The Effect of Income-shifting Aggressiveness on Corporate Investment

Transcription

1 The Effect of Income-shifting Aggressiveness on Corporate Investment Lisa De Simone Stanford Graduate School of Business Kenneth J. Klassen University of Waterloo Jeri K. Seidman McIntire School of Commerce, University of Virginia August 2018 Keywords: transfer pricing, income shifting, investment, investment efficiency The authors appreciate outstanding research assistance from Gurpal Sran. Thanks to Dhamika Dharmapala, Saskia Kohlhase (discussant), Kevin Markle, Joyce Tian, Joseph Piotroski (discussant), Brady Williams (discussant), Betty Xing and participants at the 2017 Berlin Vallendar Tax Conference, 2017 National Tax Association Annual Conference on Taxation and 2018 University of Munster Tax Symposium. Klassen thanks the Social Sciences and Humanities Research Council of Canada for generously supporting this research.

2 The Effect of Income-shifting Aggressiveness on Corporate Investment We investigate whether intra-firm tax-motivated income shifting affects investment decisions. We model the complex interaction between two affiliates when the tax transfer price is related to an external market price. Our model predicts that increasing tax aggressiveness is associated with greater affiliate investment but lower investment efficiency. We use affiliate-level data on multinational corporations to develop a firm-specific measure of sensitivity to tax incentives to identify income-shifting aggressiveness. Using this measure, we estimate a positive relation between income-shifting aggressiveness and affiliate investment levels and a negative relation between income-shifting aggressiveness and affiliate investment efficiency. By empirically testing the theory that income-shifting aggressiveness alters equilibrium production decisions, we document that global after-tax profit maximization alters investment levels and extend the literature on investment distortions.

3 I. Introduction The vital importance of investment decisions to future firm productivity and financial health has spawned extant research across macroeconomics, corporate finance, public economics, and industrial organization, among other fields (Hubbard, 1998). In order to align divisional managerial incentives and thus facilitate optimal investment and production outcomes within decentralized firms, management accounting has historically advocated specific methods for allocating costs and profits among organizational units, including the price at which goods, services, and the rights to intangible property are charged across internal units. 1 However, for multinational firms, these intercompany prices also have significant tax implications; transfer prices set to reflect tax incentives allow multinational firms to reduce their global tax burdens by shifting income out of high-tax and into low-tax jurisdictions. 2 In setting transfer prices, multinational firms therefore may be required to trade off aligning managerial incentives that impact internal investment decisions with material tax benefits. We study this tradeoff. Prior literature examines how taxes influence transfer prices and multinational production and investment behavior. Because the prices that are optimal for incentive alignment are rarely acceptable for tax purposes, Samuelson (1982) and Halperin and Srinidhi (1987, 1991) model how the transfer pricing methods themselves (as prescribed by tax regulations) change firm-level investment decisions, even absent corporate choices to reduce global taxes. Multinational companies can also use the well-documented ambiguity in transfer pricing regulation to reduce worldwide income taxes, rather than to optimally trade off aligning managerial incentives and the 1 For example, Cook (1955), Dean (1955), and Hirshleifer (1956, 1957). 2 Hanlon and Heitzman (2010) summarize tax-motivated transfer pricing research in Section Tax-motivated transfer pricing generally shifts income by selecting a price for goods, services, interest rates, or rights to exploit intangibles that allocates low revenue or high cost to a high tax affiliate and high revenue or low cost to a low tax affiliate. Section II provides additional details. 1

4 most conservative application of tax laws, as described more fully below. Further, the ambiguity in transfer pricing regulation allows firms responses to vary: some firms are more conservative, while others are more tax aggressive (Klassen, Lisowsky, and Mescall, 2017). Finally, income shifting incentives have reportedly overtaken the managerial decisionmaking objective for many firms. For example, EY (2013) document that 83 percent of firms identified a tax motive as their highest priority in setting transfer prices. In contrast, only 15 percent of respondents chose management accounting motives as the highest priority in setting transfer prices. Thus, due to the transfer pricing regulations themselves, heterogeneity in firms responses to ambiguity in these regulations, and the tax opportunity provided by intercompany transactions, affiliate-level financial results of multinational firms reflect tax differentials across jurisdictions. These same financial results are frequently used internally for affiliate-level investment decisions. 3 Little is known about whether investment decisions continue to efficiently respond to investment opportunities in this environment. Our research question is whether internal affiliatelevel financial data that reflect aggressive multinational tax-motivated income shifting alters affiliate-level investment decisions. We investigate the level of investment and investment efficiency of a broad sample of affiliates of multinational firms, including many with global parents that are not tax residents of a country with a worldwide tax system. We model and test whether income shifting, in response to a tax rate differential, affects the level and efficiency of investment decisions. Our paper is therefore related to Amberger, Markle, and Samuel (2018) who study the effect of repatriation taxes on investment efficiency. They estimate that within-firm agency conflicts due to decentralization exacerbate the negative effect of repatriation taxes on 3 Though firms may keep two sets of affiliate-level financial results one for regulatory filings and another for internal decision-making this is costly. As described further below, Hiemann and Reichelstein (2012) show that even with two sets of books, cross-affiliate tax differences lead to changes in equilibrium transfer prices as the firm optimally trades off tax and operational efficiency considerations. See also Sansing (2014). 2

