The Impact of CFC Legislation on Multinational Firms Evidence from a Two-dimensional Regression Discontinuity Approach

Transcription

1 The Impact of CFC Legislation on Multinational Firms Evidence from a Two-dimensional Regression Discontinuity Approach Peter Egger Valeria Merlo and Georg Wamser October 14, 2010 Abstract Controlled foreign company (CFC) rules are frequently imposed by states as part of their anti tax-avoidance legislation. This paper investigates how the German CFC rule affects the foreign activities of German multinationals. The investigation makes use of two discontinuities with respect to the application of the German CFC rule. This allows us to apply a two dimensional regression discontinuity approach to obtain estimates for the impact of the CFC provisions. Using a unique database of the universe of Germany s affiliates abroad for the empirical analysis, we find that subsidiaries which are affected by the CFC rule use a significantly lower amount of fixed assets. Key words: CFC Rule; Regression Discontinuity Design; Tax Avoidance; Multinational Firms; Micro-Level Data; Corporate Income Tax JEL classification: H25; F23 Acknowledgements: We gratefully acknowledge support by members of the research center of the German Bundesbank. Affiliation: ETH Zürich, CEPR, CESifo, Leverhulme Centre for Research on Globalisation and Economic Policy (GEP) at the University of Nottingham, and Oxford University Centre for Business Taxation (OUCBT). Address: ETH Zürich, KOF, Weinbergstrasse 35, WEH E6, 8092 Zürich, Switzerland. egger@kof.ethz.ch. Affiliation: ETH Zürich and MGSE. Address: ETH Zürich, KOF, Weinbergstrasse 35, WEH E7, 8092 Zürich, Switzerland. merlo@kof.ethz.ch. Affiliation: ETH Zürich and CESifo. Address: ETH Zürich, KOF, Weinbergstrasse 35, WEH C10, 8092 Zürich, Switzerland. wamser@kof.ethz.ch.

2 1 Introduction The profit-shifting activities of multinational firms have attracted a lot of attention both in policy debate and in economic research. Several recent studies provide evidence that multinational firms shift income to low-tax jurisdictions in order to save taxes (see, for example, Huizinga and Laeven, 2008, or Weichenrieder, 2009). Since this implies that countries are constrained in their abilities to raise corporate tax revenue, tax legislation responds by establishing anti tax-avoidance measures. The two most important of such measures in international tax law are thin-capitalization rules and controlled-foreign-corporation (CFC) rules, which have been established by many countries during the last years. There is little empirical evidence, however, on how such rules affect the behavior of multinational firms. Some evidence on the effectiveness of thin-capitalization rules has been provided by Buettner, Overesch, Schreiber and Wamser (2008). Their findings indicate that tighter rules cause a reduction in loans from foreign subsidiaries and effectively remove the incentive to use such loans for tax-planning purposes. Weichenrieder and Windischbauer (2008) as well as Overesch and Wamser (2010) investigate a reform of the German thin-capitalization rule and confirm that firms respond by using less internal debt. In a recent study, Ruf and Weichenrieder (2009) explore the consequences of the German CFC rule. Their study shows that the German CFC provisions are quite effective in restricting passive investments of German multinationals in low-tax jurisdictions. CFC rules specifically aim at taxing foreign income that is sheltered from home-country taxation. If the home country of the multinational has established a tax exemption system (and is a high-tax country), the benefit of low taxes at foreign locations is obvious: foreign income is taxed at a low rate and exempt from high domestic taxes. But the multinational can also benefit if the home country of the multinational firm applies a tax credit system (as, for example, the U.S.), in particular if income is not repatriated to the home country. In the case that foreign income is generated from productive (active) operations, tax-shelter activities are usually accepted. Yet this exemption privilege is often no longer granted if income is associated with passive investment (tainted-income approach). Once multinational firms exploit international tax-rate differentials to avoid taxes, such income typically arises at entities of the multinational firm located in low-tax jurisdictions, which often design their tax systems in a way that it attracts passive investments or paper prof- 2

