CESifo Economic Studies

CESifo Economic Studies Advance Access first published online on May 22, 2007
This version published online on June 4, 2007
CESifo Economic Studies, doi:10.1093/cesifo/ifm009

This Article Abstract FREE Full Text (PDF) All Versions of this Article: 53/2/263    most recent ifm009v2 ifm009v1 Alert me when this article is cited Alert me if a correction is posted Services Email this article to a friend Similar articles in this journal Alert me to new issues of the journal Add to My Personal Archive Download to citation manager Request Permissions Google Scholar Articles by Garretsen, H. Articles by Peeters, J. Search for Related Content Social Bookmarking          What's this?

© The Author 2007. Published by Oxford University Press on behalf of Ifo Institute for Economic Research, Munich. All rights reserved. For permissions, please e-mail: journals.permissions@oxfordjournals.org

Capital Mobility, Agglomeration and Corporate Tax Rates: Is the Race to the Bottom for Real?

Harry Garretsen and Jolanda Peeters¹

	Abstract

Top Abstract 1 Introduction 2 Theoretical background and... 3 Data set 4 Estimation results 5 Conclusions Appendix: The interpretation of... Appendix: Additional estimations Data Appendix References

 
Based on a data set for 19 OECD countries for the period 1981–2001, we estimate the impact of FDI on corporate tax rates, where changes in FDI are a measure for changes in capital mobility. So far the literature has been concerned with the related but rather different question as to the sensitivity of FDI to tax rates. Our article takes an opposite perspective and asks what the impact of capital mobility is on corporate tax rates. In doing so, we explicitly take the role of agglomeration into account. In theory, core countries can afford a higher tax rate compared to peripheral countries. In our estimation strategy, we instrument capital mobility to deal with reverse causality. The main conclusion is that increased international capital mobility, measured by FDI flows, implies a lower corporate tax rate. But we also find that agglomeration matters: core countries have a higher corporate tax rate than peripheral countries. If there is a race to the bottom, it seems that it is more real for some countries than others. (JEL code: H25)

Key Words: corporate taxes • capital mobility • agglomeration • new economic geography

	1 Introduction

Top Abstract 1 Introduction 2 Theoretical background and... 3 Data set 4 Estimation results 5 Conclusions Appendix: The interpretation of... Appendix: Additional estimations Data Appendix References

 
In recent years capital has become increasingly mobile internationally: most of the remaining capital controls and restrictions on, for instance, the activities of multinational firms have been removed by now. One side effect of the increased international capital mobility is that it provides opportunities for multinational firms to minimize or even avoid paying corporate income taxes. In this setting, many people fear that globalization, and in particular increased capital mobility, forces national governments to decrease the tax burden on the mobile production factor, capital. In doing so, tax competition across governments is intensified. This may result into a "race to the bottom" in corporate income taxes. Some critics even go a step further and believe that ultimately the redistributive function of the welfare state is threatened by footloose capital.

These concerns are, at least to some extent, fuelled by the actual development of the most obvious and readily available measure of corporate income taxes, the statutory tax rate. In recent years almost every OECD country has reduced the statutory tax rates. Also economic theory shows that the concern for a race to the bottom is not unfounded. A review of the tax competition literature leads to the main result that (under some restrictive assumptions) an increase in capital mobility could go along with a decrease of the corporate tax rate and an under-provision of public goods. That is not to say that a race to the bottom is inevitable though. Several other theoretical contributions relax one or more restrictive assumptions of the basic model and reach different or even opposite conclusions. In particular, when countries are asymmetric, because of agglomeration or clustering of economic activity, an increase in capital mobility could instead result into a "race to the top" (Baldwin and Krugman 2004).

Given these opposite predictions from theory, we agree with Krogstrup (2003, 2004a, 2005) that the question how capital mobility affects corporate income taxation is ultimately an empirical one. The aim of this article is therefore to investigate the impact of capital mobility on corporate tax rates. In doing so, we take FDI as our—admittedly imperfect—indicator for capital mobility. Only a limited number of studies investigate the impact of FDI on tax rates, most of the literature is concerned with the related but rather different issue as to the sensitivity of FDI to tax rates. This article contributes to the literature in several ways. First, we investigate explicitly the role of agglomeration effects in the determination of a country's tax rate. The basic idea is that higher capital mobility may lead to agglomeration of economic activity; and this may go along with a higher corporate tax rate. Second, several indicators for agglomeration are used: varying the distance decay and including a "real" measure of trade costs improves on the existing analysis of agglomeration effects. Finally, because there is a potential source of endogeneity in these types of regressions, we use an instrumental variable approach, in which the so-called Golub-index on capital restrictions will serve as an instrument for international capital mobility. Our first main conclusion is that increased international capital mobility, approximated by FDI flows, puts a downward pressure on corporate tax rates. This conclusion also holds when agglomeration variables are added to our empirical model. Our second main conclusion is that agglomeration does matter: we find evidence that more centrally located or agglomerated countries have a higher corporate tax rate.

The article is organized as follows. We start in section 2 with a brief discussion of the background literature and then introduce the main hypotheses that will subsequently be tested in the empirical part of the article. Section 3 introduces our data set and provides some descriptive statistics for our main variables of interest. Section 4 discusses estimation issues (notably the endogeneity of capital mobility), presents the empirical model and discusses our main findings. Finally, section 5 concludes.

