GAAR at a Crossroads – Part II: The ‘Tax Benefit’ Mechanism and Battling Counterfactuals

By Dr Hendri Herbst & Dewald Pieterse

Introduction

In the first instalment of this series, we traced the development of the South African general anti-avoidance rule from its origins under section 103(1) of the Income Tax Act, where the sole or main purpose test was assessed through a subjective lens, to the restructured provisions of sections 80A to 80L, which introduced a framework ostensibly governed by the objective characteristics of the arrangement itself. We noted that the prolonged absence of judicial consideration under the new GAAR allowed the South African Revenue Service’s (SARS) enforcement theories to develop largely unchecked. As the Constitutional Court deliberates on the merits of Absa Bank Limited v Commissioner for the South African Revenue Service (hereinafter Absa Bank), those theories now face their most significant judicial test.

This second article moves from the historical overview to the substance of the litigation in Absa Bank itself. The aspects of the GAAR inquiry relevant to Absa Bank are two distinct but interlocking questions. The first, to be examined in Part III, concerns the definition of a ‘party’ to an avoidance arrangement and thus determines who may be assessed. The second, which is the subject of this article, concerns the definition of a ‘tax benefit’ and thus determines the basis on which liability arises. Below, we will unpack the mechanics of the tax benefit enquiry, dissect the anatomy of the controversial Macquarie structure, and interrogate the profound divergence between the two schools of thought currently vying for judicial endorsement.

The factual matrix: the anatomy of the Macquarie structure

To contextualise the different approaches to the ‘tax benefit’ enquiry, it is necessary to survey the intricate structure that provoked SARS’s ire and gave rise to a GAAR assessment against Absa. The Absa Bank litigation arises from a series of preference share subscription agreements entered into between 2013 and 2015, involving approximately R1.9 billion. While these appeared to be standard ‘back-to-back’ funding arrangements from Absa’s perspective, SARS contends they were merely the visible funding leg of a broader, ‘impermissible avoidance arrangement’ orchestrated by the Macquarie Group.

The ‘arrangement’ comprised the following sequence of steps:

1. Absa (and its subsidiary United Towers) subscribed for preference shares in a South African company, PSIC 3, expecting tax-exempt dividends.
2. PSIC 3 used these funds to subscribe for preference shares in another South African entity, PSIC 4.
3. PSIC 4 made a capital contribution to an offshore entity, the D1 Trust, incorporated in the Isle of Man.
4. The D1 Trust lent capital back to Macquarie Securities South Africa (MSSA) via floating rate interest-bearing notes. The trust then used this interest income to acquire an interest stream arising from Brazilian government bonds.
5. Under the South Africa-Brazil Double Taxation Agreement, the interest on these bonds was effectively tax-exempt in South Africa.
6. The D1 Trust distributed this exempt income to PSIC 4 (via conduit principles), which paid dividends to PSIC 3, and ultimately to Absa.

The economic result was the conversion of taxable interest (paid by MSSA) into tax-exempt dividends for Absa. The tax liability that arguably should have accrued on the interest stream was stripped out through the offshore arbitrage. It is on this factual matrix that SARS seeks to establish the principle that any party deriving a financial benefit from such a structure, regardless of that party’s own tax position and whether that party directly obtained a ‘tax benefit’, may be assessed under the GAAR.

The statutory foundation: defining the trigger

The ‘tax benefit’ requirement is the primary jurisdictional trigger for the GAAR. Section 80L of the Income Tax Act (the Act) defines an ‘avoidance arrangement’ as any arrangement that results in a ‘tax benefit’. A ‘tax benefit’, in turn, is defined in section 1 of the Act as including ‘any avoidance, postponement or reduction of any liability for tax’.

While these words appear deceptively simple, their application requires careful interpretive work, which has afforded room for the Absa Bank dispute. The inquiry traditionally asks whether the taxpayer has ‘stepped out of the way’ of a tax liability that would otherwise have been incurred. This necessarily involves the construction of a counterfactual, namely, a hypothetical scenario against which the actual transaction is measured.

The historical position

It was first in CIR v King[i] (1947) that Watermeyer CJ established that ‘avoiding’ liability must refer to ‘anticipated liabilities’ rather than accrued ones. This concept was refined in Smith v CIR[ii] (1964), where Steyn CJ held that avoiding tax means to ‘get out of the way of, escape or prevent an anticipated liability’.