3 its (see Nicodème, 2009). As this is considered as harmful tax practice (see OECD, 1998), the OECD recommendation is to tax passive income following a residence-based approach of taxation. The OECD specifically emphasizes that countries that do not have such rules [should] consider adopting them and that countries that have such rules [should] ensure that they apply in a fashion consistent with the desirability of curbing harmful tax practices (OECD, 1998, p ). Although CFC rules seem to be perceived as increasingly important, apart from the aforementioned literature on German multinationals, only a few studies have looked at the impact of such rules. Altshuler and Hubbard (2003) examine the effects of the U.S. Tax Reform Act of 1986, which included changes in the U.S. CFC legislation (Subpart F). They show that tighter rules made it more difficult for firms to defer U.S. taxes on financial services income held in low-tax jurisdictions. In later studies, however, Altshuler and Grubert (2006) as well as Mutti and Grubert (2006) argue that hybrid entities have been increasingly used by U.S. multinationals, allowing the firms to avoid anti-abuse provisions. They further argue that the U.S. Treasury, for this reason, faces new challenges in taxing income from passive investments. The lack of research in this field may be owed to the problem that evaluating the impact of anti-abuse provisions is generally a difficult task. This is particularly true in the context of multinational firms, because the nature of such firms (i.e. the fact that they operate on a worldwide basis) and the complex interrelations between foreign subsidiaries (e.g. in terms of financing) makes it difficult to identify causal effects of national regulations such as CFC rules. Other difficulties with respect to empirical identification are related to how such rules are designed. For example, whether or not the German CFC rule is applied depends on whether the host country is defined as a low-tax jurisdiction. But comparing investments in low-tax (or tax haven) countries with those in high-tax countries can easily lead to wrong conclusions. The goal of this study is to provide estimates for the effects of the German CFC rule on the investment activity of foreign subsidiaries. Weichenrieder (1996) demonstrates in a theoretical model that if a CFC rule allows for some investment in tax preferred passive investment, the multinational may be able to reduce its cost of capital. The analysis emphasizes the case where the application of the CFC rule depends on whether the share of passive income in total income exceeds a certain threshold level. Since more real investment simultaneously eases the restriction and allows for more tax preferred passive income, the cost of capital may be reduced. While this marginal analysis provides insights on how the optimal investment behavior 3

4 of subsidiaries not affected by the CFC rule (because firms take into account the CFC threshold) is changed, we are interested in the case where a foreign subsidiary just violates the optimality condition, so that, according to the German CFC legislation, foreign income arising from passive investment becomes immediately taxable in Germany. In consequence of the high German tax level, we would expect that this has a negative impact on subsidiaries assets. Our investigation makes use of two discontinuities with respect to the application of the German CFC legislation to identify effects. From an empirical point of view, we would want an experimental setting where assignment to the group of subsidiaries that is affected by the CFC rule (the treatment group) and to group that is not (the control group) is random. While such a random experiment is not available, we exploit two important features of the German CFC rule for identification. To be precise, we make use of the fact that treatment is based on two discontinuous functions. First, the German CFC legislation applies only to investments in low-tax countries, defined as countries where the statutory tax rate lies below 30% (25% since 2001). Second, CFC legislation does not come into operation if the underlying gross return from passive investment is below 10% of the overall gross return (see Section 2 for more details). The mentioned thresholds give rise to a regression discontinuity design, allowing us to compare outcomes of foreign subsidiaries that are just below the tax and above the passive-assets cutoff values, with those just above and below the respective thresholds. This novel identification design enables us to provide causal evidence on the impact of CFC rules on the foreign activities of multinational firms. Using a unique dataset provided by the Deutsche Bundesbank (the German Central Bank) which captures virtually all activities of German multinational firms in foreign countries, we find that CFC-rule treatment is associated with significantly less investment in fixed assets. This is what we would expect as the CFC provisions imply that relevant income of a treated foreign subsidiary is immediately taxable in Germany. The negative effect, to be specific, may be explained by the high German tax level, because the overall cost of capital may increase significantly in the treatment case. Our preferred estimation specification implies that the effect amounts to e9.66 million. The size of this effect increases, the closer we get to the threshold levels. The paper is structured as follows. In Section 2 we describe the relevant aspects of the German CFC legislation. Section 3 proposes the empirical identification strategy. Section 4 presents the data used. The results and 4

