Corporate tax effects on the quality and quantity of FDI

European Economic Review 56 (2012) 1495–1511

Contents lists available at SciVerse ScienceDirect

European Economic Review journal homepage: www.elsevier.com/locate/eer

Corporate tax effects on the quality and quantity of FDI Johannes Becker a, Clemens Fuest b,n, Nadine Riedel c a

Institute of Public Economics I, University of Muenster, Wilmergasse 6-8, 48143 Muenster, Germany Centre for Business Taxation, Saı¨d Business School, University of Oxford, Park End Street, Oxford OX1 HP, UK c Institute of Economics, University of Hohenheim, Oxford University Centre for Business Taxation and CESifo Munich, Germany b

a r t i c l e i n f o

abstract

Article history: Received 10 December 2010 Accepted 6 July 2012 Available online 23 July 2012

This paper measures the relative importance of quality and quantity effects of corporate taxation on foreign direct investment. Quantity is affected if corporate taxes reduce the equilibrium stock of foreign capital in a given country. Quality effects arise if taxes decrease the extent to which investment contributes to the corporate tax base and the capital intensity of production. Depending on the sign of the quality effects, the detrimental welfare effects of corporate taxation are either mitigated or aggravated. We derive a number of hypotheses about how corporate tax changes may affect the quality of investment. Our hypotheses are then tested using data from a large sample of European multinationals. With regard to corporate tax effects on the corporate tax base, we find that quality effects account for up to 40% of the total effect. With regard to corporate tax effects on labour income, our results suggest that quality effects mitigate the negative quantity effect by nearly 60% (as corporate taxes strongly increase the labour intensity of production). An important implication is that governments should not exclusively care about the size of inbound FDI flows but also about their specific characteristics, i.e. their quality. & 2012 Elsevier B.V. All rights reserved.

JEL classification: H25 F23 Keywords: Corporate taxation Foreign direct investment Multinational firms

1. Introduction In the last decades, the world economy has experienced an unprecedented boom in foreign direct investment (FDI). In 2008, the worldwide stock of inward FDI was USD 14,909 billion, only slightly less than its alltime high in 2007, compared to only USD 1457 billion in 1988 (source: UNCTAD).1 While multinational firms are seen to be the main driving force behind FDI growth (see, e.g., Bernard and Jensen, 2007; Bernard et al., 2007), governments throughout the world have supported this trend by improving the policy environment for border crossing investment. From a policy perspective, attracting FDI is widely considered to be beneficial for the host country because this investment is thought to provide new growth opportunities, higher wages and employment, larger tax revenues—in short, a higher welfare level. Among the policy instruments which can be used to attract investment, taxes play a key role. In recent years, many countries have reduced their corporate tax rates significantly in order to attract FDI (see, e.g., Devereux et al., 2002; Loretz, 2008).2 Empirical research has measured the effects of these tax reforms and finds that there is indeed a strong and robust

n

Corresponding author. Tel.: þ44 1865 614846. E-mail addresses: [email protected] (J. Becker), [email protected] (C. Fuest), [email protected] (N. Riedel). 1 See the UNCTAD Handbook of Statistics 2009, available on www.unctad.org. For purpose of comparison, the subgroup of developed countries had a stock of inward FDI of USD 1,000 billion in 1988 and of USD 10,212 billion in 2008. All numbers are in current prices and exchange rates. 2 Effective average tax rates, which are likely to be the relevant indicator for firm location have been reduced from 37.4% in 1982 to 23.9% in 2007, see Loretz (2008). 0014-2921/$ - see front matter & 2012 Elsevier B.V. All rights reserved. http://dx.doi.org/10.1016/j.euroecorev.2012.07.001

1496

J. Becker et al. / European Economic Review 56 (2012) 1495–1511

impact of corporate tax cuts on the quantity of inbound FDI.3 However, as an indicator of success, the quantity of assets attracted by a corporate tax rate cut may be misleading or at least incomplete. The reason is that the welfare of the host country will depend on the impact of FDI on tax revenue and labour income generated by the investment and much less on the mere investment quantity. In this paper, we argue that taxes do not only affect the quantity of (inbound) FDI but also its quality, i.e. the degree to which FDI creates jobs and how it contributes to tax revenue in the host country. We will show empirically that taking into account the quality aspect of FDI has considerable consequences for optimal tax policy. For purposes of illustration, consider the following example. A multinational firm sets up two new affiliates, one in Ireland, the other in Spain. The two investment projects are equal in terms of quantity, i.e. the amounts of capital invested in each country are identical. However, due to the low Irish corporate tax rate, the highly profitable financing department is located in Ireland while the industrial assembly lines, with low profitability and a large payroll, which is deductible from the corporate tax base, are located in Spain. How are these investments to be evaluated in welfare terms? According to the conventional quantity-based success indicator, both countries were equally successful in attracting FDI. However, the investment in Ireland boosts the Irish corporate tax base while the investment located in Spain only marginally adds to the Spanish corporate tax base. It increases labour income in Spain, though, and boosts Spanish payroll tax revenue. Thus, the Spanish and the Irish investment are equal in quantity but very different in quality. Both investment characteristics, quantity and quality, are potentially affected by the level of corporate tax rates (Ireland attracts the profitable investment, Spain the labour-intensive one). Measuring the quantity and quality effects of corporate taxes on FDI is in the focus of this paper. In the first part of the paper, we propose a simple model of corporate taxation and investment in a small open economy in order to explain and define precisely our notion of quantity versus quality. We derive three hypotheses regarding the effects of corporate tax changes on the quantity and quality of FDI. Hypothesis 1 states, not surprisingly, that a higher corporate tax rate decreases the quantity of investment. Hypothesis 2 states that a higher corporate tax rate reduces the contribution of each unit of capital to the corporate tax base. Hypothesis 3 claims that a higher corporate tax rate increases the labour income generated per unit of investment. While the first hypothesis is theoretically straightforward and empirically well established, the second and third hypotheses require some further discussion. To begin with hypothesis 2, the reason for a negative quality effect may be transfer pricing or thin capitalization, as has been extensively shown by the literature.4 These studies interpret profitability differences between high- and low-tax economies as evidence for shifting of accounting profits and usually find that tax differences across locations of a multinational group are a strong predictor for the location of accounting profits. It may also be that this pattern is caused ¨ by the mobility of firms with different levels of profitability (Haufler and Stahler, forthcoming) or project selection within multinational firms (Becker and Fuest, 2007). Note, though, that taxation may theoretically also increase the average profitability of investment. This is the case if taxes mainly drive out the least profitable investment projects, so that the average profitability of investment projects remaining in the country increases. Consider finally hypothesis 3, which states that a higher corporate tax rate increases the amount of labour income generated per unit of investment. A reason for this could be that the production technology allows for the substitution of capital by labour. Alternatively, capital intensive tasks may be relocated to a foreign affiliate within the multinational group.5 This effect is economically important because a key benefit governments expect from inbound FDI is increased wages and/or reduced unemployment. If the third hypothesis is supported by the data, the negative welfare effect of a taxinduced reduction of investment is partly mitigated by the higher labour intensity of production. However, it cannot be excluded that the decrease in the capital intensity of production reduces wages. In this case, the efficiency cost of corporate taxation may even be aggravated by the quality effect. Taken together, a change in the host country’s corporate tax rate may affect the quantity of inbound investment as well as the quality of each incoming unit of investment. It is intuitively plausible that the size and the sign of these effects on quantity and quality will affect the optimal corporate tax rate in an open economy. To make this more precise, we add a normative section to our theoretical analysis and derive the optimal corporate tax rate for the economy under consideration. We show that, depending on the sign and the size of the quality effects of corporate taxation, neglecting these quality effects implies that the efficiency cost of taxation is either underestimated or overestimated. In the second part of the paper, we test these hypotheses using micro-level firm data from a large panel of European corporations. We find evidence for the quantity effect and both quality effects. Firstly, a higher corporate tax rate decreases the quantity of foreign capital invested. Secondly, an increase in corporate taxes reduces the corporate tax base per unit of capital. With respect to the corporate tax base, our results suggest that the quality effect reinforces the quantity effect and accounts for around 40% of the overall effect. Thirdly, increasing corporate tax rates ceteris paribus increases the labour income generated per unit of corporate investment. This compensates the negative quantity impact of corporate taxes on payroll by around 80%. So far, the literature has mainly treated quantity and quality effects of taxation separately (see the literature cited above). Recently, however, there is increased attention to welfare implications of corporate taxes. Gruber and Rauh (2007) 3

These studies are surveyed and discussed in de Mooij and Ederveen (2003) and Devereux (2007). See e.g. Grubert and Mutti (1991), Hines and Rice (1994), Clausing (2003), Bartelsman and Beetsma (2003), Huizinga and Laeven (2008), Huizinga et al. (2008), Weichenrieder (2009) and Dischinger and Riedel (2011). 5 See Becker and Fuest (2007) and, in a tax-unrelated context, Grossman and Rossi-Hansberg (2008). 4