5 investment efficiency. We also study the effect of a firm-level decision on investment efficiency, but the decision we study is tax-motivated (the level of income-shifting aggressiveness the firm chooses) rather than related to the level of affiliate monitoring chosen. We first model the profit-maximization decision for a firm. We combine and build on the models in Baldenius, Melumad, and Reichelstein (2004); and Hyde and Choe (2005). Baldenius et al. (2004) examine optimal transfer prices across two affiliates in different countries under both cost-plus and market-based transfer prices. In their analyses that use an external market price, Baldenius et al. (2004) model the internal transfer price as a discount from the market price, reflecting tax incentives to under-report taxable income in the relatively high-tax country, which we adopt. However, the market price rather than the transfer price is used for tax reporting in their model; decoupled prices allow the transfer price to generate efficient production quantities. Prices are not decoupled in our model. Hyde and Choe (2005) create a model with many similarities to that in Baldenius et al. (2004), and also allow the price for tax reporting to be independent of the transfer price. Hyde and Choe (2005) allow both affiliates to sell the product to their local markets, which we adopt. Combining the two models allows us to introduce a domestic market for the manufacturing affiliate, thereby creating a benchmark market price. We extend the extant models by constraining the intra-firm transfer price to be the external sales price in the domestic market with a discount. The discounted price is used for both internal decisions and taxation (i.e., prices are coupled). Increasing the discount is costly because it increases the probability of revenue authority challenge, but doing so decreases the firm s worldwide taxes, as in Hines and Rice (1994). These innovations allow us to explore the effects of varying, and endogenously arising, degrees of tax planning aggressiveness on the equilibrium investment behavior of both affiliates, and within an internal decision-making system in which the transfer price is used for internal assessment as well as for 3

6 taxation. Our model predicts firms that more aggressively respond to tax-motivated income shifting incentives arrive at an alternative investment equilibrium described by increased affiliate investment levels but decreased affiliate investment efficiency, measured as the responsiveness of investment to investment opportunities. To test these predictions, we develop a firm-specific measure of the sensitivity of affiliate profitability to tax incentives. We use the income shifting prediction model from De Simone, Klassen, and Seidman (2017), which extends the Hines and Rice (1994) and Huizinga and Laeven (2008) empirical models of reported profitability as a function of capital, labor, productivity, and tax incentives to allow for the inclusion of unprofitable affiliates. We augment the model by including firm fixed effects interacted with affiliate weighted-average tax incentives C (Huizinga and Laeven 2008) and historical annual ownership information to match affiliates to a multinational firm each year. We estimate the model using separate entity (unconsolidated) financials from Bureau van Dijk (BvD), resulting in a sample of 151,083 multinational affiliateyears over the period 2002 to Our resulting firm-specific proxy for income-shifting aggressiveness, PS, is increasing in the extent to which a multinational firm exhibits a greater sensitivity of affiliate profitability to tax incentives, averaged across all affiliate-years for that multinational firm. Next, we match BvD multinational firms to parent firms in Compustat North America and Compustat Global to include our firm-specific measure of income-shifting aggressiveness in an investment framework. Following Badertscher, Shroff, and White (2013), we model the change in fixed and total assets as a function of cash flows and investment opportunities. As our model predicts, we find that income-shifting aggressiveness is positively related to the level of affiliate investment and negatively related to affiliate investment efficiency. In fact, though we estimate the expected positive relation between investment opportunities and investment level for firms 4

7 with below-median income-shifting aggressiveness, we find no statistical relation for firms with above-median aggressiveness. We contribute to the literature in several ways. First, we examine an important consequence of tax motivated income shifting global investment decisions and investment efficiency. We contribute to the stream of research examining the role of investment decisions in firm productivity and financial health by highlighting the impact of managerial tax planning decisions on multinational affiliate investment decisions. Extant research focuses on the former system of worldwide taxation in the U.S. For example, Pinkowitz, Stulz, and Williamson (2006) and Foley, Hartzell, Titman, and Twite (2007) suggest the foreign cash holdings of U.S. multinational corporations are not efficient due to agency conflicts created by tax incentives of the former U.S. worldwide tax system, and Hanlon, Lester, and Verdi (2015) and Edwards, Kravet, and Wilson (2016) study foreign investment and find that firms with tax incentives to hold excess foreign cash engage in more and less efficient foreign acquisitions. 4 Thus, this literature suggests tax systems can lead to less efficient investment decisions. Our work is distinct from this literature in two ways. First, we incorporate heterogeneity in firm s response to the tax incentives and ambiguity in transfer pricing. Second, our sample includes multinational firms subject to a territorial system, thus highlighting that income-shifting aggressiveness itself, and not excess cash locked out by repatriation taxes, optimally alters investment decisions. Second, we extend and test the relations outlined analytically in Samuelson (1982) and Halperin and Srinidhi (1987). Modeling resource allocation decisions of multinational firms both 4 Prior literature also documents that firms respond to income taxes in deciding where to locate investments (e.g., Grubert and Mutti, 2000; Hines and Rice, 1994; Devereux and Griffith, 1998; Becker, Egger, and Merlo, 2012), however this literature implicitly assumes firms have neutral transfer prices for decision making purposes. Buettner, Overesch and Wamser (2018) examine the sensitivity of property, plant and equipment (PP&E) and employment of German multinationals affiliates to tax rates, predicting that rules limiting profit shifting increase the effective tax rate of the jurisdiction, leading to lower investment. 5