5 several robustness checks are shown in Section 5. Section 6 concludes. 2 German CFC Legislation The German CFC legislation was introduced in 1972 under the German Foreign Transactions Tax Act (Außensteuergesetz, AStG). 1 It generally aims at preventing the sheltering of passive investment income in low-tax countries from higher home-country taxation. Profits arising at foreign subsidiaries of German multinationals are usually exempt from German taxation. But on the basis of 7 14 AStG the statutory body of the German CFC rule income from investment with capital investment characteristics may be subject to home taxation. 7 (6a), AStG states explicitly that such income includes interest and dividend received from financial investments. 8 AStG further defines active business operations that are not within the scope of the CFC rule, including the activities of banking and insurance firms. For a foreign subsidiary to be treated by the CFC rule, two basic criteria have to be fulfilled. First, according to 8 (3) AStG, the tax level of the host country lies below 30% (since 2001, 25%). Second, according to 9 AStG, income from passive investment has to exceed 10% of total income. If these requirements are fulfilled, relevant income as defined by 7 AStG is immediately taxable in the residence country (see 10 AStG). While passive income is included in the taxable income of the German residents, taxes paid at the foreign locations may be deducted or credited against the German tax. Since the German tax level is significantly higher 2 compared with the tax level faced by the subsidiary at the low-tax location (below 30% or 25%, respectively), the implication is straightforward: in case the German CFC legislation applies to a foreign subsidiary, the effective tax burden of the subsidiary is expected to increase as the high German tax becomes definite. For this reason, we expect that CFC-rule treatment leads to adverse effects with respect to subsidiaries usage of fixed assets. 3 1 Ten years earlier, under the Revenue Act of 1962, the U.S. enacted Subpart F provisions to hamper the opportunities of multinationals to avoid U.S. taxes. Subpart F income is defined as income from certain types of passive investments, such as interest or dividend income from securities (see Altshuler and Hubbard, 2003). 2 Note, though, that Germany significantly reduced its corporate tax rate (including the local business tax) during the time period considered (from approx. 53% in 1996 to approx. 33% in 2007). 3 Note that this result would be in line with the analysis of Weichenrieder (1996). To be precise, the negative treatment effect would be associated with multinationals violating 5

6 3 Empirical Approach We aim at estimating the impact of the German CFC rule on the holdings of fixed assets at foreign subsidiaries of the universe of German multinational enterprises the German Bundesbank collects data for. Since the CFC legislation provides two discontinuities determining treatment in compliance with the CFC-rule as outlined in the previous section, a regression discontinuity (RD) design based on two discontinuities appears to be a natural identification strategy. To introduce the approach, it is useful to define the binary indicator CF C it, which is unity if foreign subsidiary i is treated according to the criteria as defined by the CFC provision in period t, and zero otherwise. Whether an entity is treated or not depends on the following two criteria: CF C it = 1 if T ax it < 30% (25%) and P ass it > 10%, 0 if T ax it 30% (25%) and P ass it 10%. (1) T ax it is the statutory tax rate in year t of the host country subsidiary i is located in and P ass it is the passive-assets-to-total-assets ratio of subsidiary i at time t. When thinking of CF C it as being determined by two rules, one about T ax it and one about P ass it, we might specify Rule T ax it = Rule P ass it = 1 if T ax it < 30% (25%), 0 if T ax it 30% (25%), 1 if P ass it > 10%, 0 if P ass it 10%, so that we can define CF C it = Rule T it ax Rule P it ass (see Section 2). We postulate that fixed assets of subsidiary i at time t, F ixedassets it depend on both T ax it and P ass it in a continuous way. A higher level of T ax it is a natural determinant of foreign direct investment, since it affects German firms profits abroad directly and affects their incentives to shift profits through transfer pricing, debt shifting, and royalty payments. A higher the optimality condition in the Weichenrieder (1996) model (see also the discussion in Section 5. (2) 6

