Tax Competition in Europe 1980-2007 – Evidence from Dynamic Panel Data Estimation Michael Overesch∗
Centre for European Economic Research (ZEW), Mannheim
University of Munich
Abstract Previous studies on international tax competition have focused on the contemporaneous interaction of the choice of corporate tax rates, whereas the dynamics of tax competition have been largely ignored. We use dynamic panel data models that relate current tax rates to lagged values of a country’s own as well as other countries’ tax rates. Using data on statutory, effective average and marginal tax rates for European countries from 1980 to 2007, we show that countries strongly compete over statutory and effective average tax rates. In contrast, we do not find interaction in effective marginal tax rates. This relates our results to different dimensions of tax competition and supports the view that countries compete for firms as well as paper profits rather than for marginal investments. We also show that, once the dependence of tax rates over time and tax competition is controlled for, country characteristics such as population size, GDP or openness do not contribute much to explaining corporate tax rates. Keywords: Tax competition, corporate taxes, dynamic panel data estimation JEL Classification: H20, H25, H71
Centre for European Economic Research (ZEW), L7,1, D-68163 Mannheim, +49 621 1235 394, [email protected]
∗∗ University of Munich, Department of Economics, Seminar for Economic Policy, Akademiestr. 1/II, D-80799 Munich, +49 89 2180 6753, [email protected]
Over the past 25 years, both statutory and effective corporate tax rates in Europe show a remarkable downward trend. In 1983, the mean statutory corporate tax rate of 13 western European countries accounted to 49.2%. Twentyfive years later, the mean statutory tax rate of these 13 countries has eroded to 28.3%. The most popular explanation for the striking decline of corporate tax levels is the existence of tax competition for mobile tax bases such as capital, firms, and paper profits. However, the empirical evidence regarding the nature and effectiveness of tax competition among European countries is still rather limited. While an extensive literature has confirmed the presumption that more open economies tend to have lower corporate taxes (See, e.g., Slemrod, 2004; Winner, 2005; Ghinamo, Panteghini, and Revelli, 2007), more direct evidence on the strategic choice of tax rates is scarce. In a recent study, Devereux, Lockwood, and Redoano (2008) find that OECD countries respond with their tax policies to both the effective tax levels attributable to marginal investments as well as to the statutory tax rates of other countries. Other studies confirm the finding that corporate tax rates are positively linked to the tax level of other countries (e.g., Redoano, 2007; Egger, Pfaffermayr, and Winner, 2007). This paper starts from the fact that previous studies on international tax competition have focused on the contemporaneous interaction in governments’ tax policies, whereas the dynamics of tax competition have been largely ignored. We argue that, regarding their choice of corporate tax schemes, national governments are heavily restricted by the tax system that is actually in place. This makes the inherited level of corporate taxes an important factor shaping actual tax policies. We account for the inertia in governments’ tax setting behavior by setting up a fully dynamic empirical model of tax competition. In this model, a given country’s current tax rate is allowed to depend on the tax rate inherited from the previous period as well as other countries’ lagged 2
tax rates. Using data on statutory, effective average and marginal tax rates for European countries from 1980 to 2007, we estimate various specifications of a dynamic panel data model of European tax competition. Our results differ from previous findings in several respects. Firstly, while we can show that governments strongly compete over statutory and effective average tax rates, we do not find interaction in effective marginal tax rates. This relates our results to different dimensions of tax competition and supports the view that countries compete for firms as well as paper profits rather than for marginal investments. Secondly, our findings suggest that, once the dependence of tax rates over time and the impact of tax competition is controlled for, country characteristics such as population size, GDP or openness do not contribute much to explaining corporate tax rates. Thirdly, we find that the evidence regarding the presence of tax competition strongly depends on how the interaction among countries is actually modeled. In fact, our results regarding the presence of tax competition do only hold once we assume that countries react to geographically close neighbors more strongly than to more distant countries. Our evidence is thus in line with studies pointing to the importance of geographical distance for the location of FDI (Carr, Markusen, and Maskus, 2001; Blonigen, Davies, and Head, 2003; Markusen, 2002). The paper is organized as follows. In Section 2, we provide some facts about the evolution of corporate tax rates in Europe over the past 25 years. Section 3 reviews the empirical literature dealing with international tax competition. The description of the estimation approach and the data follows in Section 4. Section 5 presents the results, and Section 6 concludes.
Trends in corporate taxation in Europe
Figure 1 summarizes the trends in corporate taxation in Europe since 1980. In the top left-hand panel, the graph shows the statutory corporate tax rate 3
Figure 1Trends in corporate taxation, 1980-2007
40% 35% 30% 25%
Average STR qa qa qa qa qa qa qa All countries exqa qa cluding eastern Europe qa qa qa q a a a a A a q q q q aA U q qa a a qqaa qqaaa qqq All countries
Average EATR 50% 45% 40%
Average EMTR 50%
25% 20% 15%
qa qa qa qa qa qa qa qa countries exqa qa q All cluding eastern Europe a qa q [email protected]
q q q [email protected]
R q q a qa qa a a a a q qq q q qa All countries
45% 40% 35% 30% 25% 20%
All countries ex-
aq q cluding eastern Europe a qa q a a A a aq q q a a AU qqaaa qq aaa qqq aa * qqq All countries
qa aq aq aq aq aq aq aq aq
EATR, selected countries aa aa a a q q qa q qa a a a a a a a a DE a aaaa q q ×××××× aaaaaaa UK × qqqqq q q q q q q ×PL q q ×q q q q q q q q a a a a a a a a a a a a a a a a a a a a ××× aaaaa CH a a ×a ×××
Notes: Graphs for average tax rates show unweighted averages. Countries in sample (period) are: AT (81-07), BE (80-07), CY (91-07), DK (80-07), ES (91-07), FI (80-07), FR (80-07), DE (80-07), GR (90-07), IE (80-07), IS (90-07), IT (80-07), LU (80-07), MT (94-07), NL (80-07), NO (90-07), PT (90-07), TR (96-07), SE (82-07), CH (80-07), UK (80-07). Eastern Europe: BG (93-07), CZ (92-07), ES (95-07), HR (95-07), HU (92-07), LV (95-07), LT (95-07), PL (92-07), RO (94-07), SI (95-07), SK (92-07).
(STR). This is the statutory headline tax rate of the corporate income tax adjusted for surcharges and the average of local income tax rates. Furthermore, the figure shows cross-country averages for the effective marginal tax rate (EMTR) and the effective average tax rate (EATR). The effective tax rates are calculated following Devereux and Griffith (2003). Unlike the statutory tax rate, effective tax rates reflect all of the relevant income and non income taxes imposed on corporate investments as well as all the rules that determine the tax bases.1 Figure 1 shows the striking decline of both statutory and effective tax rates over the past 25 years. However, the evolution of corporate taxes was not char1
We refer the reader to section 4.3 for a detailed description of effective tax rates.
acterized by a continuous downward trend. Until 1983, an increase in average tax levels can be observed. A similar trend can be found for the mid 1990s, when average tax rates remained rather constant or even increased slightly. In particular, big economies such as Germany, France and Italy increased their tax levels during this period. However, in the long-run, a substantial decline in tax rates can be observed. Starting in the mid 1980s with tax reforms in the UK and Ireland, statutory tax rates went down considerably, whereas the effective marginal tax rates only decreased slightly due to tax base broadening. During the early 1990s, countries in northern Europe reduced their corporate tax levels by introducing some form of dual-income tax systems, which impose significantly lower tax rates on capital income relative to labor income. Since the fall of the iron curtain, the former communist countries in eastern Europe have become increasingly popular as locations for multinationals’ activities. The graphs show that the countries in eastern Europe have significantly contributed to the overall decline of tax rates.2 In 2007, the average statutory tax rate of the eleven considered former transition economies in eastern Europe amounted to 18.9%. In comparison, the average of the remaining European countries was 26.8%, a difference of 8 percentage points. In the lower left-hand panel, the graph denotes a smaller decline in the effective marginal tax rates in comparison to the other tax measures. This is due to several base broadening tax reforms which are important determinants of effective marginal tax rates and which compensate for cutting statutory tax rates. However, several countries have reduced the effective marginal tax rates by means of an abolishment of non-income taxes. A prominent example for this type of reform are the tax reforms in Germany in the late 1990s. The negative trend in effective tax rates reveals that cuts in the statutory tax rates have only partly been compensated by base broadening activities. Accordingly, a clear downward trend of effective tax rates can be observed in 2 Note that the effective tax rates of the eastern European countries do not reflect the various tax incentives such as tax holidays available before joining the EU.