The concept of anticipated liability was further refined in Hicklin v SIR[iii] (1980), where Trollip JA clarified that ‘anticipated liability’ connotes ‘liability for tax that the taxpayer anticipates will or may fall on him in the future.’ The court noted that ‘such a liability may vary from an imminent certain prospect to some vague, remote possibility’, though importantly, the court expressly declined to determine the precise boundaries of this concept, stating that ‘it is unnecessary and hence unadvisable to decide here whether a vertical line should be drawn somewhere along that wide range of meanings in order to delimit the connotation of ‘an anticipated liability’.’ This latter qualifier has often been glossed over in commentary on the ‘tax benefit’ enquiry. Hicklin also established that the liability must be that of the ‘taxpayer concerned’, an entity-specific focus that remains a central pillar of Absa’s defence today.

The mechanical ‘but-for test’ was articulated by Corbett JA in CIR v Louw[iv] (1983). The inquiry was framed as: but for the arrangement, would equivalent or even lesser amounts probably have been received by the respondent in a taxable form? In Louw, the court interrogated whether loans granted to shareholder-directors were a substitute for taxable salaries or dividends. This introduced the necessity of analysing ‘probability’, forcing courts to speculate on what a taxpayer would probably have done in a tax-neutral world.

The Supreme Court of Appeal (SCA) recently applied these principles in Sasol Oil Proprietary Limited v CSARS[v] (2018). The court rejected SARS’s attempt to hypothesise an alternative procurement structure for crude oil as opposed to the actual structure. The majority held that because the supply chain had been established for non-tax reasons years prior, Sasol did not ‘anticipate’ and then avoid the specific liability SARS sought to impose.

The Absa conflict: transferable benefit vs entity-specific guardrails

As noted above, the Absa Bank litigation before the Constitutional Court has exposed a significant divergence in how these traditional tests should be applied under the ‘new GAAR’.

SARS’s position: the commercial equivalent and arrangement-centric view

SARS advances an ‘arrangement-centric’ view. It argues that section 80A of the Act does not require the specific taxpayer being assessed to have personally avoided a tax liability. Instead, it is the arrangement that must produce a tax benefit for someone. In this view, once an ‘impermissible avoidance arrangement’ exists, section 80B empowers the Commissioner to determine the consequences for ‘any party’.

SARS posits that the ‘normal commercial equivalent’ of Absa’s investment was an interest-bearing loan. SARS argues that but for the elaborate structure involving the D1 Trust and Brazilian bonds, Absa would have received taxable interest from the borrower. SARS therefore relies on the broad powers in section 80B(1)(e) to re-characterise the income as such, defining the ‘tax benefit’ as the difference between the tax-free dividends received and the taxable interest that should have been received in a ‘normal’ setting.

Absa’s position: non-conflation and a taxpayer-centric view

Absa counters that SARS’s theory is a fundamental error of law. Relying on King and Hicklin, Absa argues that ‘liability’ must mean the liability of the taxpayer being assessed. Absa maintains it had no anticipated tax liability to ‘step out of the way of’; it simply made an investment in an asset class (preference shares) that is by default exempt by law.

Absa argues that SARS is attempting to conflate an ‘economic benefit’ (a higher return) with a tax benefit (the avoidance of a specific tax liability). If Absa had not entered into this specific transaction, it would have either done nothing or invested in a different exempt asset. Absa contends that section 80B cannot be used to transfer the liability of a conduit (D1 Trust) to a financier (Absa) simply because the financier has ‘deeper pockets’.

Interrogating the competing schools of counterfactual thought

Two competing schools of thought now contend for judicial acceptance on how the ‘but-for test’ ought to be applied.

The narrow view

This school postulates that the ‘but-for’ inquiry should be limited to a simple removal of the impugned steps or arrangement. In other words, the appropriate counterfactual is that of ‘no transaction at all’ in that one asks: if the taxpayer had simply not entered into this specific transaction, would a tax liability have resulted? If the answer is ‘no’, then no tax benefit was created by the arrangement.

This approach prioritises legal certainty and adheres to the ‘literal rule’ described by Lord Cairns in Partington v The Attorney General,[vi] namely that ‘[i]f the person sought to be taxed comes within the letter of the law, he must be taxed … on the other hand, if the Crown cannot bring the subject within the letter of the law, the subject is free’. It respects the principle that there is no equity in tax (per Cape Brandy Syndicate v IRC[vii]), and the law must be applied as written, however harsh the result. Critics, however, argue that this is too narrow and allows sophisticated ‘pre-packaged’ avoidance schemes to escape the GAAR by ensuring that the ‘taxable alternative’ is never even contemplated by the parties.