7 level of P ass it entails a higher return than other investments so that, due to lower capital costs, firms can ceteris paribus afford higher investments in F ixedassets it. Apart from a continuous relationship between F ixedassets it and T ax it on the one hand and F ixedassets it and P ass it on the other hand, treatment status, CF C it, is determined by T ax it as well as P ass it, which entails a discontinuity about the thresholds specified in (1). The regression discontinuity (RD) design compares outcome F ixedassets it of foreign subsidiaries that are just above the CFC treatment (at CF C it = 1) with those just below it (at CF C it = 0). Unlike in the textbook case (see Imbens and Lemieux, 2008; Angrist and Pischke, 2009; Lee and Lemieux, 2010), the two rules determining treatment status entail a two-dimensional threshold where the continuous control functions for units with versus without treatment are surfaces rather than curves. To identify the average treatment effect as a parameter on CF C it in the neighborhood of T ax it = 30% (25%) and P ass it = 10%, we specify the following model: F ixedassets it = α + β CF C it + f(x it ) + ϵ it, (3) where α is a constant term and ϵ it a possibly heteroscedastic disturbance term. The function f(x it ) is a control function which may be estimated parametrically or non-parametrically. The former relies on a polynomial approximation and the latter may be estimated by additive nonparametric regression models (see Härdle and Linton, 1994). In order to ensure that β measures the average treatment effect in the neighborhood of T ax it = 30% (25%) and P ass it = 10%, f(x it ) should not be based on T ax it and P ass it but on transformed data T ax it = T ax it τ t and P ass it = P ass it π t, where τ t and π t are averages of T ax it and P ass it in the neighborhood of T ax it = 30% (25%) and P ass it = 10%, f(x it ). Then, f(x it ) may be based on a vector X it which is based upon a polynomial function (or an additive nonparametric function) about T ax it and P ass it. In practice, it is advisable to not restrict the control function to be the same for untreated observations with CF C it = 0 and treated ones with CF C it = 1, akin to single-threshold models as described in Imbens and Lemieux (2008), Angrist and Pischke (2009), or Lee and Lemieux (2010). Otherwise, the treatment effect could assume effects which pertain to differences in functional form of the control function for the treated and the untreated units. Moreover, identification may require focusing on observations within a certain window around treatment thresholds so that the control function in the neighborhood of T ax it = 30% (25%) and P ass it = 10% is not distorted by influential observations far off the critical 7

8 values of T ax it and P ass it associated with Rule T ax it and Rule P it ass. In our application, we use 3 rd -order polynomial functions about T ax it and P ass it in a parametric control function. Moreover, we consider estimates of β at various windows (6-percentage points and 8-percentage points) around threshold levels of T ax it and P ass it. However, it should be noted that a design as ours allows for a heterogeneous treatment effect of CF C it, since f(x it ) contains interaction effects of CF C it with T ax it and P ass it. f(x it ) is normalized to be zero only in the neighborhood of T ax it = 30% (25%) and P ass it = 10%, but it may differ from zero if only one of the two applies. 4 Data The information on the foreign activities of German multinationals comes from MiDi (MIcrodatabase Direct Investment), a balance-sheet database provided by the Deutsche Bundesbank. One remarkable feature of this data is that it includes information on all foreign investments of German firms above a certain threshold level (see Lipponer, 2009, for a detailed documentation), because firms are legally bound to report to the Deutsche Bundesbank according to the German Foreign Trade and Payments Regulation (Außenwirtschaftsverordnung). In combination with information on countries tax rates, MiDi allows us to identify subsidiaries that are affected by the German CFC rule according to the requirement as defined in Section 2. Since the relevant passive assets of a foreign subsidiary are not directly reported in the data, we follow Ruf and Weichenrieder (2009) who define passive assets as the total financial assets net of equity in affiliated firms and lending to affiliated firms. 4 We use micro-level data from MiDi to calculate the passive-to-total-asset threshold. The second threshold is simply given by the statutory tax rate of the host country of the foreign subsidiary. Since the first threshold does not precisely refer to income from passive investment relative to total income according to 9 AStG, we have to assume that every unit invested yields the same return. All subsidiaries active in the banking and insurance sector are excluded from the analysis, as these observations are explicitly not covered by the 4 Ruf and Weichenrieder (2009) choose this definition (net of lending) since the data show that internal lending is usually refinanced by external bond issues. Such cases are explicitly excluded by the German CFC legislation in 8 AStG. 8

