The effects of transfer pricing regulations on multinational income shifting

ABSTRACT

This study examines the effectiveness of transfer pricing regulations on income shifting. The results indicate that domestic (foreign) transfer pricing regulations reduce the tax and non-tax motivations of affiliates to shift income out of (into) their home country. The study further identifies that regulations restrict the income-shifting behavior of small multinational affiliates to a greater extent than that of larger multinational affiliates. The results confirm that aggressive income-shifting activities are generally conducted by large multinational corporations, which are capable of exploiting sophisticated tax planning using legitimate tax loopholes.

KEYWORDS:

• Base Erosion and Profit Shifting (BEPS)
• income-shifting behavior
• multinational corporations
• tax rate difference
• transfer pricing regulations

Acknowledgments

I would like to thank Giho Choi at University of Seoul, Chung-Huey Huang at National Taipei University, and Marcel Olbert at University of Mannheim for their helpful comments and suggestions.

Disclosure Statement

No potential conflict of interest was reported by the author.

Notes

1. During the early 2010 s, it was revealed that highly profitable MNCs such as Google, Apple, and Starbucks used complex company structures to shift income from high-tax-rate jurisdictions to low-tax-rate jurisdictions and consequently reduced their worldwide tax burden. For instance, Google enjoyed an effective tax rate of 6 percent on its offshore profits in 2014 (Jonathan Chew, Fortune, 11 March 2016), and Apple paid a tax rate of only 2.3 percent on its offshore profits (McIntyre, Phillips, and Baxandall 2015).

2. Thirty-one countries introduced transfer pricing regulations requirements during the 2000s and 2010s (Zinn, Riedel, and Spengel 2014).

3. The BEPS project basically addresses 15 Action Plans that provide solutions for narrowing the gaps in existing rules for international taxation. The BEPS project was officially launched in 2012. Final BEPS reports were released on 5 October 2015 and endorsed by G20 financial ministers on 8 October 2015. There is an ongoing discussion across countries on how to implement BEPS Action Plans into their legislation.

4. It would be ideal to analyze the effects transfer pricing regulations on income shifting using the data of actual intercompany transactions of MNCs. Because intercompany transactions are undisclosed data in most of the countries, however, this study estimates income shifting based on the empirical model developed by Hines and Rice (1994) and Huizinga and Laeven (2008). Most of the papers investigating the income shifting of MNCs employ similar empirical models for their estimations.

5. This study predicts that strengthened regulations increase the pre-tax income of high-tax-rate affiliates by limiting their outward income shifting, even after offsetting.

6. The sample of Klassen and Laplante (2012) is limited to U.S. firms for 1988–2009. The sample of Riedel, Zinn, and Hofmann (2015) is limited to European affiliates between 1999 and 2009. Saunders-Scott (2014) uses worldwide multinational affiliates for 2003–2011. A considerable number of countries have reinforced their transfer pricing regulations since the early 2010s to combat aggressive tax avoidance by MNCs. In particular, the launch of the OECD’s BEPS Action Plan in 2012 has escalated the strict enforcement of transfer pricing regulations. Therefore, prior studies cannot observe the effects of such significant changes in transfer pricing regulations.

7. MNCs intentionally underprice (overprice) an international transaction between two controlled entities to shift income from the selling (purchasing) party to the purchasing (selling) party. International trade involves not only the transfer of goods but also the transfer or utilization of services, capital and intangibles (intellectual property), etc.

8. The U.S. has applied the arm’s length standards to determine appropriate transfer prices since 1934.

9. The OECD has provided global guidance on the application of the arm’s length principle to coordinate the design of national transfer pricing regulations since 1995.

10. A few studies attempt to measure the strictness of transfer pricing regulations of countries. Saunders-Scott (2014) measures the strictness of transfer pricing regulations following the measure developed by Mescall (2011). In Mescall (2011), transfer pricing experts’ perceptions of transfer pricing risks for countries are regressed on various components of transfer pricing regulations to obtain coefficients for each component. Although the Miscall index can capture dimensions of transfer pricing enforcement and exhibits greater variation over time, however, a method of regressing experts’ perceptions of transfer pricing regulation factors may be subjective once the number of countries to be examined is increased. Zinn, Riedel, and Spengel (2014) defines six categories to evaluate regulations based on the (i) existence of transfer pricing regulations, (ii) placement of transfer pricing regulations, (iii) transfer pricing documentation requirement, and (iv) the submission deadline for documentation. Riedel, Zinn, and Hofmann (2015) approximates the tightness of transfer pricing regimes based on the level of the transfer pricing documentation requirement. The regulation indicators used in Zinn, Riedel, and Spengel (2014) and Riedel, Zinn, and Hofmann (2015) are highly segmented to finely measure the strictness of transfer pricing regulations. However, segmentation may distort the significance of each category. The development stage of transfer pricing regulations used in this study is relatively a simple but clear and intuitive measure compared to those used in previous literature.

