TNMM benchmarking study
for Brazil.

The board-level question: can the group benchmark be defended in Brazil?

A benchmarking study answers a transaction question: what return would independent parties have agreed for these functions, assets, risks, market conditions and period? A list of companies in the same industry is only input. The technical output is the demonstrable link between the accurately delineated transaction and the economic result used in the test.

For TNMM, Article 11(IV) of Law 14,596/2023 compares the controlled transaction’s net margin with the net margins of comparable transactions using an appropriate financial indicator. Article 41(1) of the regulation turns that into operating profit over a denominator aligned with the functional profile.

The decision file must identify the transaction, tested party, method, PLI, search universe, filters, accepted and rejected candidates, adjustments, range and legally available response. Q1, median and Q3 without that chain are unsupported numbers. The arm’s length principle applies to the actual transaction, not superficial sector similarity.

Brazil requires delineation before the search

Article 6 of Law 14,596/2023 fixes the sequence: first delineate the controlled transaction, then perform comparability analysis. Article 7 considers contractual terms, functions, assets, economically significant risks, product or service characteristics, economic circumstances and business strategies, tested against actual conduct.

Before opening a database, the team should close the facts: what is bought, sold, licensed, financed or supplied; who makes decisions and controls risk; which valuable assets are used; where the tested party operates; and whether economically different flows require segmentation. Agreements, invoices, ledgers, policies, approval matrices, interviews and systems are evidence.

The Brazilian seven-step audit trail

Article 21 of the consolidated IN RFB 2,161/2023 sets seven stages: period, internal comparables, external sources, method/PLI/tested party, potential comparables, adjustments and interpretation. Paragraph 1 allows stages to be repeated when the search exposes a weak assumption. A serious search is iterative.

The 2022 OECD Guidelines, paragraphs 3.4 and 3.5, describe nine accepted-practice steps and the same need to loop back. Global alignment is useful; the Brazil file still needs its legal anchor in the Law and regulation.

Each stage leaves an artifact. The method rationale links to the overview of Brazilian OECD methods; search strings, database version, screenshots and the candidate log remain with the study.

Comparable quality is not a headcount test

Under Article 5 of the Law, transactions are comparable when no difference materially affects the indicator or reliable adjustments remove that effect. Related parties are not comparables merely because both are in Brazil (Article 22, sole paragraph); a non-ordinary or fabricated transaction is unreliable (Article 24); and aggregated accounts may need segmentation (Article 26(2)–(4)).

There is no universal minimum number of comparables. Article 21(3) addresses one narrow external-database situation under the resale price, cost plus or TNMM methods. If fewer than four remain after proper screens, the independence threshold may move from 20% to 25% only when that improves range reliability. It is not a command to retain a weak company merely to reach four.

The log records the initial population, automated filter, manual review, decision, reason, evidence, reviewer and date. Counts must reconcile. A one-company discrepancy means the trail is broken.

Domestic and foreign comparables: read the regulation and profile together

The regulation separates two axes. Internal versus external is tested case by case, with internal transactions considered before an external search. Domestic versus non-domestic follows the tested party’s geographic market — not the location of the Brazilian taxpayer by default.

Article 23(2) normally starts in the market where the tested party operates. If reliable information is unavailable, paragraph 3 allows other markets when material differences can be addressed by reasonably accurate adjustments. Brazil’s OECD country profile, updated January 2026, confirms there is no absolute domestic-comparable preference.

A regional set is a starting point, not a conclusion. Market conditions, accounting, working capital, country risk, functional intensity and data availability still need testing. The decision belongs in the broader Brazil transfer pricing framework.

The tested party may be in or outside Brazil

Article 46 selects the party for which the method can be applied most appropriately and reliable data exist. It is generally the less complex entity, but paragraph 4 expressly allows a tested party in Brazil or abroad.

A defensible selection records both parties’ functions and risk control, assets and unique intangibles, unique contributions, financial segmentation, access to foreign-party information and comparable availability. Selecting the Brazilian entity because it files the ECF reverses the test. Selecting the least profitable entity does the same.

Choose a PLI that measures the function — not the desired answer

Article 42(1) links the denominator to the functional profile. Sales often fit routine distribution; relevant costs can fit services or contract manufacturing; operating assets can fit asset-intensive functions. The denominator should be reasonably independent from controlled pricing, measurable and consistently defined — Article 42(2)–(4).

A cost-based PLI is especially sensitive where a large cost share comes from intercompany purchases. Differences in operating-profit definitions between the tested party and comparables can also move the range without any change in economics.

The Berry ratio is not a fallback to improve the answer. Article 42(5)–(6) reserves it for exceptional circumstances involving functions proportional to operating expenses, no material link to product value, no other significant functions and sufficiently similar expense composition.

Multi-year data do not create a three-year safe life

Article 30 permits multi-year data when they improve reliability. OECD paragraph 3.75 likewise treats them as useful context, not as a universal mandatory period.

When a multi-year range is built, Brazil requires the indicator average for each comparable before the set is formed, weighted by the PLI denominator. The starting period is generally the current or latest available year plus the prior two; another period requires justification. A comparable with a negative weighted average or a negative indicator in more than one period is rejected.

This does not make a search valid for three years. Tested-year data and documentation stay current. The regulation also has no universal annual full-rebenchmark rule; material changes in transaction, functions, risks, market, method, PLI, source or set determine the required refresh.

