Transfer Pricing: Arm's Length Principle, Methods and Disputes

Concept

Transfer pricing answers one question: when two companies under common control trade with each other — money, goods, IP, services, guarantees — did they price the deal the way independent parties would have priced it? If not, the tax authority of the disadvantaged state recalculates the taxable profit as if the arm's length price had been charged. The contract stays valid, the cash flows stay where they went; only the tax base moves.

The rule exists because a group can set any internal price it likes. An operating company paying 9% interest to a group lender instead of 5% moves four points of profit, every year, from its own tax base to the lender's — without any external event. States respond by imposing a benchmark: the OECD Transfer Pricing Guidelines (2022 edition, published 20 January 2022) elaborate how the arm's length condition of Article 9 is found, and domestic law gives it teeth — in the UK through TIOPA 2010 Part 4, which must be read consistently with the OECD Guidelines (s.164), in the US through IRC §482 and its regulations, which run a parallel system with the same logic.

A parent lends its foreign subsidiary 10,000,000 at 9% (the numbers here and throughout are fictional teaching inputs). The subsidiary's state asks: delineate the actual transaction — is this really a loan of that size, on those terms, to this borrower? Then: what would an independent lender have charged this borrower, with its actual credit standing including the comfort of belonging to the group? If the evidence supports 5%, the subsidiary's deduction is cut to 5% on the tested amount; the excess four points are taxed. Whether the lender's state then reduces its own tax on the interest it already taxed at 9% is a second, separate question — a corresponding adjustment under Article 9(2) that arrives through agreement between the two authorities, not automatically. Until it arrives, the same four points are taxed twice.

Three features of this test are routinely misread. First, it is bilateral in design but unilateral in operation: each state adjusts in its own favour, and the UK version is explicitly a one-way street — s.147 TIOPA 2010 recalculates profits only where the actual provision confers "a potential advantage in relation to United Kingdom taxation," never in the taxpayer's favour. Second, it is indifferent to intent: no avoidance motive is required, and a fully commercial group with sloppy pricing is adjusted exactly like an aggressive one. Third, it is evidence-hungry in a specific way: the decisive material is the record of who did what — which people performed the functions, which entity had the capacity to control the risk, what an independent party in that position had available as alternatives. The invoice and the board resolution do not carry that weight.

The Controlled Transaction: Delineation Before Pricing

The unit of analysis is the controlled transaction — a transaction between two associated enterprises: enterprises where one participates directly or indirectly in the management, control or capital of the other, or the same persons participate in both (Article 9(1) of the OECD Model; the UK equivalent is the "participation condition" in Part 4, the US trigger is two or more organisations "owned or controlled directly or indirectly by the same interests" under §482).

Before any price is tested, the transaction must be accurately delineated: what was actually agreed and actually done. The OECD Guidelines (Chapter I, Section D) direct the analysis to five comparability factors — the contractual terms; the functions performed, assets used and risks assumed by each party (the FAR analysis, in US regulatory language the comparability factors of Treas. Reg. §1.482-1(d)); the characteristics of the property or services; the economic circumstances of the parties and the market; and the parties' business strategies. Where the parties' conduct diverges from the contract, conduct controls: a contract that assigns market risk to a subsidiary that has no people, no information and no financial capacity to bear it does not move the risk for tax purposes.

Risk allocation follows a specific discipline. A risk belongs, for pricing purposes, to the party that controls it — that has the capability and actually exercises the decision-making to take on, decline and manage the risk — and has the financial capacity to bear it. A contractual risk-bearer that neither decides nor could absorb the loss is stripped of the risk and of the return that was supposed to compensate it. This single rule decides most modern IP and financing disputes before any benchmark is opened.

Delineation can also change the object being priced, not just the price. In exceptional cases a transaction that makes no commercial sense between independent parties can be disregarded or replaced for tax purposes; short of that, a purported loan can be examined on quantum as well as rate — whether an independent lender would have advanced that amount to that borrower at all — with the excess treated under the applicable domestic rules rather than priced as debt.