the long-run as well. The graph on the lower right-hand panel depicts the evolution of effective average tax rates of selected countries. The graphs show that a typical country only changes its tax rate a few times during the considered period. For most of the years, we do not observe any significant tax rate variations. A closer view on tax reform policies at the country-level suggests some form of geographical clustering of tax reform activities. While at the end of the 1980s significant tax reforms were undertaken in the UK and Ireland, in the early 1990s the focus of tax reform activities moved to northern European countries. By the end of the 1990s, significant tax rate cuts took place in the eastern European transition economies. Recently, several significant tax rate cuts have taken place in Western Europe such as in Austria, Belgium, Denmark, the Netherlands, Spain, and Germany. A striking example for the recent reaction of countries to reform activities in other countries is Austria, where the statutory tax rate was reduced from 34% to 25% in 2005. One year before, four neighboring countries–the Czech Republic, Slovakia, Hungary, and Italy–had significantly reduced their statutory tax rates. Among other things, Figure 1 suggests that actual tax policies are to some extent framed by inherited tax levels. To substantiate the claim that it is important to account for the impact of lagged levels when explaining actual tax rates, we report a preliminary analysis of statutory tax rates in 32 European countries between 1982 and 2007 in Table 1. We display estimation outcomes of a simple auto-regressive model for the statutory rate, τit , as a function of the rate lagged by one period, laggedτi,t−1 . The estimation equation is τit = λ τi,t−1 + ci + θt + uit ,
where we account for time-invariant unobserved country effects, ci , and period effects, θt . Of course, this model cannot serve to identify the driving forces behind observed tax rates. It is rather meant as a preliminary test of the 6
Table 1: Preliminary Data analysis: Auto-regression in statutory tax rates Dependent variable: Statutory corporate income tax rate, τit (1) (2) Estimator Fixed effects Anderson-Hsiao τi,t−1 0.813??? 0.911??? (0.030) (0.101) 2 R (within) 0.88 -
(3) Bruno 0.889??? (0.037) -
Sample includes 588 observations (years 1982-2007, 32 countries). Estimations include a full series of period effects. Standard errors in parentheses. ??? 1% significance level.
importance of lagged tax levels in setting up a more refined model of corporate taxation in European countries. The first column reports a linear fixed effects estimation of Equation (1). The outcome points to a striking impact of inherited tax levels for current tax policies. In particular, the R2 (within) indicates that, after accounting for observed and unobserved time-constant effects, we are able to explain as much as 88% of the variation in statutory tax rates by conditioning on τi,t−1 as the only explanatory variable (apart from year effects). Hence, not accounting for the impact of lagged taxes in an empirical model of tax competition seems to be difficult to justify. Given that the fixed effects estimator is generally not well suited for a model incorporating lagged dependent variables, we also present outcomes for estimations that account for the Nickell-bias. However, irrespective of whether we use a simple 2SLS approach following Anderson and Hsiao (1981) or the more sophisticated approach suggested by Bruno (2005), we can confirm the strong impact of τi,t−1 on τi,t .
A review of previous empirical literature
The main hypothesis to explain the recent downward trend in corporate tax rates is that of tax competition. During the last decades, FDI stocks as well as the number and size of multinational firms have dramatically increased (see 7
Markusen, 2002). Furthermore, several empirical studies confirm significant effects of company taxation on multinationals’ allocation of subsidiaries, investment projects and taxable profits.3 Hence, shareholders and companies clearly do respond to tax incentives. Governments may have anticipated the behavioral response of firms and may have used tax rates as an instrument to compete for FDI or taxable profits. Then, the increasing mobility of capital may have led to a competition between countries for FDI as well as paper profits. Several studies aim at identifying tax competition at the country level as a means to explain the downward trend in company tax rates. The empirical literature in this field can be divided into two major strands. The first part deals with the general question whether an increasing openness of economies, in particular increasing capital mobility, leads to lower corporate tax rates. This would be indirect evidence for the presence of tax competition. A second strand proceeds by directly identifying strategic interaction in countries’ tax rates. The former strand of literature explains the level of tax rates by means of country characteristics. With respect to the tax competition proposition, a negative effect of increasing capital mobility on company tax rates is expected. The literature employs different measures of capital mobility. Slemrod (2004) uses a discrete indicator of trade openness provided by Sachs and Warner (1995). Using a broad sample of 58 countries, he analyzes whether increased openness or the convergence of country characteristics helps to explain trends in tax rates. Given the discrete character of the employed openness indicator, he can only estimate statistically significant effects on the statutory company tax rate if no country-fixed effects are used.
3 De Mooij and Ederveen (2006) find a median tax rate elasticity of FDI of -2.1 across 31 previous studies. Comprehensive surveys concerning empirical evidence on several aspects of companies’ international tax planning behavior are provided by Hines (1999); Gresik (2001); Blonigen (2005) as well as Devereux (2006).
A few studies use indices of liberalization of international capital transactions based on the legal framework. These studies find stronger evidence that increased openness is associated with lower taxes. For example, Swank and Steinmo (2002) find significant negative effects of an indicator of capital account liberalization provided by Quinn (1997) on statutory corporate tax rates. Schwarz (2007) confirms the negative effect of this indicator on the ratio of effective average company tax rate to the tax rate on labor income. Recently, Ghinamo, Panteghini, and Revelli (2007) estimate a model explaining simultaneously FDI inflows and the statutory corporate tax rate. Using a rich panel of 114 countries, they find significant negative effects of an index for capital account liberalization provided by Chinn and Ito (2002) on statutory tax rates. Significant negative effects of indicators of capital mobility on company tax rates are also confirmed when tax rates based on tax revenue data are used.4 Using samples of OECD countries, Rodrik (1997) as well as Winner (2005) find negative effects caused by capital mobility on company tax rates computed in accordance with the methodology suggested by Mendoza, Razin, and Tesar (1994). Unlike the other studies, Winner (2005), e.g., employs a measure of openness for capital flows based on the relation between total savings and capital formation and finds robust negative effects on company tax rates. In contrast, Swank and Steinmo (2002), Bretschger and Hettich (2002), and Bretschger and Hettich (2005) estimate only insignificant effects while using indices of capital market liberalization. To summarize, this strand of literature provides some important indirect evidence that increasing capital mobility leads to lower corporate tax levels. The second strand of literature focuses directly on the identification of pos4 An exception are studies focussing on the effects of openness on tax revenues as percentage of GDP. Some of these studies find that increasing capital mobility is associated with increasing tax revenues as a share of GDP (Quinn, 1997; Garret, 1998; Swank, 1998). These results are, however, compatible with the tax competition hypothesis since high tax base elasticities can lead to higher tax revenues despite a decline of tax rates.
