automatic decision-making in terms of legal certainty and
efficiency?
Adv LLM thesis
submitted by
Matheus Peixoto Behrends
in fulfilment of the requirements of the
‘Advanced Master of Laws in International Tax Law’
degree at the University of Amsterdam
supervised by
Dennis Weber
co-supervised by
PERSONAL STATEMENT
Regarding the Adv LLM Thesis submitted to satisfy the requirements of the ‘Advanced Master of Laws in International Tax Law’ degree:
1. I hereby certify (a) that this is an original work that has been entirely prepared and written by myself without any assistance, (b) that this thesis does not contain any materials from other sources unless these sources have been clearly identified in footnotes, and (c) that all quotations and paraphrases have been properly marked as such while full attribution has been made to the authors thereof. I accept that any violation of this certification will result in my expulsion from the Adv LLM Program or in a revocation of my Adv LLM degree. I also accept that in case of such a violation professional organisations in my home country and in countries where I may work as a tax professional, are informed of this violation.
2. I hereby authorise the University of Amsterdam and IBFD to place my thesis, of which I retain the copyright, in its library or other repository for the use of visitors to and/or staff of said library or other repository. Access shall include, but not be limited to, the hard copy of the thesis and its digital format.
3. In articles that I may publish on the basis of my Adv LLM Thesis, I will include the following statement in a footnote to the article’s title or to the author’s name:
“This article is based on the Adv LLM thesis the author submitted in fulfilment of the requirements of the ‘Advanced Master of Laws in International Tax Law’ degree at the University of Amsterdam.”
4. I hereby certify that any material in this thesis which has been accepted for a degree or diploma by any other university or institution is identified in the text. I accept that any violation of this certification will result in my expulsion from the Adv LLM Program or in a revocation of my Adv LLM degree.
signature:
name: Matheus Peixoto Behrends date: 17/04/2020
List of abbreviations used ... IV
Executive Summary ... V
Main Findings ... VI
1. Introduction ... 1
2. What is the framework to apply GAAR based on ADM?... 3
2.1. ATAD’s GAAR: legal certainty and efficiency issues... 3
2.2. ADM conformity with ATAD’s GAAR and CJEU case-law ... 4
2.3. CJEU case law and the ATAD’s GAAR tests ... 7
2.3.1. Motive test ... 7
2.3.2. Defeat of object or purpose test ... 8
2.3.3. Artificiality test ... 10
3. What are the limitations for ADM application in the ATAD’s GAAR context? ... 13
3.1. Tax administration legal boundaries to ADM ... 13
3.1.1. ADM decisions legality ... 14
3.1.2. Legal limitations to the burden of proof and general proceedings against tax avoidance ... 16
3.2. From algorithms and machine learning towards ADM models: the system capabilities ... 17
3.2.1. Tax documentation: quality and quantity ... 18
3.2.2. Identifying tax liability reduction based on abusive arrangements in a consistent and efficient way ... 20
4. Will ATAD’s GAAR benefit from ADM? ... 22
4.1. Applying ADM in the ATAD’s GAAR context ... 22
4.1.1. First objective: identify the relevant criteria used to recognise fact patterns in legal documents ... 22
4.1.2. Second objective: recognise the fact patterns from tax documentation and categorise each as evidence that contributes to taxpayers fulfilling the test ... 25
4.1.3. Third objective: deciding if the undertaking passed or failed the test ... 26
4.2. Does ADM have a spot in the ATAD’s GAAR context? ... 27
5. Conclusion ... 28
Bibliography ... 29
Statute or Regulation ... 32
Court Decision ... 32
List of abbreviations used
AG Advocate General
ADM Automatic decision-making
AI Artificial intelligence
AITTA Artificial intelligence tax treaty assistant
ATAD Anti-Tax Avoidance Directive
Art, Arts Article, articles
C-, T- Court case number
CFC’s Controlled foreign companies
CJEU Court of Justice of the European Union
ECHR European Convention for the Protection of Human Rights and Fundamental Freedoms
EU European Union
GAAR General anti-avoidance rule
GDPR General Data Processing Regulation
MS Member State from the EU
STIR System Teleinformatyczny Izby Rozliczeniowej (Information Technology System of the Clearing House)
Executive Summary
This thesis recognises the deficiencies in the application of the Directive (EU) 2016/1164, the Anti-Tax Avoidance Directive (ATAD), considering the introduction of the general anti-avoidance rules (GAAR). The core issues addressed relates to the lack of legal certainty and efficiency by tax administrations. It is identified that tax administrations and other sectors have begun to address these issues using automatic decision-making (ADM) models. The thesis’s goal is identifying whether ADM models applied by tax administrations would enhance efficiency and legal certainty.
The paper is structured by, first, examining the legal context in which tax administrations (should) apply the ATAD’s GAAR where the ADM model would exercise its decisions. Second, what should be the underlying limitations to set up ADM models, in terms of the legality of its structure and technological limitations to the objective pursued. Third, the gathered ADM limitations are confronted with the ATAD’s GAAR application to check which legal certainty and efficiency consequences may exist.
The framework analysis under the ATAD’s GAAR recognises the lack of clarity and certainty thereof, foremost in art 6. It considers the CJEU case-law as a vital interpretation source, particularly the abuse-doctrine, to provide more precise criteria to apply the ATAD’s GAAR. It draws the main criteria and identifies the objectiveness of the three tests: (1st) motive; (2nd) defeat of object or purpose; and
(3rd) artificiality.
The legal and technological limitations pertinent to ADM models are analysed. In the first case, lack of transparency and other rights from the taxpayers’ perspective is found to limit the use of ADM mechanisms and the misuse of taxpayers’ data. On the other hand, the CJEU case-law prescribes parameters considering the burden of proof and general proceedings which may potentially conflict with the ADM model, depending on how it is designed, and whether it does not adhere with certain transparency standards. In the second case, technological limitations from the artificial intelligence (AI) perspective are scrutinised, and some cases of the ADM model are analysed. It is found that a hybrid approach with the use of machine learning and traditional computer mechanism is a preferred choice for ADM models that focus on predictive pattern and have both sufficient data and a good understanding of the model.
Finally, the framework of the ATAD’s GAAR and the limitations to the ADM model show that the motive and the artificiality tests are more prone to automatisation, whereas the defeat of object or purpose is less likely, in the current stand of CJEU case-law. Supported on a practical approach, after extracting fact patterns from more than 22 CJEU cases, it is found that automatisation of the artificiality test can directly benefit from CJEU case-law. However, it has potential also to be used to assess the other two tests of the ATAD’s GAAR. Besides, it shows that the use of ADM model may not necessarily result in advantages for taxpayers if the model is under a black-box. Meaning, a decrease in legal certainty may result from a lack of clarity on the inner decision-making framework. Finally, the usefulness of ADM under the ATAD’s GAAR is found to be in line with the goals of greater legal certainty and efficiency, provided certain conditions are met.
Main Findings
The current framework of the ATAD’s GAAR is comprehensive, and the application of its rules require high levels of discretion and interpretation. The three tests in the ATAD’s GAAR (motive; defeat of object or purpose; and artificiality) lacks sufficient degree of clarity and certainty. The CJEU abuse-doctrine aids in the application of the ATAD’s GAAR with more specific fact patterns for the creation of an ADM model to test artificiality. Further developments of ATAD’s GAAR legislation or more case-law criteria could also allow the extension of the ADM approach to the other two criteria, the motive test and the defeat of object or purpose test.