J. Becker et al. / European Economic Review 56 (2012) 1495–1511

1497

and Dwenger and Steiner (2008) measure the tax base elasticity with regard to corporate tax rate changes. This elasticity can be interpreted as a summary measure of corporate tax effects. Due to data limitations, though, quantity effects cannot be differentiated from quality effects. Arulampalam et al. (2007) deal with the incidence of corporate taxes on wages. They find that an increase in corporate taxes is associated with a significant wage decrease. Here, quantity effects of taxes are only accounted for as far as they affect the wage rate. To the best of our knowledge, our paper is the first to explicitly analyse quantity and quality effects of corporate taxation from a welfare perspective. From a policy point of view, the distinction between quality and quantity effects of corporate tax rate changes is of key importance. For example, it may well be that countries with high taxes attract primarily low profitability investment or even investment which generates tax losses. The quantity of investment may not differ much from that of other countries, but the welfare effects of this investment would be different because the investment diminishes the tax base. Thus, the welfare cost of tax distortions may be higher than suggested by studies focusing on the quantity aspect alone. In contrast, if FDI is mainly supposed to increase employment and/or wages, our findings suggest that a tax-induced reduction in FDI is mitigated by a simultaneous increase in labour intensity. In general, policies which aim at attracting foreign direct investment may have to pay more attention to the qualitative dimension of this investment. The rest of the paper is set up as follows. In Section 2, we derive the hypotheses which guide the empirical analysis in Sections 3–5. Section 6 illustrates the implications of our empirical findings for the size of quality and quantity effects and for optimal corporate tax policy. Section 7 concludes. 2. A simple model of corporate tax effects on FDI quality and quantity In the following, we develop a highly stylised model of corporate taxation in an open economy in order to explain the difference between tax effects on the quantity and quality of investment. In the first part of this section, quantity and quality effects are defined in a reduced form model. Three hypotheses are derived, which will be tested in the empirical section below. The second part sets out in how far these effects have implications for optimal tax policy and welfare. 2.1. Positive analysis Consider an open economy with many identical households and two production sectors. The first sector is an international industrial sector employing internationally mobile capital and internationally immobile labour. The second one is a domestic service sector where labour is the only input. To make things very simple, we assume that total payroll is given by P þ sðNPÞ, where P is the payroll in the industrial sector and sðNPÞ is the payroll in the domestic service sector. Accordingly, N is the total payroll if all labours are employed by the international sector and sN is total payroll if there is only a domestic sector. If s¼1, total payroll is given by N and an increase in payroll in the international sector does not affect total payroll. However, if wages are higher in the international sector (for which there is considerable evidence, see e.g. Aitken et al., 1996; Griffith and Simpson, 2004; Heyman et al., 2007), a decline in payroll generated in the international sector by one unit will be replaced by s o 1 units of payroll in the domestic service sector. The underlying idea is that an increase in labour income generated, for instance, by more investment in the international sector, will also increase labour income in the economy as a whole, rather than just replacing jobs in the national service sector by similar jobs in the international sector. Given this, governments will be interested in attracting foreign direct investment even if their objective is just to maximize the income of the domestic workforce. Thus, this stylized formulation allows for capturing the interest of governments to boost the domestic labour market by attracting FDI (e.g., by setting corporate tax rates optimally, see Section 2.2). Disposable income of domestic residents, Y, consists of after-tax labour income and a share f 2 ½0,1 of after-tax corporate profits generated in the industrial sector. At f ¼ 0, we may speak of purely foreign direct investments. Domestic ownership of firms located abroad is neglected for simplicity. Profits in the service sector are equal to zero. Thus, disposable income is given by Y ¼ ðP þ sðNPÞÞð1t P Þ þ fBð1t c Þ, where P is the payroll, B stands for corporate profits in the industrial sector, and t P and t c are the proportional taxes on labour income and corporate profits, respectively. Accordingly, government revenue, G, is given by G ¼ t P ðP þ sðNPÞÞ þt c B. Foreign investment affects domestic welfare by changing payroll, profits and tax revenue. Now, denote the quantity of capital invested in the country by K and define payroll per unit of capital, p  P=K, and corporate profits per unit of capital, b  B=K. Then, the effect of an increase in the corporate tax rate t c on payroll can be c c c expressed as dP=dt ¼ p dK=dt þ K dp=dt . Dividing by P yields 1 dP ¼ eK þ ep P dt c c

ð1Þ c

where eK  ðdK=dt Þ1=K and ep  ðdp=dt Þ1=p are semi-elasticities of K and p, respectively, with respect to t c. Thus, the semi-elasticity of payroll with respect to t c has a quantitative dimension (the number of capital assets is affected, eK ) and a qualitative dimension (payroll per unit of capital is affected, ep ). What should we expect regarding the sign of eK and ep ? The sign of the quantity effect will usually be negative, i.e. eK o 0.6 For the quality effect ep , predictions are less clear. For a given non-tax cost of capital, an increase in the corporate income tax increases the cost of capital relative to the cost of labour. 6

Theoretically, a higher tax rate could also reduce the cost of capital, in particular in the presence of accelerated depreciation.

1498

J. Becker et al. / European Economic Review 56 (2012) 1495–1511

If the firm is able to substitute capital with labour input, this increases the payroll (due to higher labour input or higher wages) for a given level of investment. However, if capital and labour are technological complements, a decrease in the stock of capital reduces labour productivity which, for a given level of investment, reduces payroll (due to lower wages). It is a priori unclear which of the two effects prevails, i.e. whether the quality of investment in terms of generating labour income increases or decreases in response to a change in the corporate tax rate. In the online appendix to this paper (A1 and A3), we present a more specific model which allows to determine the conditions under which the payroll semi-elasticity is positive and negative. Generally speaking, if labour and capital are not too complementary and if marginal projects (those shifted abroad if taxes are increased) are not substantially more labour-intensive than the average project, then a corporate tax rate increase leads to a rise in payroll per unit of capital and, thus, ep 40. c c c Similarly, the effect of an increase in t c on corporate profits is dB=dt ¼ b dK=dt þ K db=dt or 1 dB ¼ eK þ eb B dt c

ð2Þ

where eb ¼ ð@b=@t c Þ1=b is the semi-elasticity of b with respect to t c. Which sign can be expected for eb ? Higher corporate tax rates are likely to increase both the marginal and the average tax burden. If only the marginal and, thus, least profitable projects are driven out of the market, the corporate tax base per capital unit may actually increase. If, however, economic rents are not bound to the country under consideration, firms will move profitable operations to other countries in response to higher tax rates. The incentives to move real economic activity to other countries created by the higher tax are stronger, the higher the profitability of the activity. This suggests that firms will move the most profitable functions first, so that the corporate tax base per capital unit may actually decrease. Moreover, firms are likely to react to the higher corporate tax by shifting accounting profits abroad through transfer pricing, debt financing and so on. Again, it is unclear which of the two effects is stronger, i.e. whether the quality of investment in terms of generating corporate tax revenue increases or declines. The online appendix (A1, A2 and A3) presents a variety of model versions which help explaining under which circumstances the corporate profit semielastacitiy has a positive or a negative sign. Generally speaking, if profit shifting opportunities are large and/or if marginal projects (those shifted abroad if taxes are increased) are more profitable than the average project, profits per unit of capital decrease in response to a tax rate increase. These considerations can be summarized as Hypothesis 1 (Quantity) An increase in t c decreases the capital stock K: eK o0: Hypothesis 2 (Quality - Tax Base Contribution) An increase in tc decreases the contribution to the corporate tax base per unit of investment b: eb o 0. Hypothesis 3 (Quality-Labour Income Contribution) An increase in t c increases the payroll–capital ratio ep 4 0. In our empirical analysis we will use firm data to directly estimate the quantity and quality effects eK , eb and ep . But before turning to the empirical analysis, it is helpful to briefly discuss the normative question of why the signs and the magnitudes of the quality and quantity effects are important for corporate tax policy. 2.2. Normative analysis This subsection shows that it is important for tax policy to account for qualitative effects of corporate taxation. Domestic welfare is given by W ¼ Y þ ZG , where G is public expenditure and Z is the marginal utility of public expenditure (or the marginal cost of public funds). With Y ¼ ðpK þ sðNpKÞÞð1t P Þ þ fbKð1t c Þ and G ¼ t P ðpK þsðNpKÞÞ þ t c bK the country’s welfare function can be expressed as W ¼ asN þðað1sÞp þ bbÞK

ð3Þ

where a  1 þðZ1Þt P and b  f þ ðZfÞt c are the social marginal utility of labour income and profits, respectively. The impact of a change in the corporate tax on welfare can now be expressed as @W @K @p @b ¼ ðZfÞbK þðað1sÞp þ bbÞ c þ að1sÞK c þ bK c @t c @t @t @t

ð4Þ

The effect of the tax change on the quantity of investment is captured by the second term on the right-hand side of (4) while the effects on the quality of investment are given by the third and the fourth term. By setting the right-hand side of (4) equals to zero we can derive a formula for the optimal corporate tax rate: t cn ¼ 

1

eK þ eb



að1sÞ p ðeK þ ep Þ f  ðZfÞ b ðeK þ eb Þ ðZfÞ

ð5Þ

The above expression for the optimal tax rate can be used to demonstrate how qualitative effects may have an impact on optimal policy making. If policy makers ignored quality effects, i.e. they assumed that p and b stayed constant if taxes are changed, ep ¼ eb ¼ 0, the optimal tax rate would be higher or lower than the true optimal tax rate described by equation (5)—depending on the signs and magnitudes of ep and eb .

J. Becker et al. / European Economic Review 56 (2012) 1495–1511

1499

Below in Section 6, we evaluate (5) given our empirical estimates of eK , eb and ep from Section 5 for different assumptions on Z, s and f. Thus, we are able to provide a quantification of the importance of quality effects for optimal corporate taxation.