	2 Theoretical background and main hypotheses

Top Abstract 1 Introduction 2 Theoretical background and... 3 Data set 4 Estimation results 5 Conclusions Appendix: The interpretation of... Appendix: Additional estimations Data Appendix References

 
Fears about the negative effects of increased capital mobility on the capital tax rate and the provision of public expenditures can be given a theoretical foundation. Using the standard tax competition literature, it is quite easy to show that an increase in capital mobility could go along with a decrease of the source based capital tax rate and an under-provision of public goods. The seminal contribution here is the model by Zodrow and Mierzkowski (1986); see also the survey paper by Wilson (1999). Since we have next to nothing to add to the literature on tax competition, we restrict ourselves here to a few observations that have a bearing on our empirical analysis. The standard tax competition model is based on some rather strong assumptions (see also Krogstrup 2002; Sørensen 2006). When one allows for instance for international tax policy coordination, tax revenues (Baldwin and Forslid 2002), foreign ownership (Huizinga and Nicodème 2003) or, more generally, asymmetries between countries to play a role, it is not clear why a race to the bottom should result. Asymmetries between countries may arise for various reasons, but here we focus on the implications of differences in country size that are due to agglomeration effects for the relationship between capital mobility and tax rates. As shown extensively in the so-called new economic geography (NEG) literature, see Baldwin et al. (2003, chapter 15) and Krogstrup (2004b), agglomeration leads to a so-called agglomeration rent and the existence of such an agglomeration rent implies that larger countries, that is the country where (most) capital is located, can allow themselves to have a relatively higher tax rate. In such a model and even with full capital mobility, a race to the bottom will not materialize.² Agglomeration is, however, not the only equilibrium outcome in such a NEG model. Typically, it is only when trade or transportation costs fall below a certain threshold level that agglomeration results. When trade costs are above this threshold level, a spreading equilibrium occurs where the mobile factors of production—here: capital—are evenly spread across locations. In that case, the NEG model leads to a similar conclusion as the standard tax competition model with respect to the effects of capital mobility on capital tax rates (see also Figure 1 in Sørensen 2006).

View larger version (18K): [in this window] [in a new window] [Download PowerPoint slide]   Figure 1 Descriptive statistics EATR

 
The main message of the NEG literature is that the relationship between capital mobility and capital tax rates is more complicated than the standard tax competition model predicts. In particular, and assuming that trade costs are not too high, this relationship will depend on the degree of agglomeration. The NEG view on capital mobility and tax competition is not merely a theoretical curiosity. The stylized facts on capital mobility and capital tax rates are not or, at least, not always in accordance with the notion of a race to the bottom as predicted by the standard tax competition model. For the case of the EU countries, Baldwin and Krugman (2004) even argue that a race to the top seems to be more in line with the facts. There is no conclusive evidence, however, to suggest that this has indeed happened. What is clear is that for almost every OECD country, statutory income tax rates have come down from the 1980s onwards. This decline in statutory tax rates has, however, been accompanied by a broadening of tax bases. As a result, other measures of corporate taxes show a rather blurred picture with respect to their development over time. Devereux, Griffith and Klemm (2002) conclude, for instance, that effective marginal rates have remained rather stable. This mixed empirical evidence suggests that the relationship between capital mobility and tax rates might be more complicated than the standard tax competition model suggests. It is for this reason that the main objective of our empirical inquiry will be to estimate the impact of capital mobility on capital tax rates while controlling for agglomeration effects that are stressed by the NEG literature.

There is no shortage of empirical studies on the relationship between FDI and capital tax rates (Hines 1999). These studies typically find that lower corporate tax rates, be it statutory/effective or marginal/average corporate tax rates, are associated with higher levels of FDI. But note that these studies try to establish whether the corporate tax rate has an effect on FDI.³ The implied causality runs from tax rates to FDI. The tax competition literature, and also the "race to the bottom" debate are, however, basically concerned with the opposite question: what is the effect of increased capital mobility—measured by increased FDI—on tax rates? This question is also the starting point for the present article and in line with Krogstrup (2004a, pp. 5–6) we observe that the abovementioned finding of a significant negative tax elasticity of FDI is a necessary but not sufficient condition for the empirical analysis of tax competition. This makes clear that any empirical analysis of the effects of capital mobility on the tax rate has to take the problem of reverse causality into account. In section 4, we will explain how we dealt with this crucial issue. The main hypotheses that follow from our brief overview of the literature are as follows:

• An increase in the degree of capital mobility will lead to a decrease of corporate tax rates (standard tax competition effect)
  
• Agglomeration matters: more agglomerated countries have higher corporate tax rates (agglomeration effect)
  
• The effect of capital mobility on corporate tax rates may vary across time or subgroups (core versus periphery); the level of trade costs will be important here.
  

To date there are still very few studies that try to assess the effect of changes in capital mobility on corporate tax rates and that, in doing so, take agglomeration effects into account. Krogstrup (2003, 2004a) is a notable exception. Our study not only differs from her work in terms of the sample and methodology employed but also in terms of the hypotheses tested. Krogstrup (2003, 2004a) also tests whether there is a downward pressure of capital mobility on corporate tax rates (our first hypothesis) but pays less attention to agglomeration effects and the relevance of trade costs (our second and third hypothesis).

	3 Data set

Top Abstract 1 Introduction 2 Theoretical background and... 3 Data set 4 Estimation results 5 Conclusions Appendix: The interpretation of... Appendix: Additional estimations Data Appendix References

 
The sample used for the panel regression analysis includes annual data for 19 OECD countries and covers the 1981–2001 period. A more detailed description of the full data set, including the control variables and the countries included in our sample, is given in the data appendix. In this section, the attention is on the main variables of our empirical analysis. The dependent variable in our empirical inquiry is the corporate income tax rate. Measures of the corporate income tax rate differ widely in the literature.⁴ The most obvious and readily available measure is the statutory tax rate, i.e. the tax rate which can be directly constructed from tax codes. This can be a misleading indicator, however, because it pays no attention to the tax base. To deal with this problem, several studies use the average tax rate as a measure for the corporate income tax rate. The main disadvantage of this approach is that it is an ex post measure and as such it is not the appropriate decision variable of the government. Because governments can decide about both the tax rate and the tax base, effective tax burdens are to be preferred in our analysis. Effective tax rates are computed from tax codes and measure the tax burden on a hypothetical corporate investment project using tax legislation. When evaluating tax competition pressures empirically, the average effective corporate tax rate is the most appropriate measure, because it is this tax rate rather than the marginal one which ultimately matters for the location decision of a firm (Gorter and De Mooij 2001).