Tax administrations may deploy ADM models to power up their capacity of analysis to a more significant number of taxpayers and potentially raise tax revenues. The legal certainty can only be achieved if the taxpayers could foresee which legal business path they can pursue without having their arrangements reassessed under the GAAR. Therefore, abiding by the principles of legal certainty, tax administrations must design ADM models that are as transparent as possible, forfeiting attempts at black-box models.
The regulation in the EU level restricts the way ADM models may be applied in the tax administration context. Black-box models may be one of the obstructions to ADM. Nevertheless, the construction and application of safer ADM models may be addressed including the human in the loop to ensure the model is consistent with the legal boundaries, while it also guarantees it works concerning its technical limitations.
1. Introduction
New advance and harmonisation for fighting against abusive tax practices are causing turbulences in the European Union (EU) law. The ATAD1 introduced general minimum standards that Member States
(MSs) are required to follow2 to deter tax avoidance3 and to provide a minimum level of protection for
the internal market against tax avoidance.4
The GAAR5 contained within the ATAD serves a vital role to cover for the breaches left by the
specific anti-avoidance rules present in this Directive. 6 Where detailed legislation cannot hold back tax
avoidance, the GAAR serves as a jack of all trades, but it also compromises a higher discretion power towards tax authorities. The harmonisation of the GAAR and the SAARs within the EU added another layer of complexity to EU tax law, raising the bar for efficient tax compliance and supervision.7 The
interaction between ADM8 and the ATAD’s GAAR can provide new insights on how tax administrations
could reduce tax avoidance and raise efficiency in this new context. Also, an ADM approach is a tool that can bring advantages to taxpayers in terms of legal certainty and efficiency.
The use of artificial intelligence (AI) and ADM systems is rising,9 including in several tax
administrations around the world, particularly in the MSs.10 The matter at hand is whether an ADM
applied by tax administrations could efficiently achieve the objectives of the ATAD’s GAAR and produce positive results for taxpayers and States.
This thesis analyses GAARs through the perspective of legal certainty and efficiency. The ATAD’s GAAR is a provision with a high degree of openness on its definitions, creating indeterminacy,11
which results from the low degree of certainty related to its application.12 Divergences in the application
of the ATAD’s could stimulate mismatches in the MS’s legal treatment of the existing national GAARs due to several variables: different tax procedures, diverse national courts’ decisions,13 and statutory
rules that provide high-level of discretion without presenting objective criteria to determine an abusive
1 Council Directive (EU) 2016/1164 of 12 July 2016 laying down rules against tax avoidance practices that
directly affect the functioning of the internal market (ATAD) [2016] OJ L193/1.
2 ibid, preamble paras 2, 3, 6, 16 and art 3. 3 ibid, preamble para 3.
4 ROOM DOCUMENT # 4 18 March 2016, Anti-Tax Avoidance Directive (ATAD) (Working Party on Tax
Questions – Direct Taxation), 1 and 4.
5 Council Directive (EU) 2016/1164 of 12 July 2016 laying down rules against tax avoidance practices that
directly affect the functioning of the internal market (ATAD) [2016] OJ L193/1 (n 1), art 6.
6 ibid, arts 4, 5, and 7-9.
7 Błażej Kuźniacki, ‘The Artificial Intelligence Tax Treaty Assistant: Decoding the Principal Purpose Test’ (2018)
72(9) , Bulletin for international taxation, pp.524–534, at p.524
<https://research.ibfd.org/#/doc?url=/document/bit_2018_09_int_2> accessed 13 May 2020.
8 Automatic Decision Making (ADM) is a branch of AI-decision systems. Therefore, based on artificial
intelligence, ADM uses the algorithms to establish criteria and assess determined data, ultimately reaching a decision. See Section 3.2.2.
9 The OECD correctly identifies some of the drivers for the general rise in the use of AI: faster computers,
improved data availability, and improved AI software. Ross D King and Stephen Roberts, ‘Chapter 5. Artificial intelligence and machine learning in science’ in OECD (ed), OECD Science, Technology and Innovation Outlook 2018: Adapting to Technological and Societal Disruption (OECD Publishing 2018), pp.121–136, at p.122.
10 World Bank, World Development Report 2016: Digital Dividends (The World Bank 2016), p.330 , at p.6
<https://openknowledge.worldbank.org/bitstream/handle/10986/23347/9781464806711.pdf> accessed 15 May 2020. For example, the Polish STIR. See Section 3.
11 See Section 2.
12 On what consists the objective of the ATAD see ROOM DOCUMENT # 4 (n 4), 1 and 4.
13 Alfredo G Prats and others, ‘EU Report: Anti-avoidance measures of general nature and scope - GAAR and
other rules’ (2018) 103A , IFA Cahiers, pp.5–35, at p.31
situation.
Taking these issues into consideration, this thesis asks: under which conditions MS’s application of the automatic decision-making – as a system of artificial intelligence – could bring legal certainty and efficiency to the ATAD’s GAAR?
Based on preliminary research, this author’s hypothesises are that there are certain core elements of the ATAD’s GAAR that are not subject to automation within the ADM framework. Foremost, due to the high complexity of arrangements, tax law, and lack of objectivity in the concepts of this GAAR. Nevertheless, other GAAR core elements could potentially rely on an ADM model. Even in these cases, the casuistic and direct human intervention may be necessary, reducing the automation process and, inevitability, leading to fewer efficiency advantages.14
This thesis object is restricted to the interaction between ADM and GAAR, searching for how to design a predictive model system that could produce better results in terms of legal certainty and efficiency and what would be the primary restrictions to achieve these goals.15 This process starts with
the ADM model design,16 which must reflect the legal reasoning behind the CJEU’s decisions concerning abusive tax arrangements.17 Therefore, Section 2 provides the first step to assess the current
understanding concerning abusive tax schemes and the framework for applying the ATAD’s GAAR. Section 3 introduces which ADM models are theoretically suitable for combating tax abuse, regarding especially its legal boundaries and technological limitations. Following this analysis, Section 4 answers whether ATAD’s GAAR could effectively benefit from the ADM application. This idea is better explained based on a practical example. Finally, the Conclusion draws the steps that could be taken to introduce an ADM applying – more clearly and efficiently – the ATAD’s GAAR and the issues when doing so.
14 See Section 3.2.
15 Following the approach that ‘the principle of legal certainty is the starting point in the application of the law’
and, therefore, ‘the rules of law must be made know and that situations of law governed by EU law must be foreseeable’. Dennis Weber, ‘Abuse of law in European tax law: an overview and some recent trends in the direct and indirect tax case law of the ECJ: part 1’ (2013) 53(6) , European taxation, pp.251–264, at p.251.
16 The term ‘model’ refers to an abstraction of the real world – the EU GAAR rules application in this case. It is a
support system that may augment our capabilities and requires us to make explicit the assumptions and boundaries. See M. Walport and others, Computational Modelling: Technological Futures: Government Office for Science (2018), p.124 , at p.113
<https://assets.publishing.service.gov.uk/government/uploads/system/uploads/attachment_data/file/682579/co mputational-modelling-blackett-review.pdf> accessed 15 April 2020.
