The European Commission’s appeal in the high-profile case concerning selective aid allegedly granted by Luxembourg to subsidiaries of Amazon.com may cast light on the murky role of the burden of proof in state aid cases.
The decision in Luxembourg v. European Commission and Amazon EU Sàrl and Amazon.com Inc. v. European Commission, joined cases T-816/17 and T-318/18 (GCEU 2021), concerns an advance pricing agreement granted by Luxembourg’s tax administration in 2003 and extended in 2010. (Prior coverage: Tax Notes Int’l, May 17, 2021, p. 953.) The APA approved a method for calculating the royalty payable by Amazon EU Sàrl (LuxOpCo), a Luxembourg corporation that operated as Amazon’s regional headquarters, to Amazon Europe Holding Technologies SCS (LuxSCS), a Luxembourg limited partnership. The royalty was for the license of European rights to Amazon’s technology, customer data, and trademarks, which LuxSCS acquired through a cost-sharing arrangement with U.S. group entities. The same cost-sharing arrangement was the subject of the high-profile U.S. transfer pricing case Amazon.com v. Commissioner, 148 T.C. 108 (2017), aff’d, 934 F.3d 976 (9th Cir. 2019).
The method approved by the APA gave LuxOpCo a routine return in the form of a fixed markup on its operating expenses (subject to seemingly arbitrary upper and lower bounds), determined using the transactional net margin method (TNMM). All residual profit in excess of this routine return was converted into a royalty rate and paid to LuxSCS. Because the payer was a taxable corporation and the payee was a hybrid entity, treated as a partnership in Luxembourg and a corporation in the United States, the royalty reduced the Amazon group’s taxable income in Luxembourg.
According to the European Commission’s October 4, 2017, decision on State Aid SA.38944 (2014/C) (ex 2014/NN), the royalty as determined under the APA conferred a selective benefit on the Amazon group in violation of article 107 of the Treaty on the Functioning of the European Union. The commission argued, alternatively, that the TNMM should have been applied using LuxSCS as the tested party and that the profit-split method would have been the more appropriate transfer pricing method. (Prior analysis: Tax Notes Int’l, May 24, 2021, p. 1022.) The commission’s decision was annulled by a 2021 judgment by the General Court of the European Union, and the commission has appealed that judgment. (Prior coverage: Tax Notes Int’l, July 26, 2021, p. 492.)
Some of the commission’s grounds of appeal, in particular those that try to salvage its primary TNMM-based argument, are worthy of summary rejection. The first part of the first ground of appeal, which targets the court’s rejection of the commission’s “selection of LuxSCS as the tested party on the basis that LuxSCS held legal ownership of the intangibles,” reflects a defiance of well-established principles underlying the selection of a tested party. There is no plausible argument for using the owner of intangibles that are as unique and valuable as those owned by LuxSCS as a tested party.
The second part of the first ground of appeal, which claims that the court “improperly invoked arguments on its own motion,” faults the court for citing the absence of comparables as grounds for rejecting LuxSCS as the tested party. Because the distinction between routine contributions and unique and valuable (or “nonroutine”) contributions largely turns on the availability of comparables, the court would have been unable to properly decide an issue that lies at the center of the case without observing that LuxSCS has no comparables.
The more interesting arguments contained in the application concern the rejection of the commission’s secondary argument in the case, which was that the profit-split method was more appropriate than the TNMM. These arguments cannot be quite as easily dismissed — at least not entirely — as the commission’s TNMM-based arguments, and the General Court’s judgment suggests that a modified version of the argument may have succeeded. They also bring to the fore important questions concerning the proper burden of proof and standard of judicial review in state aid cases involving transfer pricing, neither of which has been clearly established.
Proving State Aid
It’s difficult to pin down the burden that the commission bears in proving state aid in cases involving an APA or some other measure concerning transfer pricing. Although the commission is generally entitled to a degree of discretion regarding complex economic assessments, EU case law is inconsistent and often contradictory in the level of deference given to the commission’s assessments, according to Stephen Daly of King’s College London.