For these reasons we use effective average tax rates (EATR) in our empirical analysis. Devereux, Griffith and Klemm (2002) have computed the (base case) EATR for projects earning positive economic profits. Figure 1 shows for the 19 countries in our sample the unweighted annual average EATR and the cross-country standard deviation. It follows that the EATR fell from around 40 percent in 1981 to around 28 percent in 2001 and that the EATR has converged across countries over the past two decades. The latter is reflected in a gradually decline of the standard deviation. Despite this convergence, there are still significant differences between countries: in Ireland the EATR equalled 7.7 percent in 2001, while in Japan it amounted to 35.8 percent (these findings are confirmed by Devereux and Sorensen 2006, see their Figures 7 and 8).

Our main explanatory variable is the degree of international capital mobility and here, as with the corporate income tax rate, finding an appropriate indicator to measure capital mobility is difficult. Various variables are used in the literature as proxies for the degree of international capital mobility. First, under the assumption that increases in capital mobility result in increases in cross border investment, Swank (1998) and Garrett and Mitchell (2001) use the volume of FDI stocks or flows as a measure of capital mobility.⁵ In fact, this is a quite standard way of measuring the degree of capital market integration and it is analogous to measuring goods market integration with trade volumes. In the taxation-capital mobility literature to which our article belongs, FDI is often used as indicator for capital mobility. But there are obviously other yardsticks available to measure capital mobility. Following O’Rourke and Williamson (1999) one can also use a price approach (by means of covered interest rate differentials) or a macro approach (the correlation between national savings and investment following Feldstein and Horioka 1980). But both these alternatives to FDI are less useful for our present purposes. The tax competition literature is not concerned with the relationship between the (re-) location of portfolio investment and taxes but primarily with the relationship between real or "bricks and mortar" capital and taxation. And this part of international capital mobility is best summarized by FDI, which is probably why many other empirical papers in this literature also choose FDI as an indicator of capital mobility. Another approach to measuring capital mobility is followed by Krogstrup (2003) and Swank (1998). These studies use indices of the liberalization of capital markets based on the legal framework governing international capital transactions. In fact, all measures of international capital mobility suffer from conceptual or practical problems, see subsequently, which makes it impossible to find a perfect measure. Since the aim of our article is not to come up with a new measure of capital mobility we simply follow the literature and include FDI as our main indicator of capital mobility.

Actually, two measures of international capital mobility are used in the present study. First, FDI is thus used. More precisely, we use the sum of inward and outward FDI flows as a percentage of gross fixed capital formation (GFCF). Our choice to express the FDI flows in terms of GFCF is motivated by the availability of the data: the UNCTAD provides these data directly at their website. Moreover, it turns out that—apart from level differences—it makes no qualitative difference whether we express FDI flows in terms of GFCF or in terms of GDP, as both measures show a similar development over time.⁶ Figure 2 depicts the development of our (unweighted) measure of capital mobility. It confirms the general trend that international capital markets have become much more integrated in recent decades. At the same time, this measure shows that the standard deviation has increased over time, suggesting that the dispersion in the degree of international capital mobility has increased.⁷

View larger version (14K): [in this window] [in a new window] [Download PowerPoint slide]   Figure 2 Descriptive statistics FDI flows

 
Second, we use the so-called Golub-index as a measure for the (legal) restrictions placed on international capital mobility. This index was assembled by the OECD to summarize and quantify statutory barriers to inward FDI and their evolution over time. The main reason to use this index is that it is thought to be correlated with international capital flows (here, FDI), but at the same time it is unlikely that the Golub-index is determined by the corporate tax rate. In fact, in our empirical analysis we tested for this and it turns out that the impact of the Golub-index on the corporate income tax rate, EATR, is an indirect one, that is to say it is only through FDI that the Golub-index has an effect on EATR. This implies that this index qualifies as an instrument. The development of the Golub-index, depicted in Figure 3, shows that the liberalization of FDI flows has been substantial over the past two decades for most countries, except the United States and Japan (which had relatively low statutory restrictions to begin with). There are, however, significant differences between countries (a higher score on this index means more restrictions on FDI flows). The most open countries are in Europe, while—according to this index—Canada and Australia are the countries with the highest levels of restrictions. The development of the standard deviation indicates that the dispersion in the degree of openness to inward FDI flows has narrowed (data not shown). A final word on the usefulness of FDI flows as a measure for international capital mobility. As our referee was correct in pointing out, in the time dimension increased FDI may indeed signal increased capital mobility but this is not obvious for the cross-section dimension. In fact, we know from the empirical literature on the determinants of FDI (see for instance Barba Navaretti and Venables 2004, chapter 6) that cross-country variations in FDI may be due to various factors that are wholly unrelated to cross-country differences in capital mobility. Since we estimate a panel, this is a reminder that although FDI is often used, it is at best an imperfect indicator of capital mobility. For our sample, we found that the sample variance (and standard deviation) is much larger across the time dimension than across the cross-section dimension.

View larger version (20K): [in this window] [in a new window] [Download PowerPoint slide]   Figure 3 Descriptive statistics Golub-index

 
Finally, in order to be able to test for the relevance of agglomeration effects, a measure of agglomeration is needed. The empirical literature on the new economic geography suggests many variables which could function as a proxy for agglomeration. We use several versions of a market potential function as representing the agglomeration effect. The main reason to do so is that the market potential variable provides information on (relative) location by including distance and, as such, really touches upon the main idea of the new economic geography that relative location matters. The typical market potential function measures the potential of a location as a weighted sum of the purchasing power (GDP) of all other locations, with the weights being a declining function of distance: , where i and j are country indices, D_ij is the distance (in kilometres) between countries i and j and is the distance decay parameter. We add two kinds of adjustments to the typical market potential function, so that we end up with six versions of the basic market potential variable. First, we vary the so-called distance decay parameter . Even though = 1 is quite often found in market potential studies or related gravity studies of international trade (Head and Mayer 2000), there are also estimates that actually estimate the distance coefficients in a full-blown NEG model. These studies suggest that may differ significantly from one (see Crozet 2004 or Brakman, Garretsen and Schramm 2005). By varying we can vary the "punishment" countries face for being relatively isolated. To see this, note that when is set at 0.5, relatively remote countries are punished less for being relatively isolated, whereas when is set at 2 relatively remote countries are punished more for being relatively isolated. As our second adjustment, we correct the typical market potential function of country i with a measure of trade costs, namely the cif/fob ratio: TC_i = (cif/fob)_i * MP_i. The idea is that in NEG models the agglomeration effect is typically defined by the combination or interaction of some market access variable (here, MP) with transportation or trade costs.