sible strategic interactions between countries. More precisely, the question is analyzed whether a country responds with its own corporate tax rate to the tax policy of other countries. Effectively, the literature has estimated reaction functions for corporate tax rates of a specific country as a function of other countries’ tax policies. An estimation of separate effects for each of the foreign tax rates, however, is impossible because of the lack of degrees of freedom. Therefore, a usual procedure consists of compiling weighted averages of the foreign countries’ tax rates. Devereux, Lockwood, and Redoano (2008) use a sample of 21 OECD countries for the period from 1982 until 1999 to identify strategic interactions in statutory tax rates, as well as in the tax levels attributable to marginal investments. They detect a positive relationship between a company’s own tax rate and the composite tax rate of the remaining countries for both tax measures. Using a sample of 17 European countries from 1970 until 1999, Redoano (2007) confirms a positive relationship between the statutory tax rate and the composite tax rate of other countries. Haufler, Klemm, and Schjelderup (2006) analyze the effects of capital mobility on the relation between statutory corporate tax rates and the tax burden on labor income by using a panel of 23 OECD countries. While they estimate the expected negative effect of capital mobility on the ratio between corporate income tax to labor income tax, they also find that the relative level of a country’s corporate tax develops in accordance to the worldwide average of corporate tax rates over time. Egger, Pfaffermayr, and Winner (2007) analyze strategic interaction effects between countries in both the corporate income tax rate and the personal income tax rate. Based on a sample of 30 OECD countries, they show a positive correlation between the tax rates chosen by one country and the average tax rates of its neighboring countries for both corporate and personal income taxes. Unlike the other studies dealing with strategic responses to foreign tax rate adjustments, Altshuler and Goodspeed (2003) use a backward-looking tax
measure based on company tax revenues. They estimate tax reaction functions based on a sample of 17 European countries during the period from 1971 until 1996, assuming that the US acts as a Stackelberg leader in the worldwide tax competition process. Their results suggest that a country’s tax revenues from the company sector as a percentage of GDP are negatively affected by tax cuts in both the US as well as its European neighboring countries. In sum, the results of previous empirical studies support the hypothesis that tax competition has contributed to the decline in corporate tax rates.
The empirical model General discussion
Based on the above discussion of previous literature, the main contribution of this paper is to provide evidence on tax competition among European countries, taking account of the dynamics in the countries’ tax setting behavior. By treating corporate taxes as being determined by a dynamic process, we depart from a tradition in the empirical tax competition literature which has extensively worked with models that relate current levels of the tax rate to average current tax rates of neighboring countries or jurisdictions.5 There are three reasons for this departure. Firstly, a dynamic model explicitly accounts for a certain time lag in the reaction of governments to neighboring countries’ tax policies. Arguably, compared to local jurisdictions, such time lags are more reasonable in applications with countries as the units of observation. Secondly, a dynamic model seems to be a natural choice given the pronounced 5 See Hayashi and Boadway (2000) and Dreher (2006) for noteworthy exceptions. To our knowledge, only Dreher (2006) explicitly considers both the own tax rate as well as the weighted average of the other countries’ tax rates as lagged variables. However, using a sample of 18 OECD countries he is unable to identify any statistically significant effect of the weighted average tax rates of other countries.
persistence of corporate tax rates over time. Our model is thus designed to fully capture the fact that when choosing the current tax rate, national governments seem to be heavily restricted by the established tax system that is actually in place in general and by the inherited level of corporate taxes in particular. Thirdly, the empirical tax competition model that has been the main focus of the literature so far does not seem to be particulary well suited for our data. Note that the bulk of empirical literature has effectively estimated reaction functions for some versions of the tax competition model developed by Zodrow and Mieszkowski (1986) and Wilson (1986). This approach has the clear advantage that the empirical model is closely linked to the theoretical literature. A disadvantage of this approach, however, lies in the fact that a jurisdiction’s own tax rate and neighboring jurisdictions’ taxes are treated as being simultaneously determined. Due to this simultaneity, (functions of) neighbors’ taxes are endogenous in the tax setting equation that is usually being estimated. Parameter identification must therefore rely on instruments which are strongly correlated with tax rates in neighboring jurisdictions, but uncorrelated with the residual in the tax setting equation. The usual approach is to construct instruments based on exogenous country characteristics appearing as controls in the tax setting equation. Effectively, weighted averages of neighbors’ exogenous characteristics are used to instrument neighbors’ tax rates. While the spatial IV approach has been widely and productively applied to explain the choice of tax rates and other fiscal variables of local jurisdictions, there is less experience in applying it to fiscal choices at the macro level. The literature has found it particularly difficult to identify a set of exogenous country characteristics that might be used to derive instruments for neighbors’ taxes in tax reaction functions. Slemrod (2004), e.g., provides a detailed discussion of different factors which may determine corporate tax rates. In fact,
the only consensus of previous studies on the determinants of corporate tax rates at the national level seems to be that the personal top income tax rate and measures of country size such as population or GDP are strongly (positively) correlated with corporate tax rates (Altshuler and Goodspeed, 2003; Slemrod, 2004; Winner, 2005; Devereux, Lockwood, and Redoano, 2008).6 The problem is further aggravated by the fact that the exogeneity of both variables in a corporate tax setting equation seem to be questionable.7 In sum, we believe that there are strong conceptual and methodological reasons to consider a dynamic specification in order to explain the choice of corporate taxes at the national level. Apart from the fact that a dynamic model avoids the simultaneity problem discussed above, it accounts for inertia in countries’ tax policies and for lags in the reaction to other countries’ policies that are due to the complexity of the political decision making process at the national level. In the following, our estimation approach is formally discussed.
A dynamic empirical model of tax competition
The structural equation of our estimation approach relates the current tax rate τit of country i = 1, . . . , N in period t = 1, . . . , T to lagged values of the country’s own and neighbors’ tax rates, τit = λ τi,t−1 + φ τ−i,t−1 + xit β + ci + θt + uit ,
Slemrod (2004), e.g, finds that in particular the personal income tax rate is associated with level of the statutory company tax rate. Almost all other country characteristics such as population and measures for trade and capital openness are insignificant if country fixed effects are considered. Devereux, Lockwood, and Redoano (2008), e.g., find weak significance of country-specific control variables. 7 Note that the alternative to using an IV-approach, i.e. maximum likelihood estimation, does not solve the problem, as it implicitly relies on the same instruments as the spatial IV approach. See Baicker (2005) for a recent discussion of the different estimation techniques in a related context.
where τ−i,t−1 =
wij τj,t−1 is a linear combination of other countries’ tax
rates in t − 1 with weights wij ≥ 0 if i 6= j and wij = 0 if i = j, xit is a vector of country characteristics, ci is an unobserved country-specific effect, θt is a period effect common to all countries, and uit is a residual. The approach to identify the vector of coefficients (λ, φ, β 0 )0 follows Arellano and Bond (1991). To get rid of the unobserved effect ci , we take the first differences of the structural equation (2) to obtain ∆τit = λ ∆τi,t−1 + φ ∆τ−i,t−1 + ∆xit β + ∆θt + ∆uit .
We note that ∆τi,t−1 = τi,t−1 − τi,t−2 is correlated with ∆uit = uit − ui,t−1 . As discussed by Arellano and Bond (1991), ∆τi,t−1 can be instrumented by τi,t−p for p ≥ 2 as long as the u’s are not serially correlated, E(uis uit ) = 0 for all t 6= s,
since τi,t−p (for p ≥ 2) is correlated with ∆τi,t−1 but uncorrelated with ∆uit . As with ∆τi,t−1 , we need to specify a set of instruments for all other explanatory variables in the differenced equation. Denoting explanatory variables by z, we have to distinguish between variables which are strictly exogenous, meaning that E(zis uit ) = 0 for all t and s, those which are predetermined, meaning that E(zis uit ) = 0 for s ≤ t and E(zis uit ) 6= 0 for s > t, and those which are endogenous, i.e. E(zis uit ) = 0 for all s < t and E(zis uit ) 6= 0 for all s ≥ t. Let us first turn to ∆τ−i,t−1 = τ−i,t−1 − τ−i,t−2 and consider country j’s contribution to the first term, τj,t−1 = λ τj,t−2 + φ τ−j,t−2 + xj,t−1 β + cj + θt−1 + uj,t−1 .