3. Data We use the commercial database AMADEUS compiled by Bureau van Dijk. The database contains detailed information on firm structure and accounting data of national and multinational corporations in European countries. Our sample comprises information on affiliates of multinational firms in 22 countries (Austria, Belgium, Czech Republic, Denmark, Finland, France, Germany, Great Britain, Greece, Hungary, Iceland, Ireland, Italy, Luxembourg, Netherlands, Norway, Poland, Portugal, Slovakia, Spain, Sweden, Switzerland) for the time period of 2000–2006.7 Firms are included in the analysis if either their parent company or one of their wholly owned subsidiaries is located in a foreign country. The distribution of our sample affiliates across countries is presented in Table A4.1 in the online appendix.8 The observational unit in our analysis is the multinational affiliate per year. In total, our sample comprises 100,697 observations from 26,332 affiliates for the years 2000–2006. Hence, we observe each affiliate for 3.8 years on average. Besides the rich set of accounting information available in AMADEUS, we enlarge our data set by merging information on the country’s tax system, i.e. statutory corporate tax rates, effective marginal tax rates and the present values of capital allowances, obtained from European Commission (2006), KPMG (2006) and Loretz (2008). Moreover, we add information on GDP as a proxy for market size, GDP per capita as a proxy for a country’s income and development level, the unemployment rate as a proxy for the state of a country’s economy and the corruption index as a proxy for the state of governance institutions. The corruption index is obtained from Transparency International while the other country data is retrieved from the World Development Indicator Database. Last, we add information on the labour tax wedge obtained from OECD’s tax data base which is measured as the marginal tax burden of a worker who earns the country’s median income (the results are robust against evaluating the labour tax wedge at other points of the income distribution). Note that not only the host country’s corporate tax rate but also the tax rate differential to other firms within the same multinational group may have an impact on affiliate behavior. For instance, profit shifting or the choice of locations for projects may depend on these tax rate differentials, rather than tax rates of individual countries. We use our data set to construct a tax measure which captures these effects. As the AMADEUS data contains information on the ownership structure of multinational groups on a worldwide basis (while the accounting information is restricted to firms located in Europe), we observe all group affiliates worldwide and their host countries. To construct the tax difference measure, we identify a firm’s global ultimate owner and consider all its majority owned subsidiaries to be affiliates within the multinational group. Then, we calculate an unweighted average tax rate differential between the considered affiliates and all other majority owned firms in the multinational group (including the parent if the considered firm is not an independent company itself).9 Table A4.2 in the online appendix displays basic sample statistics. The affiliates in our sample have an average fixed asset investment of USD 166.4 million, employ 545 employees and have average annual costs per worker of USD 62.5 thousand. Labour intensity, defined as labour-to-fixed asset-ratio and payroll-to-fixed asset-ratio, is determined with 0.14 and 8.13 respectively. Corporate pre-tax profitability (measured as pre-tax profits over total assets) is 12.63%. Pre-tax profitability is used to proxy the firm’s corporate tax base. Although accounting profits may diverge from the the firm’s tax book profit (e.g. Hanlon and Maydew, 2009), they are nevertheless widely considered to be a good proxy for the the corporate tax base, especially for European countries in which the divergence between accounting and tax books is less pronounced than for example in the US (see e.g. Huizinga and Laeven, 2008). Moreover, the firms in our data base on average face a corporate tax rate of 32.5% which varies strongly between 10% and 51.6%. Note that the variable also exhibits a considerable longitudinal variation. In total, we observe 41 changes in the corporation tax within our sample period. Seventeen out of our 22 sample countries enacted at least one corporate tax rate change, whereas the average reform triggered a relative change in the corporate tax rate by 8.8% (in absolute terms). The unweighted average tax rate differential between the affiliate and other firms within the multinational group is determined with 0.6% but equally exhibits a strong spread between  31.1% and þ35.6%. There is again substantial longitudinal variation in the variable as the tax rate difference is affected by both, tax reforms in the host country and at other group locations. The effective marginal tax rate is calculated with 11.31% varying between 0.1% and 17.27% while the average present value of capital allowances is 66.0% ranging from 53.4% to 74.6%. Last, the marginal labour tax wedge for employees with the mean country income is 52.1%, varying between 33% and 77%. 7 The AMADEUS data is in principle available since 1999. However, as we observe some of our tax information for 2000 onwards only, we restrict the sample to the years 2000–2006. 8 Note that the country statistic broadly corresponds to the distribution of economic activity in Europe. Nevertheless, in line with previous work, we also find that some countries are underrepresented in the AMADEUS data, for example Germany and Switzerland, which suggests that some caution is warranted when drawing conclusions from our results for the population of firms in these countries. 9 An alternative procedure would be to calculate a size weighted average tax differential. However, as the size information is unavailable for a large number of group affiliates in our data (especially those located outside Europe), this would imply that we lose a considerable number of observations in the calculation of the tax rate difference variable. Thus, we decided to employ the unweighted measure in our baseline estimations. In robustness checks, however, we reestimated the regression models using size weighted averages and did not find our qualitative results to change.

1500

J. Becker et al. / European Economic Review 56 (2012) 1495–1511

Additionally, our sample firms’ host countries exhibit an average GDP of USD 991.1 billion, an average GDP per capita rate of USD 30,244.6, an average unemployment rate of 7.5% and an average corruption index of 7.5 (on a scale from 0 (poor) to 10 (excellent) indicating the quality of governance institutions). 4. Estimation approach In order to disentangle quality and quantity effects of corporate taxation, we determine the impact of corporate tax rate changes on the size of a firm’s capital stock (quantity effect) and on the profitability and payroll intensity of investment (quality effects). Let K, b and p denote the investment quantity, profitability and payroll intensity. Quality and quantity effects are estimated in a static and dynamic investment framework where the static model reads x

x

x

x

log xi,t ¼ b0 þ b1 txi,t þ b2 Z xi,t þ rxt þ fi þ Exi,t

with x 2 fK,b,pg

ð6Þ

with xi,t representing the fixed asset stock, profitability or payroll intensity of affiliate i at time t. As the distributions of all three variables are strongly skewed, we use logarithmic transformations as regressands.10 txi,t is a measure of the corporate tax burden of affiliate i at time t. The superscipt x 2 fK,b,pg indicates the variable definition in the regressions using fixed assets, profitability and the payroll intensity as the dependent variable. In the baseline specifications, the variable txi,t takes the form of the statutory corporate tax rate t i,t . The profitability and payroll intensity regressions also account for a second tax measure, which is the tax rate differential between the affiliate and other firms in the multinational group as defined in the previous section. Including the tax rate differential takes into accout that, conditional on the affiliate location, the multinational’s decision where to locate its profitable and/or labour intensive investment projects may depend on the tax structure of the whole multinational group. Z xi,t comprises time varying firm-specific and country-specific control characteristics. Country-specific control variables are used to absorb variations in market size (proxied by GDP), income level (proxied by GDP per capita), state of the economy (proxied by the unemployment rate) and quality of governance institutions (proxied by the corruption index). Firm-specific control variables are used depending on the regressand x. Most importantly, the regressions which employ the affiliates’ profitability and payroll intensity as dependent variable include a control variable for affiliate size (measured by sales and the corporate capital stock). rxt depicts year dummies to capture shocks over time which are common to all affiliates. In some specifications we additionally include a full set of industry-year dummies to account for industryspecific shocks over time. Exi,t describes the error term. Moreover, we add fixed effects for multinational affiliates to control x for non-observable time-constant firm specific characteristics fi . A shortcoming of the static estimation approach described in equation (6) is that it does not take into account the costs associated with adjusting capital or labour demand at the firm level and consequently neglects that the subsidiary’s capital and labour stock is dependent on the stocks of previous periods (an aspect which is ignored in the theoretical model outlined in Section 2 as well). Therefore, we additionally estimate a dynamic model. Since including the first lag of the dependent variable into the set of regressors in a fixed effects framework leads to biased coefficient estimates, we follow Arellano and Bond (1991) in estimating a first difference generalized method of moments (GMM) model which instruments for the first difference in the lagged dependent variable by deeper lags of the level of the dependent variable.11 The estimation equation then takes on the following form:

D log ðxi,t Þ ¼ gx1 D log ðxi,t1 Þ þ gx2 Dtxi,t þ gx3 DZ xi,t þ Drxt þ DExi,t

with x 2 fK,b,pg

ð7Þ 12

where D is the difference operator and the variable definitions correspond to the ones in Eq. (6). Because the model is estimated in first differences, the equation will be characterized by the presence of first order serial correlation. The validity of the GMM estimator relies on the absence of second order serial correlation. The Arellano and Bond (1991) test statistics for first and second order serial correlation will be reported at the bottom of the result tables. We further check for the exogeneity of the instrument set by employing a Sargan/Hansen overidentification test. Note moreover that the GMM estimations acknowledge that the corporate capital stock, profitability and payroll intensity may determine each other. The capital stock regressions thus control for profitability and payroll intensity, and the profitability (payroll intensity) regressions include control variables for the affiliate’s labour intensity (profitability) and size as measured by the affiliate’s capital stock or sales. To avoid potential endogeneity problems, the firm variables are instrumented with deeper lags of their level. In the profitability regressions, we find—in line with previous studies—only a shallow serial correlation in the pre-tax profit variable and thus refrain from presenting dynamic specifications in the paper (gb1 ¼ 0). As the firm level regressors 10 As this is also true for the distribution of profitability (see, e.g., Huizinga and Laeven, 2008), the sample is restricted to affiliates with positive pretax profits which we consider to be sensible in our context as tax considerations plausibly play a significant role for firms with a positive tax base. 11 Note that the difference in the lagged dependent variable correlates with the differenced error term. However, deeper lags (starting from the second lag) of the dependent variable (in levels) are available as valid instruments as they are orthogonal to the error term. The set of instruments moreover comprises all exogeneous regressors (including the industry-year fixed effects where applicable). The number of instruments employed is stated in the notes below the result tables. It varies across specifications, depending on the number of instrumented and exogeneous regressors. 12 With panel data on more than two time periods, it is not equivalent to use a fixed effect and first differencing approach. Both models give unbiased and consistent estimates although the relative efficiency of the estimators may differ, depending on the model structure. Precisely, the fixed effect estimator is less sensitive against violating the assumption of strict exogeneity of the regressors while the first difference estimator is less sensitive against violating the assumption of serially uncorrelated error terms. We find similar results using fixed effects and first difference estimators.

J. Becker et al. / European Economic Review 56 (2012) 1495–1511

1501

Table 1 Quantity Effect - Impact of Taxes on Corporate Investment Dep. Variable: Log Fixed Asset Investment, Panel 2000–2006. Model

Fixed effects model

Explanat. Var.:

(1)

(2)

GMM model (3)

(4)

(6) nnn

Lag Log Fixed Assets Stat. Corporate Tax

(5)

 1.449nnn (0.225)

EMTR

 0.469nnn (0.141)  1.882nnn (0.315) 1.351nnn (0.104) 0.033 (0.037)  0.027 (0.071)  0.178 (0.370)

 0.638nnn (0.186) 1.172nnn (0.072)  00.024 (0.033)  00.078 (0.043)  0.446nn (0.231)

 1.245nnn (0.247) 0.664nnn (0.096) 0.002 (0.022)  00.018 (0.051) 0.192 (0.305)

O

O

Industry-Year D.

O

O

O

O

O

O

O

Sargan/Hansen AR(1) Test AR(2) Test # Obs. # Firms R2 Within

 – – 100,697 26,332 0.1301

– – – 95,509 25,060 0.1328

– – – 96,259 25,031 0.7097

– – – 91,161 23,786 0.7085

0.599 0.000 0.762 66,148 19,338 –

0.423 0.000 0.423 54,818 16,722 –

0.579 0.000 0.535 51,906 15,774 –

Corruption Unemp. Rate PV Capital Allowances Log Profitability Log Payroll Intensity

 0.076nnn (0.003)  0.737nnn (0.010) O

O

0.615nnn (0.094) 0.003 (0.029)  0.072 (0.050) 0.272 (0.298) 0.587nn (0.288)  0.086 (0.062)  0.122 (0.119) O

0.382nnn (0.066)

Year D.