Distances are drawn from the CEPII database. In this data set, distances are calculated following the great circle formula, which uses latitudes and longitudes of the most important agglomerations (in terms of population). The distances incorporate internal distances as well, which implies that the average distance between producers and consumers within a country is taken into account. By incorporating these internal distances we are able to include the GDP of the respective country as well in the market potential. For internal distance we use the proxy 0.667 in which, area is the size of a country in square kilometres (see Head and Mayer 2000 for a discussion of this measure for internal distance). The cif/fob ratios are calculated from the Direction of Trade Statistics.⁸ Because importing countries report the value of imports from partner countries inclusive of carriage, insurance and freight (cif) and exporting countries report their value free on board (fob), we first calculated the cif/fob ratio between each pair of countries: t_ij = cif_ij/fob_ij. Next, we averaged for each country the resulting 18 cif/fob ratios: (cif/fob)_i = t_ij/18. Figure 4 depicts the development of our (unweighted) measure of trade cost. It clearly shows that since the 1980s trade costs between countries have fallen.

View larger version (19K): [in this window] [in a new window] [Download PowerPoint slide]   Figure 4 Descriptive statistics cif/fob ratio.

 
Table 1 gives the ranking of the countries for two of our market potential variables in 2 years. Also this table includes the order of ranking of countries’ size uncorrected for distance (measured simply by GDP). For all variables the ranking is in ascending order of magnitude (least agglomerated/smallest countries are ranked first in the table). Several conclusions follow. First, distance, and hence location, is important, as it crucially affects the ranking of countries: the United States and Japan are clearly the largest countries in terms of GDP, but when corrected for distance, Belgium and the Netherlands turn out to be the most agglomerated countries, thanks to the proximity of these two countries to other countries with a relatively high GDP.⁹ Second, Table 1 shows that the distance decay parameter is potentially a relevant parameter, as it crucially affects the ranking of countries. Consider for instance the United States (or Canada): if distance matters more ( rises from 0.5 to 2), the lower is its ranking of the market potential. Finally, Table 1 shows, unsurprisingly, that the ranking of the countries’ market potential is relatively stable over time.

View this table: [in this window] [in a new window]   Table 1 Ranking for several agglomeration measures

 
Not shown in the table is the ranking for our preferred measure of the tax rate, EATR. To get a first idea about the relation between the market potential variables and this tax rate, we computed Spearman's correlation coefficients. It turns out that the correlation between the EATR and the various market potential variables in 1981 was weakly negative, whereas in 2001 the correlation is strongly positive. This can be explained by the fact that the degree of capital mobility was fairly low in 1981, but high in 2001. The positive correlation coefficient between the EATR and the various market potential variables is a first sign of the agglomeration effect.

	4 Estimation results

Top Abstract 1 Introduction 2 Theoretical background and... 3 Data set 4 Estimation results 5 Conclusions Appendix: The interpretation of... Appendix: Additional estimations Data Appendix References

 
4.1 Specification and estimation strategy
Before we present our main estimation results, we first explain our estimation strategy. In general terms and in line with section 2, the equation to be estimated is¹⁰:

(1)

where i and t are the country- and time-index, and where TAXRATE_i,t, FDIFLOWS_i,t, and NEG_i,t stand for the corporate income tax rate, capital mobility, and the new economic Geography (NEG) or agglomeration proxies, respectively, and X_i,t is a group of control variables.

As we explained in the previous section, the EATR is our preferred measure for TAXRATE but we have experimented with other corporate tax burden variables as well. Similarly, our preferred measure for capital mobility is the sum of a country's FDI inflow and outflow scaled by gross fixed capital formation (FDIFLOWS).¹¹ Here too, we have experimented with alternatives (only in- or outflows; FDI stocks versus flows). Given our emphasis on agglomeration, several forms of the market potential are included as NEG variables. The first three market potential variables are assumed to differ only in the distance decay parameter , with MP1 assuming that = 1; MPH assuming that = 0.5 and MPD assuming that = 2. In addition, a final version of the market potential variable (TC1) includes a direct measure of trade costs, namely the cif/fob ratio while setting the distance decay parameter at 1. Finally, for the set of control variables X we opted for:

• SIZE, the GDP of country i as a share of total GDP in our sample;
  
• NEIGHBOUR, the average statutory corporate tax rate in all other countries;
  
• LEFT, left party cabinet portfolios as percent of all cabinet portfolios;
  
• GOVINV, fixed government investment as a percent of total disbursements.
  

Out of a larger group of control variables, we picked these control variables for our basic specification because each of these variables represents a different channel of influence on the corporate tax rate. The size variable is fairly straightforward: even the standard tax competition literature predicts that larger countries have higher equilibrium tax rates. The (statutory) tax rate from other countries is included because countries might react to changes in each other's tax rates. This simple notion of tax competition implies that we expect a decrease of NEIGHBOUR to lead to a decrease of the tax rate in country i. The political determinants of the corporate tax rate are proxied by LEFT, the idea being that left-wing governments ceteris paribus are characterized by higher corporate tax rates. Finally, GOVINV is included because it captures directly the impact of government actions on the attractiveness of a location. The more the government invests, the higher the corporate tax rate. Underlying is a notion that governments ceteris paribus react to changes in FDI flows in the sense that more capital mobility inclines them to (effectively) lower corporate income tax rates. National governments set the (statutory) tax rates and determine the (legal) tax base so actual changes in EATR (partially) reflect government decisions.

