The Role of Transfer Prices in Profit-Shifting by U.S. Multinational Firms: Evidence from the 2004 Homeland Investment Act

Finance and Economics Discussion Series Divisions of Research & Statistics and Monetary Affairs Federal Reserve Board, Washington, D.C. The Role of Transfer Prices in Profit-Shifting by U.S. Multinational Firms: Evidence from the 2004 Homeland Investment Act Aaron Flaaen 2017-055 Please cite this paper as: Flaaen,Aaron(2017). “TheRoleofTransferPricesinProfit-ShiftingbyU.S.Multinational Firms: Evidence from the 2004 Homeland Investment Act,” Finance and Economics Discussion Series 2017-055. Washington: Board of Governors of the Federal Reserve System, https://doi.org/10.17016/FEDS.2017.055. NOTE: Staff working papers in the Finance and Economics Discussion Series (FEDS) are preliminary materials circulated to stimulate discussion and critical comment. The analysis and conclusions set forth are those of the authors and do not indicate concurrence by other members of the research staff or the Board of Governors. References in publications to the Finance and Economics Discussion Series (other than acknowledgement) should be cleared with the author(s) to protect the tentative character of these papers.

The Role of Transfer Prices in Profit-Shifting by U.S. Multinational Firms: Evidence from the 2004 Homeland Investment Act∗ Aaron Flaaen Federal Reserve Board of Governors April 19, 2017 Abstract Usinguniquetransaction-levelmicrodata,thispaperdocumentsprofit-shiftingbehaviorbyU.S. multinationalfirmsviathestrategictransferpricingofintra-firmtrade. Asimplemodelreveals how differences in tax rates, both the corporate tax rates across countries and the dividend repatriation tax rate over time, affect the worldwide profit-maximizing transfer-prices set by firms for intra-firm exports and imports. I test the predictions of the model in the context of the2004HomelandInvestmentAct(HIA),aone-timetaxrepatriationholidaywhichgenerateda discreetchangeintheincentivesforU.S.firmstoshiftprofitstolow-taxjurisdictions. Matching individual trade transactions by firm, product, country, mode-of-transport, and month across arms-length and related-party transactions – following Bernard, Jensen, and Schott (2006) – yieldsameasureofthetransfer-pricewedgeatapointintime. Adifference-in-differencestrategy reveals that this wedge responds as predicted by the model: In the period following passage of theHIA,theexporttransferpricewedgeincreasedinlow-taxrelativetohigh-taxcountries,while the import transfer price wedge exhibited the opposite behavior. Consistent with the form of tax avoidance known as “round-tripping”, the results imply $6 billion USD of under-reported U.S. exports, nearly $7 billion USD of over-reported U.S. imports, and roughly $2 billion USD in foregone U.S. corporate tax receipts. JEL Codes: F23, H26, F14, H25, H32 Keywords: Profit-Shifting, Transfer Prices, Intra-firm Trade, Corporate Taxes ∗Arthi Rabbane and Emily Wisniewski provided excellent research assistance. Support for this research at the MichiganRDCfromNSF(ITR-0427889)isgratefullyacknowledged. Anyopinionsandconclusionsexpressedherein arethoseoftheauthoranddonotnecessarilyrepresenttheviewsoftheU.S.CensusBureau,theBoardofGovernors, oritsresearchstaff. Allresultshavebeenreviewedtoensurenoconfidentialinformationisdisclosed. FederalReserve Board of Governors, 20th and Constitution Ave, NW Washington D.C. 20036. aaron.b.flaaen@frb.gov.

1 Introduction Alargeshareofinternationaltradetakesplacewithintheboundariesofthefirm,andasaresultthe pricing of these transactions does not necessarily reflect market forces. Although trade regulations often require the price to be comparable to those undertaken at arms-length, the firm has many incentives to deviate from the arms-length principle. To take one well-known example, underpricing the exports to and over-pricing the imports from a lower tax jurisdiction will lower the overall global tax burden faced by a multinational firm. Apart from the difficulties this presents to national tax authorities, for economists the result is a large fraction of international trade that may not behave as leading models would predict. While a large literature looks at the many aspects of international transfer pricing – with profit-shifting motives being a prominent one – the challenge facing researchers is generally one of measurement. Obtaining the substantial amount of firm-level information required to assess whether intra-firm transaction prices mirror those at arms-length is typically only possible via costlyauditproceedings. Andevenwhenrichdataontransactionsorbalancesheetdataatthefirm level is available, most existing studies must assume that unobserved country-level heterogeneity that may affect intra-firm pricing decisions — potentially orthogonal to profit-shifting motives — is uncorrelated with country-level tax rates. After all, intra-firm prices could differ from those at arms length for many reasons that are unrelated to corporate tax rates. This paper uses unique data to measure firm-level transfer-price differentials while leveraging a policy change which temporarily increased the incentives of profit-shifting. The Homeland Investment Act (HIA) – part of the 2004 American Jobs Creation Act (AJCA) – granted a one-time tax holiday for U.S. firms to repatriate foreign profits from their controlled foreign corporations. Although U.S. firms are taxed on their global income, they can defer their U.S. tax liability on foreign profits – after deducting foreign tax credits – until such funds are repatriated back to the United States. The tax rate for repatriated earnings is therefore intended to remove tax rate differentials from affecting cross-country investment behavior. By lowering the taxes faced by profits held abroad to enter the U.S., the HIA increased the value of such profits to U.S. firms. Indeed, several sources (i.e. Bradley (2016), Drucker (2011)) have highlighted the potential role of the HIA in the so-called “round-tripping” of domestic profits, in which a firm transfers profits out of the United States – typically to low-tax countries – and then moves them back under special tax pro- 1

visions (in this case, those granted under the HIA) for a lower overall net tax obligation. Strategic use of transfer prices is one such method of moving profits to low-tax jurisdictions. To accurately measure the transfer-pricing behavior of firms during this policy change, I utilize thetransaction-leveltransfer-price“wedge”firstusedinBernard, Jensen, andSchott(2006). Using restricted access data from a partnership between the U.S. Census Bureau and U.S. Customs Bureau, Bernard, Jensen, and Schott (2006) matched individual arms-length (AL) and relatedparty (RP) transactions along a number of detailed dimensions available in the data, such as firm, country, detailed product, month, and method of transport. When used in the cross-section, the AL-RP wedge between the prices of these transactions amounts to an empirical analogue of the “comparable uncontrolled price” (CUP) method used by regulators to evaluate intra-firm prices against those at arms-length. This paper advances this measure by matching these wedges with new indicators of the nationality of the ultimate parent company, thereby identifying the U.S. multinational firms –rather than the U.S. affiliates of foreign multinationals – that would likely benefit from the HIA. I first formalize the logic underlying the optimal profit-maximizing transfer pricing behavior following a change in the repatriation tax rate with a simple two-period model that allows for transfer-pricingonbothexportandimporttransactions. Followingothermodelsoftransfer-pricing behavior (see Cristea and Nguyen (2016)) the firm is free to set any finite transfer-price, but with some nonzero probability is subject to an audit which assesses fines proportional to any deviations from a comparable arms-length transaction. The model demonstrates that the optimal export transfer price is increasing in the repatriation tax rate, whereas the import transfer price is decreasing in the repatriation tax rate.1 To test these predictions in the data I estimate the differential response of the AL-RP transfer price wedges in the post-HIA period in countries with low relative to high corporate tax rates. This difference-in-difference strategy has a number of attractive features. The variation in statutory tax rates determines which countries would be most profitable for changes to transfer pricing behavior following passage of the HIA. In addition to the appealing measurement of transfer-price wedges in the cross-section, the use of firm-by-product-by-country fixed effects in conjunction with examining changes in the time-series serves to mitigate any time-invariant measurement issues of 1Alternatively,theAL-RPexportwedge isdecreasing intherepatriationtaxrate,andtheAL-RPimportwedge is increasing in the repatriation tax rate. 2

the transfer-price wedge that may compromise its use in a static setting to capture profit-shifting.2 The correlation of the change in the transfer-price wedges with low vs high tax countries identifies whether profit-shifting motives are present. I find that the export transfer-price wedges of U.S. multinationals increase in the post-HIA periodincountrieswithrelativelylowtaxrates, consistentwiththemodel’spredictionsforoptimal profit-shifting motives. Moreover, the import transfer-price wedges of these firms move in the opposite direction: decreasing in the post-HIA period in low-tax countries, once again consistent with theory. The classification of multinationals according to the country of tax is critical in these calculations; recalculating these effects for the U.S. affiliates of foreign multinationals yields no evidence of profit-shifting. Applying these coefficients to the magnitude of relevant intra-firm trade yields rough estimates of the impact on trade and corporate tax receipts. Under the preferred specifications, I estimate that the HIA resulted in 6 billion USD of under-reported U.S. exports, 6.8 billion of over-reported U.S. imports, and roughly 2 billions USD of foregone corporate tax receipts. These results contribute to an expanding awareness of the differential behavior of intra-firm trade flows among international trade economists. A large fraction of this recent research has focused on the role of these trade flows in profit-shifting.3 The current paper benefits heavily from the data construction methodology outlined in Bernard, Jensen, and Schott (2006). While the authors show that their export transfer-price wedge is correlated with country-level corporate tax rates in a manner consistent with profit-shifting towards low-tax jurisdictions, the measure is also correlated with several other firm and country characteristics, such as indicators of market power and exchange rate movements. Corporate tax avoidance has received a large increase in attention by journalists, academics, and national/international institutions in recent years. Zucman (2014) documents that the share of corporate profits booked in tax havens has increased ten-fold since the 1980s to 20 percent of the total. Looking at the mechanisms for profit-shifting activities by firms, a meta-analysis of the literature by Heckemeyer and Overesch (2013) finds that the manipulation of transfer prices can accountforovertwo-thirdsofthemeasuredtotaleffect. Desai,Foley,andHines(2006)demonstrate 2Chief among these measurement concerns are whether the arms-length and related-party transactions are truly comparable products, despite the narrow firm-HS10-country-month-method of transport match. 3For an example of this differential behavior apart from profit shifting, see Neiman (2010), which demonstrates the differences in the timing and duration of price spells between arms-length and related-party trade. 3