Log GDP

0.550nnn (0.092) 0.020 (0.022)  00.129nnn (0.042)  0.750nnn (0.222) 0.417 (0.304)

0.380 (0.067)  0.984nnn (0.197)

0.984nnn (0.0667)  0.010 (0.033)  00.132nnn (0.042)  0.773nnn (0.223)  0.210 (0.210)  0.077nnn (0.003)  0.738nnn (0.010) O

Log GDP pC

1.023nnn (0.104) 0.061 (0.049)  00.183nnn (0.069)  0.750nn (0.359) 0.004 (0.367)

0.384 (0.059)  1.122nnn (0.161)

(7) nnn

 0.075 (0.058)  0.134 (0.124) O

Notes: Heteroskedasticity robust standard errors adjusted for firm clusters in parentheses. n nn nnn , , indicates significance at the 10%, 5%, 1% level. The GMM estimates in specifications (3)–(5) employ 71–75 instruments, varying with the number of endogeneous regressors (and comprising the excluded instruments plus all exogeneous regressors).

may nevertheless be prone to reverse causality problems, we employ the levels estimator proposed by Anderson and Hsiao (1982) which suggests to control for time constant affiliate effects by taking the first differences of the estimation equation and to instrument for the difference in the endogenous regressors by employing lagged levels of the variables. Our result tables will report the Kleibergen/Paap test statistic for the relevance of the instruments at the first stage and the Sargan/ Hansen test of overidentifying restrictions to assess the validity of the instruments employed.13 Further note that our baseline specifications determine the corporate tax effect on the firm’s capital stock, pre-tax profitability and payroll intensity in a separate equation approach. In robustness checks, we will assess the sensitivity of our findings to estimating a system of three equations (see Section 5).

5. Results The regression results are reported in Tables 1 to 7. All specifications include a full set of industry-year fixed effects and time-varying country controls as described in the previous section (the host country’s GDP, GDP per capita, the corruption index, unemployment rate and present value of capital allowances). Heteroskedasticity robust standard errors which account for clustering at the firm level are calculated and presented in brackets below the coefficient estimates. In Table 1 we identify the quantity effect of corporate taxation and thus measure the impact of corporate rate changes on the firm’s capital stock. Specification (1) presents the results for the static fixed-effects model described by Eq. (6) in the previous section. The estimations show a strongly and significantly negative semi-elasticity of capital investment with respect to changes in the corporate tax rate. A one percentage point increase in the corporate tax rate decreases the capital stock by 1.4%. Specification (2) reestimates the model merging the information on the statutory corporate tax rate and depreciation allowances into one measure, the effective marginal tax rate on an investment project. This regression confirms the significantly negative impact of the effective marginal tax rate on capital investment. Specifications (3) and (4) 13 In robustness checks, we also reran the specification using the GMM estimator proposed by Arellano and Bond (1991) described above and found comparable results.

1502

J. Becker et al. / European Economic Review 56 (2012) 1495–1511

Table 2 Quality Effect—Impact of Taxes on Investment Profitability Dep. Variable: Log of Pre-tax Profitability, Panel 2000–2006. Dep. Variable

Pre-tax profit per Total Assets

Explanat. Var.:

(1)

(2)

(3)

(4)

(5)

(6)

(7)

 1.163nnn (0.263) 0.367nnn (0.180)  0.285nn (0.133) 0.126 (0.090) 2.523nnn (0.422)

 1.252nnn (0.283) 0.380nn (0.164)  0.226nn (0.108) 0.094 (0.094) 3.297nnn (0.443) 0.142nnn (0.008) 0.210nnn (0.014)

 1.347nnn (0.496) 0.506 (1.259)  0.731 (1.232) 0.239 (0.204) 2.129nnn (0.838) 0.408n (0.216)  1.106nnn (0.292)

 1.018nnn (0.432) 1.291 (1.174)  1.218 (1.107) 0.183 (0.186) 1.531nn (0.726)  0.632nnn (0.206)

 1.237nnn (0.296) 0.341nnn (0.179)  0.219n (0.121) 0.022 (0.098) 3.564nnn (0.472) 0.818nnn (0.010) 0.137nnn (0.016)

 1.260nnn (0.472)  0.340 (1.181)  0.157 (1.156) 0.067 (0.195) 2.437nnn (0.785) 0.887nnn (0.200)  0.647nn (0.274)

 1.024nnn (0.428) 1.193 (1.161)  1.089 (1.088) 0.117 (0.184) 2.130nnn (0.705) 0.008 (0.205)

(a) Corporate Tax Rate Log GDP p. C. Log GDP Corruption Index Unempl. Rate Log Payroll Intensity Log Sales

Pre-tax Profit per Fixed Assets

Year D.

O

O

O

 1.047nnn (0.247) O

O

O

 1.108nnn (0.255) O

Industry-Year D.

O

O

O

O

O

O

O

Model Sargan/Hansen Kleinbergen/Paap # Observations # Firms R2 Within

FE – – 100,697 26,332 0.0073

FE – – 88,776 23,046 0.0313

IV 0.1757 0.0000 31,244 12,091 –

IV 0.0936 0.0000 33,275 12,925 –

FE – – 88,776 23,046 0.2568

IV 0.4834 0.000 31,265 12,108 –

IV 0.2329 0.000 33,276 12,925 –

 0.863nnn (0.179) 0.493nnn (0.185)  0.308nn (0.135) 0.189nn (0.094) 2.635nnn (0.448)

 0.772nnn (0.189) 0.481nnn (0.175)  0.252nn (0.114) 0.177n (0.099) 3.305nnn (0.469) 0.149nnn (0.008) 0.209nnn (0.014)

 1.115nnn (0.361) 2.149 (1.341)  1.960 (1.275) 0.315 (0.204) 2.674nnn (0.897) 0.590nnn (0.230)  0.684nn (0.314)

 0.941nnn (0.328) 2.372nn (1.215)  2.048 (1.151) 0.252 (0.191) 2.400nnn (0.810)  0.220 (0.212)

 0.768nnn (0.198) 0.510nnn (0.188)  0.243nn (0.124) 0.125 (0.103) 3.646nnn (0.501) 0.810nnn (0.011) 0.135nnn (0.017)

 1.072nnn (0.348) 1.272 (1.299)  1.314 (1.230) 0.219 (0.198) 2.900nnn (0.862) 1.086nnn (0.228)  0.295 (0.319)

 0.996nnn (0.328) 2.248n (1.226)  1.834 (1.156) 0.218 (0.190) 2.811nnn (0.805) 0.367n (0.219)

Log Fixed Assets

(b) Tax Rate Difference Log GDP p. C. Log GDP Corruption Index Unempl. Rate Log Payroll Intensity Log Sales

Year D.

O

O

O

 0.684nnn (0.292) O

O

O

 0.792nnn (0.317) O

Industry-Year D.

O

O

O

O

O

O

O

Model Sargan/Hansen Kleinbergen/Paap # Observations # Firms R2 Within

FE – – 86,297 21,992 0.0082

FE – – 76,076 19,202 0.0333

IV 0.5204 0.0000 25,519 9,603 –

IV 0.2594 0.0000 27,262 10,311 –

FE – – 76,076 19,202 0.2434

IV 0.7731 0.000 25,538 9,618 –

IV 0.4512 0.000 27,263 10,311 –

Log Fixed Assets

Notes: Heteroskedasticity robust standard errors adjusted for firm clusters in parentheses. n, nn, nnn indicates significance at the 10%, 5%, 1% level. Specifications (3)–(4) and (6)–(7) employ 50 instruments (comprising the excluded instruments plus the set of exogenous regressors).

further add control variables for the firm’s profitability and payroll intensity which again leaves the qualitative results unaffected, while the corporate tax effect diminishes in size though. In specifications (5)–(7) we reestimate the effect of corporate taxation on a firm’s capital investment in a dynamic framework as specified in Eq. (7) in the previous section. All three specifications indicate that capital investment is sticky as the coefficient estimate for the lagged dependent variable is positive and quantitatively large. Most importantly, the estimations confirm the negative effect of corporate taxes on subsidiary investment whereas the coefficient estimate for the corporate tax effect remains sizable. Column (5) suggests that a one percentage point increase in the corporate tax rate at the subsidiary’s host location reduces capital investment by 1.1% in the short run. The long run effect is larger and

J. Becker et al. / European Economic Review 56 (2012) 1495–1511

1503

Table 3 Quality Effect—Impact of Taxes on Labour and Payroll Intensity Dep. Variable: Col. (1)–(4): Log labour Intensity, Col. (5)–(9): Log Payroll Intensity; Panel 2000–2006. Model

Fixed effects model

GMM model

Explanat. Var.:

(1)

(3)

(2)

Lag Dep. Variable Corporate Tax Rate Log GDP per Capita Log GDP Corruption Index Unempl. Rate Labour Tax Rate PV Capital Allowances Log Sales

(4) nnn

1.675nnn (0.239)  0.589nnn (0.112)  0.057 (0.054) 0.420 (0.078) 0.554 (0.385)  0.013 (0.104)  0.239 (0.428) 0.083 (0.019)

1.556nnn (0.199)  0.448nnn (0.108) 0.013 (0.069) 0.130 (0.062)  1.345nnn (0.313)  0.053 (0.088)  0.986nnn (0.304) 0.080nnn (0.016)

0.409 (0.065) 1.679nnn (0.203)  0.590nnn (0.106) 0.009 (0.014) 0.819nnn (0.080) 0.061 (0.330) 0.102 (0.098) 1.127nn (0.468) 0.190nnn (0.057)

Fixed effects model

GMM model

(5)

(7)

(6)

nnn

0.303 (0.068) 1.584nnn (0.232)  0.666nnn (0.156) 0.030nn (0.017) 0.674nnn (0.193)  0.296 (0.340) 0.176n (0.096) 0.735 (0.494) 0.325nnn (0.096)

(8) nnn

1.300nnn (0.236)  0.309nnn (0.117)  0.071 (0.066) 0.051 (0.072)  0.710n (0.371)  0.024 (0.103)  0.468 (0.351) 0.198nnn (0.017)

1.436nnn (0.200)  0.367nnn (0.109) 0.010 (0.070) 0.017 (0.062)  1.648nnn (0.316)  0.076 (0.090)  1.048nnn (0.298) 0.110nnn (0.016)

0.376 (0.088) 0.658nnn (0.196)  0.245nnn (0.100)  0.029 (0.021) 0.114nn (0.059)  0.491 (0.333) 0.062 (0.098)  0.110 (0.331) 0.401nnn (0.049)

(9) nnn

0.313 (0.091) 0.730nnn (0.238)  0.160nnn (0.128)  0.024 (0.035) 0.057 (0.058)  0.543 (0.330) 0.087 (0.099)  0.139 (0.348) 0.257nn (0.120)

Log Fixed Assets

0.505nnn (0.080) 0.618nn (0.319)  0.012 (0.209)  0.033 (0.032) 0.049 (0.065)  0.427 (0.390) 0.083 (0.111)  0.020 (0.410)

0.038 (0.204)

Year D.