Assuming that (for j 6= i) E(xjs uit ) = 0 for all s ≤ t,
E(cj uit ) = 0 for all t,
E(uis ujt ) = 0 for all t and s,
note that τj,t−1 depends on τ−j,t−2 and thereby, via τi,t−2 , on ui,t−2 , but not on ui,t−1 nor on uit . Hence, under the given assumptions, τ−i,t−1 is uncorrelated with ∆uit . A similar argument shows that τ−i,t−2 as the second term entering ∆τ−i,t−1 is also uncorrelated with ∆uit . In short, under the assumptions (A.1) to (A.4a), the average tax rate of neighbors in t − 1 should be treated as a predetermined explanatory variable. Thus, τ−i,t−1 as well as further lags can be used as instruments for ∆τ−i,t−1 in the Arellano-Bond GMM procedure. As mentioned above, (A.1) represents the basic assumption that the ArellanoBond estimator produces consistent estimates without interaction among different cross-sectional units. While it is a strong assumption, it can easily be tested and is therefore unproblematic in practice. Assumptions (A.2) to (A.4) restrict the cross-correlations between different units and have to be made to ensure that the variables which are used as instruments in the differenced equation are uncorrelated with the differenced residual. Among them, (A.4) is certainly the most restrictive as it precludes spatial error correlation in the structural equation. Can we relax this assumption? Consider τ−i,t−2 and further lags as instruments for ∆τ−i,t−1 . Given that (A.1) to (A.3) hold, we need to impose E(uis ujt ) = 0 for all t 6= s
to make sure that the instruments are uncorrelated with ∆uit . This is weaker than (A.4a) as we allow the residuals of different cross-sectional units to be correlated contemporaneously. To implement the Arellano-Bond estimator under (A.4b), we now have to exclude τ−i,t−1 from the set of instruments. The linear combination of neighbors’ taxes is thus treated as an endogenous explanatory variable.
As mentioned above, the choice of suitable instruments for the elements in ∆xit depends on the assumptions regarding the exogeneity of the included explanatory variables. We refer the reader to the results section for a detailed discussion. Apart from the lagged dependent variable and neighbors’ tax rates, we include a vector of country characteristics as control variables. In selecting these variables, we draw from the empirical literature on the determinants of corporate taxes at the country level. However, we expect the importance of the controls (in terms of their power as predictors and statistical significance) to be lower than in studies using static empirical models. The reason is that the impact of slowly moving country characteristics will partly be reflected in the lagged dependent variable. Furthermore, the estimation equation is in first-differences. Accordingly, time-invariant country characteristics are controlled. In empirical studies of corporate tax setting it is common to include measures of country size among the control variables. One reason for this is that a larger local market tends to attract more market-seeking FDI (Brainard, 1997; Carr, Markusen, and Maskus, 2001; Blonigen, Davies, and Head, 2003). Empirical studies suggest that this horizontally motivated FDI is less tax sensitive compared to vertically motivated FDI (Mutti and Grubert, 2004). Therefore, an increasing market size may lead to less tax sensitive FDI, and thus, tax rates might be higher. We use total population and the GDP as measures for country and market size. Furthermore, we account for GDP growth, because governments can be thought of using tax policies to respond to the business cycle and the corresponding effects on tax revenue. Furthermore, governments may also be more aggressive in competing for international capital in phases of low economic growth. As described in Section 3, previous literature finds evidence for an increasing capital mobility to be associated with lower corporate taxes. This can be interpreted as indirect evidence for tax competition. Therefore, we will control for 16
capital mobility while focussing on direct interaction among countries in setting their tax policies. Following Devereux, Lockwood, and Redoano (2008), we use the ratio between the sum of annual inward and outward FDI flows and GDP as our measure of capital mobility and openness. The advantage of this indicator compared to alternative measures is that an FDI-based indicator directly reflects the mobile tax bases for which countries may compete. Finally, we include among our control variables the top personal income tax rate (PITR ). An increasing gap between personal and corporate income taxes may lead to an incentive to defer taxes by means of excessive retention of capital income at the corporate level. Consequently, a corporate income tax may serve as a backstop for the personal income tax level within the tax system of a country (Slemrod, 2004). We therefore expect a higher personal income tax rate to have a positive effect on the corporate tax level, as suggested by numerous other empirical studies.
Our database covers up to 32 European countries for the period from 1980 until 2007. Basically, our sample size depends on the availability of reliable tax data. Therefore, during the 1980s the sample consists of western and northern European countries. Thereafter, the sample grows significantly. Since 1996, it covers 32 European countries, including all current 27 EU member states. This database constitutes by far the most extensive country panel among all existing studies on interactions in corporate tax policies. Note also that our study is the first one that tests for tax policy interactions including Central and Eastern Europe. The choice of meaningful tax measures is essential for an empirical analysis of strategic interaction between countries’ tax policies. If governments engage in competition for mobile capital, firms, and paper profits, they should use those
tax instruments which affect the behavior of the relevant economic agents. Firms typically consider expected future tax payments rather than historical tax payments when deciding on investments or profit assignments. Therefore, we use so-called forward-looking tax rates which convey information on expected future tax payments. These indicators are based on information regarding the effective tax legislation. With regard to our analysis it is of crucial importance to use forward-looking tax measures that directly reflect changes in the tax legislation.8 We use three forward-looking measures of corporate taxation to identify tax competition: The statutory tax rate (STR), the effective marginal tax rate (EMTR), and the effective average tax rate (EATR). The STR is the simplest forward-looking indicator. However, it neglects any difference in the tax base and the existence of non-income taxes. We utilize the statutory headline tax rate of the corporate income tax adjusted to surcharges and typical local income taxes, which are imposed on the same or a similar tax base.9 In addition, we calculate effective tax rates according to the methodology proposed by Devereux and Griffith (2003). Effective tax rates are more complex and compress various aspects of the legal tax code at a respective location. The underlying idea is to determine effective tax levels of a hypothetical, standardized investment project. An advantage of using effective tax rates is that several relevant components of the tax system of a given country can be considered within one indicator. These tax measures reflect all relevant income 8
There are also backward-looking measures based on overall tax revenues (e.g. Mendoza, Razin, and Tesar, 1994), or firm-level tax payments (e.g. Desai, Foley, and Hines, 2001). These measures are not ideal for an evaluation of a country’s tax policy response for several reasons. The most important objection is the fact that backward-looking tax indicators are based on historical tax payments and, therefore, do not necessarily reflect the tax incentives on tax payers’ decisions. Consider a country which has a high statutory corporate tax rate. Then, a backward-looking tax indicator may measure a low tax level for the reason that there are strong incentives to shift taxable profits out of that country. Consequently, backwardlooking tax measures can give misleading indicators of tax policy. Devereux and Lockwood (2006) discuss different tax indicators and point to various conceptual problems of measures based on historical tax payments. 9 Local income taxes such as the IRAP in Italy or the Gewerbesteuer in Germany are taken into account.
and non-income taxes imposed on corporate investments, as well as all the rules determining the tax bases such as depreciation rules. Our specifications for computing the effective tax rates are very similar to the assumptions in a comprehensive study about company taxation by the European Commission (2001).10 Our three tax measures are suitable indicators for different dimensions of tax competition. The STR is a reliable measure for corporate taxation in cases where the determination of the tax base is not relevant for firms. Therefore, the STR is a well-suited indicator for competition for mobile paper profits. Since the EMTR indicates the tax burden attributable to marginal investments, this measure is especially important when countries compete for marginal capital investments. Finally, the EATR is the relevant indicator of the tax burden of profitable projects that generate economic rents due to firm-specific assets.11 Consequently, the EATR should be the relevant tax measure if countries compete for complete firms or subsidiaries.12 Tax policies can asymmetrically affect these three tax measures by choosing different tax types, statutory tax rates and rules determining the tax bases. Accordingly, governments are able to engage in different dimensions of tax competition. We will analyze the different dimensions of tax competition separately, which include the competition for taxable profits, firms, as well as investments. Therefore, we alternatively use the STR, the EMTR, and the EATR as dependent variables in our empirical model. 10
The standardized project contains investments in the following five asset types: industrial buildings, machinery, intangible assets, inventories, and financial assets. The project is equally financed by retained earnings, the issue of new shares, and debt. We assume an incorporated company. Only domestic taxes and only income and non-income taxes imposed at the corporate-level are considered. Due to data limitations for the time span of 28 years, specific property taxes on real estate are not included. We cannot take into account special tax regimes available only for specific firms. With regard to the definition of the taxable bases, we consider the relevant rules concerning depreciation allowances, valuation of inventories and interest deductibility in case of debt financing. 11 In accordance with European Commission (2001), we assume a pre-tax rate of return of about 20%. 12 Note that previous empirical studies confirm that the EATR is a suitable indicator in case of location decisions (Devereux and Griffith, 1998; Devereux and Lockwood, 2006; B¨ uttner and Ruf, 2007).
