10 th Norwegian-German Seminar Public Sector Economics

Transcription

1 0 th Norwegian-German Seminar Public Sector Economics Munich, 7 8 November 07 Profit Shifting of Multinational orporations with Loss-Making Affiliates Marko Koethenbuerger and Mohammed Mardan

2 Profit shifting of multinational corporations with loss-making affiliates Marko Koethenbuerger ETH Zurich, ESifo and NoeT Mohammed Mardan ETH Zurich and NoeT Michael Stimmelmayr ETH Zurich, ESifo and NoeT 7 November 07 Abstract Despite the prominence of losses for affiliates of multinational enterprises, the implications of these losses for the investment behavior have not been considered in the literature. We set up a model which allows for the possibility of lossmaking affiliates and analyze the effects on affiliate investment and government policy. We show that an inverted-type of transfer pricing from the low-tax to the high-tax jurisdiction might also promote investment incentives in the low-tax as well as the high-tax jurisdiction. In equilibrium, affiliates might overinvest. The finding contrasts existing literature where transfer pricing incentives are insulated from investment incentives. Finally, we highlight the implications of our findings for the efficiency of government policy where the loss-related investment effects might generate too high taxes in equilibrium. Keywords: tax competition, profit shifting, corporate losses, investment. JEL lassification: H5, D, H87 We are grateful to seminar and conference participants at the 09th annual conference of the National Tax Association, the 73rd annual congress of the International Institute of Public Finance, and the 4th annual MaTax onference. In particular we are grateful for the valuable comments made by Eric Zwick, James Albertus and Aija Polakova. Department of Management, Technology and Economics, Leonhardstrasse, 809 Zurich, Switzerland; koethenbuerger@kof.ethz.ch; phone: Department of Management, Technology and Economics, Leonhardstrasse, 809 Zurich, Switzerland; mardan@kof.ethz.ch; phone: Department of Management, Technology and Economics, Leonhardstrasse, 809 Zurich, Switzerland; stimmelmayr@kof.ethz.ch; phone:

3 Introduction Affiliates of multinational enterprises (MNEs) frequently report losses. orporate tax losses and the associated loss carry-forwards have grown in importance and, expressed as a percentage of GDP, show large differences among countries, with some as high as 5 percent (OED, 0). The fiscal implications of tax losses are of concern to policy makers. The tax codes in most countries do not grant an immediate tax rebate in response to losses and also limit the extent to which losses can be deducted against future income or, in the context of MNEs, against profits of an affiliate in a foreign country. While these rules attempt to reduce tax fraud and excessive use of tax deductions, thereby protecting the domestic corporate tax base (Ashulter et al., 009), MNEs have the potential to bypass these waiting rules through the strategic use of transfer pricing, as documented in recent empirical research (Hopland et al., forthcoming; DeSimone et al., 05). In the same vein, empirical research has documented a tendency of MNE affiliates to bunch around zero reported profits, which is consistent with MNEs profit shifting behavior (Grubert et al., 993; Johannesen et al., 07; Habu, 07). 3 Bunching might occur from below zero, to use losses that allow shifted profits to be taxed at a zero rate, or from above zero, to shift profits out of profitable affiliates. In this paper, we analyze how transfer pricing strategies of MNEs respond to losses within the MNE network and how this intertwines with investment choices. We set up For the U.S., Altshuler et al. (009) relate this trend to different explanations, including an increase in the dispersion of the distribution of profits and a general decline in the rate of profit. For instance, in 00, U.S. corporations that lost money reported losses that amount to more than 60 percent of profits reported by profitable corporations (Edgerton, 00). In Germany, aggregate unused losses from the past were more than four times larger than taxable corporate profits in the economy. For 004, roughly 40 percent of incorporated firms suffer a loss (Dwenger, 009), with a similar magnitude for German MNEs only (Overesch, 009). The issue of tax-induced transfer pricing has recently gained momentum in the Base Erosion and Profit Shifting (BEPS) initiative of the OED (OED, 03). Despite various attempts to regulate transfer pricing (OED, 00), limiting this type of tax avoidance behavior continues to be a taxing task for tax authorities. The notion of arm s-length pricing, which underlies most of the transfer pricing regulation, is difficult to implement with, e.g., intellectual property rights due to the highly idiosyncratic nature of this service and the associated difficulty in finding comparable transactions that are not influenced by tax-savings considerations. 3 Such a situation is consistent with highly-publicized cases of aggressive transfer pricing behavior of Apple, Google and Starbucks, which leaves almost zero taxable profits in high-tax jurisdictions (Levin and Mcain, 03).