O

 0.718nnn (0.020) 0.286nnn (0.005) O

O

 0.232 (0.234) 0.089nn (0.042) O

O

0.287nnn (0.005) O

O

0.068 (0.052) O

Industry-Year D.

O

O

O

O

O

O

O

O

O

AR(1) AR(2) Sargan/Hansen

– – –

– – –

0.000 0.120 0.794

0.000 0.419 0.643

– – –

– – –

0.000 0.530 0.188

0.000 0.870 0.581

0.000 0.798 0.457

# Observations # Firms R2 Within

92,241 24,034 0.0786

88,776 23,046 0.3904

61,086 17,735 –

50,894 15,389 –

88,074 22,901 0.0305

88,074 22,901 0.2595

58,370 17,061 –

50,350 15,268 –

54,135 16,561 –

Log Wage Log Pre-tax Profitability

0.086 (0.082) O

Notes: Heteroskedasticity robust standard errors adjusted for firm clusters in parentheses. n, nn, nnn indicates significance at the 10%, 5%, 1% level. Specifications (3)–(4) and (7)–(9) employ 78-90 instruments (comprising the excluded instruments as well as the exogeneous regressors).

estimated with 1.8%.14 Specification (6) adds control variables for the firm’s pre-tax profitability and labour intensity which slightly reduces the coefficient estimate for the corporate tax variable in absolute terms. The specification suggests that an increase in the corporate tax rate by a one percentage point reduces the firm’s capital stock by 0.98% in the short run and 1.6% in the long run. Specification (7) again determines the effect of the marginal effective tax rate on corporate capital investment deriving qualitatively and quantitatively comparable results. Furthermore note that these quantitative estimates are also consistent with previous findings in the literature (see e.g. de Mooij and Ederveen, 2003 and Devereux, 2007).15 The validity of the Arellano and Bond (1991) approach relies on two assumptions: the exogeneity of the instruments with respect to the error term and the absence of second order autocorrelation. As indicated in the previous chapter, we apply a Sargan/Hansen overidentification test to check the former assumption and the Arellano and Bond (1991) test for second order autocorrelation to check the latter one. The test statistics are reported at the bottom of Table 1 and indicate that the instruments are valid. In Table 2a and b, we present the first part of our quality estimations which assess the impact of corporate taxation on the pre-tax profitability of corporate investment. Specifications (1) and (2) of Table 2a follow a standard fixed effects approach.

14 The long-run effect is estimated by multiplying the coefficient estimate for the corporate tax rate with 1=ð1g1 Þ whereas g1 depicts the coefficient estimate for the lagged dependent variable in the dynamic panel model. Note that the sample size for the dynamic estimations is smaller than for the fixed effects specifications since for firms to be included in the dynamic estimation framework, we have to observe the fixed asset variable for at least three successive periods (to construct the first difference of the dependent variable and the first difference of the lagged dependent variable). This implies that firms with less successive observations are dropped from the estimation. 15 Note also that the coefficient estimates for the control variables exhibit the expected sign. Solely the negative effect of the corruption index on the affiliate’s capital stock may require some discussion. Theoretically, the effect of corruption on capital investment is ambiguous. While corruption and the associated bribe payments may on the one hand hamper corporate investment (see e.g. Shleifer and Vishny, 1993), corruption may on the other hand facilitate economic activity as it allows agents to get around harmful regulations or delays due to inefficient bureaucratic procedures (Leff, 1964; Lui, 1985). Empirical evidence on the relationship is mixed. While some authors report a negative effect of corruption on corporate activity (see e.g. Fisman and Svensson, 2007), others derive a positive effect of corruption on inward foreign direct investments (see Egger and Winner, 2005). Our results are in line with the latter finding.

1504

J. Becker et al. / European Economic Review 56 (2012) 1495–1511

Table 4 Quality Effect—Impact of Taxes on Labour and Payroll Intensity Dep. Variable: Col. (1)–(4): Log Labour Intensity, Col. (5)–(10): Log Payroll Intensity; Panel 2000–2006. Model

Fixed effects model

GMM model

Explanat. Var.:

(1)

(2)

(3)

(4)

0.404nnn (0.124)  0.425nnn (0.107) 0.040 (0.063) 0.069 (0.064)  1.154nnn (0.333)  0.088 (0.090)  0.401 (0.309) 0.107nnn (0.018)

0.223nnn (0.103) 0.452nnn (0.119)  0.739nnn (0.109) 0.015 (0.012) 0.794nnn (0.078)  0.079 (0.334) 0.079 (0.093) 0.761 (0.447) 0.250nnn (0.052)

0.244nnn (0.064) 0.432nnn (0.122)  0.710nnn (0.158) 0.042nn (0.020) 0.637nnn (0.189)  0.672n (0.350) 0.184nn (0.093) 0.514 (0.450) 0.323nnn (0.099)

Lag Dep. Variable Tax Rate Difference Log GDP per Capita Log GDP Corruption Index Unempl. Rate Labour Tax Rate PV Capital Allowances Log Sales

0.416nnn (0.151)  0.543nnn (0.111)  0.019 (0.044) 0.395nnn (0.079) 0.885nnn (0.409)  0.144 (0.107) 0.315 (0.431) 0.112nnn (0.021)

Fixed effects model

GMM model

(5)

(6)

(7)

(8)

(9)

(10)

0.352nnn (0.125)  0.344nnn (0.109) 0.034 (0.063)  0.035 (0.065)  1.454nnn (0.335)  0.100 (0.092)  0.463 (0.304) 0.132nnn (0.017)

0.284nnn (0.094) 0.199 (0.124)  0.159 (0.135)  0.012 (0.023) 0.065 (0.059)  0.827nn (0.349) 0.114 (0.102)  0.068 (0.353) 0.273nn (0.120)

0.353nnn (0.105) 0.306nn (0.132) 0.024 (0.128)  0.006 (0.012) 0.014 (0.067)  0.260 (0.362) 0.054 (0.166)  0.144 (0.308) 0.169n (0.101)

0.428nnn (0.074) 0.241n (0.143)  0.145 (0.183)  0.020 (0.024) 0.067 (0.071)  0.766n (0.436) 0.112 (0.118)  0.156 (0.420)

0.439nnn (0.096) 0.379nn (0.171)  0.064 (0.201)  0.005 (0.012) 0.013 (0.079)  0.410 (0.442) 0.091 (0.190)  0.288 (0.386)

0.209 (0.156)

0.264 (0.171)

0.136 (0.144)  0.226nn (0.115)  0.044 (0.054) 0.018 (0.075)  0.488 (0.392)  0.102 (0.106) 0.161 (0.358) 0.221nnn (0.018)

Log Fixed Assets nnn

Log Wage

Year D.

O

 0.737 (0.022) 0.274nnn (0.005) O

O

 0.265 (0.228) 0.153nnn (0.041) O

O

0.275nnn (0.005) O

Industry-Year D. AR(1) AR(2) Sargan/Hansen

O – – –

O – – –

O 0.000 0.448 0.605

O 0.000 0.277 0.206

O – – –

O – – –

O 0.000 0.976 0.434

O 0.000 0.909 0.217

O 0.000 0.777 0.255

O 0.000 0.997 0.162

# Observations # Firms R2 Within

78,960 19,999 0.0929

76,076 19,202 0.3984

53,041 15,155 –

44,424 13,235 –

75,525 19,081 0.0330

75,525 19,081 0.2480

43,992 13,146 –

23,967 7,067 –

47,318 14,282 –

24,920 7,357 –

Profitability

0.118nn (0.054) O

0.138n (0.072) O

0.153nnn (0.061) O

0.228nnn (0.088) O

Notes: Heteroskedasticity robust standard errors adjusted for firm clusters in parentheses. n, nn, nnn indicates significance at the 10%, 5%, 1% level. Specifications (3)–(4) and (7)–(10) employ between 78 and 90 instruments (comprising the excluded instruments as well as the exogeneous regressors).

While column (1) presents the baseline estimation, column (2) additionally controls for the affiliate’s labour intensity and affiliate sales as a proxy for firm size. The corporate tax rate is found to exert a strong and significantly negative impact on corporate profitability in both specifications. Quantitatively, column (2) suggests that an increase in the corporate tax rate by a one percentage point reduces the pre-tax profit per asset unit by 1.3% which again resembles the quantitative estimates in previous studies (see e.g. Huizinga and Laeven (2008) who also report a semi-elasticity of reported profits to corporate tax rate changes of 1.3 or Devereux, 2007 for a survey of the literature).16 In column (3) we reestimate specification (2) but additionally account for potential reverse causality with respect to labour intensity and firm size applying the instrumental variables approach proposed by Anderson and Hsiao (1982). For the specifications to be valid, the chosen instrument set of lagged levels of the input factors has to be a relevant predictor of the first-differenced inputs and it has to be uncorrelated with the error term. The relevance of the instruments is confirmed by a Kleibergen/Paap test. The Sargan/Hansen test presented in the table does not reject the exogeneity of the instruments at conventional significance levels. The estimated corporate tax effect on affiliate profitability is quantitatively largely unchanged. This result is moreover robust against using other firm size controls like the corporate fixed asset stock, see specification (4).17 Specifications (5)–(7) reestimate the regressions in columns (2)–(4) but employ pre-tax profits over fixed assets as the dependent variable which yields comparable results.18

16 Similar results are also derived by papers that assess specific multinational profit shifting channels like transfer price distortions or internal debt shifting (see e.g. Clausing, 2003; Bartelsman and Beetsma, 2003; Huizinga et al., 2008). 17 Note, however, that the p-value for the Sargan/Hansen statistic is rather low. The result should thus be interpreted with some caution due to the low power of the test. 18 Note that the sample size drops in the instrumental variable estimations compared to the fixed effects specifications as for firms to be included in the instrumental variables regressions two successive observations for the regressand (and the regressors) have to be available in the data plus information on the lags of the instrumented variables.

J. Becker et al. / European Economic Review 56 (2012) 1495–1511

1505

Table 5 Wage Effect—Impact of Taxes on the Average Wage Dep. Variable: Average Annual Wage, Panel 2000–2006. Explanat. Var.:

(1)

(2)

(3)

Lag Avg. Annual Wage Corporate Tax Rate Log GDP per Capita Log GDP Corruption Index Unempl. Rate Labour Tax Rate Log Sales

nnn

 0.489 (0.117) 0.275nnn (0.052)  0.019 (0.018)  0.381nnn (0.036)  1.131nnn (0.152)  0.102nn (0.046) 0.111nnn (0.009)

Log Labour Productivity Year D.