that firms with more extensive intra-firm trade are more likely to use tax haven operations. The OECD in particular has advanced an ambitious action plan on “base erosion and profit shifting” (see OECD (2013)) in response to the increased evidence on these topics. Several recent papers on transfer-pricing also find evidence of profit-shifting behavior among multinationalfirms. Daviesetal.(2017)demonstratethegranularityoftransfer-pricingdifferentials in France, both across the set of firms and set of destination countries (specifically, those identified as potential tax havens). In their sample, about 25 firms account for 50 percent of the intra-firm trade with tax haven countries. Most recently, Cristea and Nguyen (2016) find evidence for profitshifting by multinationals using detailed firm-transaction data from Denmark. The authors utilize a triple-difference estimation strategy that, uniquely, takes into account strategic manipulation of arms-length prices in order to more closely match the optimal intra-firm transfer prices. Cristea and Nguyen (2016) find that a 10 percentage point decrease in the tax rate of a low tax country leads to a 5.7 percent drop in the export unit values of multinationals. One common challenge confronting these prior empirical studies is the difficulty of fully controlling for time-invariant, unobserved firm-specific determinants of pricing that may be unrelated to profit-shifting motives. A large literature has developed in recent years emphasizing quality differences, transport costs, and other factors influencing pricing-to-market practices by firms.4 To the extent that these other motivations for a particular country are correlated with statutory tax rates, this behavior could be conflated with profit-shifting. By looking at the changes of a firm-specific price wedge over a narrow time window within firm-product-country categories and coincident with a U.S. policy change, the current methodology largely overcomes this concern. Several existing strands of literature have used the AJCA/HIA to study firm behavior. One group of papers looks at the effectiveness of the HIA in achieving its stated aim of increasing U.S. investmentorhiring. Usingdifferentdatasourcesbutasimilarmethodology,bothBlouinandKrull (2009) and Dharmapala, Foley, and Forbes (2011) find no evidence that firms repatriating under the AJCA/HIA increased domestic investment or employment. Rather, these papers find that these firms were more likely to increase payouts to shareholders via share repurchases. Although this was not an approved use of the funds as specified under the law, both papers emphasize the fungible nature of cash once it was repatriated back to the United States. Survey evidence in 4See Manova and Zhang (2012) and Bastos and Silva (2010) for two examples. 4

Graham, Hanlon, and Shevlin (2010) confirms that share repurchases were a leading use of funds freedupbytherepatriationsundertheAJCA/HIA.Contrarytothesestudies, however, Faulkender and Petersen (2012) finds a significant investment response, but only in those firms deemed to be financially constrained. The conflicting results are due to different control samples used for the difference-in-difference regressions. Perhaps the paper most closely related to the present one is Bradley (2016), which evaluates profit-shifting activity among multinational firms in the years surrounding the HIA. Using the restricted-access microdata from the Bureau of Economic Analysis, Bradley (2016) finds that affiliate-reported earnings increase by an estimated $17 billion for the express purpose of exploiting the round-tripping benefits from the AJCA/HIA. The paper also looks for evidence of the role of transfer-pricing in particular, using a measure of the related-party trade balance of the multinational firm from the BEAs data on aggregated firm trade. The results support the explanation of transfer-pricing as a mechanism for profit-shifting during this period, but are only statistically significant when restricting the effect to firms with low levels of affiliate permanently reinvested earnings. The next section provides a brief review of the current state of transfer-pricing regulations and the statutory environment for the international taxation of multinational firms. Section 3 sketchesasimpletwo-periodpartial-equilibriummodeloftransfer-pricingtomotivatetheempirical exercises. Section 4 describes the relevant details of the Homeland Investment Act, and provides basic statistics of its effect in the aggregate. A description of the data and empirical results follows in section 5. The final section offers some related areas in need of further research. 2 Background on Transfer Pricing and U.S. International Taxation 2.1 Transfer-Pricing Recognizing the role that the pricing of intra-firm goods can play in tax avoidance efforts, most national tax authorities have put into place transfer pricing regulations. The foundation for such regulations is typically the arms-length principle, which states that intra-firm transactions should be comparable to those conducted between unrelated parties at arms-length. The OECD has been 5

deeply involved in the coordination and harmonization of the various concepts and regulations in place; there are nevertheless substantial cross-country differences in the regulation of transfer pricing. Therearevariousofficialandunofficialmethodsoffulfillingthearms-lengthprinciplesinpricing an intra-firm transaction, with the most common being the comparable uncontrolled price (CUP) method. A comparable uncontrolled transaction can be either an internal comparable if “it is between one party to the controlled transaction and an independent party” or an external comparable if it is “between two independent parties, neither of which is a party to the controlled transaction” (see OECD 2010). Fulfilling the comparable requirement is typically the most difficult to evaluate; as a result the OECD has identified a broad array factors to take into account, such as 1) characteristics (features, quality, etc) of the product, 2) functions performed (i.e. risks assumed), 3) the contractual terms, 4) the economic circumstances of the parties, and 5) the business strategies pursued by the parties. Because of this flexibility, any observed difference between arms-length and related-party prices of seemingly similar products may not necessarily indicate tax avoidance by the firm. On the other hand, it provides substantial leeway for firms in setting intra-firm prices. If no comparable transaction can be identified, several other methods are approved for use by the United States and other countries. Some, such as the resale price and cost-plus methods are margins-based methods, while others are based on transactional profits, such as the profit-split and transactional net margin methods.5 These other methods generally require a greater burden of documentation, and are consequently less popular. 2.2 A Brief Overview of U.S. International Taxation TheUnitedStatesusesaresidencebasisforthetaxationofmultinationalfirms,taxingtheforeignas well as domestic income of its residents. Although U.S. multinational firms incur U.S. tax liabilities ontheirincomeearnedabroad,theyreceivecreditsforanytaxespaidtoforeigngovernments. Thus, in principle the U.S. tax liability from foreign earnings is limited to the difference between foreign taxes paid and the tax payments that would have been paid if earnings were taxed at the U.S. rate. A number of details complicate this general setup. The first is the concept of deferral. U.S. taxation of foreign income occurs only upon repatriation, and so the income earned in a particular 5For more information on these and other regulations on transfer pricing, see Lohse, Riedel, and Spengel (2012) 6

period can be re-invested abroad and any U.S. tax liability is deferred until a later date. There are, however, some limitations to the benefits of deferral. Under sub part F provisions of U.S. tax law, certain types of foreign income – principally that from passive investments – is subject to immediate U.S. taxation, even if not repatriated as dividends to the U.S. parent firm. A second complication comes from the way foreign tax credits are used to reduce the U.S. tax liability. Because the amount of foreign tax credits is limited to the tax obligation under the prevailing U.S. tax rate (such that a firm cannot receive a “refund” for foreign taxes paid above those that would have prevailed in the U.S.), any tax credits above this limit are referred to as “excess foreign tax credits”. The current tax system allows a firm to use any excess foreign tax credits from a particular year to reduce their U.S. tax liability on foreign income in any of the two previousyearsorthesubsequentfiveyears. Inaddition,theforeigntaxcreditlimitisnotappliedon a country-by-country basis, but rather by summing total worldwide foreign tax payments against total worldwide foreign income. The combination of selective repatriation based on deferral, and the worldwide averaging of foreign tax credits give multinational firms some control over the U.S. taxes owed on their foreign earnings. Indeed, one would assume that the presence of a separate dividend repatriation tax would distort the behavior of repatriations back to the parent, as firms would find it optimal to keep income growing abroad (particularly in low-tax countries) and free of U.S. tax. However, Hartman (1985) showed that provided 1) the foreign affiliate is mature and financing investments out of retained earnings, and 2) that the dividend tax rate is constant over time, then the optimal repatriationdecisionisindependentofthedividendtaxrate. Ofcourse,theHIAviolatesthesecond condition and therefore led to strategic repatriation decisions by firms based on the conditions and timing of the HIA. See section 3 below, as well as Clausing (2005), for further discussion of the Hartman (1985) result and the optimal repatriation decision. 3 A Simple Model of Transfer Pricing and Dividend Remittances 3.1 Static Model of Transfer-Pricing Set-up Consider a multinational firm with a parent incorporated in the home country (i.e. the United 7