Keeping in mind the theoretical discussion in section 2, one immediate problem with any attempt to estimate (1) is that our variable FDIFLOWS is endogenous; in particular we must take into account that FDI may be determined by the corporate tax rate. To deal with this, we instrumented FDIFLOWS with the Golub-index. Our maintained hypothesis is that changes in the Golub-index have a direct impact on the degree of capital mobility but not on the EATR. We tested this hypothesis and this indeed turned out to be the case. The possible impact of a change of legal restrictions on EATR will therefore be assumed to be indirect, through FDI. In our view this is a reasonable assumption to make: policy makers change corporate tax rates or tax bases not because of changes in legal restrictions on international capital mobility but they do so only if the actual capital flows change. The Golub-index seems therefore a useful instrument. It also turned out to be a significant instrument. In our 2SLS estimations the first stage regression results (see appendix with additional estimations) invariably showed that fewer legal restrictions on (inward) FDI (a decrease of the Golub-index) leads to increased capital mobility.¹² Based on the empirical research on the determinants of FDI (Barba Navaretti and Venables 2004), we experimented with other possible instruments for FDI as well, like the trade to GDP ratio, but these instruments were (mostly) not significant, so we did not include them in our final specification. We also, again see the appendix with additional estimations for an example, tested whether we were right to go for an IV-approach (as opposed to OLS) to start with. The corresponding Hausman tests show that the OLS models are misspecified.

We not only instrumented FDIFLOWS, but we also used one-period lagged FDIFLOWS in our estimations. If policy makers react to changes in the degree of capital mobility by changing corporate tax rates we take it that this reaction will not be instantaneous. Our sample period is not very long (annual data for 1981–2001) and when doing subsample estimations the sample period is even shortened further. With such a relatively short time-series, tests for (co-)integration are of limited use.¹³ In our view, and different from Krogstrup (2004a), it is to be preferred to estimate (1) in levels. That is to say, it is our assumption that the hypothesis that follows from the tax competition literature as discussed in section 2 is that (policy induced) changes in the level of the corporate tax rate are a result of changes in the level of capital mobility. What about the possible endogeneity of the control variables X_i,t? In our view this is less of a concern than in the case of FDI. First, as to the control variables we are not interested in causation but merely in establishing correlation with EATR. Second, valid instruments for LEFT, GOVINV or NEIGHBOUR are difficult to find, so we re-ran all our regressions with all explanatory variables lagged for one period. This (imperfect but often used) way to control for endogeneity did not change our results. Third, as the referee also pointed out, endogeneity may be a particular concern for the control variable NEIGHBOUR, the average statutory corporate tax rate in all other countries. We made sure that our main results did not change when we excluded this variable.

Finally, in addition to the variables introduced earlier we also included fixed effects in our pooled 2SLS estimation of to take care of country-specific effects. There are two main reasons to opt for fixed effects. The first is a theoretical one. Apart from our set of explanatory variables, one can think of various (hard to measure) reasons why countries differ in their corporate tax rates: the power distribution between labour and capital (e.g. the relevance of trade unions), the fiscal regime (is the country a tax haven), (political) preferences as to (changes in) direct or indirect taxes, the effectiveness of tax collecting, etc. The second, and more important, reason is econometrical. Instead of fixed effects one could opt for random effects. We re-ran all our estimations using random instead of fixed effects. This turned out to be immaterial for our main conclusions (certainly w.r.t. the FDI and MP variables). For our main specification, Table 3 (column I) in the next section, the appendix with additional estimations shows the estimation results when random effects are used. More systematically, we tested whether fixed or random effects should be preferred by including the residuals from the fixed (random) effects estimation of (1) as explanatory variable in the random (fixed) effects estimations. The difference between fixed and random effects turns out to be small. Since it is rather unusual (and confusing) to switch back and forth between fixed and random effects throughout the analysis and since the choice does not affect our main conclusion we stick to fixed effects in the main text.

View this table: [in this window] [in a new window]   Table 3 Adding agglomeration, full sample

 
4.2 Do FDI flows put a downward pressure on corporate tax rates?
The first hypothesis to be tested is whether an increase in FDI, our approximation of capital mobility, will lead to a decrease of corporate tax rates (the standard tax competition effect). Or more specifically, does an increase of FDIFLOWS, instrumented by the Golub- index, lead to decrease of the EATR?

From Table 2 it is clear that we find some confirmation for the hypothesis that increased capital mobility puts a downward pressure on the corporate tax rate. Without the control variables (column I), the estimation results imply that a 1 percent increase in our measure of capital mobility leads to a 0.5 percent decrease of the effective corporate tax rate. The effect is somewhat smaller when we add the control variables (column II). Note that the control variables are all significant and have the expected sign. Briefly stated, they suggest that the corporate tax rate in a country is higher when the government is leftist; when the country is relatively large in terms of GDP, and when a country's level of government investment expenditures is higher. We also find evidence of tax competition taking place, as the tax rate in neighbouring countries turns out to have a positive effect on a country's own tax rate. The next question is whether and how this conclusion has to be changed when we add the agglomeration variables.

View this table: [in this window] [in a new window]   Table 2 Standard tax competition hypothesis, full sample

 
4.3 Adding NEG variables: is there an agglomeration effect?
As explained in section 2, the relationship between capital mobility and corporate income tax rates might be quite different when we allow for agglomeration effects. When we for instance split our sample in low and high density countries (where density is measured as population per square kilometres) and using the same specification as in column II of Table 2, we find that for the group of countries with a low population density the FDI coefficient is still significant, but this is no longer the case for the high density countries. Within-country population density is a rather crude measure of agglomeration (and we thus prefer other measures in our estimations), but this result suggests that it may also be interesting from an empirical point of view to take agglomeration effects into account. Table 3 gives the estimation results when we add our preferred agglomeration variable, various versions of the market potential variable.