O

nnn

 0.443 (0.103) 0.234nnn (0.052)  0.021 (0.022)  0.324 (0.034)  1.271nnn (0.142)  0.098nn (0.044) 0.106nnn (0.009) 0.067nnn (0.003) O

0.139 (0.095)  0.661nnn (0.215) 0.666nnn (0.097)  0.037nn (0.017)  0.653nnn (0.059) 0.011 (0.193) 0.114 (0.102) 0.018 (0.137) 0.072 (0.048) O

Industry-Year D.

O

O

O

AR(1) AR(2) Sargan/Hansen

– – –

– – –

0.000 0.748 0.444

# Observations # Firms R2 Within

88,074 22,901 0.4314

82,133 21,822 0.4784

51,709 15,341 –

Notes: Heteroskedasticity robust standard errors adjusted for firm clusters in parentheses. n, nn, nnn indicates significance at the 10%, 5%, 1% level. Specification (3) employs 73 instruments (comprising the excluded instruments as well as the exogeneous regressors).

So far, our estimations have only determined the effect of corporate tax rate changes on affiliate investment. However, the corporate tax difference between the affiliate and other firms within the same multinational group may also be an important predictor of firm profitability. Accordingly, we reestimate the specifications presented in Table 2a employing the corporate tax rate difference between the affiliate and other group members as a tax determinant. The results are presented in Table 2b and indicate that the profitability per capital unit is significantly larger the smaller the affiliate’s corporate tax rate relative to other group members’ tax rates. Specification (4) suggests that an increase in the tax rate differential by a one percentage point reduces the subsidiary’s asset profitability by 0.9%. The second part of our quality estimations assesses corporate tax effects on the firms’ payroll intensity. The results are reported in Tables 3 and 4. Column (1) of Table 3 presents the baseline specification which regresses the labour-to-fixed asset-ratio on the corporate tax rate and a set of control variables (affiliate fixed effects, industry-year fixed effects and time-varying country controls for GDP, GDP per capita, the corruption index, unemployment rate, labour tax wedge, capital allowances and affiliate size as measured by sales). Specification (2) additionally includes the average worker’s wage rate and firm profitability (as measured by pre-tax profits over fixed assets). The corporate tax rate exerts a statistically significant and positive effect on the labour-to-fixed asset-ratio in both specifications. Quantitatively, column (2) suggests that a one percentage point increase in the corporate tax rate enlarges the labour-to-capital-ratio by 1.6%. Moreover, in specifications (3) and (4), we additionally account for dynamic patterns in the adjustments of the labour-tocapital-ratio and reestimate regressions (1) and (2) in an Arellano and Bond (1991) GMM framework including a lagged dependent variable which leaves the coefficient estimate for the corporate tax rate variable largely unaltered.19 Quantitatively, specification (4) suggests that a one percentage point increase in the corporate tax rate raises the labour to capital ratio by 1.6% in the short run and by 2.3% in the long run. Specifications (5)–(10) in turn reassess the effect of corporate taxes on the labour intensity by using the payroll-tocapital-ratio as the dependent variable. The results suggest that the corporate tax rate exerts a significantly positive effect on the payroll-to-capital-ratio which is robust against controlling for time-varying country controls, industry-year fixed effects, affiliate size and affiliate profitability. Moreover, the effect equally appears if we estimate a dynamic GMM model as indicated by specifications (7) and (8). Compared to the regressions which use the labour-to-fixed asset-ratio as the dependent variable, the coefficient estimates for the corporate tax rate variable are quantitatively smaller. Specification (8) suggests that a one percentage point increase in the corporate tax rate raises the payroll-to-capital-ratio by 0.7% in the

19

Using the GMM approach again implies that the sample size drops, see footnote 14.

1506

J. Becker et al. / European Economic Review 56 (2012) 1495–1511

Table 6 Robustness Check - Nonlinear Effects Adding Squared Tax Measures (Col. (1)–(4)) and Allowing the Tax Effects to Vary with Firm Size (Col. (5)–(8)). Dependent Variable:

Log Pre-tax Profitability

Log Payroll Intensity

Log Pre-tax Profitability

Log Payroll Intensity

(1)

(3)

(5)

(7)

(2)

Lag Dependent Variable Corporate Tax Rate Corporate Tax Rate Squared

(4) nnn

0.427 (0.076) 1.572 (1.332)  0.881 (1.741)

2.653 (1.668)  6.433nn (2.752)

 0.842n (0.456)

 0.800nnn (0.548)

Interaction with Fixed Asset Categories 2 Corporate Tax Rate  (Fixed Assets 1–10 Mio.)

0.422 (0.072) 0.526n (0.279)

0.372n (0.198) 0.951nnn (0.310) 1.418nnn (0.447) 1.523nnn (0.515)

 0.117 (0.522)  0.833 (0.925)  1.619 (1.437)  1.323 (1.695)

Corporate Tax Rate  (Fixed Assets 10–50 Mio.) Corporate Tax Rate  (Fixed Assets 50–100 Mio.) Corporate Tax Rate  (Fixed Assets 4100 Mio.)

Control Variables

 1.201nnn (0.277) O

 0.918nnn (0.332)  1.732 (2.190)  0.779nnn (0.344) O

Sargan/Hansen # Observations # Firms

0.2286 33,275 12,925

0.2582 27,262 10,311

Tax Rate Difference Tax Rate Difference Squared

0.421nnn (0.064) 0.175 (0.324)

0.670nnn (0.240) 1.177nnn (0.416)

 0.715 (0.597)  1.490 (1.143)

Corporate Tax Rate  (Fixed Assets 4 50 Mio.)

(8) nnn

0.429 (0.074)

Interaction with Fixed Asset Categories 1 Corporate Tax Rate  (Fixed Assets 10–50 Mio.)

Fixed Assets

(6)

nnn

0.344n (0.183) O

0.228 (0.142)  0.659 (0.826) 0.209 (0.156) O

 1.258nnn (0.281) O

 1.250nnn (0.274) O

0.242nnn (0.161) O

0.119 (0.145) O

0.489 54,135 16,561

0.252 47,318 14,282

0.2821 33,275 12,925

0.3220 33,275 12,925

0.162 54,135 16,561

0.261 54,135 16,561

Notes: Heteroskedasticity robust standard errors adjusted for firm clusters in parentheses. n, nn, nnn indicates significance at the 10%, 5%, 1% level. The specifications using profitability as dependent variable employ 51–54 instrumental variables (comprising the excluded instruments and the exogenous regressors). The specifications using payroll intensity as dependent variables employ 85–108 instrumental variables (comprising the excluded instruments and the exogenous regressors).

short run and by 1.1% in the long run. The findings moreover prevail if we use the company’s fixed asset stock as a size control, see specification (9). Furthermore, we reestimate the specifications presented in Table 3 employing the tax rate differential to other group affiliates as a tax measure. The results are presented in Table 4. Specifications (1)–(4) yield that the labour intensity of production rises in the corporate tax rate differential to other group affiliates. The results are hence in line with the notion that multinational groups tend to locate their labour intensive investment projects at high-tax affiliates within the group and vice versa. In line with our discussion in Section 2, increases in the corporate tax rate and the associated rise in the costs of capital relative to the costs of labour thus indeed appear to induce firms to substitute from capital to labour inputs and locate labour intensive projects at the respective affiliate. Specification (4) suggests that an increase in the corporate tax rate by one percentage point raises the labour-to-fixed asset-ratio by 0.4% in the short run and 0.6% in the long run. Somewhat weaker results are found if we regress the payroll-to-capital-ratio on the tax rate differential between the affiliates and other firms in the multinational group as presented in specifications (5)–(10). The coefficient estimate for the tax difference variable does not gain statistical significance in the static regressions unless we include firm profitability as a control variable in specification (6). Reestimating specification (6) in a dynamic model dampens the coefficient estimate for the tax differential which now just fails to be statistically significant at the 10% level (see specification (7)). Nevertheless, dropping small firms with less than 50 employees from the regression (which are expected to react the least sensitively to tax rate changes) quantitatively increases the coefficient estimate and renders it statistically significant at the 5% level as reported in specification (8). The coefficient estimate for the tax difference variable moreover slightly increases in size and significance if we reestimate specifications (7) and (8) using the corporate fixed asset stock as a size control (instead of sales), see specifications (9) and (10). The fact that the corporate tax effects on payroll-intensity fall short of the labour-intensity estimates suggests that taxes exert a negative impact on the average wage rate earned by the affiliate’s workers which is in line with previous findings in the literature (see Arulampalam et al., 2007). This result is confirmed in Table 5 where we regress the logarithm of the average worker’s wage level on the corporate tax rate and find a significant and negative effect which suggests that

J. Becker et al. / European Economic Review 56 (2012) 1495–1511

1507

Table 7 Simultaneous Equation Model. Explanat. Var.:

(1)

Dep. Var.:

Log fixed Assets

(2) Log pre-tax Profitability

Log wage Intensity

Lag Dep. Variable Corporate Tax Rate Log GDP per Capita Log GDP Corruption Index Unempl. Rate

 0.550nnn (0.215)  0.288 (0.522) 0.888n (0.488) 0.269nnn (0.072) 1.223nnn (0.291)

 2.298nnn (0.574) 4.573nnn (1.580)  2.068 (1.550) 0.169 (0.204) 2.856nnn (0.926)

Labour Tax Rate PV Capital Allowances

0.197 (0.271)

Log Fixed Assets Log Wage Intensity Log Pre-tax Profitability

 0.371nnn (0.097)  0.203nnn

 2.989nnn (0.311)  1.335nnn (0.239)

0.518n (0.275) 1.755nnn (0.625)  1.541nnn (0.618)  0.083 (0.085)  0.896nn (0.420) 0.092 (0.079)  0.391 (0.395)  0.335nnn (0.165)

Log fixed Assets 0.197nnn (0.041)  0.652nnn (0.186) 0.774 (0.531)  0.028 (0.510) 0.104 (0.067) 0.583n (0.310)

Log pre-tax Profitability

 1.272nnn (0.522) 4.600nnn (1.472)  3.554nnn (1.425) 0.182 (0.191) 2.449nnn (0.858)

0.502 (0.317)

0.013

 0.473nnn (0.055)  0.093nnn

 0.837nnn (0.273)  0.352n (0.207)

Log wage Intensity 0.569nnn (0.070) 0.570nn (0.280) 1.706nnn (0.679)  1.435nn (0.681)  0.096 (0.088)  1.004nnn (0.402) 0.153n (0.092)  0.211 (0.412)  0.294nn (0.151)

0.013

Year D.

