International Transfer Pricing and Tax Avoidance: Evidence from Linked Trade-Tax Statistics in the UK

K.7 International Transfer Pricing and Tax Avoidance: Evidence from Linked Trade-Tax Statistics in the UK Liu, Li, Tim Schmidt-Eisenlohr, and Dongxian Guo Please cite paper as: Liu, Li, Tim Schmidt-Eisenlohr, and Dongxian Guo (2017). International Transfer Pricing and Tax Avoidance: Evidence from Linked Trade-Tax Statistics in the UK. International Finance Discussion Papers 1214. https://doi.org/10.17016/IFDP.2017.1214 International Finance Discussion Papers Board of Governors of the Federal Reserve System Number 1214 October 2017

Board of Governors of the Federal Reserve System International Finance Discussion Papers Number 1214 October 2017 International Transfer Pricing and Tax Avoidance: Evidence from Linked Trade-Tax Statistics in the UK Li Liu, Tim Schmidt-Eisenlohr, and Dongxian Guo NOTE: International Finance Discussion Papers are preliminary materials circulated to stimulate discussion and critical comment. References in publications to International Finance Discussion Papers (other than an acknowledgment that the writer has had access to unpublished material) should be cleared with the author or authors. Recent IFDPs are available on the Web at https://www.federalreserve.gov/econres/ifdp/. This paper can be downloaded without charge from Social Science Research Network electronic library at http://www.sssrn.com.

International Transfer Pricing and Tax Avoidance: Evidence from Linked Trade-Tax Statistics in the UK∗ Li Liu, Tim Schmidt-Eisenlohr, and Dongxian Guo† July 2017 Abstract This paper employs unique data on export transactions and corporate tax returns of UK multinational firms and finds that firms manipulate their transfer prices to shift profits to lower-taxed destinations. It uncovers three new findings on tax-motivated transfermispricinginrealgoods. First,transfermispricingincreasessubstantiallywhen taxationofforeignprofitschangesfromaworldwidetoaterritorialapproachintheUK, with multinationals shifting more profits into low-tax jurisdictions. Second, transfer mispricingincreaseswithafirm’sR&Dintensity. Third,tax-motivatedtransfermispricing is concentrated in countries that are not tax havens and have low-to-medium-level corporate tax rates. Keywords: transfer pricing, corporate taxation avoidance, multinational firms JEL Classification: F23, H25, H32 ∗We thank the staff at Her Majesty’s Revenue & Customs’ (HMRC) Datalab for access to the data and their support of this project. This work contains statistical data from HMRC, which is under Crown copyright. The research data sets used may not exactly reproduce HMRC aggregates. The use of HMRC statistical data in this work does not imply the endorsement of HMRC in relation to the interpretation or analysis of the information. All results have been screened by HMRC to ensure confidentiality is not breached. The views expressed are the authors’ and do not necessarily represent the views of the IMF, its Executive Board, or IMF management, nor should the views be interpreted as representing the views of the Federal Open Market Committee, its principals, or the Board of Governors of the Federal Reserve System. †Liu: International Monetary Fund(lliu@imf.org). Schmidt-Eisenlohr: Federal Reserve Board of Governors (t.schmidteisenlohr@gmail.com). Guo: London School of Economics. We would like to thank Steve Bond, RonDavies, MichaelDevereux, DhammikaDharmapala, JimHines, NielsJohannesen, JulienMartin, Joel Slemrod, Johannes Voget, and seminar participants at the IMF, Federal Reserve Board, University of Michigan, University of Oxford, CESifo Global Economy Conference, and Southern Economic Association 2016 Conference for helpful comments. All mistakes are our own. This project was initiated when both Li andTimwereresearchfellowsatOxfordUniversityCentreforBusinessTaxation. Weacknowledgefinancial support from the ESRC under grant ES/L000016/1.

1 Introduction Globalization has led to the concentration of economic activity within a small number of multinational corporations (MNCs). This development has made it more challenging for governments to raise revenue from corporate income tax, as MNCs can shift their profits across borders to reduce their tax bills.1 In recent years, policy-makers have become increasingly concerned about this issue as the extent of profit shifting has intensified.(Zucman, 2014). A key instrument that MNCs use to shift profits is manipulating transfer prices they charge on transactions between related parties within the MNC group (transfer mispricing). For example, to reduce its pre-tax profits (and hence corporate taxes), an MNC can charge a lower price when selling to a related party in a low-tax country.2 Tax-motivated transfer mispricing can take place in trade in real goods as well as in services, and in particular in the form of royalty and licensing payments on intellectual property rights held abroad.3 This paper presents new evidence on tax-motivated transfer mispricing in real goods by exploiting a unique data set that combines the tax records of UK MNCs in manufacturing andtheirinternationaltradetransactionsovertheyearsof2005-2011. Theempiricalstrategy identifies the causal effect of the corporate income tax differential between the destination country and the UK on the unit price of exports by UK MNCs. It controls for omitted variable bias by including a full set of firm–market–product fixed effects, product–market–year fixed effects, and firm–product–year fixed effects in a triple-difference regression. Identifying variation comes from the differential change in the price charged by a multinational with a foreign subsidiary relative to the price charged by a MNC without a subsidiary in the same country in response to a change in the tax rate difference between the destination country and the UK. We find strong evidence for tax-motivated transfer mispricing in manufacturing exports to low-tax destinations. A one percentage point increase in the UK-destination country tax differential reduces related-party export prices relative to arm’s-length export prices by three percent. The extent of tax-motivated transfer mispricing is larger following the 1See, among others, Harris et al. (1993), Hines and Rice (1994), and Desai et al. (2006) for evidence of general profit shifting by MNCs to low-tax countries, including tax havens. 2Conversely, it can pay a higher price when buying from a related party in a low-tax country. 3See, for example, evidence presented in Dischinger and Riedel (2011). 1

UK’s change from a worldwide to a territorial tax system.4 After the tax reform, a one percentage point increase in the tax difference reduces related-party export prices relative to arm’s-length export prices by another 1.5 percent. Transfer mispricing increases with firm R&D intensity, with the marginal effect of the tax differential rising to 6.4 percent for those undertaking most R&D. Finally, increases (decreases) in transfer mispricing associated with widening (narrowing) tax differences are concentrated in countries that are not classified as tax havens and have low to intermediate tax rates. Thesebenchmarkfindingsareinlinewiththelargeeffectsoftaxsavingsontransferprices estimated in Clausing (2003), Bernard et al. (2006), and Flaaen (2016) but are substantially larger than the effects found in Vicard (2015) and Cristea and Nguyen (2016). We show that some of the differences can be attributed to omitted variable bias, as previous studies used smaller sets of fixed effects in the empirical analysis. Our paper adds to the literature in four distinct ways. First, we show in a simple model that a shift from a worldwide to a territorial system in taxing foreign profits of MNCs leads to stronger transfer mispricing, and we provide empirical evidence that corroborates the theoretical prediction. Second, the results establish that transfer mispricing is concentrated in the most R&D-intensive firms. This finding is robust to controlling for differential effects by firm size and the type of product traded and suggests that R&D investment facilitates transfer mispricing by making goods more specific. Third, in perhaps surprising contrast with research on France by Davies et al. (forthcoming), our evidence indicates that mispricing in tangible goods by UK MNCs is concentrated in non-tax-haven countries.5 Fourth, thanks to the rich data and the relatively large number of MNCs headquartered in the UK, the regressions in this paper control for more fixed effects than previous studies, allowing for a clean identification of tax-motivated transfer mispricing.6 In addition, the 2009 UK tax reform in 4This reform fundamentally changed the taxation of foreign earnings from a worldwide regime to a territorial regime in the UK. Under the worldwide regime, foreign earnings of UK MNCs were liable to additional UK taxes when repatriating from countries with a lower statutory tax rate than the UK. In contrast,theterritorialregimeexemptsforeignearningsofUKMNCsfromUKtaxesaltogether. Wediscuss the reform in more detail in section 2. 5The fact that we do not find evidence for mispricing of exports to tax haven countries may in parts be driven by our identification strategy which relies on tax changes over time. However, the robust finding of substantial transfer mispricing to non-haven countries is still striking. We thank Niels Johannesen for this great comment. 6Bernard et al. (2006) and Flaaen (2016) use U.S. Census data that allow them to directly look at the pricewedgewithinamultinationalfirm,implicitlytakingoutmuchofthetime-invariantvariation. However, our country–product–year and firm–product–year fixed effects still go beyond their specifications in terms of controlling for time-variant factors. Further details are discussed in section 5. 2