The Arm's Length Standard in Three Systems

The concept is shared, but its implementation differs by state. The OECD framework and the contrasting UK and US domestic regimes differ on the same issues; a group facing an audit is subject to a specific statute rather than to the OECD framework itself.

The practical consequence of the comparison: the same intercompany agreement is tested twice, under two statutes, by two authorities with opposite financial interests — which is why the pricing file has to work in both directions, not just for the state where the deduction sits.

Both statutes move, and the UK one has just moved a long way. Finance Act 2026 rewrites the architecture in four places: section 47 with Schedule 6 extends the participation condition to arrangements for common management and adds a UK-to-UK exemption where a domestic provision carries no risk of tax loss; section 46 repeals Diverted Profits Tax as a separate tax and re-charges the same ground inside corporation tax as unassessed transfer pricing profits, for accounting periods beginning on or after 1 January 2026 (HMRC INTM489310); section 48 empowers HMRC to require an International Controlled Transactions Schedule by regulations — the draft regulations and HMRC notice published for technical consultation on 16 June 2026 apply it to accounting periods beginning on or after 1 January 2027, due by the company tax return filing date (30 September 2028 for periods ending by 30 September 2027), once aggregated cross-border related-party transactions reach £1,000,000, with transaction-category thresholds of £100,000 (£1,000,000 for groups within country-by-country reporting); and section 49 with Schedule 7 aligns the domestic permanent establishment rules with Articles 5 and 7 of the 2017 OECD Model. The summary of responses of 26 November 2025 also records what did not change: medium-sized enterprises keep the exemption.

Method and Comparability: How "the Market Price" Is Actually Found

"Market price" in transfer pricing is the output of a method applied to comparables. The OECD recognises five methods — three transactional (comparable uncontrolled price, resale price, cost plus) and two profit-based (transactional net margin method, profit split) — and requires the most appropriate one for the delineated transaction; the US best method rule reaches the same place by asking which method yields the most reliable measure given comparability and data quality.

Two conclusions follow from the characteristics of these methods. The method is chosen by the delineation, not by convenience: a distributor that in fact develops local marketing intangibles has outgrown the resale price method whatever its contract says. And the output of any method is a range, established by an actual comparables search with stated screening criteria — geography, independence, functional profile, loss-makers in or out. A specific percentage or interquartile band belongs in a benchmarking study with its search documented: a range asserted without its search is exactly the kind of evidence that fails an audit.

A change of method is a dispute source in its own right, and the largest one currently open turns on exactly that. The US Tax Court, in a decision of 18 November 2020 (Coca-Cola Co. v. Commissioner, 155 T.C. 145), upheld the Service after it departed from the formula fixed by a 1996 closing agreement and applied the comparable profits method to 2007–2009. The taxpayer appealed to the Eleventh Circuit (No. 24-13470), which heard argument on 25 June 2026. Whatever the appellate outcome, the case reads as a reminder about timing discipline: the method and its rationale are fixed in the local file before the return is filed, not reconstructed after an assessment.

A newer shortcut narrows the benchmarking burden at the routine end of the distribution chain. The Inclusive Framework's Pillar One – Amount B report of 19 February 2024 sets out a simplified and streamlined approach to applying the arm's length principle to in-country baseline marketing and distribution activities, and its content has been incorporated into the Guidelines; a consolidated report of 24 February 2025 added a model competent authority agreement for treaty partners that apply it. How far it reaches depends on the state: the US made it elective, and Notice 2025-4 lets taxpayers rely on the approach for taxable years beginning on or after 1 January 2025, pending proposed regulations. A simplified result is an arm's length result only for a transaction actually in scope; nothing in the approach displaces delineation.