The estimation results reported in Table 3 give rise to the following observations. First, the size and significance of FDIFLOWS is hardly affected by the inclusion of the market potential variables. Hence, even if one controls for agglomeration, capital mobility seems to exert a downward pressure on corporate tax rates. Also varying the distance parameter (columns I–III with respectively the distance decay at = 1 (MP1), = 0.5 (MPH) and = 2 (MPD), or correcting the market potential with trade costs (TC1 = (cif/fob trade costs)*MP1), see column IV, do not change this main finding. Second, the significance of the group of control variables is rather similar if compared with column II of Table 2. Third, and most important, there is some evidence that agglomeration matters to the extent that countries with a larger market potential or market access have a higher corporate tax rate because our market potential variable, irrespective of how it is defined exactly, is significant and positive in all specifications. Finally, we performed an extensive robustness analysis to check how sensitive the results are to changes in the definition of the main variables of interest and/or in the sample composition. It follows that our main results carry through when we use different measures of capital mobility (FDI stock or only FDI in- or outflows). However, when we split the sample into two subperiods, we find that the FDI coefficient is not significant in the 1980s, but is in the 1990s, while the market potential variable is hardly significant at all. These results are rather sensitive to the exact cut-off year to split the sample, though. When we restrict our sample so as to include EU countries only, we find that for the EU, the downward pressure of capital mobility on the tax rate is primarily counteracted by the size effect and somewhat less by the agglomeration effect.

4.4 Matching our empirical results with NEG theory
Because we reported the uncorrected estimation results for the MP variable—that is, the results without appropriate scaling—Table 3 hides important information regarding the ranking of the market potential coefficients across the several cases of distance decay. However, with appropriate scaling, it turns out that when distance matters more (a higher value of , hence MPD), the market potential coefficient is still relatively higher. Hence, the implied ranking that we arrive at for the estimation results in the last row of Table 3 is that¹⁴: MPH < MP1 < MPD. The question that now arises is whether this result is compatible with the NEG theory. To answer this we first have a closer look at the so-called Core-Periphery (CP) model that is the basis for the NEG models of tax competition like Baldwin and Krugman (2004). As is further explained in the appendix on the interpretation of the results for the market potential variable, the CP model predicts that the agglomeration rent will be the highest for low levels of trade costs (Baldwin et al. 2003, ch. 15). If one accepts that a lower distance parameter corresponds with lower trade costs, then a lower value for would thus imply a larger agglomeration rent and a higher tax rate (without mobile capital relocating to a less centrally located country with lower corporate taxes). The idea simply being that with a higher degree of agglomeration a country can have a higher corporate tax rate as compared to the case of less agglomeration. Hence, using the CP model, we expect that the MP-coefficient results in terms of the order of ranking would show that: MPD MP1 MPH. This ranking is, however, opposite to our estimation results in Table 3 (see Figure A1 in the appendix for a graphical illustration). Having said this, the second core model in the NEG literature (Puga 1999) predicts a different relationship between the agglomeration rent and trade costs. In this model, the agglomeration rent will be the highest for intermediate levels of trade costs, see Figure A2 in the Appendix. This is due to the fact that, when trade costs fall over time, the two country economy moves from spreading to agglomeration to renewed spreading. Our estimation results in Table 3 would be compatible with this model. Namely, if for instance with = 2 (high trade costs) the economy would be in the agglomeration range and for = 1 and = 0.5 (low trade costs) the economy would be in an equilibrium of renewed spreading.

View larger version (10K): [in this window] [in a new window] [Download PowerPoint slide]   Figure A1 Freeness of trade and agglomeration in CP model: the tomahawk.

 

View larger version (11K): [in this window] [in a new window] [Download PowerPoint slide]   Figure A2 Freeness of trade and agglomeration: the inverted U-curve.

 
The above line of reasoning is based on a thought-experiment whereby changes in the distance decay parameter are approximations of changes in the level of trade costs or, in NEG parlour, of the freeness of trade. Another and more realistic way to conduct the same experiment is by using the trade-costs-corrected-market-potential TC. It is more realistic since we can actually, through the cif/fob trade costs measure, track actual trade costs over time for every country in our sample. It is beyond dispute that these trade costs have fallen over time (Figure 4). We have thus estimated our model for the subsample 1990–99 as well and compare the "trade cost corrected market potential" coefficient in this period with the corresponding coefficient for the whole sample, the basic idea being that the 1990s are to be associated with lower trade costs. The CP model, see again Figure A1, would predict that because of the fall in trade costs, the TC coefficient will be higher (or in any case not lower) for the 1990s. A lower TC coefficient for the period 1990–99 could, however, be compatible with the other core NEG model (see again Figure A2). Table 4 summarizes the results.¹⁵ It follows that the market potential coefficient in our subsample (column II) is lower than the market potential coefficient for the full sample (column I). This is, once again, in accordance with the second core model of the NEG literature.

View this table: [in this window] [in a new window]   Table 4 Impact of falling trade costs

 
Let us finally examine the relation between capital mobility and the tax rate from a NEG perspective. The hypothesis that follows from the CP model is that the coefficient for FDIFLOWS would get smaller, or at least not larger, as trade costs fall and the chances increase that agglomeration becomes stronger. This is because in case of more agglomeration the downward pressure from capital mobility on the tax rate would ceteris paribus be weaker. In the second core NEG-model, see Figure A2, where the two country economy moves from spreading to agglomeration to renewed spreading for low enough trade costs, the correlation between the tax rate and capital mobility is less clear-cut. The estimation results in Table 4 (see second row of Table 4) can be aligned with both models: the FDI-coefficient in the period associated with low trade costs (1990–99) is lower than the FDI-coefficient for the full sample estimation. This suggests that in the 1990s agglomeration has become more important. In terms of both models this could be the case when the level of trade costs is such that the economy is in the agglomeration regime. Any definite statements on this matter are, however, not possible because in terms of Figures A1 and A2 because we simply do not know where exactly to position our estimation results on these curves! One can only exclude certain possibilities given our priors as to the change of trade costs and the market potential variables over time.

Note that the feedback from empirics to NEG theory is only meant as a first pass and certainly not the final word on this matter. Even though the hypotheses are taken from the core NEG models, it should be recalled that these are two country models and that we have simply assumed that the analytics of a two country model carry over to the multi-country case (see Brakman, Garretsen and Schramm (2006) on this (heroic) assumption). Moreover, it should be recalled that trade costs, measured by the cif/fob ratio, faces serious measurement problems (Hummels 1999). Also, our approach to vary with the distance decay variable is only an imperfect way to approximate falling trade costs. These issues certainly warrant further research.