(0.023) O

O

(0.042) O

(0.020) O

O

(0.029) O

Industry-Year D.

O

O

O

O

O

O

# Observations

21,079

20,929

n nn nnn

indicates significance at the 10%, 5%, 1% level. All equations of systems (1) and (2) are estimated in first differences and include lagged Notes: , , levels of the variables as exclusion restriction.

part of the additional tax burden on multinational firms is in fact borne by the multinationals’ employees. Note in this context that the corporate tax effect on the labour-to-capital-ratio is not driven by this wage adjustment as we control for workers’ wages in the according specifications presented in Tables 3 and 4. Furthermore note that the Sargan-Hansen overidentification test does not reject the hypothesis that the instruments are valid in the Arellano and Bond (1991) and Anderson and Hsiao (1982) specifications presented in this section. In some regressions, the p-value nevertheless turns out rather small though (in particular specifications (3) and (4) of Table 2a, specifications (3) and (7) of Table 3 and specification (10) of Table 4) and hence the associated results should be interpreted with some caution due to the low power of the test. So far our analysis has tested for linear tax effects on the firm’s capital stock, pre-tax profitability and labour intensity. Our theoretical discussion in Section 2 in turn suggests that the corporate tax effect on pre-tax profitability and labour intensity may also take on a non-linear form. To assess this possibility, we reestimate the regressions presented in Tables 2a to 4, augmenting the set of regressors by the square of the tax measure. Specification (1) in Table 6 reassesses the instrumental variable model presented in column (4) of Table 2a and regresses the affiliate’s pre-tax profitability on the corporate tax rate and its square. The coefficient estimates indeed point to a non-linear shape of the corporate tax effect which becomes stronger for larger rates. Evaluated at the sample mean, the corporate tax effect on the affiliate’s pre-tax profitability is determined with 1.52 and is thus quantitatively slightly larger than our linear baseline estimates. Specification (2) of Table 6 reestimates the relationship employing the tax rate difference between the affiliate and other firms in the corporate group as tax measure. Here, the coefficient estimate for the squared tax variable is statistically insignificant, implying that we cannot reject the hypothesis of a linear relationship. Augmenting the model by higher order polynomials is furthermore rejected by a Wald test. Analogously, specification (3) of Table 6 reestimates the GMM model in column (9) of Table 3, regressing the affiliate’s payroll intensity on the corporate tax rate and its square. Again, the coefficient estimate for the squared tax rate term turns out statistically insignificant, implying that we cannot reject the hypothesis of a linear relationship. Analogous results are derived if we employ the tax difference measure as tax determinant (see specification (4) of Table 6). Moreover, adding higher order polynomials is again rejected by a Wald test. Furthermore, we add specifications which allow the corporate tax effects on pre-tax profitability and labour intensity to vary with firm size. To do so, we classify firms in size categories using a broad definition of three size classes (capital stock below 10 million US dollars, 10–50 million US dollars and above 50 million US dollars) and a more refined one using five

1508

J. Becker et al. / European Economic Review 56 (2012) 1495–1511

size classes (capital stock below 1 million US dollars, 1–10 million US dollars, 10–50 million US dollars, 50–100 million US dollars and above 100 million US dollars). The regressions in columns (5) and (6) reestimate specification (4) in Table 2a allowing the corporate tax effect on pre-tax profitability to vary according to the broad (specification (5)) and refined (specification (6)) size categories. The coefficient estimates for the tax rate measure are negative as in our baseline estimation and, in absolute terms, increase with firm size. Although not statistically significant, the coefficient estimates thus weakly suggest that it is especially large firms whose profitability responds to corporate taxation. Analogously, specifications (7) and (8) reestimate the payroll intensity model presented in column (9) of Table 3, allowing the corporate tax effect to vary with firm size. The coefficient estimates for the interaction terms are positive, statistically significant and increase with firm size. This again suggests that it is especially large firms that take corporate taxes into account when structuring their activities. Finally note that our baseline specifications estimate the impact of corporate taxation on the capital stock, firm profitability and payroll intensity in a separate equation framework. Our theoretical considerations in turn point to potential interdependencies between the three variables. In our baseline specifications, we account for that by adding the other two left-hand-sidevariables to the set of regressors. First-differencing and specifying a two-stage least squares model which instruments for the endogeneous regressors with deeper lags of their levels already implies the estimation of multiple equations. While both, two stage least squares and the Arellano–Bond approach yield consistent estimates, our baseline analysis focuses on the Arellano– Bond approach for efficiency reasons. As a robustness check, we also reestimate the full model as a system of equations where each equation in the system contains the other two left-hand-side-variables as regressors. The results are presented in specifications (1) and (2) Table 7. In both specifications, all three model equations are estimated in first differences. In specification (2) the fixed asset and payroll intensity equations are augmented by a lagged dependent variable, analogously to our baseline estimates.20 In line with our baseline results, both specifications suggest that an increase in the corporate tax rate reduces the corporate capital stock and firm profitability while it raises the firm’s payroll intensity. Quantitatively, the point estimates for the corporate tax variable in the profitability equation and the payroll intensity equation largely resemble our baseline estimates (see for example specification (2) of Table 7, specification (7) of Table 2a and specification (9) of Table 3), while the coefficient estimate for the corporate tax variable turns out quantitatively somewhat smaller in the capital stock equation, albeit not statistically different from our baseline results. 6. Discussion of the results Our estimations indicate that corporate taxation does not only exert a negative effect on the quantity of capital investment but also affect the quality of corporate investment projects located in a country. We find that corporate taxation reduces profitability and increases ‘payroll intensity’ of investment. Plugging in our coefficient estimates from column (6) in Table 1, column (4) in Table 2a and column (9) in Table 3 yields the following semi-elasticities: An increase in the corporate tax rate by a one percentage point reduces the quantity of investment by 1.59% (eK ¼ gK2 =ð1gK1 Þ ¼ 1:59). The profit per unit of investment declines by 1.02% (eb ¼ gb2 ¼ 1:02) and the labour income generated per unit of investment increases by 1.25% (ep ¼ gp2 =ð1gp1 Þ ¼ 1:25). What does this imply for the overall effects of corporate tax changes on the profit tax base and on payroll? Note that, with B ¼bK, the semi-elasticity of the corporate tax base with respect to the corporate tax rate can be written as @B 1 @K b @b K ¼ c  þ c  ¼ eK þ eb ¼ 1:591:02 ¼ 2:61 @t c B @t B @t B

ð8Þ

An increase in the corporate tax rate by a one percentage point is predicted to lower the corporate tax base by 2.6%. Quantity and quality effect are found to significantly contribute to this result, with the quality effect accounting for 39% of the overall effect. Similarly, using P¼pK, we can calculate the effect on payroll as @P 1 @K p @p K ¼ c  þ c  ¼ eK þ ep ¼ 1:59 þ 1:25 ¼ 0:34 @t c P @t P @t P

ð9Þ

Thus, while corporate tax increases drive investment out of the country and consequently diminish payroll, this negative quantity effect is mitigated, albeit not reversed, by a positive quality effect which suggests that labour partly substitutes for capital investment. The estimations suggest that the quality effect compensates around 79% (1:25=ð1:59Þ ¼ 0:79) of the negative quantity impact on payroll. The overall semi-elasticity is calculated with  0.34, i.e. increasing the corporate tax rate by a one percentage point reduces the labour tax base by 0.34% on average.21 20

Note that we also ran specifications where each equation in the system includes a full set of firm fixed effects which yields similar results. Note that the sample size differs across the specifications used to evaluate quantity and quality effects of corporate taxation against each other which is driven by the differing regression requirements concerning the availability of lagged variables in the specifications (see also footnotes 14, 18 and 19). In a robustness check, we constructed a subsample of observations which are included in all three specifications and find comparable results. Furthermore note that the evaluation of quality and quantity effects sketched above accounts for direct corporate tax effects on the capital stock, firm profitability and payroll intensity only. Taxation might, however, also induce indirect effects. Our baseline regressions for example suggest that the corporate pre-tax profitability is negatively affected by firm size (as measured by the corporate capital stock) and payroll intensity. The direct corporate tax effect on the latter two variables may thus eventually also impact on the firm’s pre-tax profitability. Taking the indirect effects into account derives 21

J. Becker et al. / European Economic Review 56 (2012) 1495–1511

1509

For illustrative purposes, it is interesting to use these results in the formula for the optimal tax rate in (5). Consider firstly the case where the government ignores the quality effects of corporate taxation. Eq. (5) then reads t c0n ¼ 0:63

að1sÞ p f  ðZfÞ b ðZfÞ

ð10Þ

Under the extreme assumptions that all firms are foreign owned (f ¼ 0) and that a job loss in the international sector in our model will immediately be compensated by the creation of a job in the rest of the economy (s¼1), the optimal corporate income tax rate is equal to 63%. If these assumptions are relaxed, a lower rate is optimal. Assume instead that s¼ 0.8, i.e. a loss of labour income of 1 Euro in the industrial sector can only be replaced by an increase in income in the service sector of 80 Cents. Assume further that the share of domestic households in firm ownership (f) is still zero, the payroll tax (t p ) is 50%, the marginal cost of public funds (Z) is 1.25, and the payroll–profit ratio in the industrial sector (p=b) is 2/1. In this case, the optimal tax rate is equal to 27%. Now, assume that the government takes account of quality effects. The optimal tax rate reads t c1n ¼ 0:38

