How to Choose a Transfer Pricing Method: The Decision Framework (2026)

Method selection is the first substantive decision in a transfer pricing study — and the one a Transfer Pricing Officer re-litigates most often, because a “wrong” method invalidates everything built on it. The OECD’s rule is the best method rule: the method that provides the most reliable measure of the arm’s length result given the comparability, the data and the assumptions. “Most reliable” is not “easiest to compute” and it is not “what the last study used”. This page is the working framework.

The three inputs to the decision

Every method decision rests on exactly three inputs, and the order matters:

1. The transaction — what is actually being controlled (goods, services, intangibles, loans, or a combination). One transaction can carry multiple components, each potentially under a different method.
2. The comparability — what can be compared, at what level (transaction level: CUP; entity level: TNMM/CPM; residual: profit split), and how much adjustment the differences demand.
3. The data — what can actually be measured reliably (uncontrolled prices? COGS for a pool? clean net financials?). A method you cannot compute with real data is not the best method.

The classic error is inverting the order — starting from the data (“we have a Prowess export, so TNMM”), then bending the comparability analysis to fit. The framework below keeps the order right.

The decision tree

1. Is there a genuine uncontrolled price for this exact transaction?
   ├─ Yes (internal third-party sales, commodity spot prices, published prices)
   │    → CUP. Stop. Document the price identity and terms.
   └─ No
2. Is one party's contribution a pure residual (unique intangibles,
   entrepreneurial risk, innovation) that one-sided methods cannot capture?
   ├─ Yes (both parties non-routine; or one party owns the residual value)
   │    → Profit split (conventional or residual). Document DEMPE and risk control.
   └─ No
3. Is there a genuinely routine entity whose function is cleanly benchmarkable?
   ├─ Yes → identify its function:
   │    ├─ Resale/distribution where gross spread is the return AND COGS is
   │    │    available for the pool
   │    │    → Gross Margin Method
   │    ├─ Resale/distribution where COGS is thin or net data is cleaner
   │    │    → TNMM on OP/Sales (or RPM where the price logic holds)
   │    ├─ Service / contract manufacturing / routine provider
   │    │    → Cost plus / CPM / TNMM on OP/OC (mark-up on cost)
   │    └─ Capital-intensive manufacturer
   │         → TNMM on ROCE / cost plus
   └─ No (no clearly routine party; functions interlinked)
        → Profit split (two-sided), or the least-bad one-sided method with
          documented limitations

Applying the tree: three Indian fact patterns

Pattern 1 — captive software developer, cost-plus economics. No uncontrolled price (step 1: no). No residual — the captive develops to the parent’s specification and the parent owns the IP (step 2: no). The captive is the routine entity (step 3: yes), its function is service delivery on cost (→ mark-up on cost): TNMM on OP/OC (or cost plus at transaction level). This is the default Indian result — and the default exists because the fact pattern is the most common one, not because TNMM is the “default method”.

Pattern 2 — limited-risk Indian distributor of parent’s products. No uncontrolled price (step 1: no — the product is proprietary). The residual is with the parent (brand, product development) (step 2: the distributor is not the residual owner, so no split at the distributor). The distributor is the routine entity (step 3: yes), function is resale: check COGS availability in the pool — available → GMM; thin → TNMM on OP/Sales. The Amount B consideration (OECD 1.0-1.5% net cost range for limited-risk distributors) enters here as a market benchmark the analysis can reference — with the caveats that Amount B conditions (genuine limited risk, no unique assets) must be met.

Pattern 3 — Indian entity developing platform technology the group licenses worldwide. No uncontrolled price (step 1: no). The entity performs DEMPE functions on highly unique, hard-to-value intangibles (step 2: yes) → profit split (or, where the OECD HTVI conditions are met, the HTVI simplified route: benchmark the routine cost of development, allocate the residual to the development/contribution). A TNMM on this entity’s OP/OC would systematically under-price the function — the TPO knows this, and so should the study.

Data reliability: the second gate

Once the comparability logic points at a method, the data check decides whether it survives:

Check	Question	Failure means
Pool existence	Are there ≥ ~5-10 clean comparables under the FAR screens?	Fall back to the next-best one-sided method, or document the limitation
PLI measurability	Is the PLI denominator available for the pool at consistent granularity?	GMM→TNMM, TNMM→cost plus, as the data allows
Consistency	Same definition for tested party and pool?	The method is buildable but the definition must be fixed first
Stability	Does the pool’s PLI behave (low variance, no systematic outliers)?	Wider range, more adjustment work, or a narrower pool

The honest output of a failed data check is not a weaker version of the same method — it is a documented step down the tree, with the reason stated.

Documenting the choice (the exhibit that wins cases)

The method-selection section of the Local File should read as a decision record, not a conclusion. The structure:

1. The transaction — what is controlled, with the components separated.
2. The FAR summary — each party’s functions, assets, risks (the evidence for the routine/entrepreneurial classification).
3. The tree, applied — each step answered with the facts: “CUP considered and set aside because no uncontrolled price of an identical product exists; profit split considered and set aside because the tested party performs no DEMPE function and controls no unique risk; TNMM selected because…”
4. The data check — pool size, PLI availability, consistency statement.
5. The PLI choice — which indicator, why it isolates the routine contribution, the denominator definition.

A TPO challenging the method re-runs this record. If the record shows each alternative was considered and rejected on stated facts, the challenge becomes an argument about facts — which is winnable. If the record only shows the chosen method, the challenge is a gift: the TPO’s method is now the only one in evidence.

Why “last year’s method” is not a reason

Businesses change: an entity that started as a pure service provider may have started carrying inventory; a distributor may have taken on credit risk; the pool itself drifts as listed companies enter and exit. The method is a conclusion about the current facts — and the documentation must show it was re-derived this year. The benchmarking pillar guide covers the annual refresh discipline that makes this explicit.

The method is the first domino; the tested party selection is the second. Get both from the FAR, in that order, and the rest of the study follows.