The administrative frame was completed in a separate document. On 17 June 2024 the Inclusive Framework published the definition of a covered jurisdiction and the list itself — 66 jurisdictions as at June 2024, from Albania and Argentina to Viet Nam and Zambia. The term replaced "low capacity jurisdiction" and was chosen as deliberately neutral: the commitment had been extended to low- and middle-income OECD and G20 members that expressed a willingness to apply the approach by March 2024 — Argentina, Brazil, Costa Rica, Mexico and South Africa — so a reference to low capacity had ceased to be accurate. Appearing on the list does not oblige a jurisdiction to apply the approach.

The list exists for the political commitment attached to it. Inclusive Framework members committed — subject to their own legislation and administrative practice — to respect the outcome of the simplified and streamlined approach for an in-scope transaction where a covered jurisdiction applies it, and to take all reasonable steps to relieve the double taxation that may result, where a bilateral tax treaty is in force between the jurisdictions. The commitment is political, and there is nowhere to enforce it. The list is reviewed every five years by a mechanical revalidation against World Bank income classifications; the first five-year period runs from 1 January 2025 to 31 December 2029.

The OECD refreshes the calculation tooling annually. February 2026 brought the Amount B Pricing FAQs — answers on the accounts payable guardrail, working capital computation, the definition of net revenue and industry groupings — and the 2026 version of the Pricing Automation Tool with updated inputs, sovereign credit ratings among them. The matrix and its inputs change from version to version, so a given year is priced with that year's tool.

Documentation: What Must Exist, and When

Documentation does not prove a price. It decides who has to explain, and it decides the penalty. The Guidelines' three-tier model (Chapter V) separates the master file — the group's global blueprint: structure, intangibles, intercompany financing, overall pricing policy — from the local file, which carries the entity's own controlled transactions, the method chosen for each and the comparables behind it, and from the country-by-country report, which sets out revenue, profit, tax and headcount by jurisdiction at the €750m threshold and is a risk-assessment input, never a pricing conclusion.

These are dated statutory duties, not good practice. In the UK master file and local file are mandatory for a UK member of an MNE group meeting the country-by-country threshold, for corporation tax accounting periods beginning on or after 1 April 2023 and for income tax from 2024–25; an entity with no material controlled transactions, or with only transactions covered by the UK-to-UK or APA exemption, falls outside the requirement, and an information notice for the specified records cannot be appealed (HMRC INTM450010). In the US, the principal documentation must generally exist when the return is filed and be supplied within 30 days of an examination request. Documents alone do not establish the net-adjustment penalty defence: the reasonable method-selection and application requirements must also be met (Treas. Reg. §1.6662-6(d)).

The country-by-country report has had two audiences for a short while now. Directive (EU) 2021/2101 introduced public country-by-country reporting in the EU: a group whose consolidated revenue exceeded €750m in each of the last two consecutive financial years discloses revenue, profit before tax, tax accrued and tax paid, accumulated earnings and headcount in the commercial register and on its own website — named for each EU member state and each listed jurisdiction, with the rest of the world in a single aggregated line. The first EU-wide wave runs in 2026: for a calendar financial year the reporting year is 2025 and the general deadline is twelve months from the balance sheet date, but Spain cut it to six and Hungary to six, five or four depending on the entity. The group's real deadline is the shortest one in its European perimeter, because the report is one document for the whole group.

For transfer pricing the point is the consistency requirement. The public report has to reconcile with the non-public country-by-country report and must not contradict the local files. Where a local file describes a jurisdiction as carrying limited functions and risks and the public line shows profit incompatible with that description, the question can now be put by any administration, not only by one with access to the exchange. The perimeter, the breakdown, the deadlines and the commercial-harm carve-out are covered in the public country-by-country reporting article.

The same file is therefore used for three different purposes, which groups routinely conflate: the substantive rule (was the amount arm's length), the documentation duty (does the required record exist, on time, in the required form), and the penalty defence (does that record excuse being wrong). Failing the first with the second intact costs tax and interest; failing both adds a percentage on top.