8 with and without taxation, they find that the transfer pricing methods for tangible goods outlined in IRC Section 482 and related Treasury regulations lead to altered production decisions. 5 We test whether these changes are sufficient such that there is, on average, a relation between cross-border tax-planning responses and altered investment decisions by the affiliates of a multinational group. Third, we develop a firm-level estimate of transfer-pricing aggressiveness. A firm-level measure allows for more powerful cross-sectional tests than extant designs based on consolidated data (e.g., Klassen and Laplante, 2012). Using affiliate-level data to create such an estimate responds to the challenges identified with consolidation-level financial reports (see Donohoe, McGill, and Outslay, 2012). II. Background and Hypothesis Development Multinational Corporation Tax Planning In a decentralized, complex organization, the intra-firm transfer prices are important both for management control and local incentives, and for the computation of taxable income. There is a substantial literature exploring the tax planning activities by multinational corporations (see, e.g., Hanlon and Heitzman, 2010; De Simone et al. 2017). In summary, multinational corporations operate in multiple countries, and these countries tax corporate income at different rates; therefore, locating income in lower tax-rate countries reduces the global taxes of the firm. The goal of reducing taxes can be accomplished by moving high value-adding assets to lower tax rate countries, and by pricing intra-firm sales of goods, services, and rights at prices that report more income in lower tax rate jurisdictions. An example of the former is debt location (Dharmapala and Riedel, 2013), but much of the literature has focused on the latter and, for 5 The methods are comparable uncontrolled price method, resale price method and cost plus method. For further details on transfer pricing methods, see for example 6

9 example, has explored the effect of tax rate changes (Klassen, Lang, and Wolfson, 1993), tax havens (Dyreng and Lindsey, 2009), changes to accounting rules (De Simone, 2016), and customs duties (Blouin, Robinson, and Seidman, 2018). Transfer Prices Economic Theory Seminal work by Hirshleifer (1956, 1957) demonstrates that an appropriate transfer price can achieve efficient outcomes even when production decisions are decentralized. Management accounting has embraced the use of transfer prices as a means of allowing operating unit managers to make acquisition and production decisions in an efficient manner. The resulting profitability of the units is then available for resource allocation decisions by these managers. However, when the units within the firm are in different countries, the internal prices charged between the units also have significant tax implications. Halperin and Srinidhi (1987, 1991) build on limited prior research by carefully incorporating transfer prices that are constrained by tax regulation in the U.S. Their model demonstrates that relative to the efficient no-tax scenario, production quantities are distorted and transfer prices change in the face of realistic tax-based transfer price constraints. Even with the assumption that transfer prices do not respond to tax incentives (e.g., tax rates are the same in all jurisdictions), production levels in their model are still distorted by the tax system as the tax-permitted transfer prices lead to unavoidable changes to the internal production and sales coordination processes. We note, as have others, that transfer-pricing systems are used both for performance measurement of affiliate-level managers and for the computation of taxable income for countryspecific tax payments. To permit an efficient outcome in both of these separate uses, some firms adopt a so-called two-book system. 6 Hyde and Choe (2005) model such a system and find that 6 Note that this is different from the two-book system at issue in Watrin, Ebert, and Thomsen (2014) where the two books are financial (external) reporting and tax reporting. 7

10 even when they are not required to be the same, the two optimal prices are interdependent in certain situations because different tax rates alter the desired profit location. Hiemann and Reichelstein (2012) discuss both the tax compliance and the management allocation and coordination roles of transfer pricing as they survey recent theoretical research. They analyze an objective function that seeks to have decentralized affiliate managers operate in a manner consistent with the central managers value maximizing goal (i.e., goal congruence), rather than to achieve the production outcome achieved in a world without taxes. They first consider situations in which an arms-length price is given (for tax purposes) and determine what internal-only transfer price creates goal congruence. They show that in such a two-book system, the internal transfer price will depend both on the tax rate in the low tax jurisdiction and on the permitted tax transfer price. Thus, even in the situation where there is a separate transfer price for tax minimizing purposes, the internal transfer price differs from the classic Hirshleifer (1956) result: optimal centralized decisions also consider the tax benefits of reporting more profit in the low tax rate jurisdiction when determining production levels. With a fixed (given) tax transfer price, such as under an advanced pricing agreement, the decentralized firm is able to achieve the same outcome as achieved with centralized decisions. In the other extreme settings where the internal transfer is of intangible assets or the units each contribute to a common intellectual property Hiemann and Reichelstein (2012) note that the firm cannot achieve goal congruence because the investment objectives and the tax objectives cannot be effectively balanced through a transfer pricing scheme. Depending on the substitutability or complementarity of the investments made by the two units, or on other features of the investment environment, equilibrium unit behavior may lead to either underinvestment or overinvestment, relative to the central planner s equilibrium. 8