að1sÞ p f 0:13 ðZfÞ b ðZfÞ

ð11Þ

With f ¼ 0 and s¼1, the optimal tax rate is equal to 38%, compared to 63% in the case where the government ignores quality effects of corporate taxation. Thus, taking into account the impact of corporate taxation on the quality of investment leads to lower optimal corporate tax rates. Interestingly, the picture changes, if we consider the second scenario with s¼0.8. Then, the optimal tax rate with quality and quantity effects equals 33%, compared to 27% if quality effects are ignored. Thus, in the first scenario with s ¼1, neglecting quality effects increases the optimal tax rate, while in the second scenario with s ¼0.8, neglecting quality effects reduces the optimal corporate tax rate. The reason is that, in the first scenario, the payroll effects do not play a role because each loss in labour income in the international sector is offset by a corresponding gain in the domestic service sector. Only the quality effect on the corporate tax base has to be taken into account. Compared to the baseline scenario where we abstract from quality effects, this implies that countries have a higher incentive to lower their corporate tax rate. This does not only allow them to attract more capital. In addition, each unit of capital contributes more to the corporate tax base. In the second scenario, losses in labour income due to lower investment quantity or quality cannot be fully compensated by the service sector. Here, both quality effects affect the tax rate choice. The quality effect in terms of labour income generated per unit of investment acts as a break on tax cuts since reduced corporate taxes lead to less labour intensive production in the international sector. Note that, although the example assumes that 80% of the labour income lost in the international sector is replaced by employment in the service sector, the quality effect on employment is strong enough to overcompensate the quality effect in terms of contributing to the corporate income tax base. Summarizing, our analysis suggests an important link between the payroll and corporate tax base contribution of FDI and the corporate tax rate choice. The simple model sketched in this section especially underpins the quantitative importance of welfare effects related to the payroll contribution of FDI which have so far largely been ignored in the economic literature. 7. Concluding remarks Governments throughout the world have lowered corporate tax rates in order to attract FDI. In this paper, we argue that taxes do not only affect the quantity of FDI but also the extent to which FDI contributes to corporate tax revenue and labour income, i.e. the quality of FDI. We use detailed data on European multinational firms to empirically measure quantity and quality effects of corporate taxation on FDI. With regard to quantity effects, we find a semi-elasticity of the capital stock to corporate tax rate changes of approximately  1.6, which is in line with the results of previous studies. In terms of quality effects, our results suggest that corporate taxation substantially lowers the profitability of corporate investment projects and increases the labour intensity of production. The importance of tax effects on investment quality can be illustrated by considering how taxes affect the corporate tax base or labour income in general. With regard to the corporate tax base, our findings suggests that only around 60% of the corporate tax base reduction in the wake of tax increases is driven by a decline in investment size while the other 40% are due to a diminishing profitability of the investment projects. With regard to the multinational’s payroll, the estimations suggest that the quality effect compensates 79% of the negative quantity impact. These results may have important policy implications. If taxes equally distort the quality and the quantity of FDI, measures of investment distortions based on investment quantity are likely to misrepresent the efficiency cost of taxation. (footnote continued) Eb ¼ 0:15 ( ¼ 1:021:59  ð1:05Þ0:63  1:25) (see column (4) of Table 2a). The contribution of the quality effect to the decline in the corporate tax base is thus 8.3%. In the baseline capital and payroll intensity regressions, the coefficient estimates for the other two endogeneous regressors (profitability and payroll intensity in the capital stock equation and profitability and fixed assets in the wage intensity equation) turn out statistically insignificant, suggesting that indirect effects can be neglected. Using the simultaneous equation model in specification (2) of Table 7 and accounting for both, direct and indirect effects, in turn derives EK ¼ 1:44, Eb ¼ 1:06 and Ep ¼ 1:84. The quality effect thus makes up around 40% of the overall corporate tax base effect and overcompensates the quantity effect in the calculation of the payroll sum effect. In any case, these back-of-the-envelope calculations suggest the quality effects to be important relative to the quantity effect.

1510

J. Becker et al. / European Economic Review 56 (2012) 1495–1511

Our analysis suggests that policy-makers should not only be concerned about the size of FDI inflows but equally about the profitability and labour-intensity of the incoming projects. Thus, our results may alter the interpretation of certain ‘‘success stories’’. For example, from 1990 to 2000, Germany increased its stock of foreign held capital by about 520% (source: OECD) and performed much better in attracting foreign FDI than the UK (115%), France (205%), the US (180%) or Japan (410%). The standard way of reading these figures is that Germany attracted FDI despite its relatively high tax rates (before the tax reform in 2001, the corporate tax rates in Germany were between 52 and 58% and thus among the highest throughout the developed world). Our results suggest that these large inflows of capital were of minor quality in the profitability dimension but of higher quality in terms of generating labour income. Evaluated for the year 2000, our findings indicate that the profitability of German investment would have been by 18% higher and the payroll intensity by 22% lower if Germany had reduced its tax to the average of other EU countries.22

Acknowledgments We are indebted to Theo Eicher and three anonymous referees for comments on earlier versions of this paper. We would also like to thank Nadja Dwenger, Andreas Haufler, Martin Hellwig, Harry Huizinga, Christian Keuschnigg, Kai Konrad, Viktor Steiner and participants at research workshops in Berlin, Bonn, Florence, Munich and Tilburg for their helpful comments on an earlier version of the paper. The usual disclaimer applies. We gratefully acknowledge financial support from the ESRC (Grant No. RES -060-25-0033). Appendix A. Supplementary data Supplementary data associated with this article can be found in the online version at http://dx.doi.org/10.1016/j.euroecorev. 2012.07.001.

References Aitken, B., Harrison, A., Lipsey, R., 1996. Wages and foreign ownership. A comparative study of Mexico, Venezuela, and the United States. Journal of International Economics 40 (3–4), 345–371. Arulampalam, W., Devereux, M.P., Maffini, G., 2007. The Indirect Incidence of Corporate Income Tax on Wages. Oxford University Centre for Business Taxation Working Paper Series No. 07/07. Anderson, T.W., Hsiao, C., 1982. Formulation and estimation of dynamic models using panel data. Journal of Econometrics 18 (1), 47–82. Arellano, M., Bond, S.R., 1991. Some test of specification for panel data: Monte Carlo evidence and an application to employment equations. Review of Economic Studies 58 (2), 277–297. Bartelsman, E.J., Beetsma, R.M.W.J., 2003. Why pay more? Corporate tax avoidance through transfer pricing in OECD countries. Journal of Public Economics 87, 2225–2252. Becker, J., Fuest, C., 2007. Quantity Versus Quality—The Composition Effect of Corporate Taxation on Foreign Direct Investment. CESifo Working Paper No. 2126. Bernard, A.B., Jensen, J.B., 2007. Firm structure, multinationals, and manufacturing plant deaths. Review of Economics & Statistics 89 (2), 193–204. Bernard, A.B., Jensen, J.B., Redding, S.J., Schott, P.K., 2007. Firms in international trade. Journal of Economic Perspectives 21 (3), 105–130. Clausing, K.A., 2003. Tax motivated transfer pricing and US intrafirm trade prices. Journal of Public Economics 87, 2207–2223. de Mooij, R., Ederveen, S., 2003. Taxation and foreign direct investment: a synthesis of empirical research. International Tax and Public Finance 10 (6), 673–693. Devereux, M.P., Griffith, R., Klemm, A., 2002. Corporate income tax reforms and international tax competition. Economic Policy 17 (35), 450–495. Devereux, M.P., 2007. The Impact of Taxation on the Location of Capital, Firms and Profit: A Survey of Empirical Evidence. Oxford University Centre for Business Taxation Working Paper Series No. 07/2. Dischinger, M., Riedel, N., 2011. Corporate taxes and the location of intangible assets within multinational firms. Journal of Public Economics 95 (7–8), 691–707. Dwenger, N., Steiner, V., 2008. Effective Profit Taxation and the Elasticity of the Corporate Income Tax Base: Evidence from German Corporate Tax Return Data. DIW Discussion Paper No. 829. European Commission, 2006. Structures of the Taxation Systems in the European Union, Directorate Generale Taxation and Customs Union, Brussels. Fisman, R., Svensson, J., 2007. Are corruption and taxation really harmful to growth? Firm level evidence. Journal of Development Economics 83 (1), 63–75. Griffith, R., Simpson, H., 2004. Characteristics of Foreign-Owned Firms in British Manufacturing, NBER Chapters. In: Seeking a Premier Economy: The Economic Effects of British Economic Reforms, 1980–2000, pp. 147–180. Gruber, J., Rauh, J., 2007. How elastic is the corporate income tax base?. In: Auerbach, A.J., et al. (Eds.), Taxing Corporate Income in the 21st Century, 2007, Cambridge University Press, Cambridge and New York, pp. 140–163. Grubert, H., Mutti, J.H., 1991. Taxes, tariffs and transfer pricing in multinational corporate decision making. Review of Economics and Statistics 73 (2), 285–293. Hanlon, M., Maydew, E.L., 2009. Book-tax conformity: implications for multinational firms. National Tax Journal 72 (1), 127–153. ¨ Haufler, A., Stahler, F. Tax Competition in a Simple Model with Heterogeneous Firms. How Larger Markets Reduce Profit Taxes. CESifo Working Paper No. 2687. International Economic Review, forthcoming.

22 In 2000, the German corporate tax rate was 17.6 percentage points higher than the GDP-weighted EU average. Our estimates yield semielasticities for the effect of corporate taxes on pre-tax profitability and payroll intensity of  1.02 and 1.25 (see Section 6). A reduction in the German corporate tax rate by 17.6 percentage points would thus have increased the pre-tax profit per capital unit by 17.95% ( ¼ 1:02  0:176) and decreased the payroll contribution per capital unit by 22.0% ( ¼ 1:25  0:176).

J. Becker et al. / European Economic Review 56 (2012) 1495–1511

1511

¨ Heyman, F., Sjoholm, F., Tingvall, P.G., 2007. Is there really a foreign ownership wage premium? Evidence from matched employer–employee data. Journal of International Economics 73 (2), 355–376. Hines, J.R., Rice, E.M., 1994. Fiscal paradise: foreign tax havens and american business. Quarterly Journal of Economics 109 (1), 149–182. Huizinga, H., Laeven, L., 2008. International profit shifting within european multinationals. Journal of Public Economics 92 (5/6), 1164–1182. Huizinga, H.P., Laeven, L., Nicodeme, G., 2008. Capital structure and international debt shifting. Journal of Financial Economics 88 (1), 80–118. KPMG, 2006. Corporate Tax Rate Survey. KPMG International, United Kingdom. Leff, N., 1964. Economic development through bureaucratic corruption. American Behavioral Scientist 82 (2), 337–341. Loretz, 2008. Corporate taxation in the OECD in a wider context. Oxford Review of Economic Policy 24 (4), 639–660. Lui, F.T., 1985. An equilibrium queuing model of bribery. Journal of Political Economy 93 (4), 760–781. Shleifer, A., Vishny, R.W., 1993. Corruption. Quarterly Journal of Economics 108 (3), 599–617. Weichenrieder, A.J., 2009. Profit shifting in the EU: evidence from Germany. International Tax and Public Finance 16 (3), 281–297.