🇪🇺Double Taxation Despite the Treaty: MAP, Arbitration and APAs

The concept: the parties to the dispute are two states

A double tax treaty does not apply itself. It allocates taxing rights, but two administrations read it separately and regularly read it differently: one sees a permanent establishment, the other does not; one adjusts the price of an intra-group transaction upwards, the other refuses to make the corresponding downward adjustment; both treat the same individual as their resident. A domestic appeal is structurally powerless in that configuration: the argument runs against one of the two administrations, and winning at home does not cancel the assessment abroad. For precisely this situation every treaty carries a procedural article: the mutual agreement procedure (MAP) lifts the dispute to the state-to-state level and, in some regimes, ends in mandatory arbitration. What follows are four regimes of one procedure, plus the preventive track of advance pricing arrangements.

It is worth separating MAP from its neighbour at the outset. Reclaiming over-withheld tax at source and taking a unilateral credit at home are procedures inside a single jurisdiction: they work when the treaty has worked but the money left at the domestic rate. MAP is what remains when the treaty has not worked.

The frame of the procedure in figures, before each regime is taken apart on its own.

Article 25 of the OECD Model: the addressee, the window, the duty to implement

The request goes to a competent authority — usually the ministry of finance or a dedicated unit within the revenue service. Neither a court nor the tax inspectorate takes such a request. The key change is the 21 November 2017 update to the OECD Model Tax Convention: a case may now be presented to the competent authority of either contracting state, whereas before 2017 it could only go to the state of residence. Treaties concluded on the old formula and never modified keep the restriction, and that is the first thing to check in the actual text.

The window: three years from first notification

The window is three years from the first notification of the action resulting in taxation not in accordance with the treaty (Article 25(1)). Time runs from that first notification, not from the moment an assessment becomes final and not from the exhaustion of appeals: an audit report, an adjustment notice, or even the withholding itself starts the clock.

The second sentence of Article 25(2) is the reason the procedure exists at all: an agreement reached by the competent authorities shall be implemented notwithstanding any time limits in domestic law. A domestic limitation period on refunds is no obstacle to the agreement — provided the particular treaty contains that sentence.

The Action 14 minimum standard

Effective dispute resolution is one of the BEPS minimum standards: Action 14 (the BEPS final reports of 5 October 2015) set the benchmark of an average of 24 months per case and launched peer review of jurisdictions. On 2 February 2026 the OECD issued an updated Manual on Effective Mutual Agreement Procedures (MEMAP), the first full rewrite since the 2007 edition; its recommendations include separating the competent authority organisationally from audit functions and deciding on unilateral relief within four months.

Domestic appeals running in parallel: what must not be taken to judgment

MAP neither replaces an appeal nor suspends it automatically. The practical fork is to file both and freeze the domestic proceedings. The reason is that in a number of countries a final court judgment binds the competent authority: after it, the authority is no longer free to depart from the court's position, and the negotiating space collapses — the partner either swallows a foreign court's decision whole, or the double taxation stays. Suspension of collection during MAP is likewise a matter of domestic law rather than the treaty: in some jurisdictions it is automatic, in others it requires a separate application and security.

Mandatory arbitration: Part VI of the MLI

Treaty arbitration under Article 25(5) of the OECD Model is triggered where the competent authorities have failed to reach agreement within two years of the case being presented. The Multilateral Convention (MLI) of 24 November 2016 built the same logic into the treaty network: Article 19 of the MLI gives the taxpayer the right to require unresolved issues to be submitted to arbitration once the two-year period has run, and Part VI (Articles 18–26) sets out the machinery; Part VI has been in force since 1 July 2018. How the MLI overlays treaties at all — matching, Covered Tax Agreements, the PPT and LOB — is covered in the article on the Multilateral Convention; what matters here is one point: Part VI applies only between two jurisdictions that have both opted into it.

Format: final offer

The default format is final offer arbitration — also called last best offer, or "baseball" arbitration (Article 23(1) MLI): each competent authority submits its own proposed resolution to the panel, the panel picks one of the two by simple majority, and it gives no reasons. Moving to the independent opinion format — a reasoned decision — is possible only through a reservation (Article 23(2)). The difference is practical: "baseball" forces both sides to submit moderate positions, because the extreme one loses by construction. British practice records the outcome plainly: three parallel sources of arbitration exist (Part VI of the MLI, the European Arbitration Convention and treaty-specific Article 25(5) clauses), yet agreement is normally reached at the MAP stage and cases do not reach the commission (HMRC International Manual, INTM423080). Arbitration works as a threat rather than as a procedure.