“If you’re looking for consistency between the judgments, or even consistency in the use of language by the commission or the court, you’re going to be found wanting. The court says this and that, and then the commission also says this and that,” Daly told Tax Notes. “There’s a lot of doublespeak going on in both the commission’s arguments and also the General Court’s decisions.”
However, a few principles emerge from the General Court’s judgments on alleged state aid granted to Amazon, Starbucks (Netherlands v. Commission, joined cases T-760/15 and T-636/16 (GCEU 2019)), Apple (Apple Sales International and Apple Operations Europe v. European Commission, joined cases T-778/16 and T-892/16 (GCEU 2020)), and Fiat (Luxembourg v. Commission, joined cases T-755/15 and T-759/15 (GCEU 2019)). One is that the commission has the authority to apply the arm’s-length principle to assess the existence of state aid. The General Court has repeatedly held that the legal basis for the commission’s use of the arm’s-length principle in state aid cases is article 107 of the TFEU — not the OECD’s model tax convention or its transfer pricing guidelines, nor any similar provisions of a member state’s domestic law. The arm’s-length principle serves as a diagnostic tool to determine whether a controlled party received more favorable treatment than an otherwise comparable stand-alone enterprise would have received under the member state’s general corporate tax system.
Equally well established is that the commission bears the burden of demonstrating that the tax ruling or measure at issue allowed an outcome that deviates from an arm’s-length result, and that the non-arm’s-length result led to a tax benefit. Accordingly, the commission must do more than prove that the tax ruling or measure was methodologically flawed.
The commission must establish that the flaws go “beyond the inaccuracies inherent in the methodology” used to approximate an arm’s-length outcome, and that use of the flawed method reduced the alleged beneficiary’s tax liability. To infer selective benefit directly from a methodological error, the error must make “it impossible to arrive at an approximation of an arm’s length outcome and may necessarily lead to an undervaluation of the remuneration that would have been received under market conditions,” the Amazon judgment says.
Satisfying this standard of proof, which has not been explicitly challenged, has proved difficult for the commission. It was manageable in the case concerning aid granted to the Fiat group, likely because of the nature of the alleged methodological error. In that case, the commission accepted the APA’s selection of the TNMM as the most appropriate method, the group’s Luxembourg treasury center as the tested party, and the return on equity as the net profit indicator. Its objection concerned the APA’s acceptance of an arbitrarily narrow equity base for the Luxembourg treasury center.
Provided that the commission was correct in criticizing the APA’s chosen equity base, and the General Court held that it was, the standard of proof is relatively easily satisfied. When applying the TNMM using the return on assets as the net profit indicator, the tested party’s return will be proportionate to the selected asset base. Assuming that the return is positive, as it almost always will be in any real-world TNMM analysis, using an arbitrarily narrow asset base will necessarily reduce the tested party’s profitability. Such an error automatically leads to “an undervaluation of the remuneration that would have been received under market conditions,” and it would be a simple exercise to prove as much if necessary.
In contrast, the commission has struggled when its argument alleges that the wrong transfer pricing method was applied or that the method should have been applied in a fundamentally different manner. As noted in the judgments on alleged aid granted to Amazon and Starbucks, the commission is free to use a transfer pricing method other than the method endorsed by the ruling or measure at issue, but the standard of proof requires that the commission justify its choice of method.
In many cases, these struggles can’t be directly attributed to the standard of proof. The commission’s claim in the Starbucks case that the comparable uncontrolled price method is inherently preferred over TNMM is baseless, even under the 1995 version of the OECD transfer pricing guidelines. Its argument that an intangible property holding company, including LuxSCS in the Amazon case, can properly serve as a tested party in a TNMM analysis reflects a fundamental misunderstanding of the terms “routine” and “unique and valuable contribution.” And contrary to the commission’s argument in the Apple case, the authorized OECD approach to profit attribution does not support what the General Court referred to as the commission’s “exclusion approach.”