4 a model of MNE behavior that, consistent with the empirical observation, generates a tendency of MNE affiliate profits to bunch around a zero profit level, both from below and above zero, and this bunching generates investment effects in the high-tax jurisdiction as well as low-tax jurisdiction. In showing these results, we depart from the usual assumption in the literature that, when determining the optimal amount of profit shifting, the MNE weighs the marginal benefit of higher tax savings vis-a-vis the marginal change in concealment cost associated with transfer pricing (see Haufler and Schjelderup (000) and Gresik (00), for instance). With such an interior solution to the transfer pricing problem, investment incentives are generically insulated from transfer pricing, meaning that changes in transfer prices have no effect on the MNE s investment decision. This approach by definition abstracts from the possibility of zerobunching of taxable profit. In our model, a MNE affiliate might incur a tax loss. We assume that the price levels of the final goods sold by each affiliate are uncertain. A very high price implies large profits whereas a very low price results in losses. Depending on the realizations of the price levels, the MNE determines its transfer pricing strategy. In contrast to the previous literature, the MNE may be constrained in its profit shifting strategy, which depends on the profitability of its affiliates. With a loss-making affiliate (in the high-tax jurisdiction), the amount of profits to be shifted falls below the interior solution when the profitability of the affiliate in the low-tax country is too low. Bunching occurs from above zero. In the same vein, the amount of profits to be shifted falls below the interior solution when the size of the losses in the affiliate in the high-tax country are too low. Bunching occurs from below zero. These constraints are the reason why the MNE s transfer pricing and investment decisions are intertwined in our model. In case of a binding profit shifting constraint an increase in investment relaxes the respective profit shifting constraint and allows the MNE to shift more profits. In this setting, transfer pricing might not only increase investments in high-tax jurisdictions, as likewise predicted by existing literature, but for different reasons. 4 It might also spur investments in low-tax jurisdictions, possibly leading to overinvestment 4 Mintz and Smart (004) show that internal debt shifting not only allows MNEs to save on taxes, but also promotes investments in high-tax affiliates. With internal debt shifting, investment effects follow from the reduced cost of capital due to the tax-deductibility of interest payments. This link is absent with other forms of transfer pricing that we consider.

5 in high-tax and low-tax jurisdictions. 5 The interrelatedness between profit shifting and investment decision has also implications for the efficiency of government policy. The typical positive externality that is caused by governments incentive to reduce their tax rate in order to attract profits is only present in the case of highly profitable MNE affiliates. In cases where the MNE is constrained in its profit shifting strategy, there also exists a investment externality which runs in the opposite direction. This might imply too high tax rates in equilibrium. The results of our analysis have important implications for the empirical literature. In empirical analyses on transfer pricing loss-making affiliates are frequently dropped from the analysis (Klassen et al., 993; Dharmapala, 04). While this presumably reflects the prior to avoid any bias from reversed incentives under net operating losses, Hopland et al. (forthcoming) have shown that this bias will only be eliminated under the assumption of ex-post shifting. Our analysis shows that, taking into account the possibility of zero-taxable profits and thereby departing from the usual assumption of an interior solution to transfer pricing, likely results in an underestimation of the tax sensitivity of reported profits, just because affiliates reporting zero-taxable profits will not react to a change in the tax rate. 6 Arguably, these affiliates belong to the most tax aggressive MNEs (Johannesen et al., 07). Moreover, our analysis shows that the disregard of loss-making affiliate will create a downward bias in the empirical estimates of investment responses to corporate taxes. The intuition is that a change in the corporate tax at the location where affiliates report zero-taxable profits will not change investment incentives because MNEs can escape the taxation of the marginal return on investment. 7 Our analysis contributes to the small, but evolving literature on the importance of 5 Tax rates might generate investment changes across affiliates, implying that investment levels in the two affiliates might co-move in response to a higher profit tax in one country. The implication is consistent with empirical findings in Becker and Riedel (0) who show that the tax rate of the foreign jurisdiction negatively affects domestic investment of MNEs. 6 This is consistent with Harbu (07) who finds that UK corporate tax rate cuts did not affect the ratio of taxable profits to total assets reported by foreign multinational subsidiaries and domestic stand-alone firms differently. 7 The implication is in line with empirical estimates of the user cost elasticity of investment in Dwenger and Walch (04), which increases in absolute value from 0.37 to 0.5 when accounting for tax losses. 3

6 losses for income shifting. 8 For instance, Overesch (009) shows that the statutory tax differential, the conventional measure of transfer pricing incentives, becomes less important in explaining transfer pricing in the presence of corporate tax losses. Using data on intercompany payments between affiliated firms in Norway, Hopland et al. (forthcoming) find evidence that, under losses, transfer pricing provides flexibility to adjust profit shifting ex post, i.e. after the financial performance of the affiliate has become known but before the tax return is filed. Similarly, De Simone et al. (07) show that affiliates of MNEs alter the distribution of reported profits to take advantage of losses. An increase of one percent in the tax incentives of a profitable affiliate are associated with a drop in reported profit by 8.8 percent whereas for an unprofitable affiliate the same tax incentive leads to an increase in reported profit by 6. percent. Theoretical contributions on corporate losses in an international context are scarce. Mardan and Stimmelmayr (07) analyze the implications of losses for the decision to switch from separate accounting to formula apportionment. Kalamov and Runkel (06) investigate the implications of an uncoordinated introduction of cross-border loss-offset provisions when countries compete for capital and profits of MNEs. Both studies rely on interior solutions to characterize transfer pricing incentives, which imply that investment behavior and transfer pricing incentives do not intertwine. To the best of our knowledge, existing literature has not yet established a link between transfer pricing and investment. The current paper tries to fill this gap. The paper is organized as follows. In Section we introduce the basic framework. In Section 3 we analyze transfer pricing behavior and investment choices of the MNE. We turn to government policy and characterize the efficiency of tax policy choices in Section 4 and discuss two assumption of the basic model in Section 5. Finally, we provide a summary of the results and offer some concluding remarks in Section 6. 8 Altshuler et al. (009) summarize the literature on investment effects of tax loss carry-forward provisions. This literature, however, abstracts from transfer pricing. 4