Four further systems: the documentation threshold does not set the price

Singapore, Hong Kong, the UAE and Cyprus add four domestic implementations to the UK/US comparison. A filing exemption, a pricing rule and protection against penalties answer different questions. The statutory thresholds in these jurisdictions are not market rates.

Read the category before the number. Singapore's current guidelines, Table 2 raise the services documentation threshold to S$2m from YA 2026; receipts and payments are separate categories. Hong Kong's FAQ requires tax-free/offshore income and unrealised gains reflected in the accounts to count for the revenue test. Neither jurisdiction grants a blanket exemption merely because the payment comes from abroad.

Loan illustration, Singapore. For an eligible related-party loan obtained or provided during 2026 and not exceeding S$15m, the optional IRAS margin is 1.80 percentage points above the chosen risk-free reference rate. It is not a 1.80% all-in interest rate; special domestic-loan rules and legacy IBOR conversions require separate treatment (IRAS).

UAE records. The FTA guide, §6 separates master/local files from the return disclosure and other supporting evidence. Requested local-file documents ordinarily have a 30-day response period, extendable by the FTA. The large-company file thresholds do not turn a related-party price into an arm's-length price.

UK change from 2026. TIOPA s.164A now exempts qualifying UK-to-UK provisions for chargeable periods beginning on or after 1 January 2026. Among the conditions, both parties are UK-resident companies, the relevant profits bear the same corporation-tax rate and the reference currency agrees. Patent-box provisions and listed excluded companies are outside this route; elections and HMRC notices can disapply it. This is distinct from the SME exemption and the records exemption described above.

Pricing Is One Test Among Several

An arm's length price makes the transaction correctly priced. It does not make the payment deductible, treaty-relieved or respected for every other purpose — each of those is a separate test on separate facts, and clearing one clears none of the others.

In real disputes the taxpayer wins the pricing argument and loses the case under one of the other tests — a benchmarked management fee disallowed because the service was never evidenced, an arm's length royalty denied treaty relief because the IP company was not the beneficial owner. Reading a transfer pricing study as a general clearance is a recurring failure to distinguish the group's tax tests.

The arm's length principle also applies to a second, separate analysis rather than to a different question. Article 9 prices transactions between two separate legal persons. Article 7 attributes profit inside one legal person to its permanent establishment, and under the authorised OECD approach (2010 Report on the Attribution of Profits to Permanent Establishments) the branch is first hypothesised as a separate and independent enterprise: assets and risks follow the significant people functions actually performed there, "free" capital is attributed to support them, and the resulting internal dealings are then priced by analogy with the Guidelines — with no contract anywhere, because a company cannot contract with itself. Two consequences are practical. A group can win the Article 9 argument on its subsidiaries and still lose profit to a PE attribution in the same state. And the evidence that defends a branch is the record of what people did there, not the intercompany agreement that defends a subsidiary.

Shareholder Loans: Risk, Rating and Implicit Support

Intra-group debt is the most frequent subject of disputes because every parameter of a loan affects pricing: amount, term, currency, seniority, security, covenants, fixed or floating. The OECD's financial-transactions guidance (Guidelines Chapter X) structures the analysis in two stages: first delineate — would an independent lender have advanced this amount to this borrower on these terms, and is the instrument in substance debt? — then price, by building the borrower's credit standing and finding comparable lending.

The borrower's credit standing is where groups most often go wrong in their own favour. The rating that matters is not the group's rating, and not the subsidiary's rating as an imaginary orphan: it is the subsidiary's standing including implicit support — the market's expectation that the group would not let a strategically important member fail. Implicit support is passive association: it improves the borrower's terms, it lowers the arm's length rate, and it is not a service anyone in the group can charge for. Only an explicit, legally binding guarantee is a potentially compensable transaction of its own.