17 This methodology resembles the Taxman early project, where it had a dual theoretical purpose: clarifying the
structures of the law area modelled, abuse of law in this case, while it tries the limits of the legal concepts behind this area of law. L. T McCarty, ‘Reflections on TAXMAN: An experiment in artificial intelligence and legal reasoning’ (1976) 90 , Harvard Law Review, pp.836–893.
2. What is the framework to apply GAAR based on ADM?
The ATAD’s GAAR fills the gaps of existing MS’s GAAR and creates a harmonised minimum standard.18
Moreover, it is secondary EU law and, thus, the first restriction of the ATAD’s GAAR is given by the CJEU, whose primary role is being a guardian of the EU treaties guaranteeing the application of the EU law across MSs in accordance with the treaties.19 Hence, an ADM model compatible with the EU legal
system needs to respect the EU treaties as provided by the CJEU case-law, particularly the abuse-doctrine.20 Even though the CJEU takes precedence in comparison to Directives, the analysis of the
ATAD’s GAAR is still relevant as it directs MSs on how to apply the abuse-doctrine.
Thus, in this Section, the aim is to establish what is the framework and relevant criteria that ADM would have to follow in order to set up a model that is compatible with EU law. It begins by outlining how the current ATAD’s GAAR relates with legal certainty and efficiency principles and, in the sequence, it describes the correlation between this GAAR and EU case-law.
2.1. ATAD’s GAAR: legal certainty and efficiency issues
The GAAR enacted by ATAD goes beyond domestic tax law situations. It also covers cases involving more than one MS (cross-border intra-EU), along with cases dealing with MSs and third countries.21
Although the majority of the MSs already had different GAARs,22 no Directive had fully harmonised it
before the ATAD.23 Thus, this Directive tries to address some of the gaps and other loopholes of MS’s
legislation that could lead to mismatches amongst the States,24 besides the loss of revenue to MSs, and
lack of legal certainty to taxpayers.25 Notwithstanding the best intentions and objectives of the ATAD’s
GAAR and despite extensive case-law in the EU against abusive practices,26 the existing decisions and
applicable law are still not capable of reducing the growing number of controversies surrounding its application.
Therefore, the correct application of anti-avoidance provisions is a valuable field of study in need of further clarifications.27 Amidst the EU legal scenario, in demand of a solution and the
harmonisation of these rules across MSs, it is understandable the severe criticism directed at the
18 Oana Popa, ‘An Overview of ATAD Implementation in EU Member States’ (2019) 59(2/3) , European taxation,
pp.120–122, at p.121 <https://research.ibfd.org/collections/et/printversion/pdf/et_2019_02_e2_4.pdf> accessed 13 May 2020.
19 ‘Treaty on European Union and of the Treaty Establishing the European Community: TEC’, OJ C 325 (2002),
33–184, Article 220 <http://data.europa.eu/eli/treaty/tec_2002/oj> accessed 27 April 2020.
20 Błażej Kuźniacki, ‘The C.J.E.U. Case Law Relevant to the General Anti-Avoidance Rule (G.A.A.R.) Under the
Anti-Tax Avoidance Directive (A.T.A.D.)’ (2019) 4(2) , University of Bologna Law Review, pp.261–282, at p.266 <https://bolognalawreview.unibo.it/article/view/10023> accessed 13 May 2020.
21 Council Directive (EU) 2016/1164 of 12 July 2016 laying down rules against tax avoidance practices that
directly affect the functioning of the internal market (ATAD) [2016] OJ L193/1 (n 1), preamble para 11.
22 Only two MSs, Latvia and Slovenia, did not have any kind of GAAR before the ATAD. See Popa (n 18), at
p.121.
23 As noted by Kuźniacki (2019), the GAAR had already been in the EU legal framework within partially
harmonised areas of direct tax law. See Council Directive (EU) 2011/96 of 30 November 2011 on the common system of taxation applicable in the case of parent companies and subsidiaries of different Member States [2011] OJ L345/8, art 1 (2)-(4); Council Directive (EU) 2003/49 of 3 June 2003 on a common system of taxation applicable to interest and royalty payments made between associated companies of different Member States [2003] OJ L157/49, art 5; Council Directive (EU) 2009/133 of 19 October 2009 on the common system of taxation applicable to mergers, divisions, partial divisions, transfers of assets and exchanges of shares concerning companies of different Member States and to the transfer of the registered office of an SE or SCE between Member States [2009] OJ L310/34, art 15 (1) (a).
24 Popa (n 18), at p.121. On the GAAR functionalities see Prats and others (n 13), at p.6.
25 Several aspects of the GAAR and abuse of law doctrine are susceptible to uncertainty as further discussed
under Section 2 and as noted by Prats and others (n 13), 5; 9; 18; 20; and 25.
26 See case-law in Section 2.2.
GAAR.28 The criticism reflects the openness of the GAAR application and lack of objective or explicit
criteria described in the law.29 This GAAR is too broad and, even though its goal is to tackle tax
avoidance, its design also targets tax optimization and tax planning. So, the current GAAR provision may somewhat hinder the objective to efficiently combat tax avoidance and the lack of legal certainty.30
From the stakeholders perspective, tax administrations have limited resources to efficiently access which taxpayers are engaged in abusive tax practices. The taxpayers, on the other side, do not want to extrapolate in their right to have the optimal tax organisation,31 since such actions could lead to
unpredicted higher tax liability if the GAAR rules trigger the reassessment of its business arrangements.32
2.2. ADM conformity with ATAD’s GAAR and CJEU case-law
The ATAD is a basic level of protection rule which can be complemented by MS’s domestic legislation with ‘a higher level of protection for domestic corporate tax bases’.33 MS’s national legislation containing
narrower GAAR provisions are not under our purview, as the harmonised EU view takes precedence in this study.34
The ATAD’s scope is restricted ‘to all taxpayers that are subject to corporate tax in one or more Member States, including permanent establishments in one or more Member States of entities resident for tax purposes in a third country’.35 Under the GAAR’s scope, the ‘corporate tax’ restriction is reinforced
by the wording ‘corporate tax liability’ in art 6 (1).36
Supported by these provisions, one would expect that the GAAR only serves situations where undertakings are concerned, excluding individuals from its scope. In reality, individuals may also be included, as long as MSs opt to broaden the scope of the provision complementing the GAAR with more
28 Michael Lang, ‘BEPS Action 6: introducing an antiabuse rule in tax treaties’ [2014] , WU International Taxation
Research Paper Series, pp.655–664, at p.663; Daniel W Blum, ‘The Proposal for a Global Minimum Tax: Comeback of Residence Taxation in the Digital Era?: Comment on Can GILTI+ BEAT= GLOBE?’ (2019) 47(5) , Intertax, pp.514–522, at p.517
<https://library.ibfd.org/custom/web/SD_PDF/scans/2019/G-I/INTERTAX/5_514-522.pdf> accessed 28 April 2020.
29 See Section 2.1.
30 Kuźniacki, ‘The C.J.E.U. Case Law Relevant to the General Avoidance Rule (G.A.A.R.) Under the
Anti-Tax Avoidance Directive (A.T.A.D.)’ (n 20), at p.263.
31 On the right to optimal tax organization see Section 2.2.
32 Besides higher tax liability derived from the ignored arrangements and re-assessment from the tax
administration, the taxpayer may also incur in high costs with interests and penalties usually levied under MS’s domestic legislation to punish abusive practices.