When the Burden of Proof Matters
One of the relatively rare examples in which the burden of proof was decisive concerns the commission’s secondary argument in Amazon, which is that the profit-split method was more appropriate than the TNMM. This argument suffers from some of the same flaws as the commission’s TNMM-based argument that nearly all LuxOpCo’s functions qualified as “unique and valuable” contributions. Although the General Court rightly rejected that characterization in most cases, it found that the commission was correct in two cases. Specifically, the court found that LuxOpCo’s leading role in developing Amazon’s European fulfillment network and its accumulation of European customer data both constituted unique and valuable contributions.
Not only did the Amazon court find that LuxOpCo’s contributions — both routine and nonroutine — were not fully compensated by the APA’s pricing method, it went as far as to endorse the commission’s selection of the profit-split method. “In the present case, owing to the fact that both LuxSCS and LuxOpCo performed some unique and valuable functions, the Commission cannot be criticised for finding that, overall, the profit split method may have been appropriate for examining the controlled transaction,” the judgment says.
If the commission was right, at least in part, that the profit-split method would have been more appropriate than the TNMM, why did it still lose?
Part of the reason is that the commission’s persistent misunderstanding of the concept of a unique and valuable contribution led it to insist that the nebulous “contribution analysis” variant of the profit-split method, and not the more commonplace “residual analysis,” was the most appropriate method. According to the commission’s original decision, a contribution analysis was necessary because neither party made any routine contributions at all. “In this case, where both LuxSCS and LuxOpCo are considered to perform unique and valuable functions in relation to the Intangibles, [the contribution analysis] is preferred over the residual analysis, where one party is also remunerated for its routine functions in addition to the remuneration it receives for its unique and valuable contributions to the transaction,” the commission’s 2017 decision says.
Referring to its finding that most of LuxOpCo’s contributions were routine, the General Court held that the commission failed to justify its selection of a contribution analysis over a residual profit split. According to the judgment, this was because the commission did not establish that there were no routine contributions that could be separately benchmarked in the first step of a residual profit split or show that the information necessary to perform a contribution analysis was available.
Although the court considered the commission’s inability to account for its rejection of the residual profit-split method to be sufficient grounds for rejecting this secondary argument, the judgment identifies another basis for rejection. By selecting the profit-split method based on a contribution analysis without attempting to compare the result to the result of the APA-approved method, the commission had failed to prove that applying its chosen method would have increased LuxOpCo’s taxable profit, according to the court:
Even if it did not actually apply the profit split method on the basis of the contribution analysis, the Commission should at least have identified a number of factors from which it could be concluded that LuxOpCo’s remuneration calculated pursuant to the tax ruling at issue was necessarily lower than the remuneration that a company operating on the free market would have received had that company performed functions comparable to those identified by the Commission in its functional analysis.
The court rejected the commission’s unverified assertion that a contribution analysis, which is based on the premise that both parties make nonroutine contributions, must increase LuxOpCo’s profit relative to the TNMM, which assumes that the tested party’s only contributions are routine. At the very least, the commission should have identified the correct profit-split allocation key or explained why any of LuxOpCo’s contributions would necessarily entitle it to a greater return under a contribution analysis, the judgment says.
As the court put it, “the Commission has not demonstrated the existence of an advantage, but has, at most, demonstrated the probability of the existence of an advantage.”
The Correct Burden of Proof
The commission could likely sidestep both of the General Court’s objections by simply acknowledging that LuxOpCo performed some routine functions, and therefore that the residual profit-split method may be applied. Doing so would align with the court’s findings regarding LuxOpCo’s many routine contributions, and it would likely also eliminate the need to separately show that a residual profit split would have increased LuxOpCo’s profit. The court found that not even LuxOpCo’s routine functions were adequately compensated by the APA’s formula, and a higher routine return plus any share of residual profit (assuming there was a residual profit and not a residual loss) would necessarily be higher than the amount determined under the APA.