Who is inside Part VI

The OECD maintains individual arbitration profiles only for jurisdictions that have adopted Part VI; published profiles include the Netherlands and New Zealand (as at 29.09.2025), Finland (26.05.2025), the United Kingdom (31.05.2024), Canada (16.12.2022), France (16.09.2022), Austria and Malta (28.06.2022) and Barbados (25.03.2021). The exact number of Part VI participants shifts and is reported inconsistently in secondary commentary — it has to be verified against the current OECD Signatories and Parties table. What is stable is something else: Part VI was chosen by a minority of MLI signatories, predominantly developed economies. India objects consistently to mandatory arbitration, and Russia has taken no Part VI position at all — for Russian treaty pairs there is no arbitral exit from MAP in principle.

Directive (EU) 2017/1852: the only hard timetable

Inside the EU the procedure is regulated by the calendar: Council Directive (EU) 2017/1852 of 10 October 2017 puts a deadline on every step.

The timetable cuts both ways: a deadline missed by the taxpayer closes the procedure, one missed by an authority opens the road to the commission.

The finish line has a price. The final decision takes effect only if the taxpayer accepts it and renounces domestic remedies, declaring this within 60 days of being notified of the decision (Article 15). Before signing, the directive outcome has to be weighed against the chances in a domestic court — with no right to reverse the choice afterwards.

The directive applies to complaints submitted from 1 July 2019 onwards, concerning income or capital for tax years beginning on or after 1 January 2018 (Article 23). The standard rules of functioning of the advisory commission and the arrangements for publishing final decisions are set out in Commission Implementing Regulation (EU) 2019/652 of 24 April 2019. The European Commission publishes annual statistics across the four stages of the procedure — complaints, MAP, arbitration and final decisions; data are available for 2020–2024.

The outer limits of the directive are currently being tested before the Court of Justice: on 21 July 2025 a preliminary reference from the Administrative District Court of Latvia (Administratīvā rajona tiesa) was lodged — Case C-497/25, UAB Maxima LT v Valsts ieņēmumu dienests (details verified against the official publication of the reference, CELEX 62025CN0497) — with two questions — whether a competent authority is obliged to set up an advisory commission under Article 6(1)(b) where no agreement is reached in time, and whether the directive covers the situation in which the same income enters the tax base of two states but is exempt in one of them. As at 20 August 2026 the case is undecided.

The timetable itself has been proposed for a shift. An amendment to Article 4 of the directive would require the competent authorities to notify the taxpayer without delay once it becomes clear that agreement cannot be reached, rather than waiting for the two-year period to expire — opening the arbitration phase earlier. Three procedural changes sit alongside it: "simultaneous filing" is replaced by a 30-calendar-day window; where several persons are affected, the complaint is filed either by each in its own state or by one on behalf of all; and an alternative dispute resolution commission may also be constituted on admissibility questions. All of this carries one status: a Commission proposal of 24 June 2026 under Article 115 TFEU, requiring unanimity in the Council. Transposition is set for 31 December 2028 and application for 1 January 2029; until adoption the present text of the directive applies and the two-year period runs as before. The package is covered in the article on the EU tax omnibus and the DAC recast.

The Russian track: Chapter 20.3 of the Tax Code and Order No. 102n

In Russia the procedure arrived late and in truncated form. Methodological guidance came before the statute: the Ministry of Finance published its Guidance on Conducting the Mutual Agreement Procedure on 30 January 2019. Federal Law No. 325-FZ of 29.09.2019 then inserted Chapter 20.3 (Articles 142.7–142.9) into Part One of the Tax Code, and the procedure and deadlines were fixed by Order of the Ministry of Finance of Russia No. 102n of 11.06.2020, in force from 11 October 2020: the application goes to the Ministry of Finance within three years of service of the tax audit report or an equivalent document, the ministry considers it within 90 days and either decides to run the procedure or issues a reasoned refusal.

What is left of it. Presidential Decree No. 585 of 8 August 2023 suspended the distributive articles in treaties with 38 states — in the various schedules this means Articles 5–22, 23, 24 and individual protocol provisions — while the mutual agreement procedure article did not make it into most of those schedules (there are exceptions: for Switzerland the schedule captured Article 25b, for Cyprus Articles 27 and 29). Formally the procedure is alive; substantively there is little left to argue about, because the suspended provisions are precisely the ones from which characterisation conflicts arise. The schedule-by-schedule detail is in the analysis of the suspension of Russian tax treaties.