11. The effects of other transfer pricing elements, such as consistency with the OECD guidelines, the existence of a transfer pricing penalty, and the availability of APA, are also examined in the robustness analysis.

12. While the paper tests the effectiveness of transfer pricing regulations on income shifting, tax-motivated and non-tax-motivated income shifting estimated in this paper comprise income shifting using other methods including intercompany debts. This problem is not expected to significantly affect the results of this study because transfer pricing is dominantly used for income shifting purposes compared to others means of income shifting in practices. Nevertheless, an additional analysis is conducted to confirm whether the results of this study are robust to the effects of other income shifting strategies.

13. For example, Google set up a legitimate tax planning structures known as a ‘Double Irish’ and a ‘Dutch Sandwich’ in 2004 to avoid U.S. income taxes or foreign withholding taxes on offshore profits (http://creativecommons.org/licenses/by-nc/4.0/https://www.bloomberg.com/news/articles/2018-01-02/google-s-dutch-sandwich-shielded-16-billion-euros-from-tax)rly 2010.

14. Controlling is defined as greater than 50 percent ownership.

15. If a GUO has more than one foreign subsidiary in one country, unconsolidated financial data of those subsidiaries are aggregated and treated as a single entity to control the effects of varying organizational structure of MNCs, consistent with Markle (2016). If a GUO owns a domestic subsidiary, unconsolidated financial statements of the GUO and the domestic subsidiary are also aggregated.

16. Nearly all empirical research on income shifting eliminate loss affiliates from the sample because multinational corporations use different income-shifting strategies for loss affiliates (De Simone, Klassen, and Seidman 2017). Specifically, loss affiliates will cause bias of reverse incentives for losses (Klassen, Lang, and Wolfson 1993). However, Hopland et al. (2018) address the problem of dropping loss affiliates in the empirical analysis of income-shifting. Specifically, they argue that dropping loss affiliates will overestimate the level of income shifting, while underestimating the true tax sensitivity of income-shifting strategies. This paper follows the conventional research of omitting loss affiliates from the sample to remove the bias of reverse incentives for losses. According to Hopland et al. (2018), however, tax and non-tax incentives for income shifting estimated in this study may be subject to measurement errors due to omitted loss affiliates.

17. The sample obtained from the Orbis data is subject to certain limitations. First, since the Orbis data provide only static ownership information of MNCs for the most recent year, it is impossible to decipher the ownership structure of the MNCs for all years. Therefore, this study is conducted under the assumption that the corporate structures of the MNCs are constant over the analysis period of 2009–2015. Any changes in corporate structures according to changes in equity holding ratios between affiliates of corporate groups will not influence the results of this study because equity holding ratios do not affect the variables used in this study. However, changes in affiliate members as a result of mergers and acquisitions, reverse takeover, liquidations, etc., can not be accurately reflected in the data. Although business restructuring is not generally a recurring event in every corporate group, any potential bias of omitted subsidiaries in the corporate group may influence the results of this study. Prior studies using the Orbis database, including Saunders-Scott (2014) and Markle (2016), also highlight this limitation of the Orbis database. Second, financial information for a significant number of affiliate years is not available in the database. These affiliate years are excluded from the database due to a lack of necessary information for the variables. These limitations on data availability may influence the results of the study to a certain extent. The results of this study are expected to be robust from this problem because affiliates with unavailable financial information are generally unlisted small affiliates that would not account for a significant portion of income shifting in their corporate group.

18. For all regression analyses conducted in this study, the null hypotheses for Hausman tests are rejected (P> chi2 = 0.000), supporting the application of the fixed-effect model.

19. Sensitivity tests are further conducted using total fixed assets as a proxy for capital.

20. The amount of intercompany transactions between related parties enables the study to measure the true magnitude of income shifting. Unfortunately, many countries do not require firms to disclose the amount of intercompany transactions in their financial statements. Although firms in some countries disclose the amount of intercompany transactions in financial statements, manual collection of data for worldwide firms is not feasible due to limited data accessibility and language difference. For this reason, the amount of intercompany transactions is not included as a variable in the empirical model.

21. Tax rate differences are averaged by a weight of a ratio of the affiliates’ true income over the total true income of their MNCs. Consistent with the prior literature (e.g., Hines and Rice 1994; Huizinga and Laeven 2008; and Markle 2016), revenue is used as a proxy for true income. Sensitivity tests are further conducted using total assets as a proxy for true income.

22. Each foreign country’s transfer pricing enforcement is averaged by a weight of a ratio of domiciling affiliate’s true income over the aggregate total true income of the corporate group. Revenue is used as a proxy for the true income to reflect the magnitude of importance of foreign enforcement.

23. Instead of fixed assets, total assets of the corporate group is used to determine the affiliate size to consider the effects of intangibles that provide diverse opportunities for tax planning. Total assets of the parent company can also be used to define small MNC affiliates. Consistent results are observed when total assets of the parent company is used instead of aggregated total assets of the corporate group.