33 Council Directive (EU) 2016/1164 of 12 July 2016 laying down rules against tax avoidance practices that
directly affect the functioning of the internal market (ATAD) [2016] OJ L193/1 (n 1), art 3, and preambles para 3, 16. See Daniël Smit, ‘Chapter 12 - The Anti-Tax-Avoidance Directive (ATAD)’ in P.J Wattel and B.J.M Terra (eds), European tax law (Seventh edition, Abridged student edition. Wolters Kluwer 2019), pp.245–276, at p.247.
34 MS’s domestic GAAR may not be relevant for the delineation of the EU harmonised abuse framework but the
inverse is not equally applicable. The ATAD transposes the developed CJEU concept of the abuse-doctrine into MSs, forcing MS’s hand on submitting to sovereignty even on internal tax matters related to the abuse concept. Luc de Broe and Dorien Beckers, ‘The General Anti-Abuse Rule of the Anti-Tax Avoidance Directive: An Analysis Against the Wider Perspective of the European Court of Justice’s Case Law on Abuse of EU Law’ (2017) 26(3) , EC Tax Review, pp.133–144, at p.140
<https://library.ibfd.org/custom/web/SD_PDF/scans/2017/E-F/ECTARE/3_133-144.pdf> accessed 5 May 2020.
35 Council Directive (EU) 2016/1164 of 12 July 2016 laying down rules against tax avoidance practices that
directly affect the functioning of the internal market (ATAD) [2016] OJ L193/1 (n 1), art 1.
36 Prats and others (n 13), at p.16. Also, there is room to discuss if the ATAD’s scope is restricted to those
taxpayers that are liable to tax or those who are simply subject to tax. See Smit (n 33), sec 12.2.3. Cf. Prats and others (n 13), at p.16. See C-448/15 Wereldhave Belgium and others [2017] EU:C:2017:180, 1.
restrictive conditions,37 which a feature of the ATAD ‘minimum standard’ provision.38 In this context, the
CJEU case-law confirms its competence to analyse such situations. It accepted the competence to decide matters related to harmonised EU law when the internal legislation of MSs regulates purely internal situations adopting the same solutions as the EU law in order.39
The tax abuse-doctrine also broadens the scope of the anti-avoidance provisions. It does not restrict rights or advantages by legal source, as seen in Cussens:40 ‘the principle that abusive practices are prohibited is applied to the rights and advantages provided for by EU law irrespective of whether those rights and advantages have their basis in the Treaties (...) in a regulation (...) or in a directive’.41
Once determined the general scope of the ATAD’s GAAR, it is necessary to assess the GAAR tests under this Directive. Accordingly, art 6 reads as follows:
Article 6 - General anti-abuse rule
1. For the purposes of calculating the corporate tax liability, a Member State shall ignore an arrangement or a series of arrangements which, having been put into place for [i] the main
purpose or one of the main purposes of obtaining a tax advantage that [ii] defeats the object or purpose of the applicable tax law, are [iii] not genuine having regard to all
relevant facts and circumstances. An arrangement may comprise more than one step or part. 2. For the purposes of paragraph 1, an arrangement or a series thereof shall be regarded as non-genuine to the extent that they are not put into place for valid commercial reasons which reflect economic reality.
3. Where arrangements or a series thereof are ignored in accordance with paragraph 1, the tax liability shall be calculated in accordance with national law.42
If the conditions under art 6 (1) are met, the arrangements will be ignored, and the tax liability shall be calculated in accordance with national law, as required by art 6 (3).43 MSs may ignore said arrangement
when the three elements emphasised on art 6(1) are concurrently fulfilled.44 Said elements can be
abbreviated as (1st test) motive test; (2nd test) defeating object or purpose test; and (3rd test) non-genuine
test or artificiality test.45
37 Prats and others (n 13), footnote 76. 38 See n 18.
39 C-28/95 Leur-Bloem [1997] ECLI:EU:C:1997:369, paras 32-34. See also C-48/07 Les Vergers du Vieux
Tauves [2008] ECLI:EU:C:2008:758, para 27; C-439/07 and C-499/07 KBC Bank and Beleggen, Risicokapitaal, Beheer [2009] ECLI:EU:C:2009:339, paras 59-60.
40 C-251/16 Cussens and Others [2017] EU:C:2017:881, 1.
41 In ibid 7, referencing to case-law concerning Treaties. C-33/74 Van Binsbergen v Bedrijfsvereniging voor de
Metaalnijverheid [1974] ECLI:EU:C:1974:131, 1, para 13; C-212/97 Centros [1999] ECLI:EU:C:1999:126, para 24.
42 Council Directive (EU) 2016/1164 of 12 July 2016 laying down rules against tax avoidance practices that
directly affect the functioning of the internal market (ATAD) [2016] OJ L193/1 (n 1), art 6(emphasis added).
43 This norm is in line with the CJEU practice, where '[T]he transactions involved in it must be redefined so as to
re-establish the situation that would have prevailed in the absence of the transactions constituting that abusive practice'. C-103/09 Weald Leasing [2010] ECLI:EU:C:2010:804, paras 48 and 52.
44 Literature diverges on how to divide and label art 6 (1) criteria, still there is a consensus that all these
elements are essential in order to trigger the GAAR. See Prats and others (n 13), sec 3.1.3; Smit (n 33), sec 12.5.3; Kuźniacki, ‘The C.J.E.U. Case Law Relevant to the General Anti-Avoidance Rule (G.A.A.R.) Under the Anti-Tax Avoidance Directive (A.T.A.D.)’ (n 20), sec 1.1.
45 Article 6 (1) only provides the wording ‘not genuine’, but the CJEU case-law mostly refers to ‘artificiality’ test.
The CJEU was not presented with cases where it had to analyse whether these linguistic differences between the CJEU case-law and art 6 (1) are relevant. Smit (n 33), at p.272. Cf. Prats and others (n 13), at p.20. Nevertheless, there are predictions that the CJEU will neglect wording divergences. See Smit (n 33), at p.270. The very ATAD’s Directive proposal understood these terms as equivalents. Proposal for a Council Directive laying down rules against tax avoidance practices that directly affect the functioning of the internal market ]/1, recital 12.
Similarly to the ATAD, the CJEU case-law46 also uses these tests to check whether
discriminatory domestic legislation is proportional to the purpose of tackling tax abuse under the fundamental freedoms.47 Decisions wording may slightly vary; still, the CJEU uses these criteria to curb
abuse under primary and secondary EU law.48 Therefore, the justification rationale can be replicated
similarly.49
In Van Binsbergen.50 The CJEU understood that legislation may establish discriminative restriction even against the EC freedoms51 if, first, the domestic legislation objective is to tackle tax
abuse and, second, it provides proportional measures to identify abusive arrangements.
