987-1994 are always equal to 33 percentage point.
Table 3: Top marginal taxation and growth under different financing assumption, 1965-2009
(1)
(2)
(3)
(4)
(5)
(6)
(7)
(8)
(9)
TMTR
0.120*
0.149**
0.135**
0.169**
0.172**
0.181**
0.141*
0.182**
0.168**
(0.059)
(0.063)
(0.053)
(0.069)
(0.068)
(0.066)
(0.081)
(0.068)
(0.066)
0.01*TMTRsq
-0.099**
-0.118**
-0.110**
-0.128**
-0.135**
-0.133**
-0.109
-0.141**
-0.128**
(0.047)
(0.048)
(0.042)
(0.056)
(0.053)
(0.055)
(0.068)
(0.056)
(0.054)
InitGDP
-7.217***
-7.647***
-7.758***
-7.189***
-7.309***
-6.926***
-6.992***
-7.282***
-7.269***
(1.309)
(1.453)
(1.221)
(1.263)
(1.242)
(1.194)
(1.585)
(1.245)
(1.017)
INV
-0.058*
-0.072*
-0.067
-0.119***
-0.127***
-0.126***
-0.137***
-0.117***
-0.116***
(0.028)
(0.036)
(0.043)
(0.037)
(0.035)
(0.037)
(0.044)
(0.034)
(0.040)
∆Pop
0.556
0.575
0.514*
0.574**
0.616**
0.596**
0.617**
0.575**
0.566**
(0.384)
(0.372)
(0.258)
(0.243)
(0.227)
(0.245)
(0.272)
(0.244)
(0.247)
School
-0.113
-0.175
0.148
0.115
0.090
0.101
-0.098
0.106
0.115
(0.155)
(0.165)
(0.177)
(0.186)
(0.181)
(0.196)
(0.138)
(0.185)
(0.188)
Infl
-0.168***
-0.173**
-0.158***
-0.150***
-0.144***
-0.158***
-0.142***
-0.143***
-0.154***
(0.057)
(0.069)
(0.054)
(0.041)
(0.040)
(0.040)
(0.046)
(0.044)
(0.039)
Total Tax rev
-0.145**
-0.175*
(0.064)
(0.088)
Progressivity
-0.028
(0.066)
Average Inc tax
-0.114
-0.055
-0.077
-0.019
-0.035
-0.058
(0.069)
(0.071)
(0.067)
(0.098)
(0.064)
(0.069)
Corp. Tax rate
0.120
0.111
0.132*
0.117
0.129
0.111
(0.098)
(0.084)
(0.074)
(0.081)
(0.081)
(0.085)
Property tax
-0.991***
-0.779***
-0.748***
-0.784***
-0.807***
-0.787***
(0.209)
(0.232)
(0.222)
(0.232)
(0.231)
(0.228)
SSC
-0.170**
-0.076
-0.051
-0.099*
-0.119
-0.079
(0.062)
(0.064)
(0.059)
(0.057)
(0.081)
(0.063)
Goods&service
tax
-0.130
0.025
0.040
0.024
0.075
0.038
(0.155)
(0.138)
(0.138)
(0.127)
(0.150)
(0.133)
Gov.Cons.
-0.279***
-0.301***
-0.282***
-0.365***
-0.303***
-0.269***
(0.073)
(0.067)
(0.068)
(0.088)
(0.068)
(0.083)
Constant
79.938***
85.830***
83.067***
79.841***
80.892***
77.951***
80.792***
80.058***
80.767***
(13.136)
(16.272)
(14.349)
(15.022)
(14.959)
(14.473)
(18.287)
(14.993)
(11.949)
τ*
60.6%
63.1%
61.4%
66.0%
63.7%
68.0%
64.7%
64.5%
65.6%
Implicit
financing
assumption
Total exp.
& Deficit/
Surplus
Total exp.
& Deficit/
Surplus
Total exp.
& Deficit/
Surplus
Productive
exp. &
Deficit/
Surplus
Productive
exp. &
Deficit/
Surplus &
Average
income tax
Productive
exp. &
Deficit/
Surplus &
Corporate
income tax
Productive
exp. &
Deficit/
Surplus &
Property
tax
Productive
exp. &
Deficit/
Surplus &
SSC
Productive
exp. &
Deficit/
Surplus &
Taxes on
goods and
services
Observations
157
153
153
152
152
152
152
152
152
R-squared
0.66
0.66
0.71
0.74
0.74
0.73
0.71
0.74
0.74
Number of
country
18
18
18
18
18
18
18
18
18
Note: The dependent variable is the 5-year average annual growth rate of real GDP per capita. Levels of significance:
*** 1 percent, ** 5 percent, * 10 percent. Clustered robust standard errors are in parentheses. Period dummies are
always included. Referring to Eq. (1), τ*= [-β1/(2β2*0.01)].