States) and an affiliate incorporated in a foreign country. In order to separately consider the transfer-pricing implications of exported and imported goods, I will assume that both the parent and affiliate produce a good that is sold locally as well as exported abroad. Hence, the parent company exports a good from Home and also receives an import from its affiliate in Foreign. The goods are exchanged between parent and affiliate at transfer prices, and then, acting as the wholesale/distributor, theparent/affiliatesellsthetradedgoodsinthelocalmarket. Forsimplicity, I will assume that production F(K) occurs using only capital, and that a fixed fraction (θ ,θ ) of h f the parent/affiliate output is exported abroad. I define the relevant prices below: • phh: Arms-length price of parent product in Home • phfT: Export transfer price • phfF: Export price sold in Foreign • pff: Arms-length price of Foreign affiliate product in Foreign • pfhT: Import transfer price • pfhF: Import price sold in Home • rh,rf: Implied rental rate of capital in Home/Foreign This setup implies that the parent and affiliate each engages in three activities: domestic production/sale, exports at transfer prices, and distributor/wholesaler for the imported product. Thus, assuming a tax rate of τh in the Home economy, the profits of the parent company in Home can be written as: ΠH =(1−θh)F(Kh)(phh−rh)(1−τh)+θhF(Kh)(phfT −rh)(1−τh)+θfF(Kf)(pfhF −pfhT)(1−τh). (cid:124) (cid:123)(cid:122) (cid:125) (cid:124) (cid:123)(cid:122) (cid:125) (cid:124) (cid:123)(cid:122) (cid:125) πhh: netprofitson πhfT: NetHometransferprofits πfh: NetHomeprofits Homedomesticsales onexportsales onsellingForeignexportgood Similarly,assumingataxrateofτf,theprofitsoftheaffiliateinForeigncanbewrittenidentically: ΠF =(1−θf)F(Kf)(pff −rf)(1−τf)+θfF(Kf)(pfhT −rf)(1−τf)+θhF(Kh)(phfF −phfT)(1−τf). (cid:124) (cid:123)(cid:122) (cid:125) (cid:124) (cid:123)(cid:122) (cid:125) (cid:124) (cid:123)(cid:122) (cid:125) πff: netprofitson πfhT: NetForeigntransferprofits πhf: NetForeignprofits Foreigndomesticsales onexportsales onsellingHomeexportgood 8

The tax authorities state that the intra-firm transaction must be set at arms-length, contingent to a comparable uncontrolled transaction (CUP); however, there is imperfect enforcement. With probability related to λi i ∈ {h,f} the tax authorities audit the firm and, using the firms armslengthprice(phh andpff)astheCUP,assesspenaltiesbasedonthedifferencesbetweenthetransfer priceandtheCUP.Afterincorporatingthisfeature,thecompletefirm-levelprofitscanbeexpressed as: λh (cid:104) (cid:105)2 λf (cid:104) (cid:105)2 Π = ΠUS +ΠF − (phh−phfT)θhF(Kh) − (pff −pfhT)θfF(Kf) . (1) 2 2 In a static version of the model, it is sufficient to take first order conditions with respect to the two transfer price decisions: ∂phfT : −τhθhF(Kh)+τFθhF(Kh)+λh(phh−phfT)θhF(Kh)θhF(Kh) = 0 (τf −τh)+λ(phh−phfT)θhF(Kh) = 0 (τh−τf) phfT = phh− (2) λhθhF(Kh) (cid:124) (cid:123)(cid:122) (cid:125) ifτh>τf thenunderpriceintra-firmexports ∂pfhT : τhθfF(Kf)−τFθfF(Kf)+λf(pff −pfhT)θfF(Kf)θfF(Kf) = 0 (τh−τf)+λ(pff −pfhT)θfF(Kf) = 0 (τh−τf) pfhT = pff + . (3) λfθfF(Kf) (cid:124) (cid:123)(cid:122) (cid:125) ifτh>τf thenoverpriceintra-firmimports We can see that in this static case, the firm finds it optimal to under-price intra-firm exports and over-price intra-firm imports whenever the tax rate of the home country exceeds that of the foreign country. 9

3.2 Adding in the Dynamics Now I take the static structure of the transfer-pricing arrangement from below and build in dynamics in which the firm maximizes profits over multiple periods subject to deferral and dividend repatriation from Foreign to Home. Consider a firm that maximizes Home profits in a two-period version of the above framework, where the firm can repatriate foreign profits subject to a dividend tax rate of τd.6 Inadditiontoitsexistingstockofcapital, followingBradley(2016)Iassumetheforeignaffiliate has a stock of cash holdings bf invested in a passive asset earning a rate of return of ρ. I further assume that at the beginning of period 1, the firm is indifferent between moving funds between its stock of cash and its capital stock such that it has exhausted all profitable investments in its scale of production. I denote the dividends remitted to the parent as d . In the second period all foreign i operations are liquidated and sent back to the parent firm. Assuming a discount factor of β, the multinational firm’s optimization problem is now: (cid:110) (cid:111) max Πh+(1−τd)d +βΠh+β(1−τd)d −Λ 1 1 2 2 d1,ph 1 fT,pf 1 hT λh (cid:104) (cid:105)2 λh (cid:104) (cid:105)2 where Λ = (phh−phfT)θhF(Kh) +β (phh−phfT)θhF(Kh) 2 1 1 1 2 2 2 2 subject to the constraints: λf (cid:104) (cid:105)2 d ∈ [0,Πf − (pff −pfhT)θfF(Kf) +(1−τf)ρbf +bf] [no borrowing to finance dividends] (4) 1 1 2 1 1 1 1 1 (cid:110) (cid:111) phfT, pfhT | Πh ≥ 0 [no negative domestic earnings] (5) 1 1 1 6Given the worldwide tax system used by the U.S., and ignoring cross-crediting and other considerations, the statutory dividend tax rate would be the U.S. tax liability net of any foreign tax credits. This amounts to a τd = τh−τf. 1−τf 10

There are also the following definitions: λf (cid:104) (cid:105)2 d = Πf − (pff −pfhT)θfF(Kf) +ρ(1−τf)bf +bf [liquidate foreign operations at t=2] (6) 2 2 2 2 2 2 2 2 λf (cid:104) (cid:105)2 bf = ρ(1−τf)bf +Πf − (pff −pfhT)θfF(Kf) −d +bf [LOM for foreign cash holdings] (7) 2 1 1 2 1 1 1 1 1 λh (cid:104) (cid:105)2 Kh = Πh− (phh−phfT)θhF(Kh) +(1−τd)d [LOM for home capital] (8) 2 1 2 1 1 1 1 1 Kf = Kf [due to constant returns on foreign cash holdings] (9) 1 2 Solving through the first order conditions for the transfer prices in this scenario, and assuming an interior solution where the shadow values pertaining to the constraints are zero, yields the following expressions: (1−τd)(1−τf)−(1−τh) phfT = phh− 1 (10) 1 λhθhF(Kh) 1 and similarly (1−τd)(1−τf)−(1−τh) phfT = phh− 2 . (11) 2 λhθhF(Kh) 2 Relative to equation (3), the dividend tax rate affects the optimal transfer price used by the multinational firm. Specifically, the optimal transfer-price on the export side is increasing in the dividend tax rate (the gap between the arms-length and transfer-price widens as the dividend tax rate falls). On the other hand, following the logic of equation (3), on the import side the optimal transfer-price is decreasing in the dividend tax rate. Whatabouttheoptimaldividendpolicy? Thefirstorderconditionforthefirstperioddividends yields the following intuitive expression: ∂Π(cid:101) H (1−τd) 1+β 2 = β 2 [1+ρ(1−τf)] (12) ∂KH (1−τd) 2 1 where Π(cid:101) H is defined as the second period after-tax profits from home production, net of any 2 transfer price penalties. Equation (12) indicates that the optimal first-period dividend repatriation should balance the benefits from repatriation immediately (the left hand side) with the potential to reinvest earnings in the foreign passive asset and repatriate in the subsequent period subject 11

to (assuming τd < τd) a more favorable dividend tax rate (the right hand side). Going back to 2 1 the discussion above, it is clear that if the dividend tax rate was constant, where τd = τd, then 1 2 consistent with Hartman (1985) the repatriation decision would be independent of the dividend tax rate. 4 The Homeland Investment Act This section provides the background and details pertaining to the Homeland Investment Act. An important component of the empirical strategy used in section 5 relates to the timing of this tax law change, and so I will devote particular attention to the specific dates and events that affected the resolution of uncertainty for this provision. 4.1 Background The origins of the Homeland Investment Act date back to the early 1980s and the institution of foreign sales corporations (FSCs), a tax instrument intended to reduce the tax liability of U.S. corporations from profits derived from export activity. Following complaints from the European Community, the WTO ruled in March 2000 that the tax treatment of the FSCs were a form of export subsidy and therefore illegal according to current agreements. Later that year, the U.S. passed the FSC Repeal and Extraterritorial Income Exclusion Act which essentially extended the FSC tax break to all types of entities by excluding extraterritorial income from the calculation of gross income. (Foreign companies that are U.S. taxpayers could then also use the tax break.) In January of 2002, however, the WTO ruled that this modified tax treatment continued to constitute a prohibited export subsidy, and in May 2003, authorized the European Union to impose countervailing duties up to a level of $4.04 billion on certain products originating from the U.S. WiththebeneficialtreatmentoftheFSCtaxmeasureforoffshoresaleslikelytoend,anumberof U.S. based groups began lobbying for a potential replacement. Several dozen companies organized as the Homeland Investment Coalition pushed for a temporary cut in the tax rate on repatriated foreignprofits. ThiseffortreceivedasubstantialboostwhenaSeptember2003studybyJPMorgan estimated that such a measure would attract roughly $300 billion in capital inflows into the United States, and add a half-percentage point to economic growth over a two-year window (J.P Morgan Chase (2003)). The HIA was included as part of the larger American Jobs Creation Act (AJCA) 12