the taxation of cross-border corporate income provides us with a quasi-natural experiment that introduced exogenous changes in the tax incentives of profit shifting, corroborating the causal effect of taxes on transfer prices. Our findings have several implications for tax policy design. First, they show that there is substantial profit shifting through transfer mispricing in tangible goods by UK MNCs. This represents an area of revenue leakage that warrants further attention of the UK tax authority. While the quantitative evidence is UK-specific, the empirical analysis can be extended to other countries with the suitable data in order to help uncovering the extent of tax-motivated transfer mispricing elsewhere. Second, tax-motivated transfer mispricing is not uniform across firms, but concentrated in the most R&D-intensive ones. This pattern provides useful guidance for tax authorities. Third, we find that the shift to the territorial system increased the extent of tax-motivated transfer mispricing, highlighting a relevant revenue cost of moving away from the worldwide system. Finally, the finding that transfer mispricing is concentrated in non-haven countries with low to medium tax rates suggests that policy-makers should be mindful of potential revenue loss not only to tax havens but also to other trading partners that have low statutory corporate income tax rates. Literature Several papers have looked at transfer-pricing behavior of multinational firms. Early literature, including Grubert and Mutti (1991), Harris et al. (1993), Hines and Rice (1994), and Collins et al. (1998), provided indirect evidence for profit shifting, showing that pre-tax profits are systematically correlated with tax differentials across countries. Clausing (2003) was the first to provide direct evidence on manipulated prices, using U.S. industry-level data. In another seminal paper, Bernard et al. (2006) employed transactionlevel data from the U.S. Census to study a wide set of factors that can lead to manipulated transfer prices, including corporate taxes and tariffs. More recently, Flaaen (2016) employs the same data, studying transfer-price manipulation by U.S. multinationals in the context of the 2004 Home Investment Act. Closely related to our work are three papers that also use detailed trade data to study transfer-price manipulations for a set of different countries. Davies et al. (forthcoming) and Vicard (2015) exploit information on French firms, whereas Cristea and Nguyen (2016) employ Danish data. Finally, Hebous and Johannesen (2015) analyze firm-level trade data on German MNCs, providing evidence that they shift profits to tax havens through services 3

trade. The remainder of the paper is structured as follows. Section 2 provides background on transfer pricing and the 2009 tax reform. Section 3 presents a simple framework for illustrating the effect of tax differential on transfer pricing. Section 4 describes the data, Section 5 explains the empirical approach, and Section 6 presents the empirical results. Section 7 concludes with a discussion of policy implications and avenues for future research. 2 Institutional Background This section provides an overview of transfer pricing, explaining the arm’s length principle that generally guides the setting of transfer prices and several weaknesses of this approach. It then discusses the 2009 tax reform that changed the UK’s taxation of foreign profits from a worldwide to a territorial regime. 2.1 Transfer Pricing Transferpricingisthesettingofpricesforinternal(intra-firm)transactionsingoods,services, intangibles, andcapitalflowswithinanMNC.Transferpricingaffectsthepre-taxprofitsthat each party earns from a cross-border transaction and the amount of corporation tax that is due in both countries. Consider a UK pharmaceutical group that buys raw material from a subsidiary in China. How much the UK parent pays its Chinese subsidiary for each unit of the raw material — the transfer price — affects how much profit the Chinese affiliate earns and how much local tax it pays, and the amount of profit and corporation tax faced by the UK parent company. Most tax authorities, including the HMRC in the UK, use the arm’s length principle to guide transfer pricing.7 The arm’s length principle requires that a transfer price should be the same as if the two parties involved were two independent companies, that is, the same as a comparable market transaction.8 Given the nature of related-party transactions, a range 7Thearm’slengthprincipleisestablishedinArticle9oftheOrganizationforEconomicCo-operationand Development (OECD) and the UN Model Tax Conventions and is the framework for the extensive network of bilateral income tax treaties between OECD countries, and many non-OECD governments. 8Setting the transfer price involves the search for a comparable transaction and the application of an appropriatetransfer-pricingmethod. Twotransactionsareregardedascomparablewheneitherthereareno materialdifferencesbetweenthemorreasonableaccurateadjustmentscanbemadetoeliminatetheeffectof such differences (OECD, 2011). The OECD Guidelines also provide a set of criteria to assess comparability 4

of arm’s length prices may arise for the same transaction. This leeway is due to challenges in the application of the arm’s length principle. For example, a comparable transaction may not exist between independent parties. Comparable transactions may also be costly to observe for the tax authority. If comparable arm’s-length prices are not accessible, they may be difficult to infer. As a consequence, there is room for manipulation, through under- or over-pricing, to exploit tax differentials across jurisdictions in order to lower the overall tax burden of the multinational group. Many countries implement transfer-pricing regulations as a countermeasure to mitigate transfer mispricing. The tightness of these regulations varies from mere acknowledgement of the arm’s length principle to requirement of detailed transfer-pricing reports. Rigid regulations increase the cost of transfer mispricing and are found to be somewhat effective in curbing the extent of profit shifting in developed countries.9 In the UK, transfer-pricing documentation requirements are a part of the domestic tax law, specifying that documentation must be available upon request or submitted with the company’s annual tax return.10 2.2 The 2009 Territorial Tax Reform The tax treatment of foreign profits in the home country is a key consideration for MNCs when setting their transfer prices, as taxation of foreign earnings at home affects the total profit tax bill. Basicallyall countries fall intoone of two predominant categories: ‘worldwide’ or ‘territorial’. Under ‘worldwide’ taxation, an MNC pays taxes on its domestic and foreign income, though a credit is often given for foreign taxes paid.11 Under territorial taxation, an MNC only pays taxes on profits in the source country and faces no additional tax when bringing profits home.12 So, £1 profit earned or reported in a low-tax country j would imply betweencontrolledanduncontrolledtransactionsintermsofcharacteristicsofproductsorservices,functions performed by each party taking account of the assets used and risks assumed, contractual terms, economic circumstances, and business strategies. 9Forexample,Riedeletal.(2015)showthattheintroductionandtighteningoftransferpricingrulesraises (lowers)reportedoperatingprofitsofhigh-tax(low-tax)affiliatesandreducesthesensitivityofaffiliates’pretax profits to corporate tax rate changes. 10Unlike other OECD countries, the UK does not have a prescribed list of documentation requirements, and detailed disclosure are not currently required within tax returns. 11In the case of the UK, there was a restrictive-form onshore-pooling regime for excess credits, allowing ”eligible unrelieved foreign tax” on high-taxed dividends to be credited against the UK tax arising on the pooled low-taxed dividends. 12This statement only applies to active income of foreign affiliates, which is essentially income earned through business activity. Passive income is income that derives from activities in which the recipient is not directly involved, such as investment income or royalty income. Unlike with the treatment of active income, 5

£(1−τ ) after-tax profit, where τ is the statutory tax rate in the host country. j j To capture the aspect that profits located in the low-tax country eventually require costly repatriation, assume that profits located abroad are less valuable to an MNC by a factor of γ < 1. Then, under the worldwide system, a £1 profit earned or reported in a low-tax country j would be worth £γ(1−τ ) to the MNC. j The difference in the taxation of foreign profits under the two systems implies larger tax savings for MNCs from shifting profit to lower-taxed jurisdictions under territorial taxation. Specifically, the increase in the additional savings when moving from a worldwide to a territorial system are £(1 − γ)(1 − τ ) for each pound of profit reported in the low-tax j country. Assuming that foreign profits accumulated under the worldwide system ultimately returnhome, thereishenceastrongerincentiveforprofitshiftingundertheterritorialsystem than under the worldwide system. In 2009, the UK switched from worldwide to territorial taxation, which, given our discussion, is expected to increase the extent of transfer mispricing by UK MNCs. Note that it is plausible that before 2009, part of the foreign earnings were already brought back to the UK in some other complicated, non-taxable way.13 To the extent that such activities are costly, it is still the case that the amount of tax savings from profit shifting net of costs is larger under the territorial tax system.14 Finally, a great feature of the tax reform is that its exact announcement and implementation dates were not known in advance. We can therefore exploit variation in the tax incentives to shift profits generated by the reform to study transfer mispricing in a quasi-experimental setting.15 3 Conceptual Framework This section develops a simple model that shows how tax differences motivate MNCs to manipulate transfer prices. It also shows how a shift from a worldwide tax system to a taxes on passive income are due when this income is earned; no deferral of taxes is allowed. Anti-deferral rules are defined in the controlled foreign company (CFC) rules to prevent the shifting of income to tax havens. 13Forexample,theUKdidnotrestrictparentcompaniesfromborrowingfromforeignsubsidiaries,contrary to the United States’ treatment of ”deemed” dividends under I.R.C. §956. 14Inaddition,theworldwidesystemcouldalsocausesomecostlydistortionintheallocationofinvestment across countries. 15In this paper we focus on the changing incentives of profit shifting accompanying the territorial tax reform. Other studies, for example, Egger et al. (2015) and Liu (2017), examine the effect of the territorial tax reform on dividend repatriation and fixed capital investment by UK multinationals, respectively. 6