The abuse concept progressed in Emsland-Stärke,52 the CJEU clarified under which conditions there was a justification for the breach of the EC freedoms and whether the national legislation was proportional to that end. Emsland-Stärke was a milestone case to set the two-pronged test, which is the combination of the artificiality and motive tests.53 In that case, where: ‘[D]espite formal observance of
the conditions laid down by the Community rules, the purpose of those rules has not been achieved’.54
This test was the next pillar on the fight against tax abuse. Its concept was integrated into the CJEU case-law and became broadly applicable.55
As in Van Binsbergen, the CJEU followed the same rationale in the milestone case Cadbury Schweppes,56 where it decided on whether the UK discriminative provision on Controlled Foreign
Companies (CFC’s) is against the freedom of establishment. In that case, the CJEU held that discrimination was justifiable by tax abuse prevention.57 More recently, the CJEU confirmed that the
tested national legislation dealing with anti-abusive tax provisions is valid only if it sets objective criteria that assess arrangements that lack economic reality.58 As observed in the ATAD’s art 6 (2), ‘valid
commercial reasons’ and ‘reflect economic reality’ shed further light on what arrangements may be ‘non-genuine’. Besides, this wording is also in conformity with the CJEU case-law.59 The wording in art 6 (2),
called ‘to the extent approach’ also allows that only part of the arrangements, the artificial parts, to be ignored under the GAAR.60
The CJEU case-law development confirmed that taxpayers’ must concurrently trespass the GAAR tests to be considered abusive. Accordingly, such arrangements would be ignored ‘for the
46 See Section 2.2.1.
47 Whereas it is essential to set what consists the pillars of the abuse-doctrine in the CJEU, the intent is to draw
the concept based only on the milestones case-law, as the analysis of the entirety of the case-law would be impractical for the purposes set on this study.
48 C-14/16 Euro Park Services [2017] ECLI:EU:C:2016; C-6/16 Eqiom and Enka [2017] ECLI:EU:C:2017:641,
para 30; Joined Cases C-504/16 and C-613/16 Deister Holding [2017] ECLI:EU:C:2017:1009, para 60.
49 Kuźniacki, ‘The C.J.E.U. Case Law Relevant to the General Avoidance Rule (G.A.A.R.) Under the
Anti-Tax Avoidance Directive (A.T.A.D.)’ (n 20), at p.271. See also Maarten F de Wilde, ‘The ATAD's GAAR: A Pandora's Box?’ in Pasquale Pistone and Dennis Weber (eds), The Implementation of Anti-BEPS Rules in the EU: A Comprehensive Study (Books IBFD. IBFD 2018), p.13, at p.2.
50 Van Binsbergen v Bedrijfsvereniging voor de Metaalnijverheid (n 41). 51 ibid para 13.
52 Case C-110/99 Emsland-Stärke [2000] EU:C:2000, 1. 53 ibid paras 52-53.
54 ibid para 52.
55 Prats and others (n 13), at p.8.
56 C-196/04 Cadbury Schweppes [2006] ECLI:EU:C:2006. 57 ibid para 51.
58 Eqiom and Enka (n 48) para 30. Its rationale was based on the following case-law Cadbury Schweppes (n 56)
para 55; 524/04 Test Claimants in the Thin Cap Group Litigation [2007] ECLI:EU:C:2007:161, para 74; C-318/10 SIAT [2012] ECLI:EU:C:2012:415, para 40.
59 See Cadbury Schweppes (n 56) para 55; Test Claimants in the Thin Cap Group Litigation (n 58) para 74;
C-135/17 X (Sociétés intermédiaires établies dans des pays tiers) [2019] ECLI:EU:C:2019:136, para 84.
purpose of calculating the corporate tax liability’ and being re-established to conform with the underlying transaction.61 In a good part, the ATAD’s GAAR tried to assimilate the general concept behind the CJEU
abuse-doctrine. Still, it did not encompass further detail into the legislation on how to objectively appreciate which situations fall within the GAAR rule.
2.3. CJEU case law and the ATAD’s GAAR tests
The ATAD’ GAAR hazy terms lacking further details endorsed the claims of uncertainty and a high degree of discretion under art 6 tests. In this Section, it is discussed how the CJEU case-law counterbalance the ATAD’s GAAR issues. The analysis brings cases of abuse and indicates the critical fact patterns and criteria constructed by the abuse-doctrine which MSs – consequently tax administrations – must recognise for an approach in conformity with said case-law.
This closer view of each test is crucial, since failing to properly apply the tests triggers a risk of courts decisions overturning the tax authorities’ reassessments. Hence, tax administrations should do their utmost, so their tax assessments conform to CJEU criteria.
2.3.1. Motive test
The 1st test, motive test, extracted from ATAD’s art 6 (1), evaluates whether a taxpayer put an
arrangement into place for the main purpose or one of the main purposes of obtaining a tax advantage. The motive test holds in its centre the analysis of the tax advantage and the taxpayer’s purpose behind the arrangement.
Firstly, the tax advantage analysis depends on assessing objectively whether there is a reduction in the taxpayer corporate tax liability.62 Accordingly, the liability comparison must be made
'between the transaction in question and another transaction (including doing nothing), which does not provide the benefit'.63
Under the second criteria, there are two ways to determine the taxpayer’s purpose. Either the taxpayer is blunt and assumes its legitimate intention to pursue a tax advantage,64 or the tax
administration proves the arrangement purposes is to pursue a tax advantage. Under the latter analysis, the taxpayer’s ‘purpose’ can be assessed based on objective criteria, as provided by CJEU case-law:65
elements like the absence of economic substance, the derogation from the arm’s length standard, or the existence of a transaction result which is foreseeably negative but for the tax effects. These criteria are also relied on to analyse the artificiality test.66 In Cadbury Schweppes,67 the Court understood that
the undertaking existence would have to be assessed based ‘in terms of premises, staff, and equipment’.68 Whereas in the Test Claimants in the Thin Cap Group Litigation,69 the arrangement would be disregarded if the dividends distributed by the subsidiary were not at arm’s length.70
61 Repeating the correspondence between ATAD’s article 6 (3) and CJEU case-law as seen, e.g., in Weald
Leasing (n 43) paras 48 and 52.
62 Prats and others (n 13), at p.19.
63 Frederik Zimmer, ‘In Defence of General Anti-Avoidance Rules’ (2019) 73(4) , Bulletin for international
taxation, pp.218–226, at p.224 <https://research.ibfd.org/#/doc?url=/document/bit_2019_04_o2_1> accessed 5 July 2020.
64 Weber (n 15), at p.252. In these cases, taxpayers would claim the object or purpose test was not breached.
Smit (n 33), at p.272.
65 C-255/02 Halifax [2006] EU:C:2006:121, 1, para 75. See Weber (n 15), at p.252. 66 Smit (n 33), at p.272.
67 Cadbury Schweppes (n 56) para 67. 68 ibid.
69 Test Claimants in the Thin Cap Group Litigation (n 58). 70 ibid para 80.
It could be said that the taxpayer’s purpose and subjective intention ‘can be derived from the artificiality’.71 Indeed, the artificiality criterion links the motive and artificiality test,72 but this connection is
limited and not a requirement per se. That is, there is a compelling argument on the existence of a taxpayer’s purpose when a situation is found to have both a tax advantage and an artificial arrangement,73 yet it is only a presumption. The taxpayer can still present valid reasons to rebut this
presumption,74 and the GAAR would not apply where non-tax reasons take precedence against the tax
reasons.75
The motive test constituents’ elements, tax liability and artificiality, are criteria objective in nature. This test complexity is reduced thanks to two factors. First, tax administrations can objectively assess the tax advantage criteria, as it solely depends on the tax administration capabilities to proceed with the tax calculation, which is within its purpose and abilities. Second, the artificiality criterion might in certain situations be able to link the 1st and 3rd tests, reducing the burden to prove the taxpayer’s
motive, since if there are an artificial arrangement and a tax advantage the 1st and 3rd tests are likely
fulfilled. Moreover, there is a high number of inputs from the CJEU on what consists an artificial arrangement,76 reducing discretion for domestic tax courts and tax administrations; as well as promoting
certainty for assessments based on fact patterns and criteria extracted from case-law. In brief, the conjunction of the GAAR and the CJEU case-law thoroughly delineates these two factors, paving the way for the motive test automation under ADM models.