11 Further, decoupling the two prices is not without cost. Along with the cost of additional bookkeeping, a significant cost of decoupling is the possibility that the non-tax books will be included in any litigation over the firm s tax-motivated transfer prices. Klassen et al. (2017) document for a sample of 43 tax executives, only seven (16 percent) indicate yes when asked if they compute different prices for tax and assessing performance. Similar evidence is provided in, for example, EY (2003). We also examine the effects of tax-motivated transfer prices on equilibrium investment behavior. However, we believe that the tax transfer price is frequently related to the external market price (a one-book system). To study investment equilibriums in a one-book system, we expand on the models in Baldenius et al. (2004) and Hyde and Choe (2005). Specifically, the affiliates in our model face a domestic market for the manufacturing affiliate similar to these models, thereby defining an endogenous local external market price against which tax authorities compare the transfer price. Further, we explicitly tie the transfer price to this external market price, but with a possibly costly discount, something unique to this literature. Model Assume there is a multinational company with two affiliates. Affiliate A manufacturers and distributes the product in its home country, and Affiliate B purchases the intermediary good from Affiliate A and sells it to customers not in country A. Affiliate A manufactures all the raw product and ships the quantity demanded by Affiliate B. Each finishes, markets and distributes the product for their respective markets. Each sells quantities qi, where i = (A, B) and represents sales in country A and sales not in country A, respectively. A and B autonomously set their external sales quantities. Affiliates A and B incur costs of production; for tractability, we assume the costs are linear in quantity, respectively, c ( q A,q B )and the separate costs for Affiliate B are simply 9

12 included in its revenue function, RB. Prices are given by P i R i ( q i ) = P i q i ( ) q i, such that P i < 0, R i > 0, and R i < 0. ( ) and total revenue is q i Because the affiliates are in different countries, whose tax rates are and,, Affiliate A charges a transfer price to Affiliate B that is likely to be acceptable to the tax authorities. 7 To achieve a defensible transfer price, Affiliate A charges a price based on its external market price; the tax rate differential generates an incentive for Affiliate A to discount the external market price in order to shift some income to Affiliate B. 8 Thus, the price charged is, where.we interpret the size of the discount as the aggressiveness with which the firm pursues international tax planning. To summarize, we have the following model elements: P A = f (q A ), P B = f (q B ), R A = P A q A, R B = P B q B C B C A = f (q A,q B ), C B = f (q B ) though for simplicity in the analytics, we assume CA is a constant, c. The appropriateness of the discount is subject to challenge by the tax authority. Following Hines and Rice (1994) as modified Huizinga and Laeven (2008) and De Simone et al. (2017), we begin with the following equation: (1) where is the profit before shifting, is the income shifted, and is the perceived present value of the penalty in relation to the size of the shifting. Prior literature also deflates the penalty by the affiliate s profit, revenue or capital, but we abstract away from this because we are studying 7 Baldenius et al. (2004) model a central management that does not fully know the costs and revenues of the affiliates. We assume that Affiliate A makes the quantity and discount decisions based on perfect information of Affiliate B s cost and revenue functions. 8 The expectation of a penalty constrains the transfer price to be greater than zero. 10

13 the effects of varying penalties on equilibrium choices of a single pair of affiliates. To ensure each affiliate acts in the best interest of the firm, its managers are rewarded based on the total profits, including any tax penalty. Because Affiliate A chooses ξ and qa, and then Affiliate B chooses qb, the following are also true: Incorporating these assumptions, the profits for Affiliate i,, are as follows: (2) (3) Total profit, then, is. The revenue functions are multiplied by a positive factor,, to allow us to consider the effects of differing investment opportunities. Because Affiliate B responds to local market conditions and chooses the quantity to demand from Affiliate A for sale in its market, we solve for the optimal behavior of Affiliate B: (4) Thus as one would expect, the optimal value of q B * is a function of the transfer price charged, that is, a function of q A through the market price of Affiliate A, and. And, (5) (6) 11

14 Thus, the quantity demanded by Affiliate B will be increasing in both the quantity, q A, and the discount,. In short, as the price charged by Affiliate A declines, either because its market price falls through greater production or because it allows a greater tax-motivated discount, Affiliate B will increase its quantity. Before turning to Affiliate A s decisions, we consider the effect of different investment opportunities on the equilibrium behavior of Affiliate B. Starting with equation (6), we take the derivative with respect to the market condition. (7) This equation is negative if either R B = 0or R B is a log-concave function. A negative sign means that the relation between investment opportunities and Affiliate B s quantity will be less positive for affiliate groups with a higher transfer price discount (i.e., more tax-aggressive firms). If neither of the revenue function conditions hold, then the sign of the cross-derivative is ambiguous and it is an empirical question as to whether better investment opportunities lead qb to be more or less responsive to the discount. Having optimized the behavior of Affiliate B, we now turn to optimizing the total profit, the responsibility of Affiliate A. First, we determine the optimal quantity sold externally by Affiliate A, recalling that the optimal quantity of Affiliate B is a function of the choices made by Affiliate A. (8) 12