territorial tax system increases incentives for transfer price manipulation. We start from a modified version of the model in Cristea and Nguyen (2016), which is based on Bernard et al. (2006) and Hyde and Choe (2005). There are three differences to the setup in Cristea and Nguyen (2016). The most important is a new parameter γ, which captures the aspect that deferred profits abroad may be worth less to an MNC than profits at home — a key innovation to study the difference in incentives to shift profits between worldwide and territorial taxation. Secondly, to focus the analysis, we drop the incentive price p . This change is without loss of generality, as all i questions that we are interested in are independent of that variable. Finally, we modify the penalty function for transfer mispricing. The change in the penalty function is not essential but arguably makes the model more realistic. Basic setup Thereisamultinationalfirmthatsellstoanothercountryintwoways. Itsells directly to an unrelated party at the arm’s length price p , and it sells through its subsidiary a abroad at the final sales price p . For this related-party transaction, the MNC also needs to f set an internal transfer price p . We assume that arm’s length sales and related-party sales t target independent sets of consumers, so there is no competition across these two modes of selling abroad.16 Profits of the parent π and subsidiary π are given by: h f π = (1−τ )[(p −c)q +(p −c)q )], (1) h h a a t f π = (1−τ )(p −p )q , f f f t f where τ and τ are the corporate income tax rates in countries where the headquarters and h f the subsidiary are located, respectively; c is marginal cost of production per unit of output; and q and q are quantities sold at arm’s length and through the affiliate. Assume that a f differences between the transfer price and the arm’s length price imply a penalty given by:17 λ (p −p )2q , (2) a t f 2 16This assumption greatly simplifies the analysis but does not drive the results. All mechanisms and results should persist when allowing for some degree of competition across modes of exporting. 17While Cristea and Nguyen (2016) assume the penalty function is quadratic in q , in our specification f the penalty is linear in the quantity sold abroad through the affiliate. In our view, a punishment that is convexintheextentofmispricing(p −p )butproportionaltothetradevolumeisthemostintuitive. With a t thisassumption, themodelstillgivesrisetoamanipulatedarm’slengthprice, asemphasizedinCristeaand Nguyen (2016). 7

where λ captures the strength of tax enforcement. The penalty is linear in the trade volume, q , and quadratic in the difference between the arm’s length price and the transfer price, f p −p . a t MNCs have to decide where to report their book profits. Importantly, they can only pay out dividends to their equity holders or reinvest at home if profits have been repatriated to the headquarters. Under worldwide taxation, firms pay dividend taxes on any current or deferred after-tax profits that are brought home. To capture this aspect, let γ ∈ {γ ,γ } W T be the relative value of a pound of after-tax profit in a low-tax country compared to a pound of after-tax profit at home. In a worldwide system with deferral, after-tax profits abroad are less valuable to the firm, because they are still taxed upon repatriation and therefore γ < 1. In a territorial system, there are no additional taxes on foreign repatriations, so W γ = 1.18 Assume that penalties are paid at home and are not tax deductible. An MNC T therefore maximizes its overall weighted profits: λ Π = (1−τ )[(p −c)q +(p −c)q )]− (p −p )2q +γ(1−τ )(p −p )q . (3) h a a t f a t f f f t f 2 The transfer price Taking the first-order condition with respect to p , we can derive t (γ −1+τ −γτ ) h f p = p − . (4) t a λ The optimal transfer price is equal to the arm’s length price minus a correction for taxes. The lower the tax rate abroad relative to the home rate, the lower is the optimal transfer price.19 Leaving profits abroad is costly to the MNC under worldwide taxation and the optimal transfer price is hence less distorted than under territorial taxation. This effect can easily be seen by noting that p decreases in γ. t Proposition 1 (Transfer Price Manipulation) Relative to the arm’s length price p , a (i) under a territorial tax system, the optimal transfer price decreases in the tax difference τ −τ , h f 18Of course, there may still be some costs involved in moving profits back to the headquarters. The key point is that these costs should be strictly smaller under territorial taxation than under worldwide taxation — that is, γ <γ . W T 19Note that in Cristea and Nguyen (2016), the optimal transfer price also depends on q . Our change to f the penalty function makes this non-intuitive result disappear. 8

(ii) under a worldwide tax system, the optimal transfer price decreases in the adjusted tax difference τ −γ τ , h W f (iii) transfer price manipulation is stronger under a territorial tax system than under a worldwide tax system (∂(p −p )/∂γ > 0). a t Proof. Follows directly from equation (4) In equilibrium, the optimal arm’s length price p is also a function of the tax rates at home a and abroad. As the full model is not solvable in closed form, we show that the results in Proposition 1 on the transfer price (relative to p ) generalize to the absolute transfer price a p by simulating the model (Figure 1). t The arm’s length price Now, assume that there is a standard CES demand given by q = p−σA. (5) a a Then, we can derive the first-order condition with respect to p and solve for20 a 1 p = p , (6) a x 1+κ(∆τ) σ with p = c, (7) x σ −1 (cid:18) (cid:19) γ −1+τ −γτ q h f f and κ(∆τ) = . (8) (1−τ )(σ −1) q h a As pointed out by Cristea and Nguyen (2016), multinational firms have an incentive to not only manipulate their transfer price with a related party but also their arm’s length prices to unrelated parties. Equation (6) implies that the arm’s length price increases with the destination country’s tax rate. Arm’s length prices and transfer prices hence move in the same direction in response to tax rates, which keeps them closer together and thereby limits the effectiveness of transfer pricing regulation and tax penalty. Finally, note that in the data we observe whether a firm has a subsidiary in a given country instead of whether a given trade transaction is with a related party. For MNCs with a subsidiary, we hence observe in the data a combination of the arm’s length price p and a 20For γ = 1, this result is the same as in Cristea and Nguyen (2016), despite the change to the penalty function. 9

the related party price p : f q f p = s p +(1−s )p , with s = . (9) s f t f a f q +q f a Simulation results Figure 1 shows how optimal prices change with the tax difference. While the figure shows results for a specific set of parameters, the patterns in the figure discussed below are general features of the model. First, note that the transfer price, p , t strictly decreases in the tax difference, as should be expected. In addition, the firm has an incentive to charge an arm’s length price, p , that is strictly below the standard CES price a p , as discussed in detail in Cristea and Nguyen (2016). Next, Figure 2 illustrates how the x transferpricechangeswithγ, byplottingtheoptimaltransferpriceagainstthetaxdifference for γ = 1 and γ = 0.95. The optimal transfer price is clearly decreasing in γ, in line with the result in Proposition 1. 4 Data Ourdatasetisconstructedbymergingthreedatabases. Thefirstdatabaseincludestransactionlevelexportdatafrom2005to2011providedbytheHMRC.Specifically, eachrecordincludes the firm’s trader ID (anonymized), the product code (15-digit HMRC Integrated Trade Tariff Code), the destination country, the export value in British pounds, and the weight in kilograms.21 Theunitofobservationinourempiricalanalysisisafirm–product–destination–year price. We collapse the transaction data to that level, computing total export value, total quantity, number of shipments, and average unit price. The second database, also provided by the HMRC, consists of firm-level corporation tax records that provide detailed information on the tax position of each company and how it is determined. A lookup table that cross references the trader IDs and taxpayer identifiers allows us to merge the two databases. We therefore know the exact tax position for each trading company, including whether it has positive trading profits. Thethirddatabase, theFAMEownershipdatabaseofBureauVanDick, isalsoatthefirm level and provides information, for each company, on the name and location of its ultimate 21Transactions within the EU only need to be reported by firms whose exports in a given calendar year exceed a certain threshold (for example, £250,000 in 2016). Firms have to report all transactions with countries outside the EU. 10