9 CFC legislation (see Section 2). We also focus on majority-held subsidiaries, because minority-holdings are usually not covered by the CFC rule (see 7 AStG). 5 5 Results For our investigation, we first define cells according to the normalized covariates T ax it and P ass it of T ax c,t and P ass i,t, respectively. Cells are determined in steps of 1-percentage-points, such that each subsidiary is assigned to a cell C ( T axit if the following condition is satisfied:, P ass it )) C ( T axit, P ass it ) = { T ax it, P ass it.30 < T ax it < P ass it.09,.29 < T ax it < P ass it.08,...,.29 < T ax it < P ass it.90}. As a result, the two criteria span a two-dimensional space that captures 6, 000 cells where treated observations just satisfy T ax it < 0 and P ass it > 0. Note that both thresholds define treatment, and, therefore, the regression discontinuity arises around C (0,0). Subsequently, for each cell, we take the average value of the fixed assets of subsidiaries assigned to the respective cells. Following the same logic, we will further refer to windows around treatment as the treatment effect is most reliably captured by comparing observation just below with those just above the thresholds. If we define windows in percentage points, referring to a window of size of (6,6 ), for example, means that observations within the interval [.06 T ax it, P ass it +.06] are considered. Table 1 presents the findings of the RD design. The dependent variable is the average value of the fixed assets in a cell. We report results for symmetric and asymmetric polynomials. Asymmetric polynomial functions estimated separately for CF C it = 0 and CF C it = 1 are more flexible, because the control function may assume a different functional form whose restriction 5 According to 7(6) AStG, if the foreign affiliate is an intermediate entity, minority holdings may also be subject to the German CFC legislation. To investigate the robustness of our results in Section 5, we also include minority holdings. The results are quite robust and very similar to the findings reported. 9

10 might bias the estimate of the average treatment effect β in the neighborhood of T ax it, P ass it = 0. Since the number of observations (subsidiaries) which contribute to (the average in) cell C ( T axit differs across cells, regressions, P ass it ) are weighted by the number of underlying observations per cell. In column 1, where symmetric polynomials are used, a significant impact of the CFC rule on fixed assets is found. Column 2 allows for asymmetric polynomials and confirms the negative treatment effect. The magnitude of the coefficient of the treatment dummy exceeds the coefficient in column 1 by a factor of almost 2.5. According to the asymmetric specification, CFC-rule treatment is associated with e9.66 million less investment in fixed assets, which is a quite significant impact. If we look at the effects closer at the threshold, the coefficients point at even higher effects. 6 Figure 1 displays the regression discontinuity graphically. The surfaces are estimated using the asymmetric specification of the polynomial functions. The graph shows how the fixed assets vary in the two dimensions determining treatment. Each observation (cell), underlying the estimation of the surfaces, is depicted by a single point. Very pronounced in the illustration is the discontinuity around the treatment, which highlights the treatment effect at the point where tax rate (normalized) and passive/total assets (normalized) are both zero. Let us, using the graphical illustration, compare our results to the theoretical predictions made by Weichenrieder (1996). According to the optimality condition in the Weichenrieder (1996) model, a subsidiary invests until its marginal product of capital equals its marginal cost, while taking into account restrictions on passive income. In case an empirical observation stems from a maximizing (and tax-planning) multinational, we should observe it left of the threshold level given by the passive-to-total-assets ratio. If the multinational accumulates more passive investment and violates the restriction, which is the case when considering observations exceeding the threshold, the subsidiary s income becomes subject to the high German tax level. Since Weichenrieder (1996) predicts for the former subsidiaries lower cost of capital and more investment, our results are in line with the theoretical considerations. Table 2 defines placebo treatments. In this experiment, we shift the thresholds by 5-percentage points either above or below the actual thresholds. As 6 Note that a smaller window size of (4;4) would further reduce the number of observations and would no longer allow for robust inference. We provide results, however, for the (4;4) window size in Table 3, where subsidiary-level data are used. 10

11 Table 1: The Impact of the CFC Rule on Fixed Assets Specification of Polynomial Symmetric Asymmetric Treat (1.90) (2.21) Observations 2,119 2,119 Window around treatment: (6;6) Treat (4.73) (5.74) Observations Window around treatment: (8;8) Treat (3.17) (5.38) Observations Notes: Dependent variable: fixed assets. Symmetric: the polynomial functions (of 3rd order) are forced to have identical parameters to the left and the right of the threshold. Asymmetric: the polynomial functions (of 3rd order) are allowed to have different parameters to the left and the right of the threshold. Since each cell (here one observation) represents averages of subsidiary observations, we weight each cell by the number of elements (subsidiary observations) that gave rise to the average. The (6;6 ) window around the treatment is associated with observations within an interval of 6%-points below and 6%-points above the two thresholds. Hence, observations satisfy the following condition: 24% (19%) T ax 36% (31%) and 4% P ass 16%.,, denote statistical significance at the 1%, 5%, and 10% level, respectively. Robust standard errors in parentheses. 11