Some partners have gone further. The United Kingdom suspended the 1994 Convention in its entirety: announced on 4 February 2025 and implemented by the Double Taxation Relief (Russian Federation) (Revocation) Order 2025 (SI 2025/344, made 12 March 2025), which revoked the 1994 implementing order (SI 1994/3213), so the Convention ceased to have effect from 1 April 2025 for corporation tax and from 6 April 2025 for income tax and capital gains tax. Germany notified Moscow on 30 June 2026 that the Agreement of 29 May 1996 and its Protocol are suspended with effect from 1 January 2027 (Federal Ministry of Finance statement of 9 July 2026).

In both pairs Article 25 goes off with the rest of the treaty: no mutual agreement procedure is left. There are no published Ministry of Finance statistics on Chapter 20.3, and as at 20 August 2026 no official clarification has been found on whether Article 25 operates in pairs where the partner responded with a mirror suspension.

Four regimes of one procedure

One procedure exists in four versions, and what separates them is procedural: the addressee, the window, the clock and whether an arbitral exit exists. The treaty track first.

The same five parameters in the two regimes set by regional and national law.

Three disputes — one procedure, different odds

A dispute over an individual's residence runs down the ladder in Article 4 of the treaty and comes to rest on MAP as the final rung, once permanent home, centre of vital interests, habitual abode and nationality have failed to untie the knot — the mechanics of that ladder are set out in the article on resolving dual residence. The distinctive feature of this category is that it usually has no arbitral exit: many Part VI jurisdictions expressly carve individual residence cases out of the arbitral scope, or make them conditional on both competent authorities agreeing.

A transfer pricing dispute is the opposite case: it runs longest of all, requires a corresponding adjustment by the second administration, and is the reason the arbitral machinery was built in the first place. A permanent establishment dispute sits in between: the facts are contested, but the size of the adjustment is usually smaller and agreement comes more easily. What all three share is that the basis for the assessment often lies in economic substance, while the procedure for disputing it is one and the same.

Disputes inside the global minimum tax

The whole procedural layer above is built on double tax treaties. GloBE top-up tax sits awkwardly inside it: it is a domestic tax, enacted by national law on agreed model rules, and whether Article 2 of a particular treaty covers it has no settled answer yet. The regime contains no dispute resolution mechanism of its own for divergences in the GloBE computation, and divergences arrived immediately: the first GloBE Information Returns for the 2024 fiscal year were filed by 30 June 2026, and what meets in them are figures computed under national implementations of the same model rules.

Three points produce divergence more often than the rest. The first is rule order: entitlement to the QDMTT safe harbour depends on whether a jurisdiction appears in the OECD Central Record of legislation with transitional qualified status, and a jurisdiction that is absent reads, from above, as an unclosed effective rate gap even where it considers itself qualified. The second is safe harbour availability: the Inclusive Framework package of 5 January 2026 remains administrative guidance, and until it is reproduced in national law the transitional CbCR safe harbour can expire on different dates in two jurisdictions. The third is the figures themselves: adjusted covered taxes and GloBE income are computed with national variations, so one jurisdiction can yield two different effective rates in two computations.

Today these divergences are resolved away from the competent authorities. A dispute about whether legislation qualifies runs through the OECD transitional qualification mechanism — an administrative process between the jurisdiction and the Inclusive Framework, in which the taxpayer is not a party. A dispute about the amount of top-up tax stays domestic and is appealed under the law of the jurisdiction that assessed it. And a dispute about who is first in the queue collapses into the first one, because the queue is set by qualification status. The Inclusive Framework is working on a common up-front compliance and risk assessment framework through the Amsterdam Dialogue format, with business and academia; as things stand that is a work programme item rather than an operating mechanism.

The reporting these divergences surface from — central filing, local notifications and exchange — is covered in the article on global minimum tax reporting.

APAs: how to avoid reaching a dispute

The preventive track is the advance pricing arrangement (APA): the transfer pricing methodology is agreed with the administration in advance, for several years ahead, often with roll-back to prior open periods. A unilateral APA protects against one administration only and does not by itself remove double taxation; bilateral and multilateral APAs do, but they run through the same competent authorities and along the same negotiating rails as MAP.