2.3.2. Defeat of object or purpose test
The 2nd test, ‘defeat of object or purpose’ test, ‘has to be analysed in depth with the view to identify the
object or purpose of the law and not only mere text’.77 This translates into having the text as the
cornerstone but also accounting for the context and purpose analysis.78
The 2nd test is fulfilled where, ‘[D]espite formal observance of the conditions laid down by
Community law’,79 said corporate tax law is defeated by obtaining a tax advantage via non-genuine
arrangement.
The purpose behind the 2nd test is to hinder the use of the EU fundamental freedoms with the
intent to defeat national legislation, as seen in Cadbury Schweppes.80 The question of what constitutes
defeat or ‘circumvention or exploitation of the corporate tax law’ must be answered either by the national courts, in case the legislation at stake is not of harmonised nature; or by the CJEU, if it is a provision of harmonised EU law.81 The CJEU competence would be expected, e.g., ‘[W]here a Member State has a
71 Weber (n 15), at p.253.
72 ibid, at p.252. See also Prats and others (n 13), at p.20. 73 Prats and others (n 13), at p.19.
74 Halifax (n 65) para 75. Cf. C-255/02 Halifax [2006] ECLI:EU:C:2005:200, Opinion of Advocate General
Poiares Maduro, 1. See also Prats and others (n 13), at pp.9–10.
75 Smit (n 33), at p.272. 76 See Section 2.2.3.
77 Broe and Beckers (n 34), at p.142.
78 As extracted from Alta Energy international case-law. Canada v. Alta Energy Luxembourg S.A.R.L.
(2020)A-315-18 (Federal Court of Appeal Canada 43). See Jonathan Schwarz, ‘Alta Energy: Treaty shopping is no abuse’ Kluwer International Tax Blog (27 February 2020), p.3 <http://kluwertaxblog.com/2020/02/27/alta-energy-treaty-shopping-is-no-abuse/?doing_wp_cron=1594103036.6193170547485351562500> accessed 5 June 2020.
79 Cadbury Schweppes (n 56) para 64. See also Emsland-Stärke (n 52) paras 52-53.
80 As postulated in the case, ‘nationals of a Member State cannot attempt, under cover of the rights created by
the Treaty, improperly to circumvent their national legislation. They must not improperly or fraudulently take advantage of provisions of Community law’. Cadbury Schweppes (n 56) para 35. See also Centros (n 41) para 24; Van Binsbergen v Bedrijfsvereniging voor de Metaalnijverheid (n 41) para 13.
GAAR in place when the ATAD becomes effective (1 January 2019) and continues to apply that GAAR’.82
Therefore, the CJEU case-law plays an essential role at dissecting which abuse situations go against MS’s domestic GAARs. Nevertheless, it should be accentuated that most cases are not comparable in the perspective of different MS’s domestic legislation. That is, identical GAARs from different MSs may have different outcomes from the object or purpose test. For example, where one MS bases its corporate tax law on private law, whereas the other MS strongly relies on the arrangement’s economic reality:
The object or purpose with respect to a tax advantage in the first case may thus be to grant the advantage whenever the private law conditions are fulfilled; while, in the latter case, a tax advantage will only be legitimately obtained if the arrangement is economically the one that the law intends to benefit.83
To proceed with a complete analysis of the corporate tax law objective or purpose, the courts should bear in mind the following elements. First, in order to fulfil the 2nd test, defeating either the ‘objective’ or
the ‘purpose’ would be enough. Second, one should question which corporate tax should be analysed: the one that grants or the one that denies the advantage? Even though Alfredo G Prats and others interpret, under the wording of this provision, that the corporate tax law should be assessed altogether,84
they concede that the 2nd test is not sufficiently clear to provide an appropriate analysis on which
provision must be under purview. Third, changes in the corporate tax law may present another layer of complexity if they result in fluctuations on its purpose or objective. In principle, the correct interpretation is most likely that legislation should be assessed when the arrangements ‘have been put into place’.85
To clarify which circumstances qualify for the application of the 2nd test, the CJEU identified
what should not be considered the defeat of the corporate tax law. That would be the case, e.g., where the arrangement profited from the mismatch of different MSs legislation.86
The major setback behind this 2nd test is that the legislation’s purpose and objective need to be
thoroughly assessed. Besides, this analysis cannot be done in the abstract.8788 The objective analysis
of the corporate tax law may not be so easy,89 firstly, due to a lack of historical documentation or simply
because the legislation motives are non-existent, or became irrelevant.90 Secondly, some aspects
behind the 2nd test are still unclear, e.g., determining which should be the tested legislation and what is
the tested legislation temporal scope.
The 2nd test threshold also has a higher chance of being met where the 1st and the 3rd tests are
also met. Due to the intrinsic link between the tests, the defeat of objective or purpose is likely to happen when the taxpayer also has artificial arrangements generated by tax motives.
Overall, the 2nd test complexity might be higher than expected. On the one hand, the role of
CJEU to set the standard for MS’s GAAR positively harmonises what situations defeat the test. On the
82 ibid.
83 Prats and others (n 13), at p.17.
84 The ATAD’s art 6 (1) wording indicates such historical interpretation should be pursued. ibid, at p.19. 85 ibid, at p.20.
86 Cadbury Schweppes (n 56) para 49. 87 Smit (n 33), at p.273.
88 ibid.
89 Prats and others (n 13), at p.18.
90 This circumstance is realistic due to economic discrepancies when the legislation was created. Broe and
other hand, the 2nd test complex application must be considered,91 foremost, the necessity to a detailed
analysis of national corporate legislation. The fact that MS’s legislation varies as much as the arrangements may also raise the degree of complexity which brings into question the feasibility to automate this test under an ADM model. Nevertheless, these constraints are repairable: future court decisions may delve into this test, developing solid bases that are sufficiently certain, forming a smoother path to the 2nd test automation.
2.3.3. Artificiality test
The 3rd test, non-genuine test, extracted from ATAD bear similarities to the ‘wholly artificial arrangement’
corollary from the abuse-doctrine.92 These terms are equivalents for the purpose to assess the CJEU
abuse concept and its repercussions after the ATAD.93
The artificiality test is composed of several key elements. It identifies (1) ‘an arrangement or a series of arrangements’ that (2) ‘are not genuine having regard to’ (3) ‘all relevant facts and circumstances’. An arrangement (4) ‘may comprise of more than one step or part’ and it shall be regarded as non-genuine to the extent that they are (5) ‘not put into place for valid commercial reasons’ which (6) ‘reflect economic reality’.94
The fact that this test must take into account ‘all facts and circumstances’ in order to validate its ‘commercial reasons’ reflecting ‘economic reality’, marks this test as an objective test.95 Albeit the
elements behind the 3rd test make it in nature an objective test, the 3rd test has a complex layer of
application. An arrangement can only be found to be ‘non-genuine’ in reference to the corporate tax law being tested. Meaning that based on this test, similar arrangements may have different outcomes if the reference corporate tax law is different. In that regard, abusive arrangements are expected to be found when all formal requirements by domestic legislation were fulfilled, but there is an absence of valid commercial reasons, which is ascertained by its economic reality.