7 The basic framework onsider a set-up in which a multinational enterprise (MNE) has affiliates in two smallopen countries which levy source-based profit tax rates t i, i =,. Although tax rates are determined endogenously, we assume, without loss of generality, that country is the high-tax country. 9 Affiliates use capital, k i, to produce a final good, y i, using a standard production technology, y i = f(k i ), which satisfies f (k i ) > 0 > f (k i ). apital, k i, can be borrowed from the world market at an interest rate r. We assume that the price levels of the final goods are stochastic and drawn from a cumulative distribution function, H i (p i ), and density, h i (p i ), with support on [p i, p i ]. The uncertainty of the price level in country affects the profitability of affiliate, but we assume that ex post the affiliate is always profitable even when the realization of the price takes the lowest value of p. 0 Instead, the realization of the output price in country can lead to actual losses in affiliate ex post. We denote the average price in country i by ˆp i = i p i p i h i (p i )dp i. Additionally, each affiliate requires one unit of an essential intangible input good, g, for production. This input could, for instance, be a patent. The legal rights for using the intangible input good are located in the affiliate in country. The arm s-length price for g is normalized to zero. Any deviation from the arm s-length price is associated with convex concealment costs (g) = γ(g )/, γ > 0. Furthermore, we assume that the MNE has sufficient flexibility to determine the transfer price ex post, i.e. after the realization of the output price in country. 3 9 One reason for this could be that country hosts a larger number of purely national firms and therefore derives comparably more corporate tax revenues from this additional source. 0 This assumption reduces complexity without qualitatively affecting our results. In general, MNEs might close down a permanently loss-making affiliate. However, at least in the short run it might well be that the MNE continues the operation even though it temporarily incurs losses in expectation. Loss-making affiliates might be valuable in order to gain access to a new market or to show strategic presence in a relevant market. The reasoning is consistent with the persistence of losses, as reported in Auerbach (987) and Dwenger and Walch (04), for instance. We could also include size-related determinants of the concealment cost, such as affiliate assets. Such an extension allows corporate investments to influence transfer pricing and vice versa. To isolate the effects of losses for investment behavior, we condition the concealment cost only on the deviation from the arm s-length price. 3 See Hopland et al. (forthcoming) for evidence that the use of intangible assets gives MNEs the 5

8 The before-tax profit of each of the two affiliates is given by revenue from (expected) sales of the output good minus the user cost of capital adjusted by the payment/income from the intangible asset, π e = π e = p p p h (p )dp y rk g, p h (p )dp y rk g. () The timing of the decision is as follows. First, the two governments non-cooperatively set tax rates. Then, the MNE determines the optimal investment levels (before the realization of affiliates output prices) and thereafter (after the realization of affiliates output prices) it chooses the optimal transfer price. We solve by backward induction. 3 The firm 3. Ex-post period Taxable profits differ from pre-tax profits in that only a share δ < of the capital costs can be deducted in case of positive taxable profits. The restriction on the deductibility rate reflects the observation that real-world tax systems only offer delayed expensing of investment outlays and no deductibility of the cost of equity capital. 4 Non-negative taxable profits in country If the output price in country is sufficiently large, p y δrk > 0 and, depending on the choice of g, taxable profits of both affiliates are given by π t = p y δrk g, { π t p y δrk g if p y δrk g 0, = 0 otherwise. () opportunity to shift income ex post. 4 An immediate expensing of investment outlays (as with a R-based cash flow tax) or a full deductibility of the cost of capital (as with an Allowance for orporate Equity system) would imply δ =. Such corporate tax provisions however do not universally exist in national corporate tax codes. 6

9 We have already made used of the fact that g > 0, given that country levies the higher tax rate, t t > 0. We distinguish two scenarios: in the first case, the realization of the output price in country is sufficiently high and the MNE is not restricted in setting its transfer price. In the second case, the output price is still sufficiently high and the MNE generates positive taxable profits in affiliate in the absence of profit shifting. The MNE is now, however, able to shift all profits earned in affiliate into the low-tax jurisdiction, leading to a zero tax base in country, π t = 0. Assuming for the moment that the MNE is unrestricted in its profit shifting possibilities, total profits of the MNE amount to 5 Π UP = ( t )p y ( δt )rk ( t )p y ( δt )rk (t t )g γ g, and the optimal transfer price is determined by (3) g UP = t t, (4) γ where the superscript denotes that the MNE is unconstrained and taxable profits in country are positive. Optimal profit shifting behaviour implies that the marginal cost of shifting profits has to be set equal to the marginal benefit (i.e., the tax savings). Since t > t the MNE shifts profits from the high-tax country to the low-tax country, i.e. g UP > 0. However, the realization of the output price in country might not be sufficiently high to ensure that the MNE is unconstrained in its optimal profit shifting behaviour. Denoting p 0 as the price for which taxable profits of affiliate are zero in the absence of profit shifting and p UP as the price for which taxable profits evaluated at the transfer price (4) are zero, the MNE is constrained in its the choice of the transfer price if p 0 p < δrk y t t γy p UP. (5) Under this condition, the MNE is still profitable and shifts profits from the high-tax country to the low-tax country, but to a smaller extent. The reason is based on 5 For simplicity, we assign the concealment cost to the net-of-tax profit stream of the headquarter. Assigning the concealment cost to the affiliates and making the cost tax deductible (partly or fully) changes the endogenous probability of the different scenarios that we consider in our analysis arise, but does not change our findings qualitatively. 7