global parent and subsidiaries, if applicable.22 Based on the ownership information, we group the population of UK companies into one of the following categories: (1) domestic;23 (2) parent companies of an MNC group with at least one subsidiary outside the UK; (3) subsidiaries of a UK MNC group; and (4) subsidiaries of a foreign parent company. Figure 3 Panel A shows the number of UK affiliates in each of the 108 countries that had UK exporting partners in 2011. For a more credible identification, we restrict our comparison to pricing differences between multinationals, focusing primarily on all the UK multinationals in groups (2) and (3), as our data are best suited to study their transfer pricing behavior.24 We use the location of foreign affiliates as a proxy for related-party trade, similar to Vicard (2015), Hebous and Johannesen (2015), and Cristea and Nguyen (2016). By definition, a UK MNC can only have related-party trade with countries where it has a related party. Of course, it may also trade with unrelated parties in these countries. Therefore, the price we observe for an MNC that has an affiliate in a given country is the weighted average of the prices charged in all intra-firm and arm’s length transactions. Importantly, this measurement error biases results against us finding any effects, as it makes it harder to identify systematic differences between pure arm’s length prices and our related-party price proxy.25 We augment the data set with additional data on destination country characteristics and statutory corporate tax rates.26 We obtain information on country-level variables from the World Bank (World Databank, World Development Indicators) and the PennWorld Table 8.1.27 The statutory tax rates are headline corporation tax rates drawn from KPMG 22The ownership data set is from the FAME website in 2015. We define ultimate parents as shareholders that have more than 50% total shareholding. The total shareholding was calculated by summing up the direct percentage of shares and indirect percentage shares in FAME; see https://webhelp.bvdep.com/ robo/projects/whdotnetownership_63_en/CalcPerc.htm. The FAME data set provides information on companies’ subsidiaries up to 10 levels. 23Domestic companies include (a) stand-alone companies, (b) parent companies of a domestic group with all subsidiaries in the UK, and (c) subsidiaries of a domestic group. 24Domestic firms or groups do not set transfer prices for cross-border transactions. Subsidiaries of foreign parents set transfer prices, but we only observe a small part of the overall activities of these MNC groups. In addition, foreign-owned subsidiaries were not affected by the UK tax reform to the same extent as UK multinationals. 25Interestingly, for France, Davies et al. (forthcoming), using data with direct information on relatedparty trade, found very similar results to Vicard (2015), who proxied related-party trade through affiliate information. 26Given that we include an extensive set of fixed effects in our baseline regression, we utilize the firm and destinationcountrycharacteristicsmainlytoreplicateandcomparewithspecificationsinexistingstudieson transfer pricing in Section 6. 27ThePennWorldTable8.1isconstructedbytheUniversityofCalifornia-DavisandtheGroningenGrowth DevelopmentCentreoftheUniversityofGroningenandisavailableathttp://www.rug.nl/research/ggdc/ 11

Corporate Tax Rate Tables.28 Define ∆τ ≡ |τ −τ | as the absolute value of the difference in the statutory tax rate jt jt UK,t between the UK (τ ) and the destination country (τ ). Furthermore, define a country as UK,t jt a low-tax destination if its statutory corporate tax rate is lower than the UK rate (∆τ < 0) jt and as a high-tax country if its statutory tax rate is equal to or higher than the UK rate (∆τ ≥ 0). Following this definition, a destination country can switch from a low-tax to jt a high-tax country when the tax rate in the foreign country or in the UK changes over time. Figure 3 Panel B shows the number of countries classified as low tax and high tax, respectively, over the sample period of 2005 to 2011. Figure 4 further illustrates variation in the tax rates. Panel A shows the time trend in the statutory tax rate in the UK and in its top five export destination markets in 2005 to 2011. Panel B shows the histogram of the corporate tax differential based on all observations. As these figures show, there is substantial variation in corporate tax rates, both in the time series and the cross section. The final data set includes 931,773 observations at the firm–product–year level for 1,256 unique companies in manufacturing during 2005 to 2011, capturing over £30 billion per year in exports, or approximately 16 percent of total UK manufacturing exports during the sample period. Figure 5 reports the annual exports and the share of intra-firm trade in total exports between 2005 and 2011. In total, around 39 percent of MNC exports are to countries where they have majority-owned affiliates. Table 1 provides summary statistics for the final data set. 5 Empirical Strategy Baseline Our baseline specification estimates transfer pricing behaviors of MNCs in a triple-difference regression. Specifically, we estimate: lnp = α +α +α +(β ∆τ ×I +β ∆τ ×I )×Aff +(cid:15) , (10) ijkt ijk jkt ikt 1 jt low,t 2 jt high,t ij ijkt data/pwt/pwt-8.1. 28The corporate tax rates from 2006 to 2011 are drawn from KPMG Corporate Tax Rate Table (2006 to 2014). The rates for 2005 are from KPMG (2006). 12

where p is the average unit price of exports of product k to country j by firm i in year t. ijkt ∆τ ≡ |τ −τ | is the absolute difference in statutory corporate tax rates between the jt jt UK,t destination country j and the UK in year t. I (I ) are indicators that take the value low,t high,t of one if the destination country has a lower (higher) statutory tax rate than the UK in year t, and zero otherwise. Aff is a dummy indicator that takes a value of one if the MNC firm ij i has at least one affiliate in country j, and zero otherwise. α is a firm–market–product ijk fixed effect, α is a product–market–year fixed effect, and α is a firm–product–year fixed jkt ikt effect.29 With inclusion of the fixed effects, identification relies on the differential change in the price charged by a multinational on exports to a foreign subsidiary relative to the price charged by a multinational without any subsidiary in the same country in response to a changeinthetaxratedifferencebetweenthatcountryandtheUK.Totheextentthatcertain exports only take place between related parties and therefore would not be captured in the above estimation, the estimated response of transfer pricing to tax differentials represents a lower bound of the true extent of transfer mispricing. Taking the full set of fixed effects is crucial for insulating the causal effect of tax differences. More specifically, α takes out the ijk average price a firm charges for a product in a given market. This fixed effect is essential, as firms often supply goods of different quality to different destination markets.30 The second fixed effect, α , controls for the average price of a product in a year across all firms, taking jkt out all shocks to the supply and demand of a product that are common across firms. Finally, α controls for the average price a firm charges for a product in a given year. This fixed ikt effect controls for all shocks to the supply or demand of a firm’s product that are common across markets. The coefficients we estimate (β and β ) therefore capture the causal effect 1 2 of tax differences on transfer prices, controlling for all of the main supply and demand factors that could confound the effect of taxes on prices. We expect β to be negative if MNCs systematically reduce the export prices for trans- 1 actions with their foreign affiliates to shift more profits into low-tax countries in response to an increase in ∆τ . Similarly, we expect β to be positive when MNCs systematically jt 2 29WeemploytheStatamodule,reghdfe,asthemodelincludesalargenumberoffixedeffects. Themodule was developed by Correia (2015), and it efficiently estimates models that include high-dimensional fixed effects. 30See,e.g.,Hallak(2006),Khandelwal(2010),andHallakandSchott(2011)ontheimportanceofproduct quality in international trade. 13

increase the export prices for transactions with their foreign affiliates to shift more profits out of high-tax countries in response to an increase in ∆τ . Our baseline regression does jt not include firm-level or country-level controls, as any variation at that level is absorbed by the fixed effects. To account for possible correlation in export prices among all the UK multinationals trading with the same destination market, we cluster the standard errors by country-year pairs. The quasi-natural experiment We exploit the change in the UK’s tax system in 2009 as a quasi-natural experiment. As discussed in Section 2, the reform from worldwide to territorial taxation created stronger incentives for UK multinationals to shift profits into lower-tax destinations since 2009. To check whether the reform indeed led to more transfer mispricing, we run the following specification: lnp = α +α +α +(β ∆τ ×I +β ∆τ ×I )×Aff (11) ijkt ijk jkt ikt 1 jt low,t 2 jt high,t ij +(β ∆τ ×I +β ∆τ ×I )×Aff ×Post +(cid:15) , 3 jt low,t 4 jt high,t ij t ijkt where Post is an indicator that takes value of one if year t is after the tax reform and zero t otherwise. Given that the reform took place in the second half of the fiscal year, we drop observations in 2009 for cleaner identification.31 The main coefficients of interest are now β 3 and β . If the reform increased incentives for transfer price manipulation, we would expect 4 a negative β and a positive β , conditional on a negative β and a positive β . 3 4 1 2 6 Evidence on Tax-Motivated Transfer Mispricing In this section, we provide direct evidence on the extent to which UK multinationals shift profit abroad via the manipulation of their transfer prices for exports. 6.1 Main results Baseline results Table 2 presents our main regression results. Column (1) reports the baseline results based on equation (10). The coefficient on the triple interaction for low-tax destinations is negative and highly significant, indicating that MNCs shift profits out of 31That is, Post is equal to zero until 2008 and equal to one from 2010 onward. t 14