24. All continuous variables are winsorized at the 5^th and 95^th percentile to mitigate the effects of outliers.

25. The finding of a positive coefficient on DomREG is in contrast to the result obtained by Saunders-Scott (2014), who uses a sample of affiliates for 2003–2011 also obtained from the Orbis database. Potential reasons for these conflicting results include differences in the calculation of the tax incentives and the strictness of transfer pricing regulations. Saunders-Scott (2014) calculates the tax incentives as the difference between the tax rate that the firm faces and the unweighted average of the tax rates faced by all affiliates. This study calculates the tax rate differences between the firm and each of its foreign affiliates and averages these tax rate differences weighted by each affiliate’s revenue to reflect the firm’s heterogeneous costs for income shifting. In addition, Saunders-Scott (2014) measures the level of transfer pricing risks using the index developed by Mescall (2011), whereas this study measures the strictness of transfer pricing regulations using three basic and objective factors for transfer pricing regulations.

26. Countries with strict transfer pricing regulations are generally developed countries with relatively high tax rates. A high DomREG of the parent company causes its subsidiary to have high ForREG and low C because the subsidiary is located in a lower-tax-rate jurisdiction compared to its parent company. ForREG, therefore, inevitably has a negative correlation with C. The reverse situation also occurs.

27. Equation (1)(1) $LogPLB{T}_{it}={\beta }_0+{\beta }_1{C}_{it}+{\beta }_{2}DomRE{G}_{it}+{\beta }_{3}ForRE{G}_{it}+{\beta }_{4}LogASSET{S}_{it}+{\beta }_{5}LogCOM{P}_{it}+{\beta }_{6}LogGD{P}_{it}+{\mu }_i+{\delta }_t+{\epsilon }_{it}$(1) is estimated using OLS with year and country fixed effects for path analysis. In this study, the panel regression estimates and OLS estimates with year and country fixed effects for equation (1)(1) $LogPLB{T}_{it}={\beta }_0+{\beta }_1{C}_{it}+{\beta }_{2}DomRE{G}_{it}+{\beta }_{3}ForRE{G}_{it}+{\beta }_{4}LogASSET{S}_{it}+{\beta }_{5}LogCOM{P}_{it}+{\beta }_{6}LogGD{P}_{it}+{\mu }_i+{\delta }_t+{\epsilon }_{it}$(1) are generally consistent.

28. In the first stage, a probit regression is conducted to predict the effects of tax rates, GDP, tax revenue, and the size of trade on domestic transfer pricing regulations. In the second stage of the Heckman estimation, equation (1) is reestimated after including Invers Mills’ ratio estimated from the first stage regression to control for the endogeneity.

29. The coefficient on REG_CHANGE (0.062) is similar to the coefficient on DomREG (0.070) estimated from equation (1) in terms of the magnitude.

30. It should be noted that intangible assets may cause endogeneity issues when included in this empirical model due to the high correlation between the level of affiliates’ intangible assets and income-shifting behaviors.

31. Specifically, two variables are used to measure the magnitude of income shifting using intercompany debts. The first variable is LogINTERDEBT, which is the natural logarithm of intercompany debts (i.e., AGG_DEBT minus CONSOLIDATED DEBT) of the corporate group. The variable is established based on the idea that the difference between AGG_DEBT and CONSOL_DEBT will approximate to the amount of intercompany debts because intercompany debts is offset when preparing consolidated financial statements of the parent company. The second variable, INTERDEBT_RATIO, is calculated as one minus the ratio of CONSOL_DEBT over AGG_DEBT. These two variables measure the magnitude of intercompany debts in terms of the amount (LogINTERDEBT) and the ratio (INTERDEBT_RATIO).

32. The coefficients on LogINTERDEBT and INTERDEBT_RATIO are significantly positive. The results confirm the presence of income shifting using intercompany debts.

33. The main analysis of this study uses panel regression with a fixed-effect model to control for time-invariant heterogeneity across affiliates. However, only ordinary least squares (OLS) is available for estimating the coefficients on C by year. Estimates using OLS may be biased by unobserved time-invariant confounding factors of affiliates. Nevertheless, the ‘trend in the coefficients’ is not expected to be significantly biased because any unobserved time-invariant characteristics would affect the dependent variable to the same extent in each year.

34. The results of this analysis is consistent with anecdotal evidence because a number of press releases have reported significant magnitude of tax avoidance conducted by multinational corporations by early 2010s (Tim Worstall, Forbes, 12 November 2012; Vanessa Barford and Gerry Holt, BBC, 21 May 2013).

35. The average pre-tax income achieved by 26,147 high-tax-rate affiliate-year observations in the sample during the ex-post period (2012–2015) is 16.9 percent higher than that achieved during the ex-ante period (2009–2011). On the contrary, the average pre-tax income achieved by 60,192 low-tax-rate affiliate-year observations in the sample during the ex-post period is 21.1 percent lower than that achieved during the ex-ante period.