Table 4: Marginal effect on growth of increasing top tax rate evaluated at the 40% and 50% top tax rate
X=40%
X=50%
1
2
3
4
5
6
TMTRit
0.066**
0.075**
0.069**
0.040*
0.048**
0.040*
(0.028)
(0.026)
(0.028)
(0.021)
(0.019)
(0.020)
0.01*(TMTRit – X)2
-0.128***
-0.133**
-0.141**
-0.128**
-0.133**
-0.141**
(0.056)
(0.055)
(0.056)
(0.056)
(0.055)
(0.056)
Note: see notes in Table 3. The specification used for this table results is equal to the one reported in
column 4-6 of Table 3. Except for TMTR, all the estimated coefficient are equal to the ones reported
in Table 3.
Table 5: Top marginal taxation and growth under different financing assumption, 1980-2009
(1)
(2)
(3)
(4)
(5)
(6)
(7)
(8)
(9)
TMTR
0.047**
0.054**
0.047*
0.054*
0.046
0.058*
0.056*
0.053*
0.056**
(0.023)
(0.022)
(0.026)
(0.028)
(0.027)
(0.028)
(0.028)
(0.028)
(0.026)
InitGDP
-9.113***
-9.156***
-8.813***
-7.853***
-7.668***
-7.968***
-7.760***
-8.120***
-7.417***
(1.292)
(1.209)
(1.911)
(1.611)
(1.597)
(1.558)
(1.610)
(1.640)
(1.116)
INV
0.001
-0.010
-0.080
-0.123***
-0.137***
-0.114***
-0.126***
-0.102***
-0.134***
(0.042)
(0.045)
(0.049)
(0.036)
(0.035)
(0.038)
(0.035)
(0.034)
(0.037)
∆Pop
0.179
0.246
0.266
0.145
0.164
0.150
0.083
0.190
0.126
(0.319)
(0.349)
(0.417)
(0.322)
(0.348)
(0.312)
(0.307)
(0.324)
(0.312)
School
0.050
0.128
0.057
0.070
0.035
0.065
0.039
0.146
0.073
(0.150)
(0.138)
(0.187)
(0.209)
(0.220)
(0.212)
(0.179)
(0.199)
(0.198)
Infl
-0.298***
-0.238**
-0.196**
-0.186***
-0.178***
-0.196***
-0.185***
-0.157***
-0.176***
(0.054)
(0.084)
(0.078)
(0.042)
(0.040)
(0.032)
(0.040)
(0.054)
(0.043)
Total Tax rev
-0.301***
-0.280***
(0.089)
(0.092)
Progressivity
0.114
(0.105)
Average Inc. tax
-0.199*
-0.111
-0.129
-0.124
-0.123
-0.099
(0.110)
(0.082)
(0.076)
(0.082)
(0.081)
(0.083)
Corp. Tax rate
0.127
0.073
0.104
0.074
0.069
0.091
(0.188)
(0.156)
(0.152)
(0.153)
(0.138)
(0.143)
Property tax
-0.479
-0.213
-0.277
-0.217
-0.217
-0.182
(0.376)
(0.392)
(0.419)
(0.403)
(0.395)
(0.391)
SSC
-0.440***
-0.279**
-0.285**
-0.278**
-0.279**
-0.278***
(0.113)
(0.098)
(0.105)
(0.103)
(0.097)
(0.093)
Goods&service
-0.285
-0.079
-0.052
-0.101
-0.065
-0.074
(0.196)
(0.177)
(0.179)
(0.168)
(0.173)
(0.184)
GovCons.
-0.380***
-0.400***
-0.387***
-0.392***
-0.450***
-0.402***
(0.091)
(0.086)
(0.088)
(0.096)
(0.094)
(0.104)
Constant.
101.544**
*
99.184***
98.331***
92.017***
90.215***
93.668***
91.230***
92.574***
87.169***
(14.143)
(11.859)
(20.733)
(18.729)
(18.678)
(17.898)
(18.701)
(19.233)
(12.330)
Implicit financing
sources
Total exp.
& Deficit/
Surplus
Total exp.
& Deficit/
Surplus
Total exp.