the UK by underpricing related-party exports to low-tax countries. In contrast, the triple interaction for high-tax destinations is insignificant. That is, there is no evidence that MNCs shift profits into the UK from higher-tax countries through transfer prices.32 Effects for low-tax destinations are large. A one percentage point larger tax difference, on average, reduces related-party export prices relative to arm’s length export prices by three percent. This magnitude is in line with estimates by Clausing (2003) but is substantially larger than the effects found in Vicard (2015) and Cristea and Nguyen (2016).33 Part of the difference could be due to genuine differences in the aggressiveness of transfer pricing by French and Danish firms as compared to U.S. and British firms. However, we show below that parts of the differences can be attributed to omitted variable bias, as previous studies used smaller sets of fixed effects in their empirical analysis. The quasi-natural experiment Column (2) shows regression results from the quasinatural experiment, by adding interaction terms between the main explanatory variables and a post-reform dummy indicator (as in equation (11)) to capture the effect of the tax reform. Theregressionshowsthattheextentofprofitshiftingthroughtransfermispricinghas increased under the territorial tax system. Before the reform, on average, a one percentage point increase in the tax difference led to a 2.7 percent decrease in the price of relatedparty exports relative to the price of arm’s-length exports. After 2009, the tax effect is more pronounced, reducing the relative export price for low-tax destinations by another 1.5 percent per percentage point of tax rate difference. The increase in the strength of transfer pricing following the UK tax reform is significant at the 5 percent level.34 32One reason we might not find transfer mispricing for exports to high-tax countries is that under the worldwide tax system, foreign taxes can be used as credits to reduce overall UK tax liability on foreign earnings. Therefore, the incentive to shift income out of high-tax countries can be much smaller than implied by the tax difference. Another reason is that UK MNCs may shift profits directly from subsidiaries inhigh-taxcountriestosubsidiariesinlow-taxcountries,makingprofitshiftingintotheUKthroughtransfer mispricing unnecessary. 33Bernard et al. (2006) andFlaaen (2016) also find relatively large effects. Their coefficients are, however, not easily comparable given the differences in data and estimation approaches. 34This finding is consistent with existing studies based on OECD countries that establish that firms with worldwide parents tend to shift less income than firms with territorial parents (Markle, 2016). Given that we only have two years of post-reform data, we are unable to examine in-depth the dynamics of transfer mispricing under the territorial tax regime. 15

6.2 Transfer Mispricing and R&D Do firms that undertake more investment in R&D engage in more transfer price manipulation? A priori, the relation could go either way. On the one hand, R&D increases the intangible capital of a firm, some of which can be allocated to low-tax jurisdictions to facilitate profit shifting. That way, R&D-intensive firms could shift sizable profits through licensing and royalty payments, reducing the need to misprice their goods. In this sense, transfer mispricing of tangible goods and intangibles can be viewed as substitutes in shifting profits. On the other hand, R&D can make a firm’s products more specialized, which makes finding comparable prices harder and in turn makes it easier to shift profits through transfer mispricing. In this case, transfer mispricing of tangible goods and intangibles are complements in shifting profits.35 Table 3 Column (1) presents the results from a regression that interacts ∆τ ×I × jt low,t AFF with three indicators of R&D intensity based on the average firm-level R&D spending ij in the sample.36 The results are striking. Firms with the highest R&D intensity strongly manipulate their transfer prices. Their coefficient is highly significant and has double the size of the average baseline effect estimated earlier. In contrast, there is no evidence for any systematic transfer price manipulations of firms outside the highest R&D group. The findings are consistent with the view that R&D makes goods more specific, facilitating profit shifting through mispricing. Profit shifting via transfer price manipulation therefore seems to be a complement rather than a substitute to shifting based on intellectual-property rights for R&D-intensive firms. It is plausible that large companies are more likely to invest in R&D so that indicators of R&D intensity may be highly correlated with firm size. To see whether the findings related to R&D are driven by firm size, the regression in Table 3 Column (2) replaces the R&D intensity indicators with indicators of firm size, defined based on the quartiles of the distribution of firm-level fixed assets in the sample. The results suggest that transfer mispricing is concentrated in medium and large firms. Column (3) includes both sets of interaction terms, and shows that, controlling for firm size, companies with the highest 35Note that the average extent of profit shifting via transfer mispricing in intangible assets by high R&Dintensive firms is controlled for by the firm-year fixed effects. 36Specifically, wecomputeatime-invariantmeasureoffirm-levelR&Dintensityastheratiobetweentotal qualifying R&D expenditure and total turnover during the sample period. We then group firms by their R&D intensity into the low, medium, and high categories. 16

R&D intensity strongly manipulate their transfer prices. Column (4) further controls for the type of goods based on the classification in Rauch (1999), by adding an interaction term between a dummy indicator that distinguishes between differentiated and undifferentiated goods and the main variable of interest ∆τ ×I ×AFF to the regression. Column (5) jt low,t ij includes both the firm size and the goods type interactions as controls. The basic finding that the most R&D-intensive firms manipulate their transfer prices more remains unchanged in these alternative specifications. 6.3 Heterogeneity in the Location of Transfer Mispricing A recent study on transfer mispricing, Davies et al. (forthcoming), found that price manipulation by French firms is concentrated in trade with tax havens and very low-tax countries. We test to what extent the same patterns hold for UK MNCs. We begin by splitting the sample into tax havens and countries that are not tax havens following the classification in Hines (2005). Results are presented in Columns (1) and (2) of Table 4. Interestingly, wefindsignificanteffects fornon-haven countries but noeffectsfor the tax-haven-only sample. These results remain unchanged when pooling the data and adding an additional interaction term with the tax haven indicator to the regression (Column 3). One potential explanation for this finding is that there are way fewer tax havens than other countries in the world, so the sample size in Column (1) is much smaller. However, the small number of tax havens did not prevent Davies et al. (forthcoming) from finding an effect concentrated in the former. An additional factor is that our empirical strategy relies on variation in the tax rates over time. The fact that tax rates were already quite low in most tax havens at the beginning of our sample thus further limits the identifying variation available. An economic reason for not finding effects is that trade volumes with tax havens are not that large (they have declined substantially in our sample period and represent slightly over 10 percent of UK exports in manufacturing since 2008). To the extent that there is sizable profit shifting to tax havens, it must therefore happen through other channels than transfer price manipulation, e.g., royalty payments on intellectual property rights located there. Columns (4)-(9) verify that this finding is robust to using alternative lists of tax 17

havens as in Dharmapala and Hines (2009) and as suggested in OECD (2000).37 In a next step, we study whether transfer price manipulation is concentrated in the lowest-tax destinations. For this analysis, we split the sample into quintiles according to the tax difference between the UK and the destination country. We then replace our main explanatoryvariable(∆τ ×I )byitsinteractionswithdummyindicatorsofeachquintile. jt low,t Results are presented in Figure 6. The results suggest that the extent of transfer price manipulation is roughly proportional to the tax difference. However, for the lowest-tax countries, standard errors are large and coefficients are only significantly different from zero for countries with midrange tax rates. To summarize, our results show that transfer price manipulation by UK MNCs is concentrated in exports to non-tax-haven countries and is most apparent in exports to countries withlowtointermediatecorporateincometaxrates. Asdiscussedabove, thisfindingislikely due to the fact that MNCs may shift profits to tax havens through other means than transfer mispricing. UK MNCs hence mostly exploit transfer mispricing in exports to non-haven countries with relatively low tax rates. 6.4 Effects on the Quantity and Total Value of Exports by UK Multinationals We now analyze the effect of the tax differential on the quantity and value of exports by UK MNCs. While transfer mispricing should mostly work through prices, the following regressions allow us to quantify the overall effect of transfer pricing on measured trade flows. Interestingly, we find some suggestive evidence that related-party trade flows may be slightly more depressed through tax incentives when taking the quantity response into account. There is also some evidence that the 2009 tax reform led to trade creation for tax purposes, that is, firms shifted exports to markets that they were shifting profits to. Table 5 Columns (1) and (2) show results with the dependent variable being the quantity of exports measured by weight, whereas Columns (3)-(4) and (5)-(6) focus on the unit price and the total value of transactions, respectively. First, note that there is no significant effect of the tax differential on the quantity of transactions, so manipulation of the transfer price is the main channel through which UK 37Table A.1 provides the list of tax havens in each publication in the Appendix. 18