introduced in the House of Representatives in July of 2003, and was widely acknowledged to be in response to the WTO rulings.7 Clausing (2005) contains a detailed summary of the features of the AJCA, but prominent among them were the repeal of the extraterritorial income (ETI) exclusion ruled illegal by the WTO, and an income tax deduction for domestic production activities.8 Debate on the AJCA was delayed until early 2004; meanwhile, in March 2004, the European Union began to impose the WTO-sanctioned countervailing duties. These tariffs were set at an initial rate of 5 percent and scheduled to increase by 1 percent per month while the ETI exclusion remained in place. Soon thereafter, the House of Representatives took up the AJCA and passed the bill on June 17, 2004 by a vote of 251-178. The House vote was mostly along party lines, with Republicans contributing 203 ayes, 23 noes, and 2 non-votes, and the Democrats consisted of 48 ayes, 154 noes, and 3 non-votes. Ignoring strategic interactions and conditional on Republican votes, a united block of Democrats against the bill would have effectively prevented passage. I interpret this fact to be evidence of unresolved uncertainty leading up to the passage of the AJCA by the House in June, 2004. The Senate approved the vote on July 15th, by a wider margin: 78-15. By the time President Bush signed the AJCA into law on October 22, 2004, the EU countervailing duties stood at 12 percent. 4.2 The HIA and the Dividends Received Deduction The HIA allowed firms a one-time deduction of 85 percent of their extraordinary dividends received from additional taxes coming from controlled foreign corporations. Firms could elect to take this deductionbeginningonorforoneyearaftertheAJCAwassignedintolaw. Thisdividendsreceived deduction (DRD) lowered the effective tax rate from the maximum statutory corporate rate of 35 percent to 5.25 percent (15% times 35%). There were a number of restrictions on the extraordinary dividends that would qualify for this deduction, and furthermore, on how the repatriated funds could be used by the parent firm. Extraordinary dividends were defined as being the excess of repatriations during the selected year over the average amount of repatriations during the previous five years, excluding the high- 7AnarticleintheSanFranciscoChroniclewithheadline“Lawmakerspushtaxbreakforbusinesses;Passageseen as possible because of WTO ruling” appeared the day after the AJCA was introduced in the House. See Lochhead (2003). 8Because it would affect the arms-length and related-party transactions equally, the deduction for domestic production activities should not influence the empirical results presented in section 5. 13

est/lowest years. The qualifying dividends were further limited to be the greater amount of 1) $500 million, 2) earnings reported as permanently reinvested on the last audited financial statement on/before June 30, 2003, or 3) 35 percent of the tax liability of those permanently reinvested earnings. Finally, the qualifying amount was reduced by any increase in the amount of related-party debt incurred by foreign subsidiaries between October 3, 2004 and the close of the tax year for which the firm claimed the dividends received deduction. This final requirement was intended to prevent the practice of “intra-group lending” whereby a parent corporation loans funds to a foreign subsidiary, who in turn remits the cash back in the form of a cash dividend.9 There were also restrictions on the uses of the repatriated funds. After requests for clarification regarding these rules, in January 2005, the IRS released further guidance on the restrictions on the uses of the money received from the extraordinary dividend. The uses of these funds was generallylimitedtohiringortrainingofworkers,infrastructureorcapitalinvestments,researchand development, and certain administrative expenses and debt repayments. Expenditures that were explicitly prohibited were executive compensations, inter-company distributions, dividends and stock buybacks. The law required the CEO to submit a domestic reinvestment plan to accompany the firm’s financial statements in the year of the repatriation. For more details on the restrictions of the DRD, see Redmiles (2008). 4.3 The Effects of the HIA in the Aggregate According to statistics compiled in Redmiles (2008), roughly 850 corporations repatriated $362 billion as part of the HIA in the years 2004-2006. Of this amount, $312 billion qualified for the deduction. As reported in Figure 1, this amounted to a quantity of net dividends that was roughly 10 times higher than the average in the years leading up to the DRD. Table1documentsthetop10sourcecountriesofthefundsrepatriatedaspartoftheHIA.Apart from the Netherlands, Canada, and the United Kingdom, the other top-10 countries had pre-2003 statutory tax rates that were significantly less than the U.S. tax rate – by 15 percentage points or more. The case of Netherlands is interesting: although the maximum statutory tax rate was 34.5 percent in 2004, the so-called participation exemption excludes capital gains, dividends from 9SuchstrategieswereattheforefrontofthemindsoftaxstrategistsasthedebateontheHIAprogressed. A2003 article (see Pulizzi (2003) ) referencing the HIA had as a headline: “Proposed US Tax Break Could Add Supply to Eurobond Mkt.” 14

qualifyingsubsidiaries,andprofit-participationloaninterestfromcorporatetax. Theseexemptions, together with a large range of bilateral tax treaties, has given the Netherlands a reputation as an international tax haven despite its high statutory tax rate.10 For the purposes of this paper, the exemption of dividend taxation by countries such as the Netherlands makes infeasible a direct mapping between the countries involved with transfer pricing transactions and the dividend repatriations. Although the optimal countries to engage in trade transactions should exhibit a low tax environment, the affiliate profits could subsequently be transferred as tax-free dividends to countries, such as the Netherlands, provided the appropriate tax treaties are in place. 5 Empirical Results 5.1 Data Description A principal contribution of this paper is the use of unique microdata that measures transfer-pricing differentialsbymatchingarms-lengthandrelated-partytransactionswithinnarrowfirm/country/product/month/ transport-mode criteria. Thus, at the heart of the analysis is the Linked/Longitudinal Firm Trade Transaction Database (LFTTD), which is a joint collaboration between the U.S. Census Bureau and the U.S. Customs Bureau. This dataset links the universe of goods trade transactions to the LongitudinalBusinessDatabase, theCensusBureau’sregisterofallestablishmentsoperatinginthe UnitedStates.11 Foreachindividualtradetransaction, theLFTTDrecords, amongothervariables, the source (destination) country, product code, value, quantity, date, transport mode, and whether thetransactionoccursatarms-length, orbetweenrelated-parties. Ontheexportside, atransaction between a U.S. producer and foreign consignee is defined to be between related-parties if “either party of the transaction owns directly or indirectly 10 percent or more of the other party.”12 On the import side, a transaction is defined as being between related parties “if any person directly or indirectly owns, controls, or holds power to vote 5 percent or more of the outstanding voting stock 10The famous “double Irish Dutch sandwich” tax strategy is a direct consequence of the exemptions allowed by theNetherlandsformultinationalfirms. Foranin-depthdescriptionofthisparticulartaxstrategy,seeInternational Monetary Fund (2013). 11For more information on the LBD, see Jarmin and Miranda (2002). 12See section 30.7(v) of the Foreign Trade Statistics regulations (https://www.census.gov/foreign-trade/ regulations/regs062004.pdf) 15

or shares” of the other party.13 A new addition to this dataset for purposes of transfer-pricing research are identifiers of the location of the ultimate parent company involved in the transaction. This information is crucial in the context of this research as it is the U.S.-incorporated firms that are relevant to receive the DRD, and the LFTTD transactions in general will contain many observations related to foreign multinationals with affiliates operating in the United States. These multinational identifiers come from a new link between two international corporate directories and the Business Register (BR) of the Census Bureau, and allows one to characterize each multinational firm as being part of either a U.S.(headquartered)orForeignmultinational. Forinformationonthesedirectoriesandthelinking procedure, see Flaaen (2014) and Appendix A.1. For the purposes of understanding how transfer prices may have interacted with country-level tax rates, I use estimates of statutory corporate tax rates from the now discontinued World Tax Database from the Office of Tax Policy at the University of Michigan. As this data only runs through year 2003, I extend it using a variety of sources, including Ernst & Young, KPMG, and estimates from Loretz (2013), taking care to ensure that definitional issues remained consistent across the datasets. 5.2 The AL-RP Transfer Price Wedge FromtherawLFTTDdataset,Ifirstremoveanytransactionthathasamissing,imputed,converted, or zero quantity. For the transactions that remain, I construct the unit value as the total value per unit of quantity, and define the AL-RP wedge as in Bernard, Jensen, and Schott (2006) where: wedge = lncup −lnrp . (13) ficmt ficmt ficmt The cup is defined as the average across all arms-length transactions for a firm f product ficmt i to/from country c in month t by transport mode m. The AL-RP wedge measures the difference between the actual related party unit value (price) and that implied by an empirically-constructed comparable uncontrolled price (specifically, using an internal comparable transaction). New to this paper is calculating the identical transfer-price wedge using import transactions. 13This definition dates back to Section 402(e) of the Tariff Act of 1930; as amended currently it is found in 19 U.S.C. 1401a: (https://www.law.cornell.edu/uscode/text/19/1401a) 16