Cash pooling applies the same discipline to balances rather than to a term loan (Guidelines Chapter X, Section C.2). A pool leader that only nets balances and passes instructions performs a coordination service and earns a service reward; it keeps the spread between debit and credit rates only so far as it actually controls the risks and could bear them, and the synergy the pool creates belongs to the participants that generate it. A "short-term" pool position that never reverses is delineated for what it is — a longer-term loan, priced as one.

Chevron Australia Holdings Pty Ltd v Commissioner of Taxation [2017] FCAFC 62 (Full Federal Court, 21 April 2017; Allsop CJ, Perram and Pagone JJ) is the reference case. An Australian subsidiary drew A$2.5 billion of facility from a US group financing company at around 9% — no security, no financial or operational covenants, no parent guarantee — while that group entity funded itself in the US commercial paper market at around 1.2%. The Full Federal Court dismissed the taxpayer's appeal under Division 13 ITAA 1936 and Subdivision 815-A ITAA 1997: the arm's length borrowing was not that of a company "shorn of all affiliation to its parent," and a borrower with Chevron's group standing — realistically with a parent guarantee — could not have obtained, and would not have agreed to, those terms. The pricing of the actual, covenant-free orphan loan was replaced by the pricing of the loan an arm's length borrower in Chevron Australia's position would have taken.

Scenario — the same loan at two risk profiles (fictional teaching inputs, no market rates asserted). ParentCo lends OpCo 10,000,000 for seven years in both versions.

In version A, OpCo is profitable and core to the group, the loan is senior, carries market-standard covenants, and group support is realistic: the analysis rates OpCo close to the group's own standing, and the comparable search targets senior corporate lending at that rating — a relatively low margin.

In version B, the same amount is subordinated to OpCo's bank debt, has no covenants, OpCo is loss-making and highly leveraged, and there is documented evidence the group would not support it: the rating is built standalone, the subordination and leverage push it further down, and the search targets junior lending to weak credits — a materially higher margin, if the delineation stage concludes an independent lender would have advanced the full amount at all.

The teaching point: the same cash movement prices differently because risk, seniority and support differ — and each pricing conclusion stands only on an actual comparables study, which is why no figures are offered here. Whether the interest survives the payer's deduction caps is then a separate test (see the matrix above).

IP and Royalties: Legal Ownership Against DEMPE

For intangibles the Guidelines (Chapter VI) split what groups habitually merge: legal ownership of the IP and entitlement to the IP's returns. The return follows the entities that perform and control the DEMPE functions — development, enhancement, maintenance, protection, exploitation — using assets and assuming the economically significant risks. A legal owner that performs none of these and merely provided the money is entitled, at most, to a financing return — and if it does not even control the financial risk of its funding, to no more than a risk-free return on it.

Scenario — the IP-owning entity with functions elsewhere (fictional). IPCo in State I holds the group's patents, registered in its name, and has funded development at 3,000,000 per year. Every researcher, every decision on what to develop, whether to litigate infringers and how to license sits with ParentCo's people in State R; IPCo has a single director who signs what is prepared for him. Under DEMPE the intangible returns are allocated to ParentCo: the royalties OpCos pay to IPCo are re-examined, and IPCo's retainable income shrinks toward a return on funding — risk-free if the director's signature is the only "control" of the financial risk. Note what the analysis did not do: it did not void IPCo's ownership, did not itself impose withholding, did not test beneficial ownership of the royalty stream — those follow their own rules. It reallocated the profit that the pricing of the royalties had placed on the legal owner.

The structural lesson runs in both directions. A group that wants the IP return in a particular entity must put the DEMPE people and the risk-control capacity there — which is an organisational decision with payroll, management and substance consequences, not a drafting decision. And a group whose functions genuinely are distributed can defend a split of the return — with a profit split method and evidence of who contributes what.