15 We assume that differences in the penalty primarily affect Affiliate A s choice of discount, rather than the quantity of its external sales. Thus if and R B = 0, 9 then the comparative statics for Affiliate A with respect to changes in the perceived penalty are as follows: (9) When we substitute the equilibrium condition,, and solve, we obtain the following: (10) Because the numerator of equation (10) is positive, and the three terms in the denominator are negative, we conclude that the discount charged by Affiliate A is decreasing in the penalty. We therefore conclude that firms with a lower perceived penalty will discount their transfer price more, thereby being more aggressive for tax purposes. And, with a larger discount, the quantity demanded by Affiliate B is higher. The Appendix provides a numerical example of our model. Summary and Hypotheses In our model, the firm produces its intermediary good in one affiliate and then finishes this good for local distribution plus sells it to a foreign affiliate for completion for the foreign markets. This set-up allows a quantity decision for both affiliates, a transfer price (which is used for both internal performance assessment and for taxation) between the affiliates, and an endogenous local 9 For simplicity, we assume that q A does not change with ; if we relax this assumption, we find that q A will be larger with a higher penalty. We also simplify the assumption in Hines and Rice (1994) in which the penalty declines in the size of Affiliate A. If the penalty declines in the size of Affiliate A, this would cause a negative relation between the penalty and q A. Overall, under reasonable conditions, the total quantity continues to decrease in the penalty. 13

16 external market price against which tax authorities compare the transfer price. Increasing the discount between the transfer price and the market price reduces the overall tax payments but comes at an expected cost due to possible tax enforcement actions. In equilibrium, more aggressive firms produce more goods, which requires greater investment. Specifically, a higher quantity demanded by Affiliate B will require additional investment by Affiliate B. Affiliate A will also have a larger investment because it must produce qa + qb. This leads to our first hypothesis. H1: The firm s responsiveness to tax-motivated income shifting incentives is positively related to its affiliate s investment. The demand for the product can vary across firms and so the investment opportunities also vary. The firms responses to these differences is muted when the transfer price discount is larger. Thus, in our second hypothesis we predict that investment efficiency is reduced by aggressive income shifting. H2: The firm s responsiveness to tax-motivated income shifting incentives is negatively related to its affiliate s investment efficiency. III. Measuring Income-shifting Aggressiveness Firm-Specific Measure of Sensitivity to Tax Incentives, We first estimate a firm-level sensitivity to tax incentives. We follow the literature (e.g., Hines and Rice, 1994; Huizinga and Laeven, 2008), using a transformation of the Cobb-Douglas production function to model affiliate-year reported profits as a function of capital, labor, productivity, and tax incentives. Specifically, we follow the model in De Simone et al. (2017) that allows for inclusion of unprofitable affiliates to retain a larger sample of multinational affiliateyears and achieve more precise estimation of tax-motivated income shifting across multinational 14

17 firms. We extend the model by incorporating firm indicator variables and interacting them with the tax incentive variables of interest: a capital-weighted differential tax rate of the affiliate relative to all related affiliates in the same multinational firm-year (Cit), an indicator variable for an unprofitable affiliate (Lossit), and the interaction between the two. We present this model in equation (11) below for affiliate i, year t and firm f. ln(πit + 1) = β0 + β1*ln(tangibleassetsit) + β2*ln(compexpit) + β3*industryroat + β4*ageit + β5* GDPt + β6* MarketSizet + β7*cit + β8*lossit + β9*lossit*cit + β10,f *Firmf + β11,f *Firmf*Cit + β12,f *Firmf *Lossit + β13,f *Firmf *Lossit*Cit + εit (11) We obtain all variables to estimate equation (11) from the BvD Osiris database unless otherwise noted. The dependent variable is the natural log of return on assets (ROA) plus one, where return on assets is measured as earnings before interest and taxes (OPPL) divided by total assets (TOAS). Our proxy for capital is tangible fixed assets (TFAS) and our proxy for labor is compensation expense (STAF). We measure productivity as IndustryROAt, defined as the median ROA by two-digit NACE industry-country-year, calculated using all affiliated and independent companies. Ageit is the natural log of one plus year t less the first year affiliate i is in our sample. We include two variables as proxies for economic shocks that could lead to unprofitability: (i) GDPt, measured as the year-over-year change in country-year GDP reported by the World Bank, and (ii) MarketSizet, measured as the two-digit NACE industry-country-year sum of affiliate and independent affiliate sales in year t less the sum in year t-1, scaled by 1,000,000. The indicator variable Lossit is equal to one if the affiliate reports EBIT less than zero, and zero otherwise. The variables of interest are the tax incentive variable, Cit, and its interaction with the firm fixed-effects vector, Firmf. The interactions between Cit and Firmf capture the firm-specific deviation from the sample average sensitivity to tax incentives measured across all affiliate-years 15