There is an important conceptual distinction between the AL-RP wedge calculated using export transactions data and that using the import transactions. On the export side, this strategy takes a product produced by a particular U.S. firm and then measures the differential observed selling price –in a given month, country, and mode of transport – to an intra-firm affiliate compared to an averageacrossarms-lengthbuyers. Ontheimportside, thisstrategyissomewhatdifferent. Ittakes aproductimportedbyaparticularU.S.firm, andthenmeasuresthedifferentialobservedpricethat productispurchased–inagivenmonth,sourcecountry,andmodeoftransport–fromanintra-firm affiliate compared to an average across arms-length sellers. The difference is subtle but important to keep in mind. Practically speaking, because the producing firm is the same, it is more likely that the export AL-RP wedge reflects identical products within the detailed HS-10 coding system. The import-based AL-RP wedge relies somewhat more heavily on the HS-code-based differentiation of product attributes.14 Bernard, Jensen, and Schott (2006) compute the average export AL-RP wedge for the years 1993-2000 (encompassing roughly 3.5 million observations) and find an average value of 0.43 log points (standard deviation of 1.77). For the same sample period, I find an average import AL-RP wedge of 0.18 log points (standard deviation of 1.39).15 Connecting these measures to country-level corporate tax rates is a first step in evaluating potential profit-shifting activity. The first two columns of Table 2 replicates results from Table 5 in Bernard, Jensen, and Schott (2006), showing that lower corporate tax rates are associated with larger export AL-RP wedges or lower intra-firm export prices. The columns (3) and (4) of Table 2 replicate this exercise for the import AL-RP wedges. Remarkably, after controlling for product-level fixed effects in column (4), the import AL-RP wedge has the opposite sign as the export AL-RP wedge specification, which is consistent with profit-shifting motives as specified in equation (3) above. 14Oneapproachthatcouldbeusedtonarrowyetfurtherthetransaction-levelmatchfortheimporttransactionsis toutilizethe“manufacturerID”variableontheLFTTDimportdatabase. Thiscode,upto15charactersinlength,is meanttocaptureidentifyinginformationfortheforeignmanufacturerofaparticularimporttransaction. Forfurther detailsofthisvariable,seeMonarch(2014)orKamal,Krizan,andMonarch(2015). Ontheonehand,itispotentially plant-specificandthereforeanimprovementoverthefirm-baseddefinitionofproductionusedontheexportside. On the other hand, the limited firm and address information in the code itself may pose issues for differentiation, and the variable is less commonly used by researchers. 15Also similar to Bernard, Jensen, and Schott (2006), this import wedge is higher for differentiated products (as measured by Rauch (1999)) at an average of 0.28 log points, than commodities (0.08 log points). 17

5.3 Sample Construction Not all firms were eligible for the DRD under the HIA, nor would many find it optimal to participate or find transfer-pricing strategies to be an effective method for minimizing their global tax burden. The ideal dataset would include only those parent corporations that took advantage of the deduction; unfortunately, the IRS does not disclose lists of firms underlying the repatriations. At present, I make no attempt to refine the sample of firms based on information on participation in the DRD, recognizing that the failure to do so may work against me in finding strategic transfer-pricing behavior in response to the DRD.16 After applying the cleaning procedures outlined above, I keep the six quarters before and after passage of the Homeland Investment Act, leaving a sample covering the years 2003Q1 to 2005Q4. In an attempt to limit one-off observations that will not provide useful information, I also limit the sampletoonlyincludethosetransactionsforwhichthefirm-product-countryoccursmorethanfour times during this two-year period.17 Finally, the baseline results will use only those firms identified as U.S. (as opposed to Foreign) multinationals. 5.4 Results The difference-in-difference specification described below exploits variation in the transfer-price wedges both across time (pre and post passage of the HIA) and across tax jurisdictions. I operationalize this specification as follows: wedge = α Post +α Post ×LowTax +µ +ε (14) ficmt 1 t 2 t c fic ficmt The Post variable is an indicator variable equal to one for the six quarters following the t passage of the American Jobs Creation Act (2004Q3 to 2005Q4).18 The LowTax variable is also c anindicator, equaltooneforthosecountrieswithastatutorycorporatetaxratebelow25percent– 10 percentage points below that of the U.S. corporate rate.19 The µ fixed effects remove average fic 16AmethodusedbyBlouinandKrull(2009)andotherstoconstructaDRD-relevantsampleistoreviewCompustat firms for FSP 109-2 disclosures about the financial statement effects of the AJCA Act, as well as 10K forms. While promising, Census confidentiality and disclosure limitations may limit the feasibility of this approach for this study. 17Alternatively,Ialsotrylimitingthesamplebyremovinglargeoutliersbasedonswingsinunitvalues–removing unit value changes greater than 200 percent. The results are qualitatively the same. 18AccordingtoRedmiles(2008),86percentofcorporationsreportedtheDRDforTaxYear2005,while7.7percent reported it for Tax Year 2004. The remaining 6.8 percent reported it for Tax Year 2006. 19For a list of the potential trading partners that meet this threshold, see Appendix Table A1. 18

differentials that may or may not be affected by profit-shifting considerations in the steady state; thus, theα coefficientprovidestheadditional effectinducedbytheHIAinpotentialprofit-shifting 2 to low-tax countries. Bertrand, Duflo, and Mullainathan (2004) argue that the failure to account for serially correlated outcomes in the computation of standard errors can potentially invalidate many difference-indifference estimates. While the problem of serial correlation in this context is mitigated via the use of high-dimensional fixed effects and a relatively short time-series dimension (3 years), plausible scenarios exist that would require correction of standard errors. For the export wedge, the introduction of new products by a firm within a detailed HS-10 product category may lead to correlated values of the AL-RP wedge within a pre/post period. In addition, country-level pricing practices apart from tax policies may also introduce serially correlated errors. To account for these possibilities, I use two-dimensional clustering – by firm-product and country – utilizing the methodology described in Cameron, Gelbach, and Miller (2011), which also allows for high-dimensional fixed effects. On the import side one may additionally worry about the variation in firm-products in a country that compose the arms-length price matched to the intra-firm price; for this reason I cluster the errors by firm-product-country on the import side. The results using the export AL-RP wedge are shown in Table 3, with unweighted regression results in column (1) and weighted (by arms-length values) in column (2). As predicted by theory, the export AL-RP wedge expands in low-tax countries during the period following passage of the HIA. Relative to the baseline effects, the wedge increases by somewhere on the order of 0.08 and 0.54logpointsinlow-taxcountriesintheperiodfollowingtheHIA.Forfurthersupportiveevidence of firm response following the HIA, I look to the evidence provided by the import AL-RP wedges. Shown in Table 4, the import-based results show the opposite sign on the α coefficient, consistent 2 with the export-based results and theory outlined above. The import wedge decreases by 0.11 to 0.25 log points in low-tax countries on average in the six-quarters following the passage of the HIA, once again relative to baseline effects. Forausefulrobustnesstesttocheckthevalidityofthesefindings, Iturntothesampleofforeign multinationals. These firms are subject to a different system of taxation, and would likely not have found the DRD applicable. According to the survey evidence from Graham, Hanlon, and Shevlin (2010), for the DRD to be an attractive option a foreign parent would need to have a structure 19

such that non-U.S. subsidiaries were themselves organized as underneath a U.S. subsidiary. This structure is not common. however, due to the associated tax disadvantages. With this in mind I re-run the specification above for the sample of foreign multinationals. The results are shown in Table 5. Relative to the sample of U.S. multinationals, on the export side I find the opposite effect: thetransfer-pricewedgescorrespondingtolow-taxcountriesinthepost-periodarerelatively smaller.20 The coefficient for the import AL-RP wedge is also negative, but insignificant. These additional results support the notion that it is the effects of the HIA, rather than something else, that are driving the results captured by the specification in equation (14). I use the coefficients from Tables 3 and 4 to gain a sense of the aggregate effects of these results ontradeflowsandforegonetaxrevenue. Tocomputetheeffectsontrade, Imultiplythecoefficients by the overall amount of related-party trade to/from low-tax countries of the firms included in the baseline samples from Tables 3 and 4. To then calculate the implied foregone tax revenue, I apply the country-specific statutory tax rate differential with respect to the United States. The precise calculation of implied foregone tax revenue is described below: Foregone Tax Revenue (Exports) = (cid:88) (cid:88) exprp αEXP(τUS −τ ) (15) ij 2 i i∈Cj∈JE Foregone Tax Revenue (Imports) = (−1) (cid:88) (cid:88) imprp αIMP(τUS −τ ), (16) ij 2 i i∈Cj∈JI where C is the set of countries in Table A1 , and JE and JI are the set of firms in the samples underlying Tables 3 and 4 respectively. Although the weighted coefficient estimates are the most appropriate for this aggregation exercise, Table 6 also displays the results using the unweighted coefficients. Panel A of Table 6 demonstrates that net exports during the 2003Q3-2005Q4 period were under-reported by roughly $12.7 billion dollars, or roughly $4 billion dollars using the unweighted coefficients. Panel B shows that the tax loss from the HIA due to round-tripping was approximately $2 billion dollars. Although this is a relatively small share of overall corporate tax revenues, it is roughly 50 percent 20ItisdifficulttoaccountforthisresultforforeignmultinationalsfollowingpassageoftheHIA,butonepotential explanation could involve strategic responses to U.S. firm behavior. Recognizing the benefits to their competitors balancesheetsfollowingtheround-trippingbehaviordocumentedinTables3and4,foreignmultinationalsmayhave felt compelled to transfer some resources to their U.S. affiliates. Because the typical foreign multinational parent is headquartered in a country whose statutory tax rate is not that dissimilar to that of the U.S., the global tax implications of this transfer would be small. A more formal test of this explanation would involve documenting the industry composition of the foreign vs U.S. multinational transfer-pricing behaviors. 20