Management Services: Evidence Before Price

For intra-group services the first question is whether a chargeable service exists at all. The benefit test (Guidelines para 7.6): an activity is an intra-group service only if it gives the recipient economic or commercial value — measured by whether an independent enterprise in comparable circumstances would have paid another party for it or performed it in-house. Three categories fail by design: shareholder activities (work the parent does in its capacity as owner — consolidated reporting, its own audit, financing its own participations) are not services to the subsidiary at all; duplication of what the recipient already does itself; and incidental benefits of simply belonging to the group, which is passive association again.

Then comes execution evidence. A service that was genuinely rendered leaves traces: agreements dated before performance, named people, time or cost records, deliverables, correspondence, an allocation key that can be recomputed. In audits worldwide the management fee fails at the stage of evidencing performance far more often than on price — the deduction is denied because nothing shows the service happened or benefited the payer.

For genuinely routine support there is a deliberate shortcut. The OECD's low value-adding intra-group services regime (Guidelines paras 7.43–7.65) covers supportive services outside the group's core business that use no unique intangibles and assume no significant risk — typical HR, accounting, IT support. For these the simplified approach applies a fixed 5% mark-up on the relevant cost pool, and para 7.61 states expressly that this mark-up "does not need to be justified by a benchmarking study" (TPG 7.61). The relief is narrow: the benefit test and cost documentation still apply, the cost pool must exclude pass-through costs, and states adopt the safe harbour with local variations. The US takes its own route to the same idea: the services cost method (Treas. Reg. §1.482-9(b)) allows eligible low-margin support services to be charged at cost with no mark-up at all, under its own conditions.

Chapter VII is itself being rewritten, and it should be read with that status attached. On 1 June 2026 the OECD released a public consultation draft of the revision; comments were accepted until 22 July 2026, and a public consultation meeting is scheduled for November 2026 in Paris. The draft rebuilds precisely the stage of assessment at which a management fee fails: the accurate delineation of an intra-group service, the benefit test as part of that delineation, the determination of the charge and the other conditions, documentation supplementing Chapter V, and new examples. The simplified approach itself is reproduced without substantive change — only cross-references were updated, and the 5% mark-up and the benchmarking exemption stay in place. The draft states its own status plainly: it does not represent the consensus views of the Committee on Fiscal Affairs, and neither taxpayers nor administrations should rely on the approaches discussed in it.

Four propositions in the draft are worth carrying now, because they describe how the benefit test reads in a dispute. The benefit is assessed as of the transaction and has to be reasonably expected: a rollout that fails to land, or a loss-making year at the recipient, does not by itself negate the service. Comparison with a local independent provider's price is irrelevant to the benefit test, which asks whether the activity gave the recipient economic or commercial value; whether the recipient could have done the same in-house more cheaply is a pricing question, answered separately. The existence of the service and the arm's length nature of the charge are expressly separated: an excessive charge is grounds to adjust the charge, not to declare that no service occurred. And where an activity is performed for several group members, the test applies to each recipient separately.

Hence the working presumption on management fees. A payment with no contemporaneous record of what was done and what value it gave the payer is treated, by default, as a profit distribution documented as a service, and the burden of rebuttal rests with the group. That rebuttal is provided by a file per service type: a description of the service and its recipient, evidence of actual performance with dates and named people, demarcation from the recipient's own functions, a separate description of shareholder activity, and an allocation key that can be recomputed. An agreement fixed before performance, and an expected benefit described in advance, work the other way — as confirmation that survives a bad year.

After the Adjustment: Where the Cash Sits, and How Much Time Remains

A primary adjustment changes a tax base. It does not move money, and that gap is a problem of its own. The subsidiary has been taxed as if it had paid 5% while the group actually paid 9%, so an amount equal to the adjustment is sitting in the wrong company. States answer with a secondary adjustment, recharacterising that stranded amount as something taxable in its own right — commonly a constructive dividend or a deemed loan — with withholding consequences layered on top of the primary adjustment.