18 of the firm. 10 Because we expect reported income to be decreasing in the weighted-average tax rate differential of an affiliate, we expect β7 < 0. A negative coefficient on the firm-specific interaction suggests that the firm engages in more income shifting relative to the average sample firm, and vice versa. Following the language in Huizinga and Laeven (2008), we define the total firm-level sensitivity to tax incentives = β7 + β11,f. By construction, is a time-invariant firm characteristic, reflecting that costly tax avoidance like income shifting is likely long-term ( sticky ) in nature. We multiply estimated by negative one to create the empirical proxy, PSf, which is increasing in income-shifting aggressiveness. Sample for Estimating PS The BvD Osiris database contains financial, operating, and ownership information on independent and affiliated companies worldwide. We use unconsolidated company information, including annual ownership information, from Osiris over the period 2002 to 2014 and equation (11) to estimate PSf. 11 To remain in the sample of Osiris affiliate-year observations used to estimate PSf, we require each multinational firm-year to contain at least two affiliates in different jurisdictions such that they have the ability to engage in tax-motivated income shifting. We then restrict the sample to affiliates having all variables required for estimation (ROA, tangible fixed assets, compensation expense, age, change in GDP, change in market size, C, industry code, and ROA plus one greater than zero). Following De Simone et al. (2017), we drop firms in the banking and insurance industries and affiliate-year observations belonging to multinational firms reporting 10 In STATA, firm fixed effects are coded such that the sum of the fixed effects estimates are zero and the intercept is the sample average. 11 BvD only reports current year ownership linkages; as such, most studies using BvD data assume current year ownership linkages extend back through the entire sample period (e.g., De Simone 2016; De Simone et al. 2017; Huizinga and Laeven 2008; Markle 2016; Shroff, Verdi, and Yu 2015). We use historical disks to obtain annual ownership information and ensure appropriate linkages of affiliates to multinational firms each year. 16

19 a consolidated return on sales of less than three percent, as these firms likely have significantly different income shifting incentives. Using these sample selection criteria, our resulting sample for estimating PSf contains 151,083 affiliate-years representing 5,610 unique firms. Table 1 details our sample selection process. [insert Tables 1 and 2 here] We describe the sample used for estimating equation (11) in Table 2. Panel A presents summary statistics of sample affiliate-years, including all variables used in estimation. Values not expressed as logs or ratios are reported in $USD millions. All variables are winsorized at one and 99 percent. Sample affiliate-years have mean (median) operating profits of $9.07 ($0.96) million and ROA of 7.6 (6.4) percent, and 22 percent of the sample reports EBIT less than zero. Sample affiliates report mean (median) total assets of $205.7 ($20.1) million, tangible fixed assets of $25.9 ($1.2) million, and compensation expense of $19.4 ($4.4) million. By construction, the tax incentive variable C is roughly centered around zero; mean and median C are and respectively. Panel B presents the number of affiliate-year observations by country in which the affiliate is located. France is the most represented country with 30,906 affiliate-years, followed by the United Kingdom, Spain, and Germany. High representation among European countries is consistent with the requirement in the European Union (EU) that companies publicly disclose unconsolidated financial results for affiliates with limited liability. Barbados, Egypt, Jamaica, Macedonia, and Turkey are each represented by only one affiliate-year in our sample. In untabulated descriptive analyses, we examine the frequency of affiliate-years by the location of their parent, finding greatest parent representation in France, followed by the U.S., Germany, and the United Kingdom, and the least parent representation in Argentina and Montenegro. 17

20 Results of Estimating PS We report results of estimating equation (11) to obtain the empirical proxy for incomeshifting aggressiveness PS in Table 3, Panel A. All columns in Table 3 are estimated on the full sample of 151,083 affiliate-years (representing 5,610 unique firms) to generate better estimates of the income shifting prediction model, including the sample average coefficient on C to which to compare firm-specific tax-rate sensitivities. In the first three columns, we replicate the main result from De Simone et al. (2017) Table 5, Panel B to validate our sample. These three columns report results from estimating variants of equation (11) that exclude the firm indicator variables we will use to estimate PSf. Column (1) excludes the Loss indicator variable and its interaction with the tax incentive variable C, Column (2) includes the Loss indicator variable but excludes its interaction with C, and Column (3) includes both the Loss indicator variable and its interaction with C. Across all three columns, we estimate the same sign and similar magnitudes as De Simone et al. (2017) for the coefficients on capital, labor, productivity, and change in market size as well as the variables of interest. Specifically, in Columns (1) through (3), we estimate negative and significant coefficients on C and Loss. In Column (3), we estimate a positive and significant coefficient on the interaction between C and Loss, consistent with the main finding in De Simone et al. (2017) that unprofitable affiliates exhibit less responsiveness to tax incentives than profitable affiliates, or a shift-to-loss income shifting strategy. [insert Table 3 here] Having validated their model using our sample, we estimate equation (11) including firm indicator variables and their interactions with Loss, C, and Loss*C. Column (4) reports these results. In Column (4) we include firm indicator variables, but we only include indicators for the 2,243 unique firms with at least ten affiliate-years available to generate a firm-specific 18