& Deficit/
Surplus
Productive
exp. &
Deficit/
Surplus
Productive
exp. &
Deficit/
Surplus &
Average
income tax
Productive
exp. &
Deficit/
Surplus &
Corporate
income tax
Productive
exp. &
Deficit/
Surplus &
Property
tax
Productive
exp. &
Deficit/
Surplus &
SSC
Productive
exp. &
Deficit/
Surplus &
Taxes on
goods and
services
Observations
106
105
105
105
105
105
105
105
105
R-squared
0.66
0.68
0.72
0.76
0.76
0.76
0.76
0.75
0.76
Number of cntry
18
18
18
18
18
18
18
18
18
Note: The dependent variable is the 5-year average annual growth rate of real GDP per capita. Levels of significance:
*** 1 percent, ** 5 percent, * 10 percent. Clustered robust standard errors are in parentheses. Period dummies are
always included
Table 6: Top marginal taxation and growth under different financing assumption, 1980-2009
(1)
(2)
(3)
(4)
(5)
(6)
(7)
(8)
TMTR
0.044**
0.042*
0.047
0.039*
0.047*
0.047*
0.043*
0.042**
(0.021)
(0.023)
(0.028)
(0.021)
(0.026)
(0.024)
(0.022)
(0.020)
InitGDP
-8.292***
-8.748***
-8.876***
-8.412***
-8.785***
-8.350***
-8.502***
-7.364***
(1.812)
(1.897)
(1.927)
(1.897)
(1.810)
(1.859)
(1.844)
(1.442)
Investment
-0.141*
-0.144*
-0.110*
-0.150*
-0.131
-0.146*
-0.145*
-0.160**
(0.081)
(0.078)
(0.060)
(0.076)
(0.080)
(0.080)
(0.076)
(0.073)
∆Population
0.041
0.199
0.071
0.054
0.124
-0.025
0.098
-0.035
(0.395)
(0.350)
(0.341)
(0.405)
(0.337)
(0.372)
(0.374)
(0.402)
School
0.011
0.051
0.163
0.024
0.052
-0.108
0.037
0.041
(0.305)
(0.301)
(0.285)
(0.318)
(0.311)
(0.223)
(0.293)
(0.303)
Inflation
-0.146**
-0.152*
-0.154**
-0.132**
-0.158**
-0.133*
-0.134**
-0.113
(0.063)
(0.075)
(0.070)
(0.061)
(0.061)
(0.069)
(0.061)
(0.066)
Avg. Income tax
-0.035
-0.211
-0.227
-0.177
-0.118
-0.133
-0.081
(0.224)
(0.175)
(0.146)
(0.229)
(0.230)
(0.202)
(0.233)
Corp. Tax rate
0.090
0.102
0.097
0.110
0.093
0.095
0.153
(0.172)
(0.153)
(0.159)
(0.177)
(0.164)
(0.158)
(0.147)
Property tax
-0.523
-0.672
-0.765
-0.595
-0.636
-0.609
-0.590
(0.477)
(0.581)
(0.501)
(0.572)
(0.554)
(0.513)
(0.563)
SSC
0.020
-0.134
-0.291**
0.001
-0.078
0.009
0.001
(0.184)
(0.254)
(0.114)
(0.156)
(0.199)
(0.161)
(0.193)
Good&service tax
-0.147
-0.237
-0.225
-0.160
-0.252
-0.180
-0.192
(0.168)
(0.224)
(0.214)
(0.191)
(0.173)
(0.177)
(0.177)
Productive Exp.
-0.126
-0.039
-0.106
-0.046
-0.077
-0.059
-0.102
(0.129)
(0.085)
(0.102)
(0.124)
(0.145)
(0.140)
(0.157)
Non-prod. Exp
-0.204**
-0.102
-0.186**
-0.129
-0.192*
-0.157**
-0.173
(0.075)
(0.123)
(0.085)
(0.111)
(0.094)
(0.066)
(0.102)
Other Exp.
-0.120*
-0.011
-0.106*
-0.046
-0.102
-0.068
-0.097
(0.064)
(0.076)
(0.055)
(0.079)
(0.082)
(0.064)
(0.090)
Deficit
-0.125
-0.142*
-0.043
-0.093
-0.049
-0.082
-0.053
(0.114)
(0.080)
(0.113)
(0.106)
(0.103)
(0.105)
(0.114)
Constant
94.793***
98.969***
98.657***
95.620***
99.831***
95.700***
96.378***
83.706***
(19.917)
(20.785)
(21.600)
(20.402)
(19.083)
(20.074)
(20.046)
(14.729)
Observations
89
89
89
89
89
89
89
89
R-squared
0.79
0.79
0.78
0.79
0.79
0.78
0.79
0.79
Number of cntry
18
18
18
18
18
18
18
18
Note: See Table 5.