Full range or interquartile range: Brazil makes the choice conditional

Articles 41 and 42 of the consolidated IN RFB 2,161/2023 define the TNMM margin and indicator. Article 16(2) of the Law and Article 47(1)(III)–(IV) of the regulation make the range conditional.

• Interquartile range: residual comparability or reliability uncertainty cannot be precisely identified, quantified and adjusted.
• Full range: independent observations have an equivalent degree of comparability and that residual uncertainty is absent.

If the result is within the appropriate range, Article 47(5) treats the principle as met, subject to Article 49. If it is outside, paragraph 6 points to the median for Article 48 adjustments. A spreadsheet that always prints quartiles is not a legal rationale for the interquartile range.

Brazil’s Annex V is reproducible with QUARTILE.INC

Original series: 5; 7.32; 9.04; 10; 10.57; 11.90; 12; 15
Sorted series: 5; 7.32; 9.04; 10; 10.57; 11.90; 12; 15
Number of observations: 8
Q1 position: 2.75
Median position: 4.5
Q3 position: 6.25
Results: Q1 8.61%; P50 10.29%; Q3 11.93%
Algorithm version: 1.0.0

Article 47(8) refers to Annex V. Sort observations; calculate the median position as (n + 1) / 2, Q1 as (median position + 1) / 2 and Q3 as (median position - 1) + Q1 position; interpolate fractional positions. This is equivalent to QUARTILE.INC.

For the official eight-observation series — 5.00%, 7.32%, 9.04%, 10.00%, 10.57%, 11.90%, 12.00%, 15.00% — Q1 is 8.61% at position 2.75; the median is 10.29% at 4.5; Q3 is 11.93% at 6.25.

With the first seven observations, Q1 is 8.18%, the median 10.00% and Q3 11.24%. Exact Q3 is 11.235%. Annex V does not state a universal rounding rule, so the calculator retains exact decimals and uses round-half-up to two places only to reproduce the published example.

Outside the range: the median rule is not a two-way tax election

Article 47 of the consolidated regulation(6) uses the median when the transaction is outside the range for purposes of the adjustment. Direction must be read with Article 49: a spontaneous or compensating adjustment applies where the result left the Brazilian tax base below the arm’s length base; paragraph 3 prevents reducing the base or increasing a tax loss, subject to the Article 50 compensating adjustment and treaty dispute resolution.

Aurora Distribution. Assume BRL 100 million of segmented revenue, a 1.8% tested margin, an appropriate 3.1%–6.4% range and a 4.6% median. Moving from 1.8% to 4.6% is 2.8 percentage points, or BRL 2.8 million. This illustrates magnitude only; accounting and tax implementation depend on Articles 48–50, delineation and evidence.

If the margin were above the range, the taxpayer could not simply reduce the Brazilian base to the median. Article 50(2) allows a compensating adjustment by the ECF filing when the permanent accounting entry for the transaction year and other conditions are met.

Comparability adjustments must improve reliability

Article 32 rejects the idea that more adjustments automatically produce a better study. A material difference is adjusted if and only if reliability is expected to improve. The same difference cannot be corrected twice; numerous or substantial adjustments may show that the set is not comparable.

The workpaper identifies the difference, expected PLI effect, financial and rate sources, transparent formula, before-and-after result by comparable, sensitivity and conclusion. The purpose is not to pull the tested result into the range. It is to make the compared economic conditions more comparable.

The Brazilian Local File must make the study reproducible

For the full Local File — preceding-year controlled transactions before adjustments of BRL 500 million or more — Article 59(IV) asks for the method and critical assumptions, tested party and PLI, multi-year period, external source and query screens, accepted and rejected comparables, adjustment calculations, range, conclusion and responsibility for a third-party technical study.

At BRL 15 million or more and below BRL 500 million, Article 61 governs the simplified file and still requires comparables, values or ranges, method rationale and adjustments. Article 63 permits the tax authority to require the database search to be reproduced at the taxpayer’s premises. A polished PDF without the source version, search strategy and decision history is incomplete.

The evidence pack connects to Brazil transfer pricing documentation, the simplified Local File and the overview of Brazil transfer pricing methods.

A real TNMM dispute: one accounting line changed the outcome

In SKM2020.105.ØLR, the official Danish Eastern High Court decision examined a pharmaceutical distributor whose TNMM result was below the range when annual goodwill amortisation of DKK 57.1 million was included and inside when the item was excluded.

The court treated the goodwill as an operating asset and amortisation as an operating expense. The company had not shown that removal made the comparison more reliable or adequately supported special conditions for 2006–2009. The 2010 result was inside the range even with amortisation. The case was remanded for year-by-year quantification; the later Supreme Court appeal was withdrawn.

This is not Brazilian precedent. Its control lesson is universal: do the tested party and comparables classify the same economic item consistently? Brazil’s Articles 41 and 32 require that consistency and a demonstrated reliability gain. The IBDT article supplies doctrine; the procedural statements here come from the official Danish pages.

What the TaxUp team delivers

The TaxUp team connects four workstreams: delineation and diagnosis; reproducible search; economic analysis; and a Brazil defence file. The project begins with transaction, contracts, conduct, segmentation, tested party and method hypotheses. It then records the universe, filters, query, manual review and reconciliation.

The economic layer closes the PLI, adjustments, multi-year data, range and tested result. Delivery combines report, screenshots, extracts, workbook, log, accounting reconciliation and Local File mapping. Scope follows the facts and available evidence; a fixed page count or promised number of comparables is not a proxy for robustness.