	5 Conclusions

Top Abstract 1 Introduction 2 Theoretical background and... 3 Data set 4 Estimation results 5 Conclusions Appendix: The interpretation of... Appendix: Additional estimations Data Appendix References

 
The question how an increase in capital mobility affects corporate tax rates is ultimately an empirical one. This article has provided such an empirical investigation. Using a panel of 19 OECD countries for the period 1981–2001, we have examined whether (i) an increase in the degree of capital mobility leads to a decrease of corporate tax rates, where capital mobility is (imperfectly) approximated by FDI flows; (ii) more agglomerated countries can afford higher corporate tax rates; (iii) the effect of capital mobility on corporate tax rates varies across time or subgroups, where we focus on the role of trade costs.

We find that increased international capital mobility (FDI) has a negative effect on the corporate tax rate, which is consistent with the standard tax competition literature. This result is relatively robust to alternative specifications. Second, this downward pressure on tax rates is also present when agglomeration variables are added to our empirical model. But, and in accordance with the predictions of the NEG literature, we also find some evidence that agglomeration effects matter. Compared to more peripheral countries, core countries have a higher corporate tax rate. Finally, these main results hold for alternative measures of trade costs and the market potential. The introduction of agglomeration effects is the key contribution of the NEG literature to the "race to the bottom" debate on the impact of capital mobility on corporate taxation. Our results illustrate that it is, however, quite difficult to ground the empirical findings, as the relationship between agglomeration, capital mobility and taxation depends on a particular NEG model. Further research on this is certainly needed.

	Appendix: The interpretation of the results for the market potential variable

Top Abstract 1 Introduction 2 Theoretical background and... 3 Data set 4 Estimation results 5 Conclusions Appendix: The interpretation of... Appendix: Additional estimations Data Appendix References

 
The freeness of trade in a two region NEG model between countries i and j is given by _ij (T_ij)^1-, where is the substitution elasticity between varieties of the manufactured good and T_ij is the transport cost function. In turn, T_ij = T(D_ij), with T being the trade cost parameter and D_ij is the distance function between countries i and j. In the two country case the distance between the two countries can be normalized to 1 with the result that _ij = T^1- (see also Puga 1999). In a multi-country setting it no longer makes sense to normalize inter-country distances and this is also true for our sample of 19 countries (unless we make the unrealistic assumption that all 19 countries are at equidistance to each other). Hence, we have to specify a distance function: assume that T_ij = T(D_ij) we then arrive at _ij= [T(D_ij)]^1-. Other distance specifications are of course possible but this one is quite common in for instance gravity equations of trade and in NEG applications too (Head and Mayer 2000, Crozet 2004). Defining the freeness of trade as _{ij =} [T(D_ij)]^1- shows that the freeness of trade changes when for instance the distance decay parameter changes or when actual trade or transport costs T change. In section 4 the discussion of Table 3 is based on alleged changes in whereas the discussion of Table 4 is based on changing T. In both cases the freeness of trade ceteris paribus changes.

In the CP model (which is basically the initial NEG model due to Krugman 1991) that underlies the analysis of the relationship between capital mobility and tax rates in Baldwin et al. (2003), the relationship between the freeness of trade (or trade costs) and the degree of agglomeration is as depicted by Figure A1. This so-called Tomahawk figure shows for the two region CP model that for low levels of the freeness of trade (= high trade costs) we end up with spreading and only when the freeness of trade surpasses ^B we end up with (full) agglomeration (^B is the so called break point; ^S is the sustain point). Given D_ij and , a higher (lower) distance decay parameter and/or a lower (higher) T are associated with a lower (higher) freeness of trade. This means that we can rank our three distance parameters along the horizontal axis as depicted below Figure A1. Given these respective distance parameters and the associated estimation results for the MP coefficient in Table 3 we can, however, not say where exactly we are on the horizontal axis. But, given this ranking of the distance parameters through the relationship between the distance parameter and the freeness of trade, we know that it is not possible to be in the agglomeration equilibrium for = 2 and at the same time to still be in the spreading equilibrium for the case for = 1 or = 0.5 as well (the reverse is possible). A similar reasoning leads us to expect that the MP-coefficient in Table 4 for the full sample period (1990s) a period associated with higher (lower) trade costs T is relatively low (high).

If anything, we therefore expect the coefficient for the Market Potential variable with = 0.5 (MPH) to be at least as large as the coefficient associated with = 2 (MPD) or = 1 (MP1). The idea simply being that with agglomeration a country can have a higher corporate tax rate as compared to case of spreading. Following a similar line of reasoning for our TC variables TC = (cif/fob measure)*MP, we can thus also hypothesize that the MP- coefficient for the 1990s—i.e. the period with lower trade costs—will be higher that the trade costs coefficient for the whole period.

Our conclusions in the main text as to the feedback from the estimation results in Tables 3 and 4 to the NEG theory are sensitive to underlying NEG model. The CP model is one of the two core models in the NEG literature (Puga 1999). In the other core model, the relationship between agglomeration and the freeness of trade is different, as Figure A2 illustrates (see for an in-depth explanation Puga 1999, Head and Mayer 2004 or Brakman, Garretsen and Schramm 2006). Figure A2 depicts the so-called inverted U-curve also known as the Bell Shaped curve where moving from left to right on the horizontal axis (that is from high to low trade costs), the two country economy moves from spreading to (partial) agglomeration to renewed spreading. There are now two break-points ^B_low and ^B_high. Given our previous argument about the implied ranking of the freeness of trade , Figure A2 illustrates that the restrictions to be placed on the market potential coefficient and the correlation between the tax rate (e.g. EATR) and capital mobility (e.g. FDIFLOWS) would be different from those based on the NEG model underlying Figure A1. In fact, our estimation results as shown in Tables 3 and 4 can be aligned with Figure A2, if for instance agglomeration has become more important in the 1990s.