On the OECD's 2024 figures, 80 jurisdictions allow bilateral APAs, against 73 in 2023, and 49 are actively handling cases. The average time to conclude an APA rose to 39.6 months from 36.8 months a year earlier, while the share of withdrawals and closures without agreement climbed from 12% to more than 19%. The American picture is more granular: under Announcement 2026-8 of 30 March 2026, the APMA programme executed 110 APAs in 2025 (14 unilateral, 90 bilateral, 6 multilateral) and accepted 178 complete applications (23, 153 and 2 respectively); as at 31 December 2025 the queue stood at 622 applications, of which 310 were renewals. A new bilateral APA took on average 50.0 months, with a median of 46.4; across all new and renewal APAs the average was 44.1 months, median 41.6. The largest US treaty partners were India, with 26% of applications filed and 35% of bilateral APAs executed, and Japan, with 24% and 25%. Filing and administration are governed by Rev. Proc. 2015-41, and the model APA text was left unchanged in the 2025 report.

The conclusion the numbers point to is simple: an APA moves the dispute several years forward and conducts it in negotiating mode. It is not a fast route. If the structure's horizon is shorter than four years, the economics of a bilateral APA do not add up.

ICAP: multilateral risk assessment before the dispute

Alongside APAs sits a less familiar preventive route — the International Compliance Assurance Programme (ICAP): a voluntary, co-ordinated multilateral assessment of a group's key international tax risks. The group assembles a documentation package once, a co-ordinating administration brings in the other interested ones, and each issues a letter with its risk assessment on the covered areas. No pricing agreement comes out of it and nothing binds anyone for the future; the value is that a low-risk outcome closes the subject without an audit. The pilot launched in January 2018, and since September 2021 ICAP has run as a full programme of the OECD Forum on Tax Administration.

The only published statistics are the OECD's aggregated results of 29 January 2024, covering twenty cases completed by October 2023. Twenty-two administrations were participating by then; the average number of administrations in a single case was five, the minimum three and the maximum nine.

The OECD attributes the overrun partly to the pandemic, which fell across the second pilot. Even with the overrun, 61 weeks is about fourteen months against an average MAP of 27.4 months and an average APA of 39.6 months.

Outcomes are counted by risk area. For 40% of groups every main covered area was assessed as low risk by every administration that included it in scope; for 80% the result was either entirely low risk or not low risk in just one or two areas. In roughly a third of cases at least one issue identified during the assessment was resolved inside the process, avoiding both audit and MAP. By area, the low-risk share ran: permanent establishments 95%, tangible property 90%, intra-group services 88%, financing 76%, intangible property 75%.

What the figures mean is that a completed screening works as a positive presumption. A low-risk outcome in an area binds no administration legally, but an administration that issued the letter itself will open an audit on that area with less enthusiasm, and the second administration can see that the position has already been through a multilateral review. The weak spot is where the whole preventive line is weak: intangibles and financing come out low risk less often than anything else, and they are also the areas that reach MAP most often.

How long it takes and how it ends

On the OECD's 2024 MAP statistics (published for Tax Certainty Day in October 2025) the average time to close a case is 27.4 months: 30.9 months for transfer pricing cases against 32 months in 2023, and 24.5 months for other cases against 23.4 months a year earlier. Transfer pricing is speeding up; the other categories are slowing down.

The headline figure is the outcome. Around 73% of cases closed in 2024 ended in full resolution of the issue, and only around 4% closed with no agreement at all. The inventory has stopped being a swamp too: cases older than 2016 make up 3.3% of the closing 2024 balance, cases older than four years less than 20%, and cases under two years old more than 56%. National figures can beat the average: on HMRC's data the average time to close a UK case in the 2024/25 financial year was 24.8 months.

What that means for the go or no-go decision: MAP is not a lottery. Three quarters of cases end with double taxation eliminated, and only one in twenty-five ends in nothing. The price is two and a half years and professional support on both sides.

Common mistakes

Six mistakes recur more often than the rest.

What makes a case strong and what makes it weak

The table above is organised around procedure. Next to it sits the substantive question: what the competent authority takes into the negotiation, and what in the file works against the taxpayer before the first round.

The first row costs more than the others. A method fixed in advance moves the dispute into the question of whether the figure is right, where the competent authorities have a treaty instrument. A method reconstructed after an assessment moves it into the question of whether there was ever a position at all, where there is almost no negotiating space.