10 the lower output price and thus the lower profitability of affiliate. Since the MNE s profit shifting constraint stems from the insufficient profitability of affiliate, we refer to this scenario as the profit constraint in country. In this case, the MNE sets the transfer price in a way that reduces taxable profits in country to zero, i.e. g 0 = p y δrk, (6) where the superscript 0 denotes that taxable profits in country are zero. The MNE pursues this strategy as long as taxable profits are positive in the absence of profit shifting, i.e. as long as p δrk y p 0. Non-positive taxable profits in country If the realization of the final goods price is too low in country, that is p < p 0, the affiliate in country makes losses and the tax bases in the two countries are given by 6 { π t = 0 { otherwise. p y δrk g if p y δrk g 0, π t = 0 if p y δrk g 0, p y δrk g otherwise. (7) In setting up the conditions, we have already used the fact that g might be negative, as shown below. The tax base of affiliate is non-positive in the absence of profit shifting and, thus, the MNE has an incentive to shift profits from country to country, given that the effective tax rate in country is zero. We distinguish three scenarios depending on the levels of p and p. In the first case, none of the conditions in equation (7) is binding, i.e. taxable profits in country and losses in country are sufficiently large, such that the MNE is unrestricted in setting its transfer price. In the second case, the MNE is restricted by the size of taxable profits 6 We abstract from tax provisions to offset losses against positive profits across time or across affiliates, either because they do not eliminate incentives to engage in transfer pricing into loss-making affiliates or they are not commonly available in national tax codes (or both). Different to shifting into loss-making affiliates, loss carry forwards entail a cost of forgone interest and can only be used over a limited time span. Loss carry backs are by far less generous than loss carry-forwards. Only a small number of countries offer loss carry backs and the period in which this is possible is very short (at most three years). Many countries do not grant loss carry backs at all. Even more restricted is the possibility of cross-border loss offset. Only four countries (Austria, Denmark, France and Italy) currently allow for cross-border loss offsets (see OED, 0). 8

11 in country. This scenario implies that the MNE is forced to shift less profits from country to country, because profits of the affiliate in the low-tax country are insufficient to fully offset the loss incurred by the affiliate in country. We refer to this scenario as the profit constraint in country. In the third case, the MNE is restricted by the limited amount of losses in country. Given that the statutory tax rate in country is larger than in country, t > t, the MNE never chooses a transfer price which would result in positive taxable profits in country. Therefore, the optimal amount of profits shifted from the low-tax country to the high-tax country is confined by the size of affiliate s losses and we refer to this scenario as the MNE s loss constraint in country. Suppose scenario one holds and the MNE is unconstrained in determining its profit shifting behavior. Then total profits of the MNE are Π UN = ( t )p y ( δt )rk p y rk t g γ g. (8) and the optimal transfer price is determined by g UN = t γ. (9) The superscript indicates that the MNE is unconstrained and taxable profits in country are negative. In this unconstrained case, the MNE sets a transfer price which is negative to shift profits out of the low-tax country into the high-tax country where the effective tax rate is zero. The transfer price only depends on the level of country s tax rate. In contrast to the scenario where the affiliate in country is profitable, now, both price levels determine whether the MNE is unconstrained in setting the transfer price, or not. Denoting ( ) as the price for which taxable profits π t (π t ) evaluated at the transfer price (4) are zero, the MNE is unconstrained only if p δrk y t γy, p δrk y t γy. (0) That is, the price in country must be sufficiently high whereas the price in country needs to be sufficiently low. 9

12 One of the two conditions in (7) might, however, be binding. Namely, either the price in country could be too low (p < ) or the price in country might not be sufficiently low ( < p < p 0 ). The question arises whether the MNE then suffers from the profit constraint in country or the loss constraint in country. From (7) we infer the threshold under which both constraints are binding, i.e. the tax bases in both countries reduce to zero. This condition is given by implying a threshold for p of p y δrk p y δrk = 0, () p δr(k k ) y p y y p P, () For p below the threshold specified in (), the MNE is profit constrained in country. We note that this threshold depends on the realization of the output price p. The higher the output price in country the lower needs to be the output price in country to ensure that the MNE is still profit constrained in country. From equation (7) we infer that the optimal transfer price depends on the price induced restrictions set up by the constraints in the equation. If prices are such that the MNE is profit constrained, the optimal transfer price is given by g P = (p y δrk ) < 0. (3) Instead, if prices imply that the MNE s is loss constraint, the optimal transfer pricing strategy is g L = p y δrk < 0, (4) where the superscript of the two former transfer prices indicates whether the MNE is profit constrained or loss constrained. Figure summarizes our discussion of the MNE s profit shifting constraints. One should note that the p P threshold cuts the p -axis below the value of p 0. The explanation for this result originates from the assumption that the affiliate in country is always profitable. Therefore, the p P and p 0 lines intersect to the left of p and the hypothetical value on the p -axis would be a price of p 0 at which taxable profits in country are zero in the absence of profit shifting. 0

13 3. Ex-ante period Figure : Price dependent profit shifting In this section, we analyze the optimal investment decisions of the MNE conditional on the MNE s constraints. Taking all possible scenarios together, the MNE s expected profit is E(Π) = p p p UP p UN 0 p p P Π UP h (p )h (p )dp dp Π UN h (p )h (p )dp dp Π L h (p )h (p )dp dp p UP p p 0 UN P p p p 0 Π 0 h (p )h (p )dp dp Π P h (p )h (p )dp dp Π L h (p )h (p )dp dp. (5)