Table 2 offers some descriptive statistics of the variables used in the empirical analysis. Table 2: Descriptive Statistics Variable
j wij STR j
wij EMTR j
wij EATR j
GDP Growth Population PITR
Definition Dependent variables: Statutory corporate income tax rate adj. for surcharges and local income taxes Effective average tax rate Effective marginal tax rate Composite neighbors’ tax rate: Weighted average of other countries’ STRs.a) Weighted average of other countries’ EMTRs.a) Weighted average of other countries’ EATRs.a) Control variables: Sum of inward and outward foreign direct investment as proportion of GDP GDP in billions of purchasing power parities (PPP) Real growth rate of GDP Total number of inhabitants in millions Personal top income tax rate
Sample includes 32 countries between 1980 and 2007. The tax variables are based on own calculations. The underlying tax information is collected from several databases provided by the International Bureau of Fiscal Documentation (IBFD), Amsterdam, and from surveys annually provided by Ernst&Young, PwC and KPMG. The information on FDI flows is taken from the World Development Indicators of the World Bank. The other control variables are taken from Eurostat and the World Development Indicators of the World Bank. a) Other countries’ tax rates weighted by the squared inverse of distance in kilometers between capitals and adjusted by for population (in logs). See text for details.
When estimating our dynamic tax competition model, the weights wij have to be treated as predetermined. This poses the question of how to specify the scheme that provides us with suitable weights.13 Previous studies of international tax competition have extensively employed uniform weights, which put equal weight on each foreign country in computing the average tax rate of other countries (see Devereux, Lockwood, and Redoano, 2008; Garretsen and Peeters, 2007; Redoano, 2007; Dreher, 2006; Haufler, Klemm, and Schjelderup, 2006). One conceptual problem of uniform weights is that for P N → ∞, j wij τj,t becomes perfectly collinear to a common period effect. Thus, in general, with uniform weights we cannot identify the competition effect separately from a common period-specific shock. The same argument holds if other countries’ tax rates are weighted by some country characteristic such as GDP or population. Since common period-specific shocks such as, for instance, changing expectations regarding the world business cycle, may be important factors shaping governments’ tax policies, we are well advised to choose weights which allow for a separate identification of tax competition effects. Given the concerns mentioned above, we define weights that are based on geographical distance. The literature provides clear-cut evidence for a negative effect of distance on FDI (see, e.g., Carr, Markusen, and Maskus, 2001; Blonigen, Davies, and Head, 2003; Markusen, 2002). In case of investment decisions, geographical distance drives transportation costs for produced goods but also information costs (see Portes and Rey, 2005). Geographical distance should also negatively affect pure paper-profit shifting since the underlying intra-firm transactions such as intra-firm trading should be inversely related to geographical distance as well. Therefore, it seems to be reasonable to as-
13 A detailed discussion of different weighting schemes and their interpretation is provided by Redoano (2007).
sume that governments perceive tax policies of immediate neighbors to be more relevant than tax policies of more distant countries. To operationalize the above argument, we define two weighting schemes for other countries’ tax rates. Denoting the geographical distance between countries i and j (in kilometers) by dij and total population by pop (in millions), 1 2 we set wij = wij = 0 if j = i and
1 wij =P
1/d2ij , 2 k6=i 1/dik
ln(popj + 1)/d2ij 2 k6=i ln(popk + 1)/dik
2 wij =P
∀ j 6= i.
While the first scheme takes only account of distance, the second scheme adjusts the weights by country size in terms of total population. This ensures that the contribution of very small countries like Luxembourg or Malta is discounted relative to big countries like France or Germany (holding distance fixed). The effect of including population size on the composition of average tax rates can be nicely illustrated by looking at the vector of weights for 1 , the largest weights are those of a country like France. To begin with wF,j
Belgium (0.24), Luxembourg (0.19), UK (0.14), the Netherlands (0.09), and Switzerland (0.09). The remaining share of 0.26 is scattered among other countries (individual weights lower than 0.04). With population size included 2 according to wF,j , the largest weights are those of Belgium (0.26), UK (0.25),
the Netherlands (0.11), Switzerland (0.08), and Germany (0.04). Hence, using weighting scheme 2 essentially removes Luxembourg from the influential neighbors (potentially) affecting tax policies in France.14
14 For Germany, the largest weights are those of the Czech Republic (0.21), Denmark (0.13), Poland (0.06), Austria (0.06), Slovak Republic (0.05), the Netherlands (0.05), Luxembourg (0.04), and Belgium (0.04) for scheme 1 and those of Czech Republic (0.23), Denmark (0.11), Poland (0.10), Netherlands (0.06), Austria (0.06), Slovak Republic (0.04), Belgium (0.04), France (0.04), and Hungary (0.04).
We start the discussion of our empirical results with a set of baseline estimations reported in Table 3. It shows the outcomes for six specifications of the dynamic tax setting equation estimated by the GMM procedure proposed by Arellano and Bond (1991). Note that, apart from a full series of year effects, we do not include any control variables. Hence, we estimate equations of the type
τit = λ τi,t−1 + φ
wij τj,t−1 + ci + θt + uit ,
where τ refers to either the STR, the EMTR, or the EATR. In general, the number of potential instruments that can be used in the Arellano-Bond GMM procedure is large. In fact, the instrument count is quadratic in time dimension, T . This can cause several problems in finite samples.15 A large set of instruments can overfit endogenous variables, and it tends to weaken the usual tests for overidentifying restrictions. To avoid such problems, our strategy is to use not more than two lagged values as instruments for any endogenous explanatory variable. In particular, we use τi,t−2 as the only instrument for ∆τi,t−1 . In fact, since all our tax measures are highly serially correlated, using additional lags would add a negligible amount of information to the system. With respect to ∆τ−i,t−1 , we allow τ−i,t−2 and τ−i,t−3 to be used as instruments. Apart from limiting the number of instruments, throughout the paper we report only GMM regressions where the number of instruments does not exceed the cross-sectional dimension of the system, N . Note that the strict limit to the number of lags used as instruments leads to 15 The number of elements in the estimated variance matrix of the moments is quadratic in the number of instruments, and therefore quartic in T . In finite samples, it is often difficult to estimate this matrix with reasonable precision.