The commission will have the opportunity to adjust its argument regarding the profit-split method if it chooses. According to Daly, EU state aid procedure allows the commission to start a new case if its appeal fails.
“The commission could just go back and start a new case. It can do that because it started its investigation in 2014, and that actually stops the clock,” Daly said. “There’s no time limit, so the commission can still use 2014 as the date it started and look back 10 years from then for recovery, regardless of whenever this new case is started.”
But the commission’s focus, at least for now, appears to be on vindicating its original argument. In its second ground of appeal, the commission alleges that the General Court “erroneously invoked functions performed by U.S.-based Amazon Group entities” and failed to provide reasons for concluding that LuxOpCo’s brand-development functions were “valuable” but not “unique.” The commission also claims that “the General Court improperly invoked arguments on its own motion to reject the decision’s first subsidiary finding of advantage on the basis that the commission’s reliance on the profit split method with a contribution analysis does not demonstrate that the tax ruling necessarily leads to an advantage.”
These arguments suffer from some familiar misunderstandings. One is that functions (as opposed to assets or risks) alone qualify as “unique and valuable contributions.” Another misperception is that “unique and valuable function” is roughly synonymous with “important function.” It is difficult to see how the court can be faulted for considering intangible development functions carried out in the United States when those functions created the value of LuxSCS’s intangible property rights. And setting aside the commission’s allegation that the court “improperly invoked arguments on its own motion,” it’s hard to seriously argue that the TNMM could not leave the tested party with more profit than it would have received under any conceivable contribution analysis.
But the commission’s claim that the General Court, in evaluating the commission’s contribution analysis argument, nominally endorsed the correct standard of proof while applying “a different, stricter standard” is more plausible.
In Amazon, the commission clearly established what the court considered to be methodological flaws in the APA that went well “beyond the inaccuracies inherent” in the TNMM. The APA did not reward LuxOpCo for its unique and valuable contributions, nor did it even fully compensate LuxOpCo for all its routine functions, according to the court. The commission was seemingly successful in establishing that LuxOpCo was not appropriately compensated for all its routine contributions or any of its nonroutine contributions.
Under the burden of proof described in the Amazon judgment, this arguably should have been sufficient to establish the existence of state aid. If the commission has proved that the APA’s TNMM analysis undercompensated LuxOpCo, then whatever the most appropriate transfer pricing method is, it must increase LuxOpCo’s compensation relative to the amount determined under the APA. The court’s apparent preference for the residual profit split implies that it was aware that there was a more reliable method than the TNMM and that applying it would have almost certainly increased LuxOpCo’s taxable income.
However, the court instead held that the commission had established at most the probability of state aid because it chose the wrong kind of profit split. But nothing in the Amazon judgment or any related case law says that state aid is only state aid if the commission’s chosen method is free from flaws. Nor does the judgment or related case law specify what should happen when the commission tries to replace a flawed transfer pricing method with a less flawed method. The Amazon judgment implies that the commission’s selected method must also be more appropriate than other methods that the commission opted not to apply, a principle that is never stated in the judgment’s lengthy discussion of the standard of proof.
Faulting the commission for failing to prove the existence of an advantage through a method comparison, even after the commission seemingly proved the existence of an advantage on other grounds, also seems odd when the commission has no obligation to apply the method that it considers to be best. Although it would make sense to require that the commission actually apply the method it claims is best, especially to allow the commission or a reviewing court to reliably quantify the amount of state aid that the member state must recover, the Amazon judgment acknowledges that the commission isn’t required to do so.
The situation must be frustrating for the commission, which probably “can’t believe it lost,” according to Daly. “It’s a couple of cases now where the General Court has said, yeah, you’ve absolutely proved that mistakes were made by the tax authority, but you’ve just not gone that further step to demonstrate just how significant that mistake is in terms of the overall tax that’s been lost. It’s just one further step.”
Whether that further step is really necessary will likely remain uncertain until the commission’s appeal is decided.