multinationals engage in tax-motivated profit shifting. However, there is weak evidence for a quantity effect, which is significant at a 15 percent level with a coefficient of 0.03. As a result, Column (5) predicts a larger effect for export values than Column (3) for the unit price: a one percentage point increase in the tax differential on average depresses the value of intra-firm exports to low-tax countries by around six percent relative to arm’s length exports. Finally, there is some evidence for a tax-induced trade creation effect following the territorial tax reform. The tax reform interaction term in the weight regression is positive and significant at the 10 percent level (Column (2)). This result suggests that, after 2009, some exporters increased their exports to countries to which they were shifting profits prior to the reform. Quantification of Effects We discuss the quantitative importance of our findings, computing estimates of shifted profits and foregone tax revenues to the UK based on our estimated coefficient for transfer mispricing. We calculate total shifted profits as C (cid:88) β ×I ×∆τ ×exp , (12) 1 low,c c c c=1 whereβ isthecoefficientestimatefrombaselineequation(10), exp isthevolumeofrelated- 1 c party exports to country c, and ∆τ is the tax difference between the UK and country c. c We estimate that in 2010, UK multinationals shifted about 601 million GBP toward lowtax jurisdictions via transfer mispricing, where Ireland ranks the top country to which such transfer mispricing took place. At the 2010 tax rate of 28 percent, this finding implies foregone tax revenues of £168.2 million due to transfer mispricing in exports by UK MNCs in manufacturing. These estimates are relatively small but are in line with Davies et al. (forthcoming), who estimate that French firms would have paid about 1 percent (333 million Euro out of 36 billion Euro) more corporate income tax in the absence of tax-motivated transfer price manipulation. 6.5 Comparison to Previous Studies Several previous studies estimate the extent of the price wedge between arm’s length and intra-firmtradewithrespecttothestatutorycorporatetaxinthedestinationcountryrelative to the home country. These studies all find significant responses of the price wedge to the 19

tax rate differential in a baseline regression of log unit price on a measure of the tax wedge. A key challenge for the literature to credibly identify the effect of taxes on transfer prices is the large heterogeneity in the sets of fixed effects employed in these studies.38 The large variation in the sets of fixed effects employed make it difficult to compare results across papers. In the following exercise, we show how our results change when employing the less comprehensive fixed-effect specifications used in some of the previous papers. Table 6 reports the results based on four specifications of equation (10). Column (1) includes no fixed effects and reports a negative and highly significant coefficient estimate on ∆τ × I × Aff (β(cid:98) ) and a positive and statistically insignificant coefficient estimate jt low,t ij 1 on ∆τ ×I ×Aff (β(cid:98) ). Column (2) follows the main specification in Vicard (2015) jt high,t ij 2 by including a set of firm-product-year fixed effects and country-product-year fixed effects. Column (3) uses the main specification in Cristea and Nguyen (2016) and includes a set of country-firm-product fixed effects, and year fixed effects interacted with the low-tax country dummy indicator I . Column (4) adds firm-level and country-level controls, as in Cristea low,t and Nguyen (2016). Columns (2)-(4) show that results are highly sensitive to the inclusion of different fixed effects, with the coefficient on high-tax countries even having the opposite sign in one specification from that predicted by the theory. To make sure that the varying results in Panel A are not an artifact of using different regression samples, Panel B repeats the analysis by including observations that are used in the most comprehensive specification of equation (10). β(cid:98) is negative and significant in all specifications, though the coefficient size varies 1 substantially. β(cid:98) is yet more sensitive to inclusion of different fixed effects. We conclude 2 that even when looking at the exact same sample, failing to include the full set of fixed effects thus leads to substantially biased estimates. 38These studies also differ in the data employed. They covered different countries and years and had different levels of aggregation. Clausing (2003) uses monthly US data in 1997 to 1999. Bernard et al. (2006) uses annual data on US exports in 1993 to 2000. Davies et al. (forthcoming) uses cross-sectional data for a set of French exporters in 1999, and Vicard (2015) uses French data in 2000, 2007 to 2009, and 2014. Cristea and Nguyen (2016) analyzes trade data by foreign-owned multinationals in Denmark in 1999 to 2006. Finally, Bernard et al. (2006) and Flaaen (2016) both employ U.S. Census data and look at the price wedges within U.S. multinational firms between related-party sales and arm’s-length sales. While this leads to a clean identification of effects in some dimensions, it also leads to an additional endogeneity concern, as discussed by Cristea and Nguyen (2016). 20

7 Conclusions In this paper, we use linked trade-tax administrative records on UK multinationals in manufacturing to estimate the extent of tax-motivated transfer mispricing in exports of real goods. Our findings suggest that, on average, a 1 percentage point tax difference reduces related-party export prices to low-tax countries by 3 percent relative to the prices charged at arm’s length. The extent of tax-motivated transfer mispricing has increased in the post-2009 territorial tax regime and is substantially larger in R&D-intensive firms. The new evidence on transfer mispricing has several implications for policy and future research. First, we document compelling evidence that transfer mispricing takes place in exports of real goods in addition to any shifting based on intra-firm loans or royalty and license fee payments for the use of intellectual property. This result points out another area of revenue leakage risk and calls for tax authorities to keep paying attention to transfer pricing issues in tangible goods. Moreover, transfer mispricing is not uniform across firms but concentrated in the most R&D-intensive ones. This finding provides tax authorities with useful guidance on where to look for mispricing activities. Second, the UK’s shift from a worldwide to a more territorial tax regime in 2009 has increased the extent of transfer mispricing. This result is consistent with the view that, compared to a worldwide system, the territorial tax regime creates more incentives for profit shifting. It is also in line with previous studies that document more profit shifting by MNCs under the territorial tax system (Markle, 2016).39 The revenue costs associated with increased transfer pricing under the territorial tax reform hence represent another important aspect for consideration in the ongoing debate about international tax policy.40 Finally, in contrast to earlier research on France by Davies et al. (forthcoming), our finding suggests that transfer mispricing in goods is not concentrated in tax havens. If anything, the evidence suggests that low-tax, non-tax-haven destinations are at the center of UK multinationals’ transfer mispricing activities. One intuitive explanation for this finding is that transfer mispricing in goods requires sufficiently large trade flows to shift relevant amounts of profits internationally. Small tax havens may simply not have enough trade flows 39However,twofeaturesofthepre-2009worldwidesystem,deferralandlimitedcrosscrediting,mightblur the distinctions from the territorial system (de Mooij and Ederveen, 2003). 40This finding does not necessarily imply that worldwide taxation is preferable to territorial taxation, as the latter may have other desirable effects such as increasing the efficiency of investment allocation (see, for example, Liu (2017)). 21

coming their way. In addition, other ways of shifting profits may be easier for tax haven destinations. Future research should try to shed light on the differences between the UK, France, and other countries that might explain these differing findings. In any case, our results should caution policy-makers from focusing too much on tax havens compared with other non-haven, low-tax and medium-tax countries. References Bernard, Andrew B., J. Bradford Jensen, and Peter K. Schott, “Transfer Pricing by U.S.-Based Multinational Firms,” Working Paper 12493, National Bureau of Economic Research August 2006. Clausing, Kimberly A., “Tax-Motivated Transfer Pricing and US intrafirm trade prices,” Journal of Public Economics, September 2003, 87 (9-10), 2207–2223. Collins, Julie, Deen Kemsley, and Mark Lang, “Cross-Jurisdictional Income Shifting and Earnings Valuation,” Journal of Accounting Research, 1998, 36 (2), 209–229. Correia, Sergio, “REGHDFE: Stata module to perform linear or instrumental-variable regression absorbing any number of high-dimensional fixed effects,” Technical Report, Duke University 2015. Cristea, Anca and Daniel Nguyen, “Transfer Pricing by Multinational Firms: New EvidencefromForeignFirmOwnerships,” American Economic Journal: Economic Policy, 2016, 8 (3), 170–202. Davies, Ronald B., Julien Martin, Mathieu Parenti, and Farid Toubal, “Knocking on Tax Haven’s Door: Multinational Firms and Transfer Pricing,” Review of Economics and Statistics, forthcoming. de Mooij, Ruud and Sjef Ederveen, “Taxation and Foreign Direct Investment: A Synthesis of Empirical Research,” International Tax and Public Finance, 2003, 10 (6), 673–93. Desai, Mihir A, C Fritz Foley, and James R Hines, “The demand for tax haven operations,” Journal of Public Economics, 2006, 90 (3), 513–531. Dharmapala, Dhammika and James R. Hines, “Which countries become tax havens?,” Journal of Public Economics, 2009, 93 (9), 1058 – 1068. Dischinger, Matthias and Nadine Riedel, “Corporate taxes and the location of intangible assets within multinational firms,” Journal of Public Economics, 2011, 95 (7), 691–707. Egger, Peter, Valeria Merlo, Martin Ruf, and Georg Wamser, “Consequences of the New UK Tax Exemption System: Evidence from Micro-level Data,” The Economic Journal, 2015, 125 (589), 1764–1789. 22

