Production location of multinational firms under transfer pricing: the impact of the arm’s length principle

For comprehensive surveys, see Markusen (2004), Navaretti and Venables (2004); and Blonigen (2005).

Hebous et al. (2011) show that lower corporate tax induces inflows of foreign capital irrespective of the type of investment, such as greenfield foreign direct investment and cross-border mergers and acquisitions. Voget (2011) finds that one percentage point decline in foreign effective tax rate augments the likelihood of headquarters’ relocation by 0.22 percentage point. Karkinsky and Riedel (2012) and Griffith et al. (2014) investigate the link between corporate taxation and patent location.

Bernard et al. (2010) show that over 46% of US imports were composed of intra-firm transactions in 2000. Lanz and Miroudot (2011) report that US imports of intermediate products scored around 50% in 2009. See also Slaughter (2000) and Hanson et al. (2005) for the importance of intra-firm trade.

For (in)direct evidence on transfer pricing, see Swenson (2001), Bartelsman and Beetsma (2003), Clausing (2003), Bernard et al. (2006), Cristea and Nguyen (2016), Gumpert et al. (2016), Guvenen et al. (2017), and Davies et al. (2018).

The project is called “Base Erosion and Profit Shifting.” Further details can be found at https://​www.​oecd.​org/​g20/​topics/​taxation/​beps.​htm, accessed on 17 March 2017.

See http://​www.​oecd.​org/​ctp/​oecd-presents-outputs-of-oecd-g20-beps-project-for-discussion-at-g20-finance-ministers-meeting.​htm, accessed on 17 March 2017.

The CUP method is applied to tangible assets. The corresponding method for intangible assets is called the comparable uncontrolled transaction (CUT) method. There are other methods such as cost-plus method, resale price method, and profit split method. See OECD (2017, ch.2) for details.

The uncontrolled transaction includes both a transaction of an MNE with an independent firm (“internal comparables”) and a transaction between other independent firms (“external comparables”). Our framework captures the internal comparable transaction.

In our setting, the arm’s length price is the price that the upstream affiliate charged to the local downstream firm. As MNEs are usually not price takers, the arm’s length price in an internal comparable transaction is more or less an MNE’s choice variable and thus is subject to manipulation ( Cristea and Nguyen (2016)). Although tax authorities recognize this possibility of manipulation (see e.g., IRS\(\S \)1.482–1(d)(4)(iii)), they have to refer to internal comparable transactions in some cases due to limited information available. One recent example of such case is the transfer pricing case of Medtronic, a medical device company based in the USA, v. the Internal Revenue Service (IRS) (see for details Global Tax Alert (News from Transfer Pricing), 21 June 2016: http://​www.​ey.​com/​gl/​en/​services/​tax/​international-tax/​alert--us-tax-court-imposes-a-proper-arms-length-allocation-method-for-transfer-pricing, accessed on 2 June 2018). Medtronic US gave license to its Puerto Rican affiliate for production. The affiliate purchased components from Medtronic US, manufactured finished medical devices, and sold them to the Puerto Rican market. Whereas the IRS accused Medtronic US of profit shifting for the years at issue, 2005–2006, Medtronic US argued that the royalty rates charged to the Puerto Rican affiliate, which are a sort of transfer price of inputs, were the arm’s length royalty rates in light of the CUT method. The arm’s length royalty rates calculated by Medtronic US came from several internal comparable transactions, including a license agreement between Medtronic US and Siemens (German conglomerate company). In 2016, the US Tax Court accepted the royalty rates proposed by Medtronic US with small adjustments. See Avi-Yonah (2007) for other applications of the CUP/CUT method.

Earlier contributions include Copithorne (1971), Horst (1971), Samuelson (1982), and Kant (1988). In addition to the profit-shifting motive of transfer pricing mentioned in the text, studies such as Elitzur and Mintz (1996), Schjelderup and Sørgard (1997), and Zhao (2000) point out a strategic motive. The strategic motive comes from MNEs’ incentive to make their downstream affiliates competitive against rival firms. See Sect. 4.4. for more on this point.

In a related context, Choi et al. (2018) find a possibility of dual sourcing where an MNE buys inputs from both independent suppliers and related subsidiaries.