14 Differentiating (5) with respect to k, the optimal ex-ante investment level in country is given by 7 ( t )ˆp f (k ) = ( δt )r p UN P p p p UP p p 0 UN 0 p p P p 0 UN p t [p f (k ) δr]h (p )h (p )dp dp t [p f (k ) δr]h (p )h (p )dp dp γ(g UP g 0 )[p f (k ) δr]h (p )h (p )dp dp γ(g UP g 0 )[p f (k ) δr]h (p )h (p )dp dp γ(g UP g 0 )[p f (k ) δr]h (p )h (p )dp dp γ[p P f (k ) δr][p P y δrk t /γ]h (p P ) pp h (p )dp. (6) k In equation (6), the first effect on the right side describes the standard effect of taxation when affiliates are always successful and earn enough profits so that none of the constraints is binding. In this case profit shifting is insulated from the investment decision and capital investment is distorted by taxation in country because only a fraction δ of investment costs can be deducted from the tax base. The second and third terms show how investment is affected if the affiliate in country makes relatively large losses, i.e when the MNE is either unconstrained in its tax planning (second term) or when the MNE is profit-constrained in country (third term). These effects are negative and thus lead to an increase in the investment level in country. The reason is that the MNE knows that in this case it will pay no taxes in country and it thus has an incentive to increase its investment ex-ante. Terms four to six are also negative because g UP > g 0 and show that profit shifting has a positive effect on investment if the affiliate in country is either profit or loss constrained. In these two cases, an increase in the investment level in country relaxes either constraint and allows the MNE to shift additional profits. The last term captures how the likelihood of the MNE 7 The derivations of the first-order condition for k and k can be found in Appendix A..

15 being profit-constrained in country is affected by the capital investment in country. Hence, from (6) we can conclude that profit shifting affects investment incentives in the high-tax country by lowering the cost of capital and, thus, increasing investments. Differentiating (5) with respect to k, yields the first-order condition ( t )ˆp f (k ) = ( δt )r UN P p p UN p γ[p f (k ) δr](g P g UN )h (p )h (p )dp dp γ[p f (k ) δr](g P g UN )h (p P ) pp h (p )dp = 0. (7) k The interpretation of the first effect on the right side of equation (7) is the same as for (6). It shows that investment in country will be insulated from profit shifting if affiliates earn enough profits. The second term on the right side of (7) is negative because g P > g UN and shows that profit shifting exerts a positive effect on the optimal investment level in country if the MNE is profit-constrained in country. This is because the MNE has an incentive to increase its investment in country in order to relax the profit constraint and in turn to shift more profits. Additional profits earned through the increase in investment in affiliate can be shifted to the loss making affiliate in country without triggering additional tax payments. Similarly to (6), the last term shows how the MNE s probability of being profit-constrained in country is affected by a change in capital investment in country. Thus, our result shows that profit shifting can have a positive effect on investment even in the low-tax country, possibly leading to overinvestment (p i f (k i ) < r) in the two countries. We can summarize: Proposition If MNEs are constrained in their profit shifting activity, the possibility of profit shifting exerts a positive effect on investment. The positive investment effect can also appear in the low-tax country. From the first-order conditions (6) and (7), we can derive how taxes affect optimal capital investments. Generally, these effects are ambiguous since taxes also affect the likelihoods of being in one or the other scenario. However, when prices are e.g. uniformly 3

16 distributed on a large support, the probability of being at a specific point in the price distribution is very small. Hence, under this condition, we can establish 8 k t < 0, k t < 0, k t < 0, k t = 0. (8) As in the standard model with only profitable affiliates, taxes in the host country reduce the incentives to invest. Possibly surprising, the tax rate in country also negatively affects investment in country. The reason is that a higher tax rate in country reduces the incentives to shift profits when the MNE is profit- or loss-constrained in country. 9 However, lower incentives to shift profits also mean a lower incentive to relax the respective constraint by increasing capital investment in country. The reason why country s tax rate has no effect on capital investment in country is that, when the MNE is profit-constrained in country, the statutory rate t is irrelevant for the MNE s profit shifting incentives because the effective tax rate in country is zero due to affiliate loss position. We summarize our results in: Proposition If MNEs are constrained in their profit shifting activity, the possibility of profit shifting allows profit taxes of the low-tax country to increase investment in the high-tax country. Further, although investment in the low-tax country is increased because the return to investment is shifted to the high-tax country, the latter s (statutory) tax rate generically does not influence the investment level in the low-tax country. Proposition and stand in contrast to the existing literature where MNEs profit shifting and investment decision are insulated from each other if MNEs use license payment to shift profits. This result emerges because previous studies assumed that the optimal profit shifting is characterized by an interior solution. Instead, if the MNE is constrained in its profit shifting strategy, our model show that even transfer price manipulation of an intangible asset can create investment effect. While previous studies related to the use of internal debt showed that investment effect can emerge through internal debt shifting, these investment effect could only arise in the high-tax country 8 See Appendix A. for more details. 9 The implication is consistent with empirical findings in Becker and Riedel (0) who show that the tax rate of the foreign jurisdiction negatively affects domestic investment of MNEs. 4

17 because through the reduction in capital costs via interest deduction of internal debt. Our results, however, suggest that when MNE affiliates can make losses, the investment effect can also occur in the low-tax country. Our model provides an underpinning to the empirically-observed bunching of MNE affiliates in two different ways. We highlight economic incentives to bunch around zero profits from above as well as from below. When division is profit constrained, its profits will be reduced to zero, providing a tendency to bunch from above. Differently, when division does not have enough absorbing capacity and, thereby, is loss constrained, its profit level will be increased to a zero level. Bunching occurs from below. Investment effects will reinforce the tendency to bunch. For instance, when division is profit constrained, the marginal return to investment in division is shifted to division. As shown above, this will promote division s investment incentives and, at the same time, reinforces bunching from above, since more investment costs will be deducted in division. 4 The government In this section, we analyze the implications of our results derived in the previous section on the efficiency of governments tax policies. We assume that each government sets its corporate income tax rate to maximize corporate tax revenues. The tax base in both countries consists of expected taxable profits generated by the local affiliate. On top of that, country collects additional tax revenues through the taxation of purely national firms. These revenues are labelled by t G(t ). 0 Expected tax revenues in each country 0 We assume that the tax base of national firms is sufficiently large to ensure t > t in equilibrium. 5