The scope of which (1) ‘arrangements’ may be relevant is restricted to those arrangements that confer a tax advantage to the taxpayer promoting a tax advantage, that is, lowering its corporate tax as seen in the 1st test description.96 In line with this scope, finding what consists (2) ‘not genuine’ must take
into account only (3) ‘all relevant facts and circumstances’ in consideration to these arrangements. To assess if taxpayers did not set the arrangements for (5) ‘valid commercial reasons’, MSs may consider all valid economic reasons, including financial activities97 or ‘any other objective factors
which are ascertainable by third parties’.98 The comprehensiveness of ‘all facts and circumstances’ is
91 Prats and others (n 13), at p.19; Jakob Bundgaard and Peter K Schmidt, ‘Uncertainties Following the Final EU
Anti-Tax Avoidance Directive’ [2016] , Kluwer International Tax Blog, at p.3
<http://kluwertaxblog.com/2016/10/17/uncertainties-following-final-eu-anti-tax-avoidance-directive/>.
92 The wording ‘wholly artificial arrangement’ first appeared in the case Imperial Chemical Industries, being
repeated in several other cases on tax avoidance, as noted by Kuźniacki, ‘The C.J.E.U. Case Law Relevant to the General Anti-Avoidance Rule (G.A.A.R.) Under the Anti-Tax Avoidance Directive (A.T.A.D.)’ (n 20), at p.275. Based on the search form portal of CJEU case-law the wording ‘wholly artificial arrangement’ was used in 64 cases. CURIA, Search Form: InfoCuria Case-Law <https://curia.europa.eu/jcms/jcms/Jo2_7044/en/> accessed 20 May 2020.
93 See n 45.
94 Council Directive (EU) 2016/1164 of 12 July 2016 laying down rules against tax avoidance practices that
directly affect the functioning of the internal market (ATAD) [2016] OJ L193/1 (n 1), Article 6 (1) (2).
95 Prats and others (n 13), at p.20. 96 See Section 2.3.1.
97 As noted by Smit (n 33), at p.271 when analysing the Council Directive (EU) 2016/1164 of 12 July 2016 laying
down rules against tax avoidance practices that directly affect the functioning of the internal market (ATAD) [2016] OJ L193/1 (n 1), preamble para 3.
98 Smit (n 33), at p.271.See Test Claimants in the Thin Cap Group Litigation (n 58) para 74; Halifax (n 65) para
relatively shortened by the CJEU, as its rulings indicate a series of facts patterns that should take particular relevance by the referring court analysis.99
In the end, to assert the existence of (2) ‘non-genuine’ arrangements, the (3) ‘facts and circumstances’ that lead to the formal compliance with the corporate tax law mustn’t (6) ‘reflect the economic reality’.
The CJEU addressed in several cases the question of which (3) ‘facts and circumstances’ may appoint an arrangement that does not (6) ‘reflect the economic reality’. However, many of these rulings are not necessarily based on the corporate tax law, scope of the ATAD’s GAAR. Even if these cases deal with other fields of tax law, e.g., the value-added tax (VAT), their insights may provide crucial fact patterns to apply the artificiality test.
For example, the Halifax case directly dealt with arrangements that were constituted with the intent to attain the conditions to apply for VAT deduction. Even if that case does not concern corporate tax, it sheds light on the (3) ‘facts and circumstances’ to take into consideration for the analysis of (6) ‘economic reality’. In Halifax, the court concluded that it should be taken into account: ‘the links of a legal, economic and/or personal nature between the operators involved in the scheme for reduction of the tax burden’.100 Consequently, this conclusion made in Halifax does not carry a positive nor negative
judgement on it. Meaning that the links between operators involved only narrows the scope of what is relevant to determine the arrangements artificiality. Such links are not by themselves indicators of non-genuine arrangement.
Thus, fact patterns should be split between those that only indicate relevant (3) ‘facts and circumstances’ to be analysed and that narrow the scope of the ADM, to those that actually carry a positive or negative value, that is, they indicate when the arrangements do not (6) ‘reflect economic reality’. A fact pattern that carries a positive value confirming non-genuine arrangements would be, e.g., the existence of a ‘letter-box’ company, not carrying any business in a given State.101
Furthermore, one could question whether the artificiality test is fulfilled when artificial structures conceive only part of the transaction. In that regard, even if only part of the arrangement(s) is artificial, the artificiality test is fulfilled up to that part. The wording present in the ATAD’s art 6 (2), ‘to the extent that’, confirms this rationale by which the artificiality test, and the GAAR, would be applicable even in the presence of only partially ‘non-genuine’ arrangement(s).102 In parallel, this wording also restricts tax
administration’s power, meaning they can only ignore arrangements that were not genuine, guaranteeing that the remaining transactions stay unscratched. This interpretation suits the cases of free transfer of profits where, in that context, the CJEU recognised that no artificial arrangement existed, but still, it understood the arrangements were abusive and, in that regard, should be assessed. Also, in Oy AA103and SGI,104 the transactions that generated a shift in the profit of the undertakings were real
but deemed not in conformity with the arm’s length principle and, therefore, abusive.105
99 See Table 1 examples of the facts indicated by the CJEU to refer to ‘all facts and circumstances. 100 Halifax (n 65) para 81.
101 C-341/04 Eurofood IFSC [2006] ECLI:EU:C:2006:281, para 35. 102 Prats and others (n 13), at p.20.
103 C-231/05 Oy AA [2007] ECLI:EU:C:2007:439, 1. 104 C-311/08 SGI [2010] ECLI:EU:C:2010:26.
105 ibid paras 67-69; Oy AA (n 103) paras 58 and 59. Even though an arrangement that distances from the arm’s
length principle may be a relevant indicator of abuse, there are commercial justifications that may be accepted to such an arrangement. That would be the case where it exists commercial reasons which could result from the taxpayers status as a shareholder interested in the financial results of the invested company. C-382/16 Hornbach-Baumarkt [2018] ECLI:EU:C:2018:366, para 57.
When addressing whether a particular taxpayer factually incurred in artificial arrangements, the CJEU refers to the national court to make the assessment based on objective factors laid by third parties. Nevertheless, the CJEU tries to point out which criteria should be used by the referring court to assess the artificiality. In several cases, the CJEU determined (3) ‘relevant facts and circumstances’ as well as in which situations they most likely do not (6) ‘reflect economic reality’.106 Under this overview, due to
the numerous inputs provided by the CJEU, it is expected that this 3rd test, the artificiality, can be more
prone to the automatisation process for the ADM model.
Notwithstanding the divergences between the three tests, it should be contented that, although literature and case-law split them, they are also intertwined in nature. As seen in the 2nd test, meeting
the requirements for the 1st and 3rd test, positively influences meeting the 2nd test requirements, which
means that the interpretation of each test depends upon the other. Likewise, the understanding of artificiality is only possible by comprehending what defeats the object or purpose of the applicable law; otherwise, the artificiality is non-existent if a particular arrangement is within the law purposes and objectives.
3. What are the limitations for ADM application in the ATAD’s GAAR context?
Section 2 accomplished the first step, providing the outline for what defeats the tests put by the ATAD’s GAAR and the CJEU abuse-doctrine. The legal framework established in Section 2 intends to set guidelines of the decision-making rationale that tax administrations must follow, so their ADM models conform with EU law. The GAAR tests were analysed in terms of its objectiveness and complexity. The more fact patterns and detailed criteria integrated into ADM models, the more each test could appropriately integrate into automated ADM models.