of what Joint Committee on Taxation estimated as the cost of the bill (3.5-3.9 billion USD, see Joint Committee on Taxation (2004)). Nevertheless, the values in Table 6 should be understood as back-of-the-envelope estimates, as there are a number of factors that could lead to a bias in either direction. On the one hand, the estimates based on equations (15) and (16) are using the top statutory corporate tax rate in the U.S., whereas the likely marginal tax rate to additional U.S. earnings may indeed be lower. On the other hand, I am likely under-estimating the amount of related-party trade as a basis for the foregone tax revenue as I am only including the trade corresponding to the firms in my sample. Finally, it is important to emphasize that these estimates are the result of the additional increase in profit-shifting incentives due to the HIA, and not the effect of transfer-pricing related profit-shifting more generally. 5.5 Discussion The estimates in this paper are consistent with a growing body of research that shows that profitshifting generally – and transfer-pricing behavior in particular – is prevalent among multinational firms. The concern with prior empirical results is typically that the measurement of transfer prices is an imperfect analogue to the CUP methods used by both firms and tax authorities, or, that the regressions of such measures of transfer prices with country-level tax rates suffer from omitted variables that bias the conclusions. The approach of the current paper addresses both of these concerns. First, matching intra-firm trade prices at the firm-product-country-month level with arms-length trade prices creates a conceptually accurate measure of the object under study. Second, applying these measures to a pseudo natural experiment of a policy change removes many potentially conflating factors. Anillustrationofthechallengesconfrontingtraditionalmeasurementofprofit-shiftingviatransfer price behavior comes from considering the difficulties in accounting for product heterogeneity. Even in the context of the AL-RP transfer price wedge calculated above, relying on a narrow matching strategy that includes firm, country, and HS-10 product categories, there is still scope for product-level differences – particularly in quality and other product attributes within an HS-10 code – to distort the interpretation of the AL-RP wedge as indicating strategic behavior. Product qualityisnotoriouslyhardtocapture, andsomedisaggregatedHScodesactuallyincludealargevariety of diverse products. Further, it is possible that such discrepancies of product attributes could 21

be correlated with country-level characteristics that align with corporate tax rates, etc. What one concludes from the results in Table 2, for example, could be subject to this concern. The benefits of the strategy used in this paper is that, to influence the results we obtain such discrepancies in the products within these narrow bins would need to also be changing systematically at a point in time that is also correlated with country-level tax rates. This additional layer of detail makes a spurious connection in these results unlikely. Onefairconcernwiththeinterpretationoftheresultsfromequation(14)iswhethertheperiods used for the pre- and post-HIA environments accurately capture the timing of a firms’ strategic response. If the HIA was anticipated by firms, it would have reduced the incentive to repatriate profits ahead of time for two reasons: 1) any current repatriation would be more costly than that under the holiday, and 2) because current dividends were used to calculate the firm-specific definition of “extraordinary” dividends that qualified for the deduction. The incentive to shift money abroad via transfer-pricing, however, would go in the opposite direction: firms would find it optimal to move money abroad to low-tax environments in anticipation of an impending, low-cost repatriation, making it more difficult to pick up the expected effect in the data. And, as indicated in section 4 the House vote in June 2004 implies that substantial uncertainty existed until at least June 2004, in line with the pre/post separation in equation (14). Moreover,itshouldbenotedthatthedataunderlyingtheseresultsdonotcontaintheintangible transactions (i.e. intellectual property, patents, and the like) that are often the subject of scrutiny fortaxauthorities. Becausetheseintangiblegoodsaremoredifficulttoprice,theyareoftenbelieved to be a major source of transfer pricing manipulation. As a result, by limiting the analysis to goods trade, the results may under-state the true extent of profit-shifting induced by the HIA. On the other hand, with a limited stock of such intangibles available to any given firm, prior motives for profit-shifting may have exhausted these one-off opportunities. Trade in goods may have been a convenient alternative. 6 Conclusion Usingauniquetransaction-leveldatasetcombinedwithaone-timepolicychange,thispaperreveals the transfer-pricing mechanisms underlying cross-country profit-shifting activities of U.S. multinational firms. I find that the gap between arms-length and related-party export prices – within 22

narrowly-defined transaction pairs – increases for low-tax countries in the period following the passage of a one-time dividend repatriation tax holiday; the comparable gap between arms-length and related-party import prices decreases for low-tax countries during this same period. Both of these results point to strategic profit-shifting activity by U.S. multinationals. By moving income outside of the United States with transfer pricing, profits could be declared in low-tax environments and then brought back under the terms of the tax repatriation holiday – a strategy popularly known as “round-tripping”. While such behavior is widely believed to exist, the unique microdata used in this paper allows for an empirical test that escapes the many pitfalls of prior efforts. Apart from highlighting the importance of tax avoidance behavior by multinational firms, the results of this paper demonstrate that intra-firm trade prices do not always correspond to the allocative market values that international trade economists typically model. The implications of this can be far-reaching. To give one example, the scope for trade to adjust to external factors in an economy depends on whether the initial trade allocations were influenced by market conditions, ratherthantheaccountingpracticesofmultinationalfirms. Thedifferentialbehaviorofarms-length and related party trade flows is an area in need of further research. 23

References Bastos, Paulo and Joana Silva. 2010. “The Quality of a Firm’s Exports: Where you Export to Matters.” Journal of International Economics 82 (2):99–111. Bernard, Andrew, J. Bradford Jensen, and Peter K. Schott. 2006. “Transfer Pricing by U.S.-Based Multinational Firms.” Working paper, Yale School of Management. Bertrand, Marianne, Esther Duflo, and Sendhil Mullainathan. 2004. “How Much Should We TrustDifferences-in-DifferencesEstimates?” Quarterly Journal of Economics 119(1):249– 275. Blouin, Jennifer and Linda Krull. 2009. “Bringing It Home: A Study of the Incentives Surrounding the Repatriation of Foreign Earnings Under the American Jobs Creation Act of 2004.” Journal of Accounting Research 47 (4):1027–1059. Bradley, Sebastian J. 2016. “Round-tripping of Domestic Profits under the American Jobs Creation Act of 2004.” Tech. rep., Working Paper. URL http://works.bepress.com/ sebastien_bradley/4/. Cameron, Colin, JonahGelbach, andDouglasMiller.2011. “RobustInferencewithMultiway Clustering.” Journal of Business and Economic Statistics 29 (2):238–249. Clausing, Kimberly A. 2005. “Tax Holidays (and Other Escapes) in the American Jobs Creation Act.” National Tax Journal 55 (3):331–346. Cristea, Anca D. and Dan Nguyen. 2016. “Transfer Pricing by Multinational Firms: Evidence from Foreign Firm Ownerships.” American Economic Journal: Economic Policy forthcoming. Davies, Ronald, Julien Martin, Mathieu Parenti, and Farid Toubal. 2017. “Knocking on Tax Haven’s Door: Multinational Firms and Transfer Pricing.” Review of Economics and Statistics forthcoming. Desai, Mihir A., C. Fritz Foley, and James R. Hines. 2006. “The Demand for Tax Haven Operations.” Journal of Public Economics 90 (3):513–531. Dharmapala, Dhammika, Fritz C. Foley, and Kristin J. Forbes. 2011. “Watch What I Do, Not What I Say: The Unintended Consequences of the Homeland Investment Act.” The Journal of Finance 66 (3):753–787. Drucker, Jesse. 2011. “Biggest Tax Avoiders Win Most Gaming $1 Trillion U.S. Tax Break.” :1. Faulkender, Michael and Mitchell Petersen. 2012. “Investment and Capital Constraints: Repatriations Under the American Jobs Creation Act.” Review of Financial Studies 25 (11):3351–3388. Flaaen, Aaron. 2013. “Constructing a Bridge Between the LexisNexis Directory of Corporate Affiliations and the U.S. Business Register.” Working paper, University of Michigan. 24