	Appendix: Additional estimations

Top Abstract 1 Introduction 2 Theoretical background and... 3 Data set 4 Estimation results 5 Conclusions Appendix: The interpretation of... Appendix: Additional estimations Data Appendix References

 
(i)First stage of 2SLS estimation
In order to establish that the GOLUB index is not correlated with the EATR variable (the corporate tax rate) but is correlated (with the right sign) with FDI,¹⁶ we show for our main specification (Table 3, column I) the first stage estimation results for both the fixed effects and, see discussion in sections 3 and 4.1, the random effects specification. Besides the GOLUB-index we include the regressors as the other explanatory variables. It is clear that the GOLUB index is significant and has the expected sign. A similar conclusion holds for the other specifications.

View this table: [in this window] [in a new window]   Table A1 First stage of 2SLS estimation

 
(ii)IV or OLS?
Another issue concerns the IV approach as such. The results of the Hausman test show that OLS is not the preferred model: here we included the residuals from the OLS estimation FDI = f(Golub; other regressors) in a OLS estimation, the latter with the same specification as in Table 3, column I.

View this table: [in this window] [in a new window]   Table A2 IV versus OLS

 
(iii)Main specification with random effects
We ran our estimations with random effects as well. The results are very similar to the fixed effects outcomes (certainly for the FDI and market potential variables). Below we include the random effects estimation results (using the Wallace and Hussain estimator) for our basic specification as an example, again compare with Table 3 (column I) in the article.

View this table: [in this window] [in a new window]   Table A3 Main specification with random effects

 
With the exception of SIZE the results are qualitatively similar to those obtained by using fixed effects.

	Data Appendix

Top Abstract 1 Introduction 2 Theoretical background and... 3 Data set 4 Estimation results 5 Conclusions Appendix: The interpretation of... Appendix: Additional estimations Data Appendix References

 
The sample used for the panel regression analysis includes OECD 19 countries and covers the 1981–2001 period with an annual frequency (although some series have a shorter time period). The countries included are Austria, Belgium, Finland, France, Germany, Greece, Ireland, Italy, Netherlands, Norway, Portugal, Spain, Sweden, Switzerland, United Kingdom, Australia, Canada, Japan and United States. In the table below, the definitions and sources of data used in the empirical research are described briefly.

Variable Definition Source Dependent variables     EATR Effective average tax rate Devereux, Griffith, Klemm (2002)17     Captax Capital tax ratio based on gross operating surplus Carey and Rabesona (2002)     Cortot Corporate tax revenues as a share of total tax revenues OECD Revenue Statistics     Corgdp Corporate tax revenues as a share of GDP OECD Revenue Statistics Explanatory variables     Golub Index to measure inward DI restrictions Golub (2003)     FDIFLOWS Sum of FDI in- and outflows as a percentage of GFCF Unctad, FDI online     FDIinflows FDI inflows as a percentage of GFCF Unctad, FDI online     FDIoutflows FDI outflows as a percentage of GFCF Unctad, FDI online     MP1 Market potential (with = 1) MPi = jgdpj Dij GDP data: OECD Annual National Accounts Distance: CEPII18     MPH Market potential (with = 0.5) GDP data: OECD Annual National Accounts Distance: CEPII     MPD Market potential (with = 2) GDP data: OECD Annual National Accounts Distance: CEPII     TC1 Trade-cost-corrected-market potential (with = 1) Market potential: see above Cif/fob ratio: IMF, Direction of Trade Statistics     TCH Trade-cost-corrected-market potential (with = 0.5) Market potential: see above Cif/fob ratio: IMF, Direction of Trade Statistics     TCD Trade-cost-corrected-market potential (with = 2) Market potential: see above Cif/fob ratio: IMF, Direction of Trade Statistics Control variables     NEIGHBOUR Average statutory corporate tax rate in other countries Devereux, Griffith, Klemm (2002)     LEFT Left party cabinet portfolios as a% of all cabinet portfolios Swank, Comparative Parties Data Set19     SIZE gdp country i as a share of gdp total OECD, Annual National Accounts     GOVINV Government fixed investment as a share of total disbursements OECD, Economic Outlook

	Footnotes

 
¹ The authors are affiliated with the Utrecht School of Economics, Utrecht University, and the Economics and Research Division, De Nederlandsche Bank, respectively, e-mail: h.garretsen{at}econ.uu.nl; j.j.w.peeters{at}dnb.nl. We would like to thank Rob Alessie, Robert-Paul Berben, Jan Marc Berk, Leon Bettendorf, Maarten Bosker, Steven Brakman, Peter van Els, Joeri Gorter, Ralph de Haas, Signe Krogstrup, Marc Schramm, Job Swank, an anonymous referee, seminar participants at the Dutch central bank and seminar participants at the CESifo Venice Summer Institute workshop on "The future of capital income taxation" (17–18 July 2006), and in particular our discussant at the workshop Michael Stimmelmayr, for useful comments and suggestions. The views in this article do not necessarily reflect those of De Nederlandsche Bank.

² This observation was made previously by Kind, Knarvik and Schjelderup (2000), Ludema and Wooton (2000) and Andersson and Forslid (2003). The latter study, see also Baldwin and Forslid (2002) and Brakman, Garretsen and Van Marrewijk (2005), also includes the role of government spending.

³ Some studies go further and aim to bring out the special mechanisms through which taxes influence FDI. In this context, Razin, Rubinshtein and Sadka (2006) for instance find that the source-country tax rate works primarily on the selection process, whereas the host-country tax rate affects mainly the magnitude of FDI, once they occur. A different approach in the tax competition literature tests whether governments react to other governments’ changes in tax rates by estimating tax reaction functions (see, for instance, Krogstrup 2002).

⁴ See for instance Devereux and Griffith (2003), De Mooij and Ederveen (2001), Gorter and De Mooij (2001).

⁵ Whether stocks or flows are the conceptually preferred measure to use is disputable, see Krogstrup (2003).

⁶ The correlation coefficient between FDI flows/GDP and FDI flows/GFCF is 0.99.

⁷ Admittedly, this figure exaggerates the increase in capital mobility, as it includes the FDI data for 1999 and 2000, which were years with a very high level of FDI flows. Nevertheless, when we extend our sample period to include the years 2002–04, the degree of capital mobility, as measured by FDI flows, still shows a very clear increase over time.

⁸ We recognize that there are a n