21 coefficient. 12 For parsimony, we do not report estimated coefficients on these 2,243 firm indicator variables or the interactions. After including these firm indicator variables and their interactions, we continue to find negative and significant coefficients on C and Loss and a positive and significant coefficient on their interaction, consistent with prior results. In Panel B of Table 3, we present descriptive statistics for the resulting empirical estimate of firm-specific sensitivity to tax incentives PSf. We measure PSf as the sum of the coefficient on the tax incentive variable C and its interaction with the firm indicator variable, multiplied by negative one such that PSf is increasing in income-shifting aggressiveness. The resulting measure is positive at the mean and median, consistent with firms on average responding to explicit tax incentives when reporting affiliate profitability. In economic terms, the mean value of PSf (0.240) translates to a semi-elasticity of reported ROA of at the mean ROA. In contrast, we estimate a negative PSf for 44 percent of the firms in our sample, which suggests these firms respond more to implicit tax incentives, consistent with Markle, Mills and Williams (2017). We also report the distribution of affiliate-years included in each of the 2,243 unique multinational firms for which we estimate PSf. At the mean (median), we include 59 (28) affiliate-years for each unique firm. IV. Hypothesis Tests Research Design To examine the relation between firm income-shifting aggressiveness and the level and efficiency of affiliate investments, we estimate the following model for the determinants of investment for affiliate i, year t and firm f. Invit = δ0 + δ1 Qit-1 + δ2 CFft-1 + δ3 ln(tangibleassets)it-1 + δ4 HighPSf 12 As a result, we estimate PS for 2,243 unique firms. As discussed below, we are able to match 2,117 of these unique firms to Compustat to test our hypotheses. 19

22 + δ5 Qit-1*HighPSf + εit (12) We obtain all variables from BvD, Compustat North America, and Compustat Global. Our model specification is informed by prior research, as well our use of affiliate-level data, which imposes numerous data constraints. We specify two proxies for the dependent variable, Invit, based on affiliate-level investment in fixed assets from BvD. The first defines Invit as FixedInvit, the percent change in total fixed assets (TFASit TFASit-1/TFASit-1). The second is TotalInvit, the percent change in total assets (TOASit TOASit-1/TOASit-1). To proxy for investment opportunities, we use Q. Because the growth opportunities of affiliates likely vary within firms based on location and business, we specify two alternative country-industry-year measures of Q rather than a firm-year measure: a) mean and b) median country-industry-year Tobin s Q from the prior year (where Tobin s Q is estimated as countryindustry-year mean (median) market value of equity (PRCC_Fft*CSHOft) plus consolidated total assets (ATft) less the book value of equity (CEQft), scaled by total assets using Compustat data and the Fama-French 12 industry classifications). CFft-1 controls for differences in the firm s internal financing capability and is defined as prior year consolidated cash flow from operations (OANCFft-1, where available) scaled by beginning of year net fixed assets (PPENTft-2) using Compustat data. 13 ln(tangibleassets)it-1 controls for affiliate size following Shroff et al. (2015) and is defined is the natural log of beginning of year affiliate total assets (ln(toasit-1)) obtained from BvD. All variables are winsorized at 1 and 99 percent. 13 If OANCF is missing, cash flow from operations equals (operating income after depreciation (change in current assets change in cash) (change in current liabilities change in debt in current liabilities change in taxes payable) DEPR). Change is current assets is (ACT t - ACT t-1). Change in cash is (CASH t - CASH t-1). Change in current liabilities is (LCT t LCT t-1). Change in debt in currently liabilities is (DLC t DLC t-1). Change in taxes payable is (TXP t-txp t- 1). 20

23 We incorporate our empirical measure PSf into the model to test our hypotheses. Recall that our empirical estimate of, PSf, is the sum of the estimated coefficients on the tax incentive variable and the interaction between the tax incentive variable and the firm-specific indicator variable, all multiplied by negative one. As such, PSf is increasing in the firm s income-shifting aggressiveness. We specify our regression variable, HighPSf, as equal to one for affiliates of firms with above median PSf, and zero otherwise. Our first hypothesis therefore predicts δ4 > 0, or that higher firm-level sensitivity to tax incentives will lead to higher unexpected affiliate-level investment. Our second hypothesis predicts δ5 < 0, or that higher firm-level sensitivity to tax incentives will lead to lower affiliate-level investment efficiency, measured as the responsiveness of affiliate investment to investment opportunities. Hypothesis Testing Sample In order to estimate equation (12) and test our hypotheses, we require affiliate-year data from BvD to measure our dependent variable, country-industry-year and firm-year data for control variables from BvD and Compustat, and firm-level PS to generate our variable of interest. To generate the sample that satisfies these data requirements, we first match the 2,243 unique firms from our affiliate-year sample for which we could estimate PS to consolidated company information from Compustat North America and Compustat Global 2002 to We require an ISIN to facilitate the Osiris-Compustat Global match for international firms and a ticker to facilitate the Osiris-Compustat North America match for North American firms. We also require an industry classification (Compustat variable NAICS) code to include industry controls because investment is known to vary by industry. In addition, we require that we can calculate the variables defined above. Of the 2,243 unique firms with estimated PS from the affiliate-year sample, we are 21