———. 2014. “Multinational Firms in Context.” Working paper, University of Michigan. Graham, John R., Michelle Hanlon, and Terry Shevlin. 2010. “Barriers to Mobility: The Lockout Effect of U.S. Taxation of Worldwide Corporate Profits.” National Tax Journal 63 (4):1111–1144. Hartman, David G. 1985. “Tax Policy and Foreign Direct Investment.” Journal of Public Economics 26 (1):107–121. Heckemeyer, Jost H. and Michael Overesch. 2013. “Multinationals’ Profit Response to Tax Differentials: Effect Size and Shifting Channels.” Working Paper 13-045, ZEW. International Monetary Fund. 2013. “Fiscal Monitor: Taxing Times.” Tech. rep., International Monetary Fund: World Economic and Financial Surveys, Washington D.C. Jarmin, Ron and Javier Miranda. 2002. “The Longitudinal Business Database.” Mimeo, available at http://www.vrdc.cornell.edu/info7470/2007/Readings/jarmin-miranda- 2002.pdf. Joint Committee on Taxation. 2004. “Comparison of the Estimated Budget Effects of H.R. 4520, the “American Jobs Creation Act of 2004,” as Passed by the House of Representatives, and H.R. 4520, the “Jumpstart Our Business Strength Act,” as Amended by the Senate.” Tech. Rep. JCX-53-2004. URL https://www.jct.gov/publications.html? func=startdown&id=1634. J.P Morgan Chase. 2003. “Introducing the Homeland Investment Act.” Tech. rep., J.P. Morgan Chase Banks: Economic and Policy Research. Kamal, Fariha, C.J. Krizan, and Ryan Monarch. 2015. “Identifying Foreign Suppliers in U.S. Merchandise Import Transactions.” Working Paper CES-EP-15-11, U.S. Census Bureau Center for Economic Studies. Lochhead, Carolyn. 2003. “Lawmakers Push Tax Break for Businesses; Passage Seen as Possible Because of WTO Ruling.” San Francisco Chronicle July 26, 2003 (B1). Lohse, Theresa, Nadine Riedel, and Christoph Spengel. 2012. “The Increasing Importance of Transfer Pricing Regulations – a Worldwide Overview.” Working Paper 12/27, Oxford University Centre for Business Taxation. Loretz, Simon.2013. “CorporateTaxationintheOECDinaWiderContext.” Oxford Review of Economic Policy 24 (4):639–660. Manova, Kalina and Zhiwei Zhang. 2012. “Export Prices Across Firms and Destinations.” Quarterly Journal of Economics 127 (1):379–436. Monarch, Ryan. 2014. “It’s Not You, It’s Me: Breakups in U.S.-China Trade Relationships.” Working Paper CES-EP-14-08, U.S. Census Bureau Center for Economic Studies. Neiman, Brent. 2010. “Stickiness, Synchronization, and Passthrough in Intrafirm Trade Prices.” Journal of Monetary Economics 57:295–308. 25

OECD. 2013. “Action Plan on Base Erosion and Profit Shifting.” Tech. rep., OECD Publishing, Paris. URL http://dx.doi.org/10.1787/9789264202719-en. Pulizzi, Henry J. 2003. “Proposed US Tax Break Could Add Supply to Eurobond Mkt.” Dow Jones International News July 31, 2003. Rauch, James E. 1999. “Networks Versus Markets in International Trade.” Journal of International Economics 48:7–35. Redmiles,Melissa.2008. “TheOne-TimeReceivedDividendReduction.” Statistics of Income Bulletin Spring:102–114. Wasi, Nada and Aaron Flaaen. 2014. “Record Linkage Using Stata: Pre-Processing, Linking and Reviewing Utilities.” The Stata Journal 15 (3):672–697. Zucman, Gabriel. 2014. “Taxing across Borders: Tracking Personal Wealth and Corporate Profits.” Journal of Economic Perspectives 28 (4):121–148. 26

Table1: RepatriatedDividendsUndertheHomelandInvestmentAct,TaxYear2004-2006,Selected Countries of Incorporation Cash Dividends Controlled Foreign Corporations Country of Amount Percent Percent Incorporation (millions USD) of Total Amount of Total Netherlands 94,415 26.1 253 5.9 Switzerland 35,783 9.9 155 3.7 Bermuda 34,974 9.7 82 1.9 Ireland 27,588 7.6 112 2.6 Canada 25,541 7.1 426 10.0 Luxembourg 25,439 7.0 87 2.0 United Kingdom 22,264 6.2 330 7.8 Cayman Islands 19,894 5.5 101 2.4 Hong Kong 5,511 1.5 163 3.8 Singapore 5,518 1.5 89 2.1 All Other Countries 64,940 17.9 2,448 57.8 Total 361,866 100 4,246 100 Source: IRS Form 8895, and Redmiles (2008) Table 2: Transfer Price Wedges and Tax Rates: 1993-2000 Export AL-RP Wedge1 Import AL-RP Wedge2 (1) (2) (3) (4) Tax Rate (WTD) −4.178∗∗∗ −1.638∗∗∗ −0.0923 0.600∗∗∗ (0.665) (0.447) (0.278) (0.226) Fixed Effects No Product No Product R-Squared 0.00 0.15 0.00 0.071 Observations 3,585,777 3,585,777 12,431,800 12,431,800 1Source: Bernard, Jensen, and Schott (2006) 2Source: This Paper 27

Table 3: Difference-in-Difference Analysis: Export Transfer-Price Wedges Export AL-RP Wedge VARIABLES (1) (2) Post-Period X Low-Tax 0.080** 0.542*** (0.040) (0.099) Post-Period -0.028 -0.236* (0.026) (0.142) Constant 0.222*** 0.723*** (0.013) (0.074) Observations 149,100 149,100 R-squared 0.426 0.509 Fixed Effects Firm-Country-Product Firm-Country-Product Weighted No Yes Firms (Rounded) 1200 1200 The monthly sample period ranges from 2003Q1 to 2005Q4. The Post-Period is defined asthesixquartersfollowingtheresolveduncertaintyoftheHIA,2004Q3to2005Q4. The list of low-tax countries is given in Table A1. Standard errors clustered by firm-product andcountryfollowingthemethodologyofCameron, Gelbach, andMiller(2011). Column (2) weights the estimates using the arms-length export value. *** p<0.01, ** p<0.05, * p<0.1 28

Table 4: Difference-in-Difference Analysis: Import Transfer-Price Wedges Import AL-RP Wedge VARIABLES (1) (2) Post-Period X Low-Tax -0.110** -0.235* (0.048) (0.131) Post-Period 0.0037 0.0462* (0.011) (0.028) Constant 0.106*** 0.0327** (0.0059) (0.016) Observations 219,700 219,700 R-squared 0.300 0.415 Fixed Effects Firm-Country-Product Firm-Country-Product Weighted No Yes Firms (Rounded) 1000 1000 The monthly sample period ranges from 2003Q1 to 2005Q4. The Post-Period is defined asthesixquartersfollowingtheresolveduncertaintyoftheHIA,2004Q3to2005Q4. The list of low-tax countries is given in Table A1. Standard errors clustered by firm-productcountry. Column (2) weights the estimates using the arms-length export value. *** p<0.01, ** p<0.05, * p<0.1 29

slanoitanitluM ngieroF rof sisylanA ecnereffiD-ni-ecnereffiD :ssentsuboR :5 elbaT egdeW PR-LA tropmI goL egdeW PR-LA tropxE goL )4( )3( )2( )1( 811.0- 5830.0- ***432.1- **262.0xaT-woL X doireP-tsoP )641.0( )540.0( )604.0( )221.0( *7170.0- *4020.0- 9470.0 6350.0doireP-tsoP )830.0( )210.0( )480.0( )260.0( 7010.0- ***411.0- ***143.0 ***922.0 tnatsnoC )220.0( )600.0( )740.0( )030.0( 002,021 002,021 009,94 009,94 snoitavresbO 534.0 863.0 655.0 183.0 derauqs-R tcudorP-yrtnuoC-mriF tcudorP-yrtnuoC-mriF tcudorP-yrtnuoC-mriF tcudorP-yrtnuoC-mriF stceffE dexiF seY oN seY oN dethgieW 0041 0041 008 008 )dednuoR( smriF ,AIHehtfoytniatrecnudevloserehtgniwollofsretrauqxisehtsadenfiedsidoireP-tsoPehT .4Q5002ot1Q3002morfsegnardoirepelpmasylhtnomehT yrtnuocdnatcudorp-mrfiybderetsulcerasrorredradnats)2(dna)1(snmulocnI .1AelbaTninevigsiseirtnuocxat-wolfotsilehT .4Q5002ot3Q4002 dna)2(snmuloC .yrtnuoc-tcudorp-mrfiybderetsulcerayeht)4(dna)3(snmulocni;)1102(relliMdna,hcableG,noremaCfoygolodohtemehtgniwollof .eulav tropxe htgnel-smra eht gnisu setamitse eht thgiew )4( 1.0<p * ,50.0<p ** ,10.0<p *** 30

Table 6: Implied Effects on Trade and Taxes (millions USD) Weighted Unweighted (1) (2) Panel A: Trade Impacts Exports (Under-reported) 5,932.4 869.9 Imports (Over-reported) 6,777.6 3,716.1 Net Exports 12,710.0 4,046.0 Panel B: Implied Foregone Tax Revenue from Exports 951.1 139.5 from Imports 1,204.1 564.3 Total 2,155.1 703.7 Panel A applies the coefficients from tables 3 and 4 to adjust the related-party value of exports to and imports from low-tax countries during the post-period of 2003Q3 to 2005Q4. Panel B then multiplies the country-specific related-party trade by the difference in the statutory tax rate with respect to the U.S. rate. See equations (15) and (16) in the text. 31

Figure 1: Net Quarterly Dividends from Abroad: 1989Q1 - 2007Q4 Source: Board of Governors of the Federal Reserve, Financial Accounts of the United States (Z1). 32