Where a repatriation route exists, it works on its own terms. In the US, Rev. Proc. 99-32 lets the taxpayer establish an interest-bearing account receivable from, or payable to, the related person in the amount of the primary adjustment, deemed created on the last day of the adjusted year and bearing interest at an arm's length rate; where the account is paid within the 90-day window following the closing agreement or the return, the repatriating distribution ceases to be a dividend for federal income tax purposes. Repatriation occurs without recharacterisation as a deemed dividend — but only through the procedure, and only on time.

The taxpayer's own room to correct is asymmetric. Under Treas. Reg. §1.482-1(a)(3) a controlled taxpayer may report an arm's length result on a timely filed return even where it charged something different, but no untimely or amended return may be used to decrease taxable income on the strength of a §482 allocation. Year-end true-ups therefore run one way in the US, and a group that finds its own error after filing has lost the self-help route in that state while keeping it in the other.

MAP time limits depend on the treaty and applicable domestic law. The OECD-model period is three years from first notification of the relevant action. For UK claims, HMRC SP1/2018, paragraphs 10–11 also explains the domestic six-year rule and how a treaty can extend it. The UK does not require domestic appeals to be exhausted first. Check both states' deadlines before choosing the route.

APA and MAP: The Route When Two States Disagree

A transfer pricing adjustment in one state creates economic double taxation until the other state yields. Nothing in the system makes it yield automatically. Article 9(2) of the OECD Model directs the other state to make a corresponding adjustment — but only where it considers the primary adjustment justified in principle and in amount, and the article sends disagreement to consultation between the competent authorities. The instrument for that consultation is the mutual agreement procedure (MAP) under Article 25, and the treaty obligation there is to endeavour to resolve the case, not to resolve it; a MAP can close with double taxation intact unless the applicable treaty adds mandatory binding arbitration (as some do through the MLI's arbitration part or, within the EU, the tax dispute resolution directive). The mechanics of running a MAP, its interaction with domestic appeals and its time limits are the subject of the tax disputes article.

The forward-looking route is the advance pricing agreement (APA): an agreement fixing, before the returns are filed, the method and conditions for defined transactions over a defined term. In the UK the statutory basis is TIOPA 2010 ss.218–230, administered under Statement of Practice 2 (2010), in a programme HMRC has run since 1999 for complex pricing issues; in the US the route is the APMA programme under Rev. Proc. 2015-41, with MAP requests governed by the parallel Rev. Proc. 2015-40. An APA's protection is exactly as wide as its terms: it covers the transactions it names, for the years it names, on the critical assumptions it states — a business restructuring that breaks an assumption suspends the protection. A bilateral APA, agreed between two competent authorities, is the strongest instrument in the field because it commits both states to one answer in advance; a unilateral APA binds only the state that gave it and leaves the other side free to adjust.

The numbers behind that choice were published by the OECD on 31 October 2025 on 2024 data. The average time to close a MAP case is 27.4 months: 30.9 months for transfer pricing cases against 32 a year earlier, and 24.5 months for other cases against 23.4. Bilateral APAs are available in 80 jurisdictions against 73 a year earlier, the average time to conclude one has risen to 39.6 months against 36.8, and the share of applications rejected or closed without agreement has risen from just under 12% to more than 19%. The planning consequence is arithmetic: an application covering a large recurring flow is filed roughly three and a half years before the year from which certainty is needed, and roll-back to open prior periods is requested at the outset rather than mid-negotiation.

A bilateral APA provides categorically stronger protection than MAP after adjustment, and neither mechanism operates automatically. A group pricing a large recurring flow — a shareholder loan, a royalty stream — is choosing, implicitly, between paying for certainty in advance and litigating ambiguity later in two states at once.

Transaction Matrix

The matrix condenses the article: for each common controlled transaction, the delineation question that decides the analysis, the usual starting method, and the evidence that actually carries the case.

Delineation precedes method selection: the delineation question is decided on facts that exist before any benchmark is bought, which is why the cheapest moment to fix a transfer pricing problem is when the transaction is structured, not when it is documented.