Table 3: Baseline estimations 1982-2007 (Arellano-Bond) Spatial weights Dependent variable, τit τi,t−1 P
Test 1st order auto-cov. Test 2nd order auto-cov. Test overid. restrictions
Distance STR EMTR (1) (2) 0.942??? 0.952??? (0.097) (0.078) ?? 0.450 0.306? (0.197) (0.164) 0.000 0.007 0.569 0.531 0.812 0.455
EATR (3) 0.964??? (0.102) 0.439?? (0.193) 0.000 0.825 0.872
Distance, STR (4) 0.962??? (0.097) 0.506?? (0.234) 0.000 0.564 0.768
pop. adjusted EMTR EATR (5) (6) 0.961??? 0.976??? (0.078) (0.103) ? 0.347 0.514?? (0.196) (0.234) 0.008 0.000 0.574 0.773 0.474 0.858
Sample includes 588 observations (years 1982-2007, 32 countries). Both explanatory variables are treated as endogenous. Number of IVs=29. Estimator is two-step difference GMM. Robust standard errors (corrected for finite sample bias) in parentheses. Level equations include a full series of period effects. Numbers reported for tests are p-values. ? 10% significance level. ?? Idem., 5%. ??? Idem., 1%.
an information set for parameter identification in the Arellano-Bond GMM procedure which is very similar to the computationally simpler Anderson and Hsiao (1981) approach. In fact, we checked all Arellano-Bond estimations presented here by their corresponding Anderson-Hsiao counterparts, and found very similar results (both in terms of coefficient estimates and significance levels) in all cases. Note that for each country, the first and second year with information on tax rates are used up for generating differences and instruments, τi,t−2 and τ−i,t−2 . We can thus make use of 588 observations from the period 1982-2007. We report two-step difference GMM estimations which are generally more efficient than one-step estimates, and standard errors that are corrected for finite sample bias (Windmeijer, 2005). While the first three columns in Table 3 depict results for weights that account only for geographical distance, the regressions in Columns (4) to (6) use distance-based weights that are adjusted to the country size in terms of population. We note that the results for the first set of baseline estimations are very similar across all six specifications, both in terms of estimated coefficients and significance levels. The coefficient of a country’s lagged own tax 24
rate is estimated to be in the range of 0.94 to 0.98, and the point estimates are characterized by a remarkable precision. The estimates of the tax competition effect lie between 0.44 and 0.52 for the STR and the EATR and between 0.31 and 0.35 for the EMTR. Note furthermore that, while the effect for the STR and the EATR is significant at the 5% level for both weighting schemes, the effect for the EMTR is only weakly significant. Apart from coefficient estimates and standard errors, we report the usual tests on auto-correlation of residuals uit and the Hansen test on overidentifying restrictions. The Arellano-Bond tests on auto-correlation are applied to residuals in differences. Hence, to check for first-order serial correlation in levels, we look for a second-order correlation in differences. The numbers shown in the Table are p-values, indicating the presence of a first-order auto-correlation in differences (reflecting the shared term in ∆uit and ∆ui,t−1 ) and, more importantly, the absence of second-order auto-correlation (suggesting that no significant correlation exists between ui,t−1 and ui,t−2 ). Hence, across all baseline specifications, the assumption of no serial correlation in errors is supported.16 Finally, the tests of overidentifying restrictions (reported numbers are p-values) indicate that the validity of τ−i,t−3 as an additional instrument for τ−i,t−1 cannot be rejected at any reasonable level of significance. Note also that, although we made every effort to collect data on tax rates for as many European countries as possible, the cross-sectional dimension of our panel is still limited to 32 countries from 1996 to 2007 and even less for earlier years. Thus, the rule-of-thumb restriction that the instrument count should not exceed the number of cross-sectional units is already close to becoming binding in a model without further control variables. As a robustness check, we re-estimated our six baseline specifications using only observations from the year 1988 onwards. This increases the average number of countries per year in the sample from 24 to 28 and reduces the number of year dummies from 16 Given that first-order auto-correlation does not exist, there seems to be little reason to test for a higher order residual auto-correlation.
Table 4: Baseline estimations 1988-2007 (Arellano-Bond) Spatial weights Dependent variable, τit τi,t−1 P
Test 1st order auto-cov. Test 2nd order auto-cov. Test overid. restrictions
Distance STR EMTR (1) (2) 0.930??? 0.942??? (0.111) (0.094) ??? 0.477 0.338?? (0.183) (0.152) 0.000 0.010 0.750 0.339 0.952 0.527
EATR (3) 0.949??? (0.128) 0.497??? (0.176) 0.000 0.852 0.991
Distance, STR (4) 0.945??? (0.108) 0.509?? (0.212) 0.000 0.737 0.903
pop. adjusted EMTR EATR (5) (6) 0.950??? 0.963??? (0.094) (0.127) ?? 0.383 0.577??? (0.189) (0.219) 0.010 0.000 0.373 0.922 0.532 0.988
Sample includes 513 observations (years 1988-2007, 32 countries). Both explanatory variables are treated as endogenous. Number of IVs=23. Estimator is two-step difference GMM. Robust standard errors (corrected for finite sample bias) in parentheses. Level equations include a full series of period effects. Numbers reported for tests are p-values. ?? 5% significance level. ??? Idem., 1%.
26 to 20. The results are shown in Table 4. While the parameter estimated for a country’s own lagged tax rate is slightly reduced, the estimate of the competition effect increases slightly compared to the regressions when using all available information. Note also that the competition effect is now estimated with significantly higher precision. However, the general picture obtained from the baseline model does not depend on whether the years 1982-1985 are included in the estimations.
Estimations including control variables
We now turn to estimations with included control variables. As discussed above, we include the top personal income tax rate, total population (in logs), GDP (in logs), openness (measured as the ratio between the sum of annual inbound and outbound FDI flows and GDP), and the growth rate of GDP. The level equation thus looks like
τit = λ τi,t−1 + φ τ−i,t−1 + β1 PITR it + β2 ln pop i,t−1 + β3 ln GDP i,t−1 + β4 openness i,t−1 + β5 growth i,t−1 + ci + θt + uit . (7) 26
Table 5: Tax competition in Europe, 1988-2006 (Arellano-Bond) Spatial weights Dependent variable, τit τi,t−1 P
Top income tax rate log(Population) log(GDP) Openness Growth Test 1st order auto-cov. Test 2nd order auto-cov. Test overid. restrictions
Distance STR EMTR (1) (2) 0.958??? 0.824??? (0.330) (0.208) ?? 0.464 0.276 (0.230) (0.265) 0.123 0.533 (0.483) (0.534) 0.124 -0.984 (0.379) (0.623) 0.003 0.127 (0.127) (0.136) -0.000 -0.001 (0.003) (0.003) -0.151 -0.335 (0.256) (0.337) 0.006 0.037 0.938 0.214 0.248 0.146
EATR (3) 0.978??? (0.338) 0.443? (0.230) 0.123 (0.422) -0.072 (0.195) 0.006 (0.099) -0.000 (0.002) -0.095 (0.213) 0.008 0.522 0.293
Distance, STR (4) 0.959??? (0.333) 0.509?? (0.226) 0.142 (0.476) 0.103 (0.367) 0.014 (0.130) 0.000 (0.003) -0.176 (0.263) 0.006 0.992 0.266
pop. adjusted EMTR EATR (5) (6) 0.814??? 0.973??? (0.200) (0.334) 0.461 0.551?? (0.370) (0.248) 0.572 0.144 (0.488) (0.409) -1.04? -0.097 (0.617) (0.188) 0.120 0.013 (0.121) (0.099) -0.002 -0.000 (0.003) (0.002) -0.327 -0.117 (0.302) (0.213) 0.041 0.007 0.241 0.619 0.202 0.338
Sample includes 428 observations (years 1982-2006, 31 countries). Endogenous explanatory variables: τi,t−1 , P j wij τj,t−1 , top income tax rate, and growth. Predetermined explanatory variables: population, GDP, and openness. Number of IVs=27. Estimator is two-step difference GMM. Robust standard errors (corrected for finite sample bias) in parentheses. Level equations include a full series of period effects. Numbers reported for tests are p-values. ? 10% significance level. ?? Idem., 5%. ??? Idem., 1%.