Private actors and Governments have integrated ADM involving AI107 in numerous forms in our
life. ADM provides the right tool for advances in efficiency, but also of discrimination. For example, it can assess when individuals are more likely to commit abuse by receiving undue subsidies from the State.108
The interaction between AI and law has been under scrutiny as early as 1958.109 However, as
noted by Kuźniacki,110 the literature deemed a premature analysis at the time.111 Only recently its
application on the law area thrived. States started further developing its internal assessing methods allowing the more efficient and equal collection of taxes.112
The following step is to set the other restrictions to the ADM model. This includes determining the legal boundaries of an ADM model applied to the tax administration context. Next, it includes checking what are the ADM model technical requirements and capabilities under the ATAD’s GAAR context.
3.1. Tax administration legal boundaries to ADM
ADM systems are already a reality in the tax management sector in the majority of UN Member States.113
The Netherlands already integrates the use of ADM in its tax system, leading to an increase in decisions quality, primarily when related to complex calculations.114
While ADM measures shine for its efficiency,115 the following subsections assess crucial legal
107 Due to the unstructured concept of AI, there is no agreement on its definition. Sofia Samoili and others, AI
watch: Defining Artificial Intelligence : towards an operational definition and taxonomy of artificial intelligence (EUR vol 30117, Publications Office of the European Union 2020), p.90 , at p.7. For this thesis, the AI concept used follows the AI HLEG concept explained in Section 3.2.2.
108 This case is under scrutiny in the Netherlands, where the State started a project in 2014, the algorithm SyRI,
which was used by local authorities to identify ‘lists of people suspected of some form of housing or social security fraud’. DutchNews, ‘Government’s fraud algorithm SyRI breaks human rights, privacy law’
DutchNewsnl (13 May 2020), p.2 <https://www.dutchnews.nl/news/2020/02/governments-fraud-algorithm-syri-breaks-human-rights-privacy-law/> accessed 13 May 2020.
109 Lucien Mehl, ‘Automation in the legal world’ [1958] , National Physical Laboratory, pp.755–787
<https://aitopics.org/download/classics:97D0F0CA> accessed 13 May 2020.
110 Błażej Kuźniacki, ‘The Marriage of Artificial Intelligence and Tax Law: Past, Present, and Future’ [2019] ,
Kluwer International Tax Blog, p.9 <http://dx.doi.org/10.2139/ssrn.3323867> accessed 13 May 2020.
111 Richard L SHUEY, ‘Impact of VLSI on Artificial Intelligence’, VLSI Electronics Microstructure Science (Elsevier
1983), pp.333–353, at pp.334–337 <https://doi.org/10.1016/B978-0-12-234107-6.50013-2> accessed 11 April 2020.Cf. Edwina L Rissland, Kevin D Ashley and Ronald P Loui, ‘AI and Law: A fruitful synergy’ (2003) 150(1-2) , Artificial Intelligence, at p.6 <https://www.journals.elsevier.com/artificial-intelligence> accessed 11 February 2020.
112 159 UN MSs already put to use ICTs systems for tax managing, World Bank (n 10), at p.6. 113 ibid.
114 Marga Groothuis, ‘Applying ICTs in Juridicial Decision Making by Government Agencies’, Encyclopedia of
Digital Government (IGI Global 2007), pp.87–96, at p.94 <http://hdl.handle.net/1887/13567> accessed 13 May 2020.
115 Eugenia Politou, Efthimios Alepis and Constantinos Patsakis, ‘Profiling tax and financial behaviour with big
data under the GDPR’ (2019) 35(3) , Computer Law & Security Review, pp.306–329, at p.323 <https://www.sciencedirect.com/science/article/pii/S026736491830133X> accessed 13 May 2020.
boundaries to ADM based decisions at the hands of tax administration and then indicates other general boundaries specific to the application of anti-avoidance provisions.
3.1.1. ADM decisions legality
The first legal boundary raises the question of whether tax administrations can base its decisions on ADM. This Section dissects some of the legal limits in government utilities of data automation process of and decision making.
MSs recently introduced the General Data Processing Regulation (GDPR),116 which is a modern
protection law against harms and constraints from the misuse of data. This Directive restricts its scope to individuals.117 At first sight, the GDPR scope would not be the same as ATAD’s, in the sense that the
latter only concerns corporate taxpayers. However, as observed in Section 2.1, MSs may choose to include individuals under the ATAD scope, which, on its turn, enables the GDPR to protect individuals reached by the GAAR. On one side, if the GDPR does not cover undertakings, on the other, the principles arisen from this Directive could also be a parameter in the future to deter the use of data against undertakings.
Even though the GDPR is not the only taxpayer’s safeguard against unlawful tax proceedings,118
this Directive is the only that addresses the ADM process explicitly. The provisions under GDPR stipulates a series of requirements for lawful ADM application. They involve, inter alia, the right to explanation under arts 12 to 15;119 and the right to have a decision not rendered solely by automated
processing under art 22.120
The right to explanation grasped from GDPR arts 12 to 15 is especially sensitive. This provision only confirms and regulates the right that EU individuals have to a fair trial under art 6 of the European Convention for the Protection of Human Rights and Fundamental Freedoms (ECHR).121 The European
Court of Human Rights concluded that tax law matters might also be included under the ECHR art 6 scope.122 To admit tax law cases under its purview, ‘the Court in particular took into account (...) [if] the
surcharge was imposed under a general rule with both a deterrent and a punitive purpose.’123
In Poland, an exemplar case of measure against tax schemes is the Information Technology System of the Clearing House (STIR). That ADM system is meant to combat and avoid carousel frauds under the VAT framework.124 The system infers that certain taxpayers have a high risk of committing
frauds. However, from the taxpayers perspective, the inner system workings are unclear and non-transparent. The consequence for taxpayers is the potential blockage of their bank accounts. Hence, this non-transparent system feature, aggregated with the severe consequences, should trigger the right
116 Council Regulation (EU) 2016/679 of 27 April 2016 on the protection of natural persons with regard to the
processing of personal data and on the free movement of such data, and repealing Directive 95/46/EC (General Data Protection Regulation): GDPR [2016] OJ L 119/1.
117 ibid, arts 1 and 2. 118 See Section 3.1.2.
119 See Council Regulation (EU) 2016/679 of 27 April 2016 on the protection of natural persons with regard to the
processing of personal data and on the free movement of such data, and repealing Directive 95/46/EC (General Data Protection Regulation): GDPR [2016] OJ L 119/1 (n 116), arts 13 (2) (f), art 15 (1) (h).
120 ibid, art 22.
121 European Convention on Human Rights, as amended by Protocols Nos. 11 and 14 supplemented by
Protocols Nos. 1, 4, 6, 7, 12, 13 and 16 1952, ECHR (Council of Europe), art 6.
122 Guide on Article 6 of the European Convention on Human Rights - Right to a fair trial (criminal limb) 30 April
2019 (European Court of Human Rights), para 33.
123 ibid. See also Aleksandra Bal, ‘Ruled by Algorithms: The Use of ‘Black Box’ Models in Tax Law’ (2019) 95(12)
, Tax Notes International, pp.1159–1165, at p.1163.