Flaaen, Aaron, “The Role of Transfer Prices in Prot-Shifting by U.S. Multinational Firms: Evidence from the 2004 Homeland Investment Act,” mimeo, Federal Reserve Board. 2016. Grubert, Harry and John Mutti, “Taxes, tariffs and transfer pricing in multinational corporate decision making,” The Review of economics and Statistics, 1991, pp. 285–293. Hallak, Juan Carlos,“Productqualityandthedirectionoftrade,”Journal of international Economics, 2006, 68 (1), 238–265. and Peter K Schott, “Estimating cross-country differences in product quality,” The Quarterly Journal of Economics, 2011, 126, 417–474. Harris, David, Randall Morck, and Joel B Slemrod, “Income shifting in US multinational corporations,” in “Studies in international taxation,” University of Chicago Press, 1993, pp. 277–308. Hebous, Shafik and Niels Johannesen, “At Your Service! The Role of Tax Havens in International Trade with Services,” CESifo Working Paper Series 5414, CESifo Group Munich 2015. Hines, James R., “Do Tax Havens Flourish?,” Tax Policy and the Economy, 2005, 19, 65–99. and Eric M Rice, “Fiscal Paradise: Foreign Tax Havens and American Business,” The Quarterly Journal of Economics, 1994, 109 (1), 149–182. Hyde, Charles E and Chongwoo Choe, “Keeping two sets of books: The relationship between tax and incentive transfer prices,” Journal of Economics & Management Strategy, 2005, 14 (1), 165–186. Khandelwal, Amit, “The long and short (of) quality ladders,” The Review of Economic Studies, 2010, 77 (4), 1450–1476. KPMG, “KPMG Global Survey: An international analysis of corporate tax rates from 1993 to 2006,” Technical Report 2006. Liu, Li, “Where Does Multinational Investment Go with Territorial Taxation? Evidence from the UK,” Working Paper 1705, Oxford University Centre for Business Taxation 2017. Markle, Kevin, “A Comparison of the Tax-Motivated Income Shifting of Multinationals in Territorial and Worldwide Countries,” Contemporary Accounting Research, 2016, 33 (1), 7–43. OECD, “Towards Global Tax Co-operation: Progress in Identifying and Eliminating Harmful Tax Practices,” Progress Report to the G20, OECD 2000. Rauch, James E., “Networks versus Markets in International Trade,” Journal of International Economics, June 1999, 48 (1), 7–35. Riedel, Nadine, Theresa Zinn, and Patricia Hofmann, “Do Transfer Pricing Laws Limit International Income Shifting? Evidence from Europe,” CESifo Working Paper Series 4404, CESifo Group Munich 2015. 23

Vicard, Vincent, “Profit Shifting Through Transfer Pricing: Evidence from French Firm Level Trade Data,” Working Paper 555, Banque de France 2015. Zucman, Gabriel, “Taxing across borders: Tracking personal wealth and corporate profits,” The Journal of Economic Perspectives, 2014, 28 (4), 121–148. 24

8 Figures Figure 1. Simulated Prices in response to Tax Differential Prices (relative to CES), gamma=1 1.2 1.0 0.8 0.6 0.4 p_a p_t 0.2 p_f p_x 0.0 0.00 0.05 0.10 0.15 0.20 0.25 0.30 0.35 tax difference Notes: This figures shows simulated paths of standard CES price (p ), final sales price (p f)), x ( optimal arm’s length price (p ), and optimal transfer price (p ), for different values of tax difference a t and γ = 1. Figure 2. Optimal transfer prices for different γ Transfer Prices (relative to CES) 1.2 p_t, γ=1 1.0 p_t, γ=0.95 0.8 0.6 0.4 0.2 0.0 0.00 0.05 0.10 0.15 0.20 0.25 0.30 0.35 tax difference Notes: This figures shows the simulated path of optimal transfer price (p ) under different values t of tax difference, for γ = 1 and γ = 0.95, respectively. 25

Figure 3. Distribution of Affiliates (a) Location of UK MNC Affiliates (b) Number of Low/High Tax Countries Notes: Panel A shows the worldwide location of UK MNC affiliates in 2011, the last year in our sample period. Panel B shows the number of countries with statutory tax rate lower than the UK rate (low-tax) and higher than the UK rate (high-tax), respectively, during 2005-2011. 26

Figure 4. Tax Incentives for Profit Shifting (a) Statutory Tax Rate (%) (b) Distribution of Tax Differential Notes: Panel A shows the statutory tax rate in the UK and in its top-5 exporting partners during 2005-2011. Panel B shows a histogram distribution of the tax differential during the same time period. 27

Figure 5. Time-Series Export Values and the Share of Intra-Firm Trade (a) Time-Series (b) Share of Intra-firm Exports to Tax Havens Notes: Panel A shows the value of total exports in the final dataset during 2005-2011 (on the left y-axis), and the share of intra-firm trade in total exports within/outside the EU (on the right y-axis), respectively. Panel B shows the share of intra-firm trade to tax havens in total intra-firm trade, measured by total number of transactions (solid line) and value of exports (dashed line). 28

Figure 6. Non-linear Transfer Mispricing in Low-Tax Countries Notes: This figure plots the point estimate of the tax coefficient β as in equation (10) and the 1 corresponding 90% confidence intervals at each quintile of tax wedge ∆τ in the low-tax countries. jt 29

9 Tables Table 1. Summary Statistics Mean Std. Dev P25 P50 P75 Obs (1) (2) (3) (4) (5) (6) Product Characteristics Export Value (GBP) 115,151.10 463,230.80 956.00 4,518.00 28,388.00 931,773 Net Mass (in kilogram) 16,572.34 83,322.69 9.00 85.00 1,161.00 931,773 Average Value (per kilogram) 265.75 660.79 8.55 35.00 177.27 931,773 Number of Transactions 6.02 19.47 1.00 2.00 6.00 931,059 Firm Characteristics Log Sales 16.35 1.95 15.15 16.27 17.39 10,077 Intra-firm Trade 0.39 0.49 0 0 1 10,077 Profit Making 0.71 0.45 0 1 1 10,077 Country Characteristics Low Tax Country Dummy 0.55 0.50 0 1 1 705 Low Tax Wedge (τ −τj,%) 7.68 5.72 3.00 6.00 11.50 445,883 UK High Tax Wedge (τj −τ ,%) 5.85 4.68 2.00 5.00 8.31 485,890 UK Notes: This table lists the summary statistics for the key variables in this paper’s main analysis sample: an unbalanced panel of annual trade-tax records for all UK multinational exporters in manufacturing in years 2005-2011. i

Table 2. Effect of the Tax Differentials on Transfer Pricing by UK Multinationals ∆ × (1) (2) τjt I ×AFF -0.030*** -0.027** low,t ij (0.011) (0.011) I ×AFF -0.007 -0.000 high,t ij (0.006) (0.006) AFF ×Post 0.132*** ij t (0.043) I ×AFF ×Post -0.015*** low,t ij t (0.005) I ×AFF ×Post -0.008 high,t ij t (0.007) R2 0.973 0.974 N 387,709 315,330 Notes: This table presents regression results on the causal effect of the tax differential on transfer prices of exports by UK multinational, based on equation (10). The main variables of interests are the three-way interaction terms ∆τ ×I ×Aff and ∆τ ×I ×Aff . All other variables jt low,t ij jt high,t ij are as previously defined. ***,**,* denotes significance at 1%, 5%, and 10% level, respectively. ii