The broad view

This school, recently endorsed by the Tax Court in Mr Taxpayer G[viii] (September 2025), advocates for a broader analysis of ‘commercially equivalent’ alternatives. The taxpayer argued that he obtained no tax benefit because, had he not entered the scheme, he would have simply not transacted at all (in accordance with the narrow view). The Tax Court explicitly rejected this ‘no transaction’ defence as ‘absurd’. Drawing on Lord Hoffmann’s reasoning in the Hong Kong case CIR v Tai Hing Cotton Mill (Development) Ltd,[ix] the court held that a tax benefit arises if your liability is less than it would have been on ‘some other appropriate hypothesis’. In Mr Taxpayer G, the appropriate hypothesis was that the payments were, in substance, fees for services rendered. This judgment effectively limits the choice principle, because the implication is that a taxpayer can no longer justify a choice between commercial alternatives substance by claiming that they merely chose the least-taxed alternative available to them.

While this approach captures ‘dressed up’ transactions, it introduces significant subjectivity. We submit that if SARS has to decide subjectively what the taxpayer would have done under ‘some other appropriate hypothesis’, it enters the realm of mere speculation. Allowing a ‘wide field for speculation’ creates a serious risk of uncertainty. The ‘appropriate hypothesis’ test assumes there is a single ‘normal’ commercial way to conclude a deal, but commerciality is inherently subjective, as something may be commercially acceptable for one party but entirely unacceptable for another due to different risk appetites or strategic goals.

Furthermore, the counterfactual school ignores the reality that if a transaction is rendered inefficient due to tax and no suitable alternatives are available, the taxpayer will simply walk away. To hypothesise that the taxpayer would have proceeded with a heavily taxed ‘normal’ alternative is often a factual fallacy.

Moreover, and perhaps most alarmingly, it is not entirely clear how widely the word ‘appropriate’ in ‘some other appropriate hypothesis’ should be read – for almost any possible arrangement, one could conceive of a less tax-efficient alternative approach, and therefore, relative to such an alternative, the actual arrangement would necessarily produce a ‘tax benefit’.

A view on the correct test

In our view, given the above, the ‘correct’ test must favour the narrower school. Purposive interpretation cannot be used to override the clear letter of the law or to repair a casus omissus (a matter the legislature failed to provide for). As held in Defy v C:SARS:[x]

‘It needs to be borne in mind that a statute is not a statement of policy by the legislature that leaves the detail to be filled in by a court. It is policy that has been translated into law. If it has not been adequately translated I do not think that it is for the courts to rewrite the statute. That would seem to me to strike at the heart of the rule of law.’

If the legislature finds that certain transactions are being abused, the lawmaker can always simply amend the legislation. Legal certainty is more important for economic progression than a malleable and indeterminate application of wide-reaching provisions. This is particularly acute against the backdrop of South Africa’s very narrow tax base. Allowing the fiscus to tax based on ‘what might have been’ rather than ‘what was’ creates an unacceptable risk for large-scale commerce.

Conclusion

Despite no apparent ambivalence in our historical GAAR jurisprudence, it appears that the ‘tax benefit’ requirement has morphed from a straightforward factual trigger into a complex question. As we await the Absa Bank ruling, the divergence between the approaches to the ‘but-for test’ introduces considerable uncertainty.

If the Court endorses the ‘transferable tax benefit’ theory, the practical consequence is that a party who derives a lawful return from a structure may nonetheless be assessed for the tax avoidance of other participants in that structure. This would fundamentally alter the risk profile of structured finance. In Part III, we turn to the question of volition and whether a participant in a multi-party arrangement can properly be held liable under the GAAR for avoidance conduct of which it was entirely ignorant. We will assess the competing interpretations of ‘party’ under section 80B and the systemic consequences of the Constitutional Court’s eventual ruling for the GAAR and for taxpayers more broadly.

[i] 14 SATC 184.

[ii] 26 SATC 1.

[iii] 41 SATC 179.

[iv] 1983 (3) SA 551 (A).

[v] [2019] 1 All SA 106 (SCA).

[vi] [1869] LR 4 HL 100.

[vii] (1921) 12 TC 368.

[viii] IT 24502 [2025] SATC 12 (unreported).

[ix] (2007-08) Vol 22 IRBRD 2/2007 paragraph 14.

[x] (2010) 72 SATC 99 at 110.