18 are given by T = t t t p p p UP p UP p p 0 p 0 t UN t p p P p 0 T = t G(t ) t p p ( p y δrk t ) t h (p )h (p )dp dp γ (p y δrk p y δrk ) h (p )h (p )dp dp ( p y δrk t γ ) h (p )h (p )dp dp (p y δrk p y δrk ) h (p )h (p )dp dp p UP (p y δrk p y δrk ) h (p )h (p )dp dp, (9) ( p y δrk t ) t h (p )h (p )dp dp. (0) γ We investigate the efficiency of the governments tax policies by deriving tax externalities, i.e. the influence of one country s corporate tax rate on the other country s tax revenues. Differentiating T i with respect to t j yields T t = t t t T t = t p p p UP p UP p p P p P p p UP γ h (p )h (p )dp dp [p f (k ) δr] k t h (p )h (p )dp dp [p f (k ) δr] k h (p )h (p )dp dp, t () ( γ [p f (k ) δr] k ) h (p )h (p )dp dp. t () If both affiliates earn enough profits the model replicates the standard insights that the tax externality generated by each country is given by the tax-induced adjustment in the MNE s profit shifting behavior which yields a positive tax externality /γ (first effect in () and ()). This is because a country s tax rate has no effect on the level of capital investments in the other country. This is different when the MNE is constrained in its profit shifting strategy. In this situation, a higher tax rate in one country affects investments in the other country and thus results in additional externalities. 6

19 Interestingly, the negative own-investment effect of a higher tax t generates a tax externality for country. This happens when affiliate is either profit or loss constrained. Under a profit constraint in country, the MNE shifts all profits out of affiliate into affiliate. Because a higher t lowers the capital stock k, also profits in affiliate are reduced and therefore the tax base in country. Under a loss constraint in country, a lower capital stock k lowers affiliate s capacity to absorb affiliate s profits, which results in a higher tax base in country. As such, the tax externality might be negative in sign, contrary to standard analysis, if the investment effect is strong enough. The externality on country s tax revenues is influenced by the cross-investment effect. However, the tax rate in country affects tax revenues in country only if affiliate is profitable, that is when p is sufficiently high, because in the other cases taxable profits of affiliate are zero. Similar to before, the cross-investment effect runs counter the standard profit shifting externality because a higher t reduces capital investment in country and thus the tax base in country. Again, depending on the relative sizes of these effects the tax externality on country might also be negative. We summarize in: Proposition 3 The sign of the tax externalities is ambiguous and depends on the relative sizes of the profit shifting and the investment externalities. If the investment externality is strong tax competition can lead to excessively high tax rates. Proposition 3 shows that in addition to the standard positive profit shifting externality our model delivers an additional negative externality of profit shifting which so far has been absent in the literature due to the assumption of unconstrained profit shifting by the MNE. In the presence of loss-making affiliates tax effects on investment will affect the tax externality in the high-tax as well as in the low-tax country. As such the negative externality will counteract the positive externality and might in cases overcompensate it. Becker and Riedel (0), have shown that investment effects can counteract the standard positive externality. Although their analysis abstracts from loss-making affiliates, and instead attributes the cross-border tax effect on investment to a common input like patents, our results are in line with their finding and comple- 7

20 ment their analysis. Because even in the standard model investment effects arise, our results suggest that focussing only on the direct effect of profit shifting is insufficient to derive implications for international tax issues. 5 Discussion Ex-ante transfer pricing In analyzing the role of losses for corporate investment incentives we have assume that transfer prices are set after the output price shock has been revealed. Ex-post transfer pricing implies that the MNE will excessively report profits around zero, an implication that is consistent with recent evidence on profit shifting with loss-making affiliates (Hopland et al., forthcoming; DeSimone et al., 07). The empirical finding suggests that MNEs have enough flexibility at their disposal to fine-tune transfer prices such that reported profits become slightly positive or negative. Still, an interesting question is whether the role of losses for investment behavior vanishes when the MNE has less flexibility in setting transfer prices. In doing so, we take the opposite view and derive the implications for the case when the MNE has to choose the transfer price before output prices are realized. In a situation where the transfer price has to be set before output prices are known, the MNE bases its optimal transfer pricing behavior on the expected price levels ˆp and ˆp. The major difference compared to the ex-post transfer pricing analysis becomes manifest in the fact that the MNE now knows (in expectation) in which of the five scenarios it will end up (cf. footnote ). Thus, the MNE maximizes a simplified version of equation (5) where profits of only one specific scenario will be maximized. For example, if the expected price in country is such that p 0 < ˆp < p UP the MNE will maximize Π 0. Our previous result that capital investments will be affected by profit shifting when the MNE is constrained in its tax-planning strategies, is basically unaffected with exante transfer pricing. However, there is one difference to the previous analysis. In cases The view that MNEs have sufficient flexibility in setting transfer prices is also in line with evidence reported in Johannesen et al. (07) where MNE affiliates in developing countries with presumably less tax enforcement capacity bunch around zero reported profit levels. 8