Table 3. Heterogeneous Transfer Mispricing in R&D ∆ ×AFF × (1) (2) (3) (4) (5) τjt ij I -0.033 -0.034 low,t (0.028) (0.030) I ×R&D -0.010 -0.025 low,t low,i (0.015) (0.023) I ×R&D 0.000 -0.015 0.014 0.017 low,t medium,i (0.017) (0.025) (0.015) (0.015) I ×R&D -0.064*** -0.073*** -0.044* -0.041* low,t high,i (0.016) (0.025) (0.024) (0.022) I ×Size -0.004 0.026 low,t small,i (0.019) (0.022) I ×Size -0.037* -0.000 -0.020 low,t medium,i (0.022) (0.004) (0.026) I ×Size -0.041*** 0.009 -0.015 low,t large,i (0.015) (0.027) (0.022) I ×Diff 0.019 0.033 low,t i (0.033) (0.032) I -0.007 -0.007 -0.006 -0.009 -0.009 high,t (0.006) (0.006) (0.006) (0.006) (0.006) R2 0.973 0.973 0.973 0.973 0.973 N 387,709 373,767 373,767 331,787 321,221 Notes: Thistablepresentsregressionresultsontheheterogeneouseffectofthetaxdifferential ontransferpricesofexportsbyUKmultinationalwithlow, medium, andhighR&Dintensity, basedonequation(10). TheR&Dintensityindicatorsaredefinedinreferencetothequartiles of the distribution of average firm-level R&D expenses relative to total sales. The size indicators are defined in reference to the quartiles of the distribution of average firm-level fixed assets. The indicator for differentiated products is based on Rauch (1999). ***, **, * denotes significance at 1%, 5%, and 10% level, respectively. iii

snevaH xaT ni slanoitanitluM KU yb gnicirP refsnarT no slaitnereffiD xaT eht fo tceffE .4 elbaT )0002(DCEO )9002(seniHdnaalapamrahD )5002(seniH elpmaSlluF snevaHxaT-noN snevaHxaT elpmaSlluF snevaHxaT-noN snevaHxaT elpmaSlluF snevaHxaT-noN snevaHxaT )9( )8( )7( )6( )5( )4( )3( )2( )1( × tjτ∆ ***130.0- ***230.0- 450.0- **030.0- **920.0- 200.0 **030.0- **920.0- 100.0 jiFFA×t,wolI )110.0( )110.0( )950.0( )210.0( )210.0( )420.0( )210.0( )210.0( )420.0( 700.0- 700.0- 700.0- 800.0- 030.0 700.0- 800.0- 030.0 jiFFA×t,hgihI )600.0( )600.0( )600.0( )600.0( )231.0( )600.0( )600.0( )131.0( *601.0 100.0- 100.0jnevaH×jiFFA×t,wolI )460.0( )120.0( )120.0( 890.0 770.0 770.0 jnevaH×jiFFA×t,hgihI )412.0( )002.0( )002.0( 379.0 379.0 199.0 379.0 379.0 199.0 379.0 379.0 199.0 2R 907,783 626,573 778,4 907,783 038,233 823,22 907,783 662,233 598,22 N snevah xat ni lanoitanitlum KU yb stropxe fo secirp refsnart no laitnereffid xat eht fo tceffe eht no stluser noisserger stneserp elbat sihT :setoN dna alapamrahD ,)5002(seniH ni seirav snevah xat sa dedrager era taht seirtnuoc fo tsil ehT .)01( noitauqe no desab ,seirtnuoc nevah-non dna .ylevitcepser ,level %01 dna ,%5 ,%1 ta ecnacfiingis setoned * ,** ,*** .snevaH xaT evitarepoocnU fo tsil DCEO eht dna ,)9002( seniH iv

Table 5. Effect of the Tax Differentials on Trade Diversion by UK Multinationals Dependent variable: ln(Weight) ln(UnitPrice) ln(TotalValue) (1) (2) (3) (4) (5) (6) ∆ ×I ×AFF -0.030 -0.013 -0.030*** -0.027** -0.060*** -0.041* τjt low,t ij (0.020) (0.020) (0.011) (0.011) (0.023) (0.022) ∆ ×I ×AFF -0.014* -0.011 -0.007 -0.000 -0.021** -0.011 τjt high,t ij (0.009) (0.011) (0.006) (0.006) (0.009) (0.011) AFF ×Post 0.112 0.132*** 0.244** ij t (0.104) (0.043) (0.104) I ×AFF ×Post 0.017* -0.015*** 0.001 low,t ij t (0.009) (0.005) (0.010) I ×AFF ×Post 0.001 -0.008 -0.007 high,t ij t (0.013) (0.007) (0.013) R2 0.975 0.976 0.973 0.974 0.967 0.969 N 387,709 315,330 387,709 315,330 387,709 315,330 Notes: This table presents regression results on the effect of the tax differential on the quantity of exports (Columns (1)-(2)), the transfer prices (Columns (3)-(4)), and the total value of exports (Columns (5)-(6)) by UK multinationals, respectively. ***, **, * denotes significance at 1%, 5%, and 10% level, respectively. v

Table 6. Comparison to Existing Studies Specification: No FE Vicard Cristea & Nguyen Cristea & Nguyen (2015) (2016) (2016) ∆ ×AFF × (1) (2) (3) (4) τjt ij Panel A: I -0.061*** -0.007*** -0.012 -0.005 low,t (0.001) (0.002) (0.011) (0.012) I 0.002 -0.008*** -0.006 -0.011 high,t (0.001) (0.002) (0.005) (0.008) R2 0.0149 0.934 0.894 0.893 N 941,358 550,252 673,436 568,333 I -0.054*** -0.007*** -0.026*** -0.025* low,t (0.003) (0.002) (0.010) (0.013) I -0.001 -0.009*** -0.007 -0.014** high,t (0.003) (0.002) (0.006) (0.007) R2 0.001 0.942 0.897 0.896 N 387,709 387,709 387,709 326,816 Both panels include: Fixed Effects None Firm-Product- Year, Year, Year, Country- Country-Firm-Product, Country-Firm-Product, Product-Year Low-tax Country Low-tax Country Dummy-Year Dummy-Year Controls Firm-level and Country-level Notes: This table presents regression results on the effect of the tax differential on transfer prices of exports by UK multinational following specifications in a number of previous studies. ***, **, * denotes significance at 1%, 5%, and 10% level, respectively. vi

A Appendix Table A.1. List of Tax Havens Dharmapalaand OECDListofUncooperative Country Hines(2005) Hines(2009) TaxHavens(2000) Andorra 1 1 1 Anguilla 1 1 1 AntiguaandBarbuda 1 1 1 Aruba 1 0 1 Bahamas 1 1 1 Bahrain 1 1 1 Barbados 1 1 1 Belize 1 1 1 Bermuda 1 1 1 BritishVirginIslands 1 1 1 CaymanIslands 1 1 1 ChannelIslands 0 1 1 CookIslands 1 1 1 CostaRica 1 0 0 Cyprus 1 1 1 Djibouti 1 0 0 Dominica 1 1 1 Gibraltar 1 1 1 Grenada 1 1 1 Guernsey 1 0 0 HongKong 1 1 0 Ireland 1 1 0 IsleofMan 1 1 1 Jersey 1 0 0 Jordan 1 1 0 Lebanon 1 1 0 Liberia 1 1 1 Liechtenstein 1 1 1 Luxembourg 1 1 0 Macao 1 1 0 Maldives 1 1 1 Malta 1 1 1 MarshallIslands 1 1 1 Mauritius 1 0 1 Micronesia 1 0 0 Monaco 1 1 1 Montserrat 1 1 1 Nauru 1 0 1 NetherlandsAntilles 1 1 1 Niue 1 0 1 Panama 1 1 1 SaintKittsandNevis 1 1 1 SaintLucia 1 1 1 St. Martin 1 0 0 SaintVincentandtheGrenadines 1 1 1 Samoa 1 0 1 SanMarino 1 0 1 Seychelles 1 0 1 Singapore 1 1 0 Switzerland 1 1 0 Tonga 1 0 1 TurksandCaicosIslands 1 1 1 Vanuatu 1 1 1 Notes: This table lists the countries that are regarded as tax havens in three different publications. The tax haven dummy indicator takes the value of one if the country was regarded as a tax haven under the corresponding publication, and zero otherwise. vii