24 able to match 63,775 affiliate-years for testing our hypotheses. These 63,775 affiliate years represent 1,983 unique Compustat firms. We summarize this sample selection process in Table 4. [insert Table 4 here.] Table 5 describes the affiliate-year sample used to test our hypotheses. Values are reported in $USD millions and all variables are winsorized at one and 99 percent. Sample affiliates report mean (median) tangible fixed assets of $30.28 ($1.46) million. Mean (median) total assets are $286.0 ($25.64) million. Untabulated statistics show that affiliates of firms with above-median income-shifting aggressiveness are significantly larger, with average tangible (total) assets of $34.32 ($298.0) million as compared to $26.08 ($273.5) for affiliates of less aggressive firms. The average year-on-year percent change in tangible fixed assets (total assets) is 14.7 (9.2) percent. At the mean (median), sample affiliates have an estimated Tobin s Q of 1.56 (1.38). For the 63,775 affiliate-years included in the hypothesis tests sample, we estimate mean (median) PS of 0.19 (0.09), indicating that sample firms on average respond to explicit tax incentives when reporting affiliate profitability. Table 6 presents correlations between regression variables. [insert Tables 5 and 6 here.] Hypothesis Test Results Table 7 presents results of our hypothesis tests. The first two columns use the percent change in fixed assets to proxy for affiliate investment, and columns (3) and (4) use the percent change in total assets. Columns also vary in the proxy for growth opportunities: (1) and (3) use MeanQt-1 and (2) and (4) use MedQt-1. Consistent with H1, we estimate a positive and significant coefficient on HighPSf in three of the four columns, suggesting that tax-motivated income shifting incentives are positively related to affiliate-level investment. The relation is most significant for fixed asset investment. As described more fully in our discussion of H2 below, the magnitude of the effect varies with investment opportunities. For example, relative to a firm with low income- 22

25 shifting aggressiveness, an affiliate of a firm with high income-shifting aggressiveness and growth opportunities at the 25 th percentile of observations experiences investment that is between 6.5 and 13.2 percent higher. [insert Table 7 here.] Regarding our second hypothesis, we estimate a negative and significant coefficient on Qt-1*HighPSf in three of the four columns, consistent with H2, suggesting that tax-motivated income shifting incentives are negatively related to affiliate investment efficiency. Like the results reported above, the coefficients are highly significant for fixed asset investment and less so for total asset investment. Specifically, we estimate that the affiliates of firms with low incomeshifting aggressiveness exhibit the predicted positive relation between investment opportunities and investment but that this relation is dramatically smaller for the affiliates of firms with high income-shifting aggressiveness. Regarding economic magnitude, we estimate that a one standard deviation increase in the growth opportunity set (0.45 for MeanQt-1 and 0.38 for MedianQt-1) for an affiliate of a firm with low income-shifting aggressiveness increases affiliate investment by 8.4 to 16.5 percent. However, a similar increase in growth opportunities for an affiliate of a firm with high aggressiveness has a significantly dampened effect, changing investment by between positive and negative three percent. These findings confirm reduced investment efficiency for affiliates of firms with high income-shifting aggressiveness, consistent with H2. V. Additional Tests Human Capital Investment and Investment Efficiency To provide a more complete analysis of the types of investment decisions influenced by income-shifting aggressiveness, we next test investments in human capital. We construct two affiliate-level measures of human capital investments using BvD data. Employmentit is the percent 23

26 change in number of employees (EMPLit EMPLit-1/EMPLit-1). Compensationit is the percent change in compensation expense (STAFit STAFit-1/STAFit-1). We then re-estimate equation (12) using these measures as the dependent variable to test the responsiveness of affiliate-level human capital investment and investment efficiency to income-shifting aggressiveness. Our sample is slightly smaller than that used to test our hypotheses due to missing data on the number of employees or compensation expenses in years t or t-1. We present results in Table 8. Columns (1) and (2) use the percentage change in employees as a proxy for human capital, and columns (3) and (4) use the percentage change in compensation expense. Columns (1) and (3) use MeanQt-1 as the proxy for growth opportunities and (2) and (4) use MedQt-1. Consistent with our main results using investments in physical capital, we estimate a positive and significant coefficient on HighPSf in three of the four columns, suggesting that taxmotivated income shifting incentives are positively related to investment. Also consistent with our main results, we estimate a negative and significant coefficient on Qt-1*HighPSf in three of the four columns. These relations are most statistically significant for compensation expense. Together with results presented in Table 7, these tests suggest that tax-motivated income shifting incentives are negatively related to investment efficiency in both physical and human capital. For both types of investment decisions, affiliates of firms with low income-shifting aggressiveness exhibit the predicted positive relation between investment opportunities and investment but this relation is dramatically smaller for the affiliates of firms with high incomeshifting aggressiveness. Specification Checks We perform several additional tests to verify the robustness of our results. First, we confirm that results are robust to eliminating U.S. multinationals from the sample, and thus that our results are not attributable to repatriation-tax induced excess cash documented by prior work. Second, we 24