12 Figure 1: The Impact of the CFC Rule on Fixed Assets 12

13 Table 2: Placebo Treatments (Threshold shifted 5%-points) Specification of Polynomial Symmetric Asymmetric Threshold 5%-points within Treatment Group (Treat=1) Treat (.807) (1.29) Observations Threshold 5%-points within Non-Treatment Group (Treat=0) Treat (3.27) (10.67) Observations 1,610 1,610 Notes: Dependent variable: fixed assets. Symmetric: the polynomial functions (of 3rd order) are forced to have identical parameters to the left and the right of the threshold. Asymmetric: the polynomial functions (of 3rd order) are allowed to have different parameters to the left and the right of the threshold. Since each cell (here one observation) represents averages of subsidiary observations, we weight each cell by the number of elements (subsidiary observations) that gave rise to the average. The (6;6 ) window around the treatment is associated with observations within an interval of 6%-points below and 6%-points above the two thresholds. Hence, observations satisfy the following condition: 24% (19%) T ax 36% (31%) and 4% P ass 16%.,, denote statistical significance at the 1%, 5%, and 10% level, respectively. Robust standard errors in parentheses. a result, the placebo thresholds lie within the groups of treated and nontreated observations. The respective regressions demonstrate that there is no indication of any significant impact. 6 Using Subsidiary-level Data Since single subsidiary-level observations contribute to the average values of the fixed assets in each cell as defined above, we also present results using the micro-level data. Such subsidiary-level estimates are useful robustness checks because we may take into account the underlying subsidiary structure 13

14 Table 3: Treatment Effect using Subsidiary-level Data Window around treatment: (4;4) Treat Specification of Polynomial Symmetric Asymmetric (6.40) Observations Window around treatment: (6;6) Treat (4.18) Observations 3,717 3,717 Window around treatment: (8;8) Treat (3.57) Observations 7,963 7,963 Notes: Dependent variable: fixed assets. Symmetric: the polynomial functions (of 3rd order) are forced to have identical parameters to the left and the right of the threshold. Asymmetric: the polynomial functions (of 3rd order) are allowed to have different parameters to the left and the right of the threshold. Since each cell (here one observation) represents averages of subsidiary observations, we weight each cell by the number of elements (subsidiary observations) that gave rise to the average. The (6;6 ) window around the treatment is associated with observations within an interval of 6%-points below and 6%-points above the two thresholds. Hence, observations satisfy the following condition: 24% (19%) T ax 36% (31%) and 4% P ass 16%.,, denote statistical significance at the 1%, 5%, and 10% level, respectively. Robust standard errors in parentheses. Standard errors are robust and clustered at the subsidiary and year level using multi-way clustering as suggested by Cameron, Gelbach and Miller (2010). 14

15 when calculating standard errors. Moreover, we may move closer to the thresholds by defining even smaller windows. Table 3 shows the impact of the German CFC rule on fixed assets of foreign subsidiaries. The findings should be very similar to the findings from above. 7 Using window size (6,6 ), we see that the coefficients are very close to the findings in Table 1. The estimations are quite robust and we confirm the significant treatment effect found in Table 1. As we would expect from the results above, the closer we move to the thresholds, the more pronounced is the treatment effect. The estimated coefficient in case of the (4;4) is quite remarkable. It implies that the CFC-rule treatment is associated with e18.75 million less investment in fixed assets. 7 Conclusions This study suggests a novel approach to investigate the impact of the German CFC rule on foreign subsidiaries of German multinationals. CFC rules have been introduced by many countries to curtail the deliberate avoidance of profit taxes by multinational firms. Apart from some U.S. studies, only very little is known about the impact such rules have on foreign operations of multinational firms. Evidence using German data is provided by Ruf and Weichenrieder (2009), who argue that the German CFC rule effectively restricts investments in passive assets. They particularly show that subsidiaries in low-tax countries, defined by the tax threshold of 30% (25%), hold significantly less passive assets. In this sense, it seems that the measure prevents firms from allocating passive assets to low-tax countries in the first place; and therefore, hampers the ability of multinationals to avoid taxes. If the CFC rule applies to a foreign subsidiary, income arising in the foreign country is subject to full taxation in Germany, as if the multinational was taxed on its worldwide income (which would be the case in a real residence-based tax system). Against the background of the high German tax level, this could lead to an increase in the overall cost of capital and may deter firms from investing in real capital. According to the German CFC legislation, the rule applies to subsidiaries which satisfy two conditions. First, they have to be located in a country where the statutory tax rate lies below 30% (25% since 2001). Second, the 7 Estimated coefficients might actually differ, depending on how cells are centered around T ax and P ass. As already mentioned, standard errors may differ as well. 15