Note that we include the contemporaneous PITR, as they can be assumed to be subject to the same political decision process as the corporate income tax rates. For the remaining explanatory variables, we use lagged levels to account for the time lag between the political decision regarding taxes and their actual implementation. Table 5 reports estimation results from Arellano-Bond GMM estimations. Note that, apart from τi,t−1 and τ−i,t−1 , we also treat the PITR and the growth rate as endogenous variables. In practice, this means that in the differenced equation, ∆PITR it is instrumented by PITR i,t−2 , and ∆growth i,t−1 by growth i,t−2 . The remaining control variables are assumed to be exogenous. To keep the instrument count below the number of cross-sectional units,17
we follow the example of the baseline estimations in Table 4 and restrict the observations to those from 1988 onwards.18 Compared to the extensive literature on the determinants of corporate tax rates reviewed above, our dynamic panel data estimations yield striking results. First of all, while the standard errors of τi,t−1 are two to three times higher than in the baseline estimations, the coefficients are still about 0.82 for the EMTR and in the range between 0.96 and 0.98 for the STR and the EATR, and they are all significant at the 1% level. Secondly, we find that the inclusion of the set of control variables has no significant effect on the parameter estimates for the tax competition effect. However, the effect for the EMTR is no longer statistically significant. Hence, we still find evidence for the substantial relevance of neighbors’ lagged tax rates as long as we measure the tax burden by the statutory rate or the EATR. However, the data does not support the notion that choices regarding the EMTR are affected by neighboring countries’ choices. Another striking result of the regressions is that, irrespective of which tax measure is being considered, the control variables themselves do not seem to be partially correlated with corporate tax rates. Hence, our results suggest that, once the history of tax policies and the direct interaction among countries is controlled for, the country characteristics that have been discussed in the literature do not contribute much to explaining the variation in corporate taxes. Our results shown in Table 5 suggest that countries strategically response to neighboring countries’ choices of their tax levels in terms of the STR and the EATR. In contrast, we do not find robust interaction effects regarding the EMTR. This latter result confirms theoretical expectations described by Devereux, Lockwood, and Redoano (2008). They show that the interaction in terms of the EMTR should become very weak if the tax policy of a single country does not significantly influence the world interest rate. 17 18
The number of countries is reduced to 31 because we do not have FDI data for Malta. We do not have FDI data for 2007, making 2006 the most recent cross-section.
The different tax measures used for the empirical analysis can be linked to different dimensions of tax competition. Empirical studies dealing with behavioral response to tax incentives suggest that multinational firms allocate profits according to tax rate differentials (Hines and Rice, 1994; Huizinga and Laeven, 2008). Furthermore, there is evidence that the EATR rather than the EMTR affects multinationals’ location decisions (Devereux and Griffith, 1998; B¨ uttner and Ruf, 2007). Our results therefore support the view that countries compete for both paper profits using the STR, as well as for mobile firms and subsidiaries using the EATR.
The evolution of tax competition over time
We proceed by investigating the evolution of tax competition among European countries over time. As shown in the previous subsection, there is evidence for competition in statutory and effective average tax rates if we pool the available information over many years. The question that we ask here is: Can we say anything about trends in the intensity of tax competition? We can respond to this question in a very straightforward manner. Our dynamic tax competition model is estimated by means of a 10-year window that moves from the the first years in our (effective) sample (1982-1991) to the last years (1997-2006). With one-year steps, this gives 16 positions of the window. If the degree of tax competition has significantly changed over time, we expect to find evidence for this by changes in the estimated coefficients of τ−i,t−1 . Table 6 depicts the tax competition effects for the EATR that have been derived using the same set of control variables as in Table 5 (coefficients of own lagged tax rate and controls are not shown). With both weighting schemes, we find significant tax competition effects in the 10-year windows between 1985 and 1997. Moreover, we also find evidence for intense tax competition using the windows located between 1992 and 2003. With weights adjusted
Table 6: Evolution of tax competition over time (10-year windows) Spatial weights Dependent variable, τit 1982-1991 1983-1992 1984-1993 1985-1994 1986-1995 1987-1996 1988-1997 1989-1998 1990-1999 1991-2000 1992-2001 1993-2002 1994-2003 1995-2004 1996-2005 1997-2006
Nob 103 111 119 127 136 147 163 179 197 215 234 248 260 273 283 289
Distance EATR Coefficient 0.735 0.804 0.478 0.763 1.14?? 0.677? 0.757?? 0.530 0.480 0.401 0.453 0.350 0.425? 0.105 0.492 0.186
Std. error 0.694 0.467 0.978 0.488 0.470 0.393 0.316 0.400 0.336 0.354 0.347 0.258 0.242 0.462 0.518 0.577
Distance, pop. adjusted EATR Coefficient Std. error 0.360 0.469 0.587 0.514 0.447 1.46 ? 0.954 0.538 ?? 1.00 0.414 0.733? 0.361 0.636? 0.353 0.393 0.484 0.468 0.323 0.433 0.323 0.532?? 0.258 0.547?? 0.225 0.478?? 0.201 -0.110 0.523 -0.820 0.653 0.376 0.778
P Table shows tax competition effects (coefficients of j wij τj,t−1 ) and corresponding standard errors (adjusted to heteroscedasticity and serial correlation) for estimations using 10-year windows as indicated in leftmost column. Regressions include τi,t−1 , a full series of period effects, and control variables as reported in Table 5 (results not shown). Estimator is 2SLS (Anderson-Hsiao). ? 10% significance level. ?? Idem., 5%.
for country size, the difference between periods of more and less intense tax competition is somewhat more pronounced. In any case, the emerging picture of the evolution of tax competition over time corresponds in an interesting way to the descriptive analysis, showing that the almost steady decline of corporate tax rates in Europe since the mid-eighties was interrupted by a period of more or less constant average tax rates in the mid-nineties. Thus, the decomposition of the tax competition effect provides further evidence suggesting that tax competition has significantly contributed to the decline of corporate taxes in the past 25 years.
Table 7: Tax competition with uniform and population-based weights (1982-2006) Spatial weights Dependent variable, τit τi,t−1 P
Uniform STR (1) 0.894??? (0.089) -0.150 (0.318)
EMTR (2) 0.856??? (0.130) -0.373 (0.312)
EATR (3) 0.927??? (0.104) -0.309 (0.229)
Population STR EMTR (4) (5) 0.893??? 0.853??? (0.087) (0.133) -0.159 -0.384 (0.342) (0.348)
EATR (6) 0.921??? (0.103) -0.325 (0.240)
Sample includes 484 observations (years 1982-2006, 31 countries). Estimator is 2SLS on differenced model (Anderson-Hsiao). Standard errors (robust to heteroscedasticity and serial correlation) in parentheses. Regressions include control variables as reported in Table 5 (results not shown). Level equations also include a full series of period effects. ??? 1% significance level.
The role of geographical distance
As discussed above, we deviate from part of the empirical literature on international tax competition by using distance-based weights. Table 7 shows estimations which contrast the results reported so far to those obtained for uniform and population-based weights. With uniform weights, τ−it resumes the role of the (unweighted) average of other countries’ taxes. With population based weights, other countries’ taxes are weighted by ln(pop j + 1), where pop is population in millions. Note that we cannot control for common period specific shocks when using uniform or population-based weights as τ−i,t−1 (at least in large samples) does not vary across countries. However, since we derive our estimates from the differenced equation, we implicitly account for country-specific linear time trends. We have also included all control variables as before (coefficients not shown). The results provide striking evidence for the importance of geographical distance in specifying the weighting schemes used to compute neighbors’ average taxes. In Table 7, the null of no tax competition effect on the choice of corporate taxes in Europe cannot be rejected in any of the six specifications. This contrasts to the results presented by Devereux, Lockwood, and Redoano (2008), who find strong competition effects using uniform weights. Again, the choice of a fully dynamic model of tax competition compared to a static model
of contemporaneous interaction has significant consequences.