A Data Appendix A.1 Matching Corporate Directories to the Business Register The discussion below is an abbreviated form of the full technical note (see Flaaen (2013)) documenting the bridge between the DCA and the Business Register. A.1.1 Directories of International Corporate Structure The LexisNexis Directory of Corporate Affiliations (DCA) is the primary source of information on the ownership and locations of U.S. and foreign affiliates. The DCA describes the organization and hierarchy of public and private firms, and consists of three separate databases: U.S.PublicCompanies, U.S.PrivateCompanies, andInternational–thoseparent companies with headquarters located outside the United States. The U.S. Public database contains all firms traded on the major U.S. exchanges, as well as major firms traded on smaller U.S. exchanges. To be included in the U.S. Private database, a firm must demonstrate revenues in excess of $1 million, 300 or more employees, or substantial assets. Those firms included in the International database, which include both public and private companies, generally have revenues greater than $10 million. Each database contains information on all parent company subsidiaries, regardless of the location of the subsidiary in relation to the parent. The second source used to identify multinational firms comes from Uniworld Business Publications (UBP). This company has produced periodic volumes documenting the locations and international scope of i) American firms operating in foreign countries; and ii) foreign firms with operations in the United States. Although only published biennially, these directories benefit from a focus on multinational firms, and from no sales threshold for inclusion. Because there exist no common identifiers between these directories and Census Bureau data infrastructure, we rely on probabilistic name and address matching — so-called “fuzzy merging” — to link the directories to the Census data infrastructure. A.1.2 Background on Name and Address Matching Matchingtwodatarecordsbasedonnameandaddressinformationisnecessarilyanimperfect exercise. Issues such as abbreviations, misspellings, alternate spellings, and alternate name conventions rule out an exact merging procedure, leaving the researcher with probabilistic stringmatchingalgorithmsthatevaluatethe“closeness”ofmatch—givenbyascoreorrank — between the two character strings in question. Due to the large computing requirements of these algorithms, it is common to use so-called “blocker” variables to restrict the search samples within each dataset. A “blocker” variable must match exactly, and as a result this implies the need for a high degree of conformity between these variables in the two datasets. In the context of name and address matching, the most common “blocker” variables are the state and city of the establishment. The matching procedure uses a set of record linking utilities described in Wasi and Flaaen (2014). This program uses a bigram string comparator algorithm on multiple variables with 33

differing user-specified weights.21 This way the researcher can apply, for example, a larger weight on a near name match than on a perfect zip code match. Hence, the “match score” for this program can be interpreted as a weighted average of each variable’s percentage of bigram character matches. A.1.3 The Unit of Matching The primary unit of observation in the DCA, UBP, and BR datasets is the business establishment. Hence, the primary unit of matching is the establishment, and not the firm. However, thereareanumberofimportantchallengeswithanestablishment-to-establishment link. First, the DCA (UBP) and BR may occasionally have differing definitions of the establishment. One dataset may separate out several operating groups within the same firm address (i.e. JP Morgan – Derivatives, and JP Morgan - Emerging Markets), while another may group these activities together by their common address. Second, the name associated with a particular establishment can at times reflect the subsidiary name, location, or activity (i.e. Alabama plant, processing division, etc), and at times reflect the parent company name. Recognizing these challenges, the primary goal of the matching will be to assign each DCA (UBP) establishment to the most appropriate business location of the parent firm identified in the BR. As such, the primary matching variables will be the establishment name, along with geographic indicators of street, city, zip code, and state. A.1.4 The Matching Process: An Overview The danger associated with probabilistic name and address procedures is the potential for false-positive matches. Thus, there is an inherent tension for the researcher between a broad search criteria that seeks to maximize the number of true matches and a narrow and exacting criteria that eliminates false-positive matches. The matching approach used here is conservative in the sense that the methodology will favor criteria that limit the potential for false positives at the potential expense of slightly higher match rates. As such, the procedure generally requires a match score exceeding 95 percent, except in those cases where ancillary evidence provides increased confidence in the match.22 This matching proceeds in an iterative fashion, in which a series of matching procedures are applied with decreasingly restrictive sets of matching requirements. In other words, the initial matching attempt uses the most stringent standards possible, after which the non-matching records proceed to a further matching iteration, often with less stringent standards. In each iteration, the matching records are assigned a flag that indicates the standard associated with the match. See Table A2 for a summary of the establishment-level match rate statistics by year and type of firm. Table A3 lists the corresponding information for the Uniworld data. 21The term bigram refers to two consecutive characters within a string (the word bigram contains 5 possible bigrams: “bi”, “ig”, “gr”, “ra”, and “am”). The program is a modified version of an existing string comparator algorithmbyMichaelBlasnik,andassignsascoreforeachvariablebetweenthetwodatasetsbasedonthepercentage of matching bigrams. See Flaaen (2013) or ? for more information. 22Theprimarysourcesofsuchancillaryevidenceareclericalreviewofthematches,andadditionalparentidentifier matching evidence. 34

A.1.5 Construction of Multinational Indicators The DCA data allows for the construction of variables indicating the multinational status of the U.S.-based establishment. If the parent firm contains addresses outside of the United States, but is headquartered within the U.S., we designate this establishment as part of a U.S. multinational firm. If the parent firm is headquartered outside of the United States, we designate this establishment as part of a Foreign multinational firm. We also retain the nationality of parent firm.23 There can be a number of issues when translating the DCA-based indicators through the DCA-BR bridge for use within the Census Bureau data architecture. First, there may be disagreements between the DCA and Census on what constitutes a firm, such that an establishment matches may report differing multinational indicators for the same Censusidentifiedfirm. Second, suchanissuemightalsoariseduetojoint-ventures. Finally, incorrect matches may also affect the degree to which establishment matches agree when aggregated to a firm definition. To address these issues, we apply the following rules when using the DCA-basedmultinationalindicatorsandaggregatingtothe(Census-based)firmlevel. There are three potential cases:24 Potential 1: A Census-identified firm in which two or more establishments match to different foreign-country parent firms 1. CollapsetheCensus-identifiedfirmemploymentbasedontheestablishment-parentfirm link by country of foreign ownership 2. Calculate the firm employment share of each establishment match 3. If one particular link of country of foreign ownership yields an employment share above 0.75, apply that link to all establishments within the firm. 4. If one particular link of country of foreign ownership yields an employment share above 0.5andtotalfirmemploymentisbelow10,000,thenapplythatlinktoallestablishments within the firm. 5. All other cases require manual review. Potential 2: A Census-identified firm in which one establishment is matched to a foreigncountry parent firm, and another establishment is matched to a U.S. multinational firm. 1. CollapsetheCensus-identifiedfirmemploymentbasedontheestablishment-parentfirm link by type of DCA link (Foreign vs U.S. Multinational) 2. Calculate the firm employment share of each establishment match 3. If one particular type of link yields an employment share above 0.75, apply that link to all establishments within the firm. 23The multinational status of firms from the UBP directories are more straightforward. 24Some of these cases also apply to the UBP-BR bridge. 35

4. If one particular type of link yields an employment share above 0.5 and total firm employment is below 10,000, then apply that link to all establishments within the firm. 5. All other cases require manual review. Potential 3: A Census-identified firm in which one establishment is matched to a nonmultinational firm, and another establishment is matched to a foreign-country parent firm (or U.S. multinational firm). Apply same steps as in Potential 2. Table A1: List of Countries with Average Statutory Corporate Tax Rates Below 25 Percent: 2004- 2006 Average Corporate Average Corporate Country Tax Rate Country Tax Rate Albania 23.3 Isle of Man 6.6 Armenia 20 Latvia 15 Azerbaijan 24.5 Lebanon 15 Bahamas, The 0 Liechtenstein 15 Bahrain 0 Lithuania 15 Belarus 24 Macao, China 14 Bermuda 0 Macedonia, FYR 15 Bosnia and Herzegovina 10 Maldives 0 British Virgin Islands 15 Moldova 19 Bulgaria 16.5 Montenegro 9 Cambodia 20 Oman 24 Cayman Islands 0 Paraguay 20 Channel Islands 13.3 Poland 19 Chile 17 Romania 19 Croatia 20 Russian Federation 24 Cyprus 11.6 Saudi Arabia 23.3 Estonia 24.3 Serbia 10 Georgia 20 Singapore 20.6 Guernsey 0 Slovak Republic 19 Hong Kong, China 17.5 Switzerland 22.7 Hungary 16 Uzbekistan 19 Iceland 18 Vanuatu 0 Ireland 12.6 Yugoslavia 20 Source: Office of Tax Policy at the University of Michigan, Ernst & Young, KPMG, and Loretz (2013). Note: Inclusion in this list does not signify that a matching AL-RP transaction pair exists in the baseline dataset used for analysis. 36

Table A2: DCA Match Statistics: 2003-2007 # of DCA Matched Percent Establishments to B.R. Matched Total 2003 123,553 86,838 0.70 2004 117,639 84,450 0.72 2005 110,106 80,245 0.73 2006 110,826 79,275 0.72 2007 112,346 81,656 0.73 U.S. Multinationals 2003 25,905 17,465 0.67 2004 24,028 16,923 0.70 2005 20,870 15,191 0.73 2006 21,335 15,539 0.73 2007 22,500 16,396 0.73 Foreign Multinationals 2003 11,101 7,398 0.66 2004 10,152 7,156 0.70 2005 9,409 6,865 0.73 2006 9,981 7,243 0.73 2007 10,331 7,555 0.73 Table A3: Uniworld Match Statistics: 2003-2007 # of Uniworld Matched Percent Establishments to B.R. Matched Foreign Multinationals 2004 3,220 2,347 0.73 2006 3,495 2,590 0.74 U.S. Multinationals1 2003 3,001 2,403 0.80 2005 2,951 2,489 0.84 2007 4,043 3,236 0.80 1U.S.multinationalsincludeonlytheestablishmentidentifiedastheU.S. headquarters. 37