21 where the MNE is profit-constrained (either in country or ) the possibility of profit shifting will lead to over-investment in the country in which the constraint is binding. The intuition is that, when the MNE is profit-constrained in country, all profits of affiliate are shifted to country. As affiliate is loss-making, total profits of the MNE will be taxed at a zero rate. Moreover, since profit shifting exerts a positive effect on investment in country, it is easy to show that at the optimum p f (k ) < r holds. Instead, when the MNE is profit-constrained in country, the MNE shifts out all profit of affiliate into country. Thus, affiliate s profits will be taxed at country s rate t. However, investment in country is also affected by the MNE s profit shifting incentives. Again, it is easy to show that these incentives are strong enough such that at the optimum p f (k ) < r holds. To see whether tax externalities modify when the MNE has to commit to its profit shifting strategy ex ante, we start with the scenario in which affiliate is highly profitable. In this case, the MNE is unconstrained in its profit shifting strategy. Since capital investments in the two countries are not affected by the tax rate of the other country, the tax externalities comprise only the standard profit shifting effect and are thus given by T i t j = t i γ > 0. (3) Although a change in country s tax rate may still affect capital investment in country, this will not affect country s tax revenues in all other scenarios, just because the tax base of affiliate is zero even if capital investment in country changes. However, country s tax rate setting will have an effect on tax revenues in country. Intuitively, the externality exists not because the tax in country affects capital investment in country, but because it affects investment in country. Hence, similar to the case of ex-post profit shifting, the own-investment effect of a higher tax t generates a tax externality for country. The spillover exists when the MNE is profit constrained in country. In this case, the MNE can save on tax payments in country by shifting the marginal return on investment to country. 9

22 Asymmetric concealment cost The use of losses for tax savings leads to possibly unexpected directions in which transfer payments flow. They may flow to loss-making affiliates in otherwise high-tax jurisdictions, which are not susceptible of being the host country of transfer income and thereby inflated reported profits. Thus, fiscal authorities in low-tax jurisdictions, from where the profit flows originate, presumably do not place much monitoring effort on transfer payment of domestic affiliates to these otherwise high-tax jurisdictions. MNEs do not have to incur too much cost to prepare documents to justify the transfer payments and to restructure intra-firm transactions in order to masquerade the tax-savings strategy. All this might suggest that concealment costs are asymmetric w.r.t. deviations from the true price and that the asymmetry implies excessive deviations from the true price in the unexpected way with severe fiscal implications. Given the formal analysis above, the concern turns out to have less validity than possibly conjectured. Precisely, in the context of our model, the concealment cost function might take the form (g) = { γ(g )/ if g 0, γ(g )/ otherwise, (4) where γ > γ > 0. Overpricing of the internal service, g > 0, is more costly to the MNE as compared to tax-induced underpricing, g < 0. The parameter γ becomes relevant for MNE behavior when division is making losses, but is unconstrained in the choice of the transfer price. In this case, (9) governs the choice of the transfer price and a lower value of γ magnifies profit shifting from the low-tax jurisdiction to the high-tax jurisdiction. In all other cases, in which we observe an inverted transfer pricing, the asymmetry in concealment cost capitalizes in firm value, but does not influence MNE behavior. The capitalization effect occurs because the equilibrium is inherently asymmetric in nature and the equilibrium level of profit shifting will be non-zero. Of course, the described limited scope of asymmetric concealment cost to influence MNE behavior should be understood in a qualitative manner. When the price distribution is such that the scenario of an unconstrained transfer pricing choice in the presence of losses in division gets a sufficiently high probability mass, then the quantitative implications are important. 0

23 6 onclusion This paper examines multinational companies incentives to shift profits when their affiliates might end up in a loss position. Our theory suggests that investment and profit shifting incentives are insulated as assumed in previous studies only if multinationals are unconstrained in their profit shifting activities. However, when multinationals are constrained, the possibility of profit shifting will affect its investment decisions and gives rise to increase capital investments both in the high-tax and the low-tax country. Moreover, the existence of investment effects will also affect the efficiency of governments tax policies. In the standard model, the possibility of profit shifting only creates an incentive to lower tax rate to attract mobile profits. However, when investments also react to changes in the tax rate an additional negative externality of profit shifting arises. This is because affiliates profitability will determine the amount of profit shifting when the multinational is constrained. Our results thus suggest that focussing only on the direct effect of profit shifting is insufficient to derive implications for international tax issues.

24 A Appendix A. Deriving the first-order conditions for capital investment Differentiating (5) with respect to k yields E(Π) = k p UP p p 0 p UN P p p UN 0 p p P p 0 p p p p p UP {( t )p f (k ) ( δt )r}h (p )h (p )dp dp {( t )p f (k ) γ(p y δrk )[p f (k ) δr] ( δt )r}h (p )h (p )dp dp {p f (k ) r}h (p )h (p )dp dp {p f (k ) r}h (p )h (p )dp dp {( t )p f (k ) γ(p y δrk )[p f (k ) δr] ( δt )r}h (p )h (p )dp dp {( t )p f (k ) γ(p y δrk )[p f (k ) δr] ( δt )r}h (p )h (p )dp dp {( t )p UP f (k ) ( δt )r}h (p UP ) pup k h (p )dp {( t )p UP f (k ) γ(p UP y δrk )[p UP f (k ) δr] ( δt )r}h (p UP ) pup {( t )p 0 f (k ) γ(p 0 y δrk )[p 0 f (k ) δr] ( δt )r}h (p 0 ) p0 h (p )dp k p { UN p UN p UN p f (k ) r}h (p )h ( {p P f (k ) r}h (p P ) pun k dp ) pp h (p )dp k {( t )p 0 f (k ) γ(p 0 y δrk )[p 0 f (k ) δr] ( δt )r}h (p 0 ) p0 k h (p )dp {( t )p P f (k ) γ(p P y δrk )[p P f (k ) δr] ( δt )r}h (p P {( t ) f (k ) γ( y δrk )[ f (k ) δr] ( δt )r}h ( k h (p )dp ) pp h (p )dp k ) pun k h (p )dp {( t )p 0 f (k ) γ(p 0 y δrk )[p 0 f (k ) δr] ( δt )r}h (p 0 ) p0 h (p )dp = 0. (A.) k

