Transfer pricing benchmarking studies: methodology for defensible arm's length analysis
Dr. Raffaele Petruzzi·PhD · LL.M. · M.Sc.TL;DR
A defensible benchmarking study starts with the right method (TNMM, CUP, profit split) for the controlled transaction, applies documented search and screening criteria to a reliable comparables database, makes targeted comparability adjustments, and produces an interquartile range with a written rejection log. The interquartile range is the headline; the rejection log is the defence.
Which transfer pricing method should I use for a benchmarking study?
OECD Chapter II ranks five methods. Use the CUP method when comparable uncontrolled prices exist (commodities, financial instruments). Use TNMM (Transactional Net Margin Method) when comparability at gross-margin level is weak but operating-margin comparables are available — the default for routine services and distribution. Use profit split for highly integrated operations.
| Method | Indicator tested | Typical use case |
|---|---|---|
| CUP (Comparable Uncontrolled Price) | Price | Commodities, loans, public-market financial instruments |
| Resale Price | Gross margin | Distribution without significant value-add |
| Cost Plus | Mark-up on cost | Routine manufacturing, low-value-added services |
| TNMM (Transactional Net Margin) | Net margin (operating profit / sales or costs) | Most service and distribution structures — Europe's workhorse method |
| Profit Split (Transactional) | Allocation of combined profits | Highly integrated operations, unique contributions, financial instruments |
How do I build a defensible comparables search?
Document a five-step search: industry classification (NACE/SIC codes), geographic scope, independence test (rejecting comparables with >25% common ownership), financial completeness (3-year continuous data minimum), and qualitative screening (functional comparability, loss-rejecting). Each step needs a documented hit-rate so the rejection trail is reconstructable.
- Industry codes. Pick the most specific NACE or SIC range that matches the tested party. Document why broader or narrower codes were rejected.
- Geography. EU-wide for mid-market; narrower for jurisdictions with materially different cost bases (e.g. Switzerland, Nordic). Document the rationale.
- Independence. Exclude entities with >25% common shareholder control. The Amadeus / Orbis databases flag this automatically.
- Financial completeness. Three years of continuous reported financials, audited or equivalent. Reject incomplete records; document the rejection count.
- Qualitative screen. Manual review of activity descriptions to reject functionally non-comparable entities (e.g. a logistics distributor in a search for software distributors).
When should I make comparability adjustments?
Make targeted adjustments when a quantifiable difference between the tested party and comparables would materially distort the result and the data exists to make the adjustment reliably. Common defensible adjustments: working capital, accounting policy normalisations, country risk. Avoid over-adjustment — each adjustment is an audit handhold.
OECD para 3.50 sets the bar: adjustments must be material, reliable, and not transform comparables into the tested party. Working capital adjustments are the most common defensible move because the data is observable and the calculation is mechanical. Country-risk premiums are the most contested because the data is sparse and the methodology proprietary.
Why is the interquartile range, not the arithmetic mean, the OECD default?
The OECD prefers the interquartile range (25th to 75th percentile) because it discards comparability noise at the extremes without claiming false precision at the centre. Arithmetic means are skewed by outliers; medians lose information. The interquartile range gives authorities and taxpayers a defensible arm's length band rather than a single point.
The practical question is what to do when the tested party's result falls outside the range. OECD Annex IV permits a tested-party adjustment to the median, the closest quartile, or any point within the range, with the choice driven by the specific facts. Most jurisdictions default to the median; some (notably Italy, Germany) accept the closest quartile when the taxpayer can demonstrate why a deeper adjustment is unwarranted.
When does the 5% low-value-added services safe harbour apply?
OECD Chapter VII Section D (the LVAS regime) lets routine intra-group services be priced at cost plus 5% without a benchmarking study, provided the services meet the supportive-not-core test and the documentation requirements. June 2026 brings the first major OECD review of the safe harbour; the 5% figure may move.
Services that qualify: accounting and audit support, HR administration, IT infrastructure, legal administrative tasks, tax compliance. Services that DO NOT qualify: anything strategic, R&D, IP-related, financial advisory, manufacturing, or revenue-generating. The audit risk is in the gray zone — does centralised marketing strategy qualify? Does a procurement hub? The 2024–2025 audit pattern is that authorities are tightening on gray-zone classification.
How does benchmarking change in a business restructuring?
OECD Chapter IX requires benchmarking BEFORE the restructuring (the pre-deal arm's length state) AND benchmarking AFTER (the post-deal remuneration). The challenge is constructing a backcast — what would routine entities have earned in the pre-deal structure — when by definition no fully comparable entities exist. The Coca-Cola 11th Circuit appeal turned on exactly this question.
The defensive technique is two-track documentation: an explicit pre-restructuring economic analysis with comparables (even imperfect ones), plus a separate post-restructuring TP analysis that does not rely on the pre-deal benchmarking. Tax authorities are increasingly willing to disregard restructurings that cannot be defended on both tracks.
Veelgestelde vragen
Veelvoorkomende vragen
How many comparables do I need for a reliable TNMM study?
OECD does not set a minimum. Practice is 6–10 comparables for routine services and distribution; 4 may be defensible for narrow industries. Below 4, the interquartile range becomes unreliable and authorities can challenge the entire study. Above 20, the search is probably too broad and comparability is suffering.
Can I use a regional benchmarking study, or do I need country-specific?
EU-wide TNMM studies are accepted in Austria, Germany, Italy, Netherlands, and most other EU jurisdictions for mid-market service and distribution entities. Country-specific studies are required in some Latin American and Asian jurisdictions and are preferable when cost bases differ materially. Document the geographic scope choice.
What's the difference between an Orbis search and an Amadeus search?
Amadeus is the European subset of Bureau van Dijk's Orbis database. For purely European comparables, Amadeus is faster and more focused. For multi-region searches, use Orbis. Both pull from the same underlying data; the difference is the query interface and country coverage.
How often should I refresh a benchmarking study?
The OECD recommends a full refresh every 3 years and a comparable-set update every year. Most jurisdictions (Austria, Germany, Italy) accept the 3-year refresh cadence in practice; some (Poland) require annual updates. Major business changes (M&A, restructuring, market entry/exit) trigger an out-of-cycle refresh regardless.
Meer in deze categorie
Technical Foundations
How the DEMPE framework works for high-value intangibles: a practical guide beyond the glossary
OECD Chapter VI in practice. DEMPE is not a checklist; it is a substance test that allocates intangible returns to the entities that actually develop, enhance, maintain, protect, and exploit them.
15 juni 2026 · 11 min
Intra-group financial guarantees under OECD Chapter X: valuation, documentation, and defence
When a parent guarantees a subsidiary's external borrowing, the arm's length fee turns on three questions: how much credit uplift the guarantee actually delivers, how much of that is implicit support, and how the residual is priced.
8 juni 2026 · 12 min
FAQ