Table A1 : variables’ description, data source and summary statistics over the relevant regression period
Definition
Source
Mean
S.D.
Min
Max
Growthit
5-year average annual growth rate of real GDP per capita
PWT
2.01
1.75
-2.84
10.00
TMTRit
Top marginal tax rates on personal income. It is calculated as the
additional central and sub-central government personal income tax
resulting from a unit increase in gross wage earnings (%)
PSS
55.55
12.75
32.00
92.30
InitGDPit
Natural log of real GDP per capita at the beginning of the five year growth
period
PWT
9.99
0.37
8.62
10.80
Inv
Gross investment to GDP ratio (%)
PWT
23.86
4.73
15.04
35.68
ΔPop
Growth rate of population (%)
PWT
0.64
0.47
-0.18
2.17
Schoolit
Average years of schooling
B&L
8.86
2.28
3.07
12.91
Inflation
Inflation rate of the Consumer Price Index (%)
WDI
5.51
4.82
-1.68
22.67
Total tax rev.
Total general government tax revenue as a share of GDP (%)
OECD
33.52
8.02
15.55
50.48
Progressivity
Ration between Avg. Income tax and TMTR
OECD/PSS
6.57
4.24
2.01
25.54
Avg. Income tax
General government revenue from income taxation as a share of GDP (%)
OECD
10.57
4.82
1.91
25.87
Corp. tax rate
General government revenue from corporate taxation as a share of GDP
(%)
OECD
2.98
1.44
1.08
11.31
Property tax
General government revenue from property taxation as a share of GDP (%)
OECD
2.02
0.99
0.35
4.63
SSC
General government revenue from social security contributions as a share
of GDP (%)
OECD
7.57
5.02
0.0
18.85
Good&Service
General government revenue from taxation on good and services as a share
of GDP (%)
OECD
10.11
3.30
3.70
16.4
GovCons.
General government consumption as a share of GDP (%). It includes all
government current expenditure for purchase of goods and services, most
of expenditure on national defense and security, but excludes government
military expenditure that are part of government capital formation.
WDI
18.65
4.31
8.19
29.01
Productive Exp.
It sums consolidated central government expenditure for: general public
services, defense, education, health, housing, transport and communication.
It is defined as a share of GDP (%).
GFS
15.56
4.95
6.06
29.15
Unproductive
Exp.
It sums consolidated central government expenditure for: social security
and welfare, recreation and economic services. It is defined as a share of
GDP (%).
GFS
15.53
5.12
7.10
37.34
Other Exp.
Residual total consolidated central government expenditure not included in
either Productive Exp.or Unproductive Exp. It is defined as a share of GDP
(%).
GFS
13.39
7.13
1.11
29.91
Deficitit
Budget deficit measured as the difference between total general
government expenditure and revenues, (%)
OECD/GFS
2.11
3.48
-5.49
11.93
Note: All the variables described in this table are defined as 5-year average over the growth period. Summary statistics refer to 18 OECD countries over
the period 1965-2009, except for variables Productive Exp.; Unproductive Ex.; Other Exp and Deficit which refer to the period 1980-2009. PWT=Penn
World Tables 7.1; PSS=Piketty, Saez and Stantcheva (2011); B&L= Barro and Lee (2010); WDI= World Development Indicators; OECD/GFS= OECD
– National Accounts. In very few cases when data for the budget deficit from OECD– National Accounts were missing, data from IMF-Government
Finance Statistics are used.
Figure
Fig. 1 Evidence on the nonlinear top tax-growth nexus. Panel A shows a scatterplot of the relationship between 5-year average annual growth rates of
real GDP per capita and 5-year average top marginal tax rates. A linear regression line (dotted line), a quadratic regression line (dashed line) and a
nonparametric Lowess fit (solid line) are also plotted. In Panel B country-specific quadratic regression line are plotted along with the quadratic fit for
the pooled model (dashed line - shaded area is its 95% confidence interval). The vertical dashed lines around the value of 76 indicate the level of the
growth-maximizing top tax rate estimated by the quadratic regression. Data are from Piketty, Saez and Stantcheva (2014) and PWT 7.1