25 Rearranging the terms containing single integrals yields E(Π) = k p UP p p 0 p UN P p p UN 0 p p P p 0 UN p p p p p UP {( t )p f (k ) ( δt )r}h (p )h (p )dp dp {( t )p f (k ) γ(p y δrk )[p f (k ) δr] ( δt )r}h (p )h (p )dp dp {p f (k ) r}h (p )h (p )dp dp {p f (k ) r}h (p )h (p )dp dp {( t )p f (k ) γ(p y δrk )[p f (k ) δr] ( δt )r}h (p )h (p )dp dp {( t )p f (k ) γ(p y δrk )[p f (k ) δr] ( δt )r}h (p )h (p )dp dp γ[p P f (k ) δr][p P y δrk t /γ]h (p P γ[p UP f (k ) δr][p UP y δrk (t t )/γ]h (p UP γ[ f (k ) δr][ y δrk t /γ]h ( ) pp h (p )dp k ) pup ) pun k h (p )dp k h (p )dp = 0. (A.) Because p UP = δrk y t t γy and = δrk y t γy, the last two terms vanish. Rearranging the second to the fifth terms as if taxes at a rate t have to be paid and capital costs 3

26 can be deducted at this rate, we get E(Π) = k UN p UN P p p p UP p p 0 UN 0 p p P p 0 UN p p [( t )p f (k ) ( δt )r]h (p )dp t [p f (k ) δr]h (p )h (p )dp dp t [p f (k ) δr]h (p )h (p )dp dp γ[p f (k ) δr][(t t )/γ (p y δrk )]h (p )h (p )dp dp γ[p f (k ) δr][(t t )/γ (p y δrk )]h (p )h (p )dp dp γ[p f (k ) δr][(t t )/γ (p y δrk )]h (p )h (p )dp dp γ[p P f (k ) δr][p P y δrk t /γ]h (p P ) pp h (p )dp = 0.(A.3) k Substituting g UP = (t t )/γ and g 0 = p y δrk solving the integral and rearranging the terms leads to first-order condition as given in (6) ( t )ˆp f (k ) = ( δt )r UN p UN P p p p UP p p 0 UN 0 p p P p 0 UN p t [p f (k ) δr]h (p )h (p )dp dp t [p f (k ) δr]h (p )h (p )dp dp γ(g UP g 0 )[p f (k ) δr]h (p )h (p )dp dp γ(g UP g 0 )[p f (k ) δr]h (p )h (p )dp dp γ(g UP g 0 )[p f (k ) δr]h (p )h (p )dp dp γ[p P f (k ) δr][p P y δrk t /γ]h (p P ) pp h (p )dp. (A.4) k 4

27 Differentiating (5) with respect to k yields E(Π) = k p UP p p 0 p UN P p p UN 0 p p P p 0 UN p P p UN p 0 p P UN p 0 p p p UP [( t )p f (k ) ( δt )r]h (p )h (p )dp dp [( t )p f (k ) ( δt )r]h (p )h (p )dp dp [( t )p f (k ) ( δt )r]h (p )h (p )dp dp {p f (k ) r γ(p y δrk )[p f (k ) δr]}h (p )h (p )dp dp [( t )p f (k ) ( δt )r]h (p )h (p )dp dp [( t )p f (k ) ( δt )r]h (p )h (p )dp dp [( t ) f (k ) ( δt )r]h ( ) pun k h (p )dp { f (k ) r γ( y δrk )[ f (k ) δr]}h ( ) pun k h (p )dp {p f (k ) r γ(p y δrk )[p f (k ) δr]}h (p P ) pp h (p )dp k [( t ) f (k ) ( δt )r]h ( ) pun k h (p )dp [( t )p f (k ) ( δt )r]h (p P ) pp h (p )dp k [( t ) f (k ) ( δt )r]h ( ) pun k h (p )dp = 0. (A.5) 5

28 Rearranging the terms containing single integrals, taking into account that y δrk t /γ = 0, yields E(Π) = k p UP p p 0 p UN P p p UN 0 p p P p 0 UN p p p p UP [( t )p f (k ) ( δt )r]h (p )h (p )dp dp [( t )p f (k ) ( δt )r]h (p )h (p )dp dp [( t )p f (k ) ( δt )r]h (p )h (p )dp dp {p f (k ) r γ(p y δrk )[p f (k ) δr]}h (p )h (p )dp dp [( t )p f (k ) ( δt )r]h (p )h (p )dp dp [( t )p f (k ) ( δt )r]h (p )h (p )dp dp γ[p f (k ) δr][p y δrk t /γ]h (p P ) pp h (p )dp = 0. (A.6) k Rearranging the terms containing double integrals yields E(Π) = k UN P p p UN p p p p [( t )p f (k ) ( δt )r]h (p )h (p )dp dp γ[p f (k ) δr][t /γ (p y δrk )]h (p )h (p )dp dp γ[p f (k ) δr][p y δrk t /γ]h (p P ) pp h (p )dp = 0. (A.7) k Replacing t /γ = g UN and (p y δrk ) = g P, solving the double integral and putting terms on the right side of the derivation delivers the first order conditions as in equation (7) ( t )ˆp f (k ) = ( δt )r UN P p p UN p γ[p f (k ) δr](g P g UN )h (p )h (p )dp dp γ[p f (k ) δr](g P g UN )h (p P ) pp h (p )dp = 0. (A.8) k 6