The role of the ALP on transfer prices is also studied by Gresik and Osmundsen (2008), Bauer and Langenmayr (2013), Choe and Matsushima (2013), and Keuschnigg and Devereux (2013). However, they do not consider the location choice of MNEs.

See Keen and Konrad (2013) for a comprehensive survey.

For subsequent development in the literature on bidding for a firm, see, e.g., Bjorvatn and Eckel (2006), Haufler and Wooton (2006), Ferrett and Wooton (2010), and Furusawa et al. (2015).

We consider a decentralized decision structure where the MNE leaves quantity choice to the downstream affiliate.

If \(t \ge T\), the upstream affiliate is always located in the low-tax parent country (separate location) in both the benchmark and ALP cases. Thus, our focus is on the range of \(t < T\), where the imposition of the ALP may change the location pattern.

It is common in the literature to assume exogenous profits or profits independent of transfer price (Schjelderup and Sørgard 1997; Nielsen et al. 2003, 2008; Haufler and Mardan 2014). In these studies, the tax-manipulation effect, which we will define shortly, is so strong that the price-cost margin (and thus profits) can be negative as in our analysis. To isolate the tax-manipulation effect as clearly as possible, we follow the convention of the literature. In Appendix 8, we endogenize it by introducing a local downstream firm in the parent country.

More precisely, \(q^S > 0\) requires \(t > t^a\), while \(p^S > 0\) does \(t > t^b\).

Clausing (2003) empirically supports this result: she finds that MNEs in the USA tend to set lower export prices, as tax rates in the trading partners are lower.

These figures are derived using the following parameter values: \(c=0.3\); \({\bar{\pi }}=3\); (a)\(T=0.35\); (b)\(T=0.2\). In this numerical example, \(\underline{t}=0\) holds. We note that the qualitative results stated in Proposition 1 do not depend on these particular parameter values.

We assume that the ALP applies to both cross-border transactions (i.e., separate location) and domestic transactions (i.e., co-location). Article Nine and the OECD guidelines are fully or partly applicable to domestic transfer pricing in some member countries of the OECD such as the UK, Norway and Canada (Wittendorff 2012).

To see this formally, we calculate the difference between the coefficient of \(t-T\) in \({\tilde{g}}^S\) and that in \(g^S\):

The parameter values are the same as those in Fig. 1a: \(c=0.3\); \(\bar{\pi }=3\); \(T=0.35\). We note that the qualitative results stated in Proposition 1 do not depend on these particular parameter values.

For \(t^*\) [defined in Eq. (9)] to be positive, we assume a sufficiently high T such that \(T>{\overline{T}}=1/4\).

To see this formally, we haveThere exists t satisfying the above inequality because it holds that \(t^a< t^q < t^*\) and \(t > \underline{t} \equiv \max \{ 0, t^a, t^b \}\) from (A1): \(t>\underline{t}\).

Noting that \(g^S < g_*^S\) and \({\tilde{g}} > g_*^S\), we haveCombining this result with \(q^S +q_*^S > 2{\tilde{q}}\) gives \({\tilde{q}} < q^S\). This also implies that the price-cost margin is smaller in the ALP case than in the benchmark case: \({\tilde{q}} = {\tilde{p}}-{\tilde{g}} < p^S - g^S = q^S\).

Similar results can be found in Yao (2013) in a different setting mentioned in the Introduction.

We do not distinguish the notation of variables between the unconstrained problem in the text and the constrained problem here.

It always holds that \(t' < T\). \(t'\) is decreasing in N if N is not sufficiently large, implying \(t' > \underline{t}\) for not sufficiently large N.

\(\delta \) is bounded above by \(\overline{\delta } \equiv \min \{ {\hat{\delta }}, \hat{{\hat{\delta }}} \}\), where \({\hat{\delta }} \equiv \text {sup} \{ \delta : t^{**}>0 \}\) and \(\hat{{\hat{\delta }}} \equiv \min \{ \text {sup} \{ \delta : \Pi> 0 \}, \ \text {sup} \{\delta : \Pi ^S > 0 \} \}\).

The parameter values are the same as those in Fig. 1: \(c=0.3\); (a)\(T=0.35\); (b)\(T=0.2\). In this numerical example, \(\underline{t}=0\) holds.

The parameter values are the same as those in Fig. 1: \(c=0.3\); (a)\(T=0.35\); (b)\(T=0.2\).