Over the past 25 years, a significant downward trend in both statutory and effective tax rates can be observed. We have analyzed to what degree tax competition has contributed to this process. Our empirical analysis is based on a broad sample covering up to 32 European countries for the period from 1980 until 2007. We employ dynamic panel estimations in order to take account for the strong serial correlation of corporate tax rates. Our results differ from previous findings in several dimensions. Firstly, while we can show that governments strongly compete over statutory and effective average tax rates, we do not find interaction in effective marginal tax rates. This relates our results to different dimensions of tax competition and supports the view that countries compete for firms as well as paper profits rather than for marginal investments. Furthermore, our results suggest that, once the dependence of tax rates over time and the impact of tax competition is controlled for, often discussed country characteristics such as population size, GDP or openness as potential driving forces of corporate tax policies do not contribute much to explaining actual levels of the relevant tax rates. Thirdly, we find that the evidence regarding the presence of tax competition depends on how the interaction among countries is actually modeled. In particular, our results regarding the presence of tax competition do only hold once we assume that countries react to geographically close neighbors more strongly than to more distant countries.
References Altshuler, R., and T. J. Goodspeed (2003): “Follow the leader? Evidence on European and U.S. tax competition,” mimeo, Rutgers University, Hunter College, and CUNY Graduate Center. Anderson, T., and C. Hsiao (1981): “Estimation of Dynamic Models with Error Components,” Journal of the American Statistical Association, 76, 598–606. Arellano, M., and S. Bond (1991): “Some tests of specification for panel data: Monte Carlo evidence and an application to employment equations,” Review of Economic Studies, 58, 277–297. Baicker, K. (2005): “The spillover effects of state spending,” Journal of Public Economics, 89, 529–544. Blonigen, B. A. (2005): “A review of the empirical literature on FDI determinants,” Atlantic Economic Journal, 33, 383–403. Blonigen, B. A., R. B. Davies, and K. Head (2003): “Estimating the knowledge-capital model of the multinational enterprise: comment,” American Economic Review, 93, 980–994. Brainard, S. L. (1997): “An Empirical Assessment of the ProximityConcentration Trade-Off Between Multinational Sales and Trade,” American Economic Review, 87, 520–544. Bretschger, L., and F. Hettich (2002): “Globalisation, capital mobility and tax competition: theory and evidence from OECD countries,” , European Journal of Political Economy, 18, 695–716. (2005): “Globalization and International Tax Competition: Empirical Evidence Based on Effective Tax Rates,” Journal of Economic Integration, 20, 530–542. Bruno, G. S. (2005): “Approximating the Bias of the LSDV Estimator for Dynamic Unbalanced Panel Data Models,” Economics Letters, 87, 361–366. ¨ ttner, T., and M. Ruf (2007): “Tax incentives and the location of FDI: Bu Evidence from a panel of German multinationals,” International Tax and Public Finance, 14, 151–164. Carr, D. L., J. R. Markusen, and K. E. Maskus (2001): “Estimating the knowledge-capital model of the multinational enterprise,” American Economic Review, 91, 693–708.
Chinn, M. D., and H. Ito (2002): “Capital account liberalization, institutions and financial development: cross country evidence,” NBER Working Paper No. 8967. De Mooij, R. A., and S. Ederveen (2006): “What a difference does it make? Understanding the empirical literature on taxation and international capital flows,” European Commission Economic Papers No. 261. Desai, M., F. Foley, and J. R. Hines (2001): “Repatriation taxes and dividend distortions,” National Tax Journal, 54, 829–851. Devereux, M. P. (2006): “The impact of taxation on the location of capital, firms and profit: A survey of empirical evidence,” OUCBT Working Paper 07/02. Devereux, M. P., and R. Griffith (1998): “Taxes and the location of production: Evidence from a panel of US multinationals,” Journal of Public Economics, 68, 335–367. (2003): “Evaluating tax policy for location decisions,” International Tax and Public Finance, 10, 107–126. Devereux, M. P., and B. Lockwood (2006): “Taxes and the size of the foreign-owned capital stock: which tax rate matters?,” paper presented at the 2006 meeting of the European Tax Policy Forum, London. Devereux, M. P., B. Lockwood, and M. Redoano (2008): “Do countries compete over corporate tax rates?,” forthcoming in: Journal of Public Economics. Dreher, A. (2006): “The influence of globalization on taxes and social policy: An empirical analysis for OECD countries,” European Journal of Political Economy, 22, 179–201. Egger, P., M. Pfaffermayr, and H. Winner (2007): “Competition in corporate and personal income taxation,” paper presented at the 2007 meeting of the European Tax Policy Forum, London. European Commission (2001): “Company taxation in the internal market,” Commission Staff Working Paper COM (2001), 582 final, Luxembourg. Garret, G. (1998): “Global markets and national tax policies: collision course or virtuous circle,” International Organization, 52, 787–824. Garretsen, H., and J. Peeters (2007): “Capital Mobility, Agglomeration and Corporate Tax Rates: Is the Race to the Bottom for Real?,” CESifo Economic Studies, 53, 263–293. 34
Ghinamo, M., P. Panteghini, and F. Revelli (2007): “FDI determination and corporate tax competition in a volatile world,” CESifo Working Paper No. 1965. Gresik, T. A. (2001): “The taxing task of taxing transnationals,” Journal of Economic Literature, 39, 800–838. Haufler, A., A. Klemm, and G. Schjelderup (2006): “Globalization and the mix of wage and profit taxes,” CESifo Working Paper No. 1678. Hayashi, M., and R. Boadway (2000): “An empirical analysis of intergovernmental tax interaction: The case of business income taxes in Canada,” Canadian Journal of Economics, 34, 481–503. Hines, J. R. (1999): “Lessons from behavioral responses to international taxation,” National Tax Journal, 52, 305–322. Hines, J. R., and E. M. Rice (1994): “Fiscal paradise: Foreign tax havens and American business,” Quarterly Journal of Economics, 109, 149–182. Huizinga, H., and L. Laeven (2008): “International profit shifting within European multinationals,” forthcoming in: Journal of Public Economics. Markusen, J. R. (2002): Multinational firms and the theory of international trade. MIT Press, Cambridge (Mass.). Mendoza, E. G., A. Razin, and L. Tesar (1994): “Effective tax rates in macroeconomics: Cross-country estimates of tax rates on factor incomes and consumption,” Journal of Monetary Economics, 34, 297–323. Mutti, J., and H. Grubert (2004): “Empirical Asymmetries in Foreign Direct Investment and Taxation,” Journal of International Economics, 62, 337–358. Portes, R., and H. Rey (2005): “The Determinants of Cross-Border Equity Flows,” Journal of International Economics, 65, 269–296. Quinn, D. (1997): “The correlates of change in international financial regulation,” American Political Science Review, 91, 531–552. Redoano, M. (2007): “Fiscal interaction among European countries. Does the EU matter?,” CESifo Working Paper No. 1952. Rodrik, D. (1997): “Trade, social insurance, and the limits to globalization,” NBER Working Paper No. 5905. Sachs, J. D., and A. Warner (1995): “Economic Reform and the Process of Global Integration,” Brooking Papers on Economic Activity, 1, 1–118. 35
Schwarz, P. (2007): “Does Capital Mobility Reduce the Corporate-Labor Tax Ratio?,” Public Choice, 130, 363–380. Slemrod, J. (2004): “Are corporate tax rates, or countries, converging?,” Journal of Public Economics, 88, 1169–1186. Swank, D. (1998): “Funding the welfare state: globalization and the taxation of business in advanced market economies,” Political Studies, 46, 671–692. Swank, D., and S. Steinmo (2002): “The new political economy of taxation in advanced capitalist democracies,” American Journal of Political Science, 46, 642–655. Wilson, J. D. (1986): “A theory of interregional tax competition,” Journal of Urban Economics, 19, 296–315. Windmeijer, F. (2005): “A Finite Sample Correction for the Variance of Linear Efficient Two-Step GMM Estimators,” Journal of Econometrics, 126, 25–51. Winner, H. (2005): “Has tax competition emerged in OECD countries? Evidence from panel data,” International Tax and Public Finance, 12, 667– 687. Zodrow, G. R., and P. Mieszkowski (1986): “Pigou, Tiebout, property taxation, and the underprovision of local public goods,” Journal of Urban Economics, 19, 356–370.