16 passive-to-total-asset ratio exceeds 10%. The thresholds give rise to a twodimensional regression discontinuity approach. We show that subsidiaries which are treated by the CFC rule use significantly less fixed assets, where the impact on just treated subsidiaries (i.e. those close to the threshold values) is particularly pronounced. It seems that the jump in the cost of capital is rather dramatic, which is reasonable since the statutory tax rate faced in the source country is comparatively low, while the German tax level has been quite high during the time span considered in our analysis. It is important to note that our regression design makes sure that this finding is not confounded by the fact that these subsidiaries specialize, for instance, in financial investments and therefore exhibit less fixed assets. In fact, at the thresholds, we may interpret the effect as the true impact of the CFC rule, because subsidiaries differ only in their treatment status. Since CFC rules typically distinguish between active and passive income, it is rather questionable whether this outcome a negative impact on active investment is intended by policy makers. In a sense, it interferes with the principle of exempting foreign income from residence taxation: governments explicitly emphasize that, under a source-based system of taxation, tax exemption of active foreign income is accepted. As all investment (income) seems to be affected by the CFC rule, this finding implies that the competition between firms from different residence countries is distorted, depending on whether home countries apply CFC rules or not. 16

17 8 References Angrist, Joshua D. and Jörn-Steffen Pischke (2009), Mostly Harmless Econometrics: An Empiricists s Companion, Princeton, NJ: Princeton University Press, Altshuler, Rosanne and H. Grubert (2006), Governments and multinational corporations in the race to the bottom, Tax Notes International 41, Altshuler, Rosanne and R. Glenn Hubbard (2003), The effect of the tax reform act of 1986 on the location of assets in financial services firms, Journal of Public Economics 87, 1, Buettner, Thiess, Michael Overesch, Ulrich Schreiber and Georg Wamser (2008), The Impact of Thin-Capitalization Rules on Multinationals Financing and Investment Decisions, Bundesbank Discussion Paper 03/2008. Cameron, A. Colin, Jonah Gelbach and Douglas Miller (2010), Robust Inference with Multi-way Clustering, Journal of Business and Economic Statistics, forthcoming. Härdle, Wolfgang and Oliver Linton (1994), Applied nonparametric methods, in Robert F. Engle and Daniel McFadden (eds.), Handbook of of Econometrics Vol. 4, Amsterdam: New Holland, Huizinga, Harry and Luc Laeven (2008), International profit shifting within multinationals: A multi-country perspective, Journal of Public Economics 92, 5-6, Imbens, Guido and Thomas Lemieux (2008), Regression discontinuity designs: A guide to practice, Journal of Econometrics, 142,?, Lee, David S. and Thomas Lemieux (2010), Regression discontinuity designs in economics, Journal of Economic Literature 48, 2, Lipponer, A. (2008), Microdatabase Direct Investment - MiDi. guide, Bundesbank Working Paper, Frankfurt. A brief Mutti, John and Harry Grubert (2006), New developments in the effect of taxes on royalties and the migration of intangible assets abroad. Paper 17

18 prepared for the NBER/CRIW Conference on International Service Flows, Washington D.C., 28 April Nicodème, Gaëtan (2009), On recent developments in fighting harmful tax practices, National Tax Journal 62, 4, Organisation for Economic Co-Operation and Development (1998), Harmful tax competition: An emerging global issue. OECD, Paris. Overesch, Michael and Georg Wamser (2010), Corporate tax planning and thin-capitalization rules: Evidence from a quasi-experiment, Applied Economics 42, 5, Ruf, Martin and Alfons Weichenrieder (2009), The taxation of passive foreign investment - lessons from German experience, CESifo Working Paper 2624, Munich. Weichenrieder, Alfons J. (1996), Anti-tax-avoidance provisions and the size of foreign direct investment, International Tax and Public Finance 3, Weichenrieder, Alfons J. (2009), Profit shifting in the EU: evidence from Germany, International Tax and Public Finance 16, Weichenrieder, Alfons J. and Helen Windischbauer (2008), Thin-capitalization rules and company responses - experience from German legislation, CE- Sifo Working Paper 2456, Munich 18