If your Romanian subsidiary belongs to a larger group, one document decides most transfer-pricing disputes: the comparability analysis, known in practice as the benchmarking study. At audit, ANAF opens it, rejects part of the companies used as reference, rebuilds the range on a narrower set and brings the market margin to the median. The result: several million lei of additional taxable base, plus interest and penalties. Between an adjustment that becomes final and one that is set aside stands, almost always, the quality of this single file.
Between related parties, the price is not set by a real market. Benchmarking reconstructs one from public data: within a database of financial statements, it searches for independent transactions or companies comparable to the tested transaction, in order to establish the market range.
If the outcome of the intra-group transaction falls within the range of the comparables, the price complies with the arm's length principle (Article 11 of the Fiscal Code); if not, an adjustment follows. Comparables can be internal (transactions the same company also carries out with independent parties, under similar conditions — the best, but rare) or external (independent companies identified in commercial databases). In practice, the vast majority of studies use external comparables — and that is exactly where the quality problems begin.
The same logic applies to any type of flow: a distribution margin, a cost-plus on services, a royalty or the interest rate on an intra-group loan. Each has its own comparability analysis; none is defended by assertions alone.
A sound benchmarking exercise is not a simple database query, but a sequence in which each stage determines the validity of the one that follows.
You establish exactly what is being tested and which entity has the simpler profile, easier to compare. Choosing the wrong tested party compromises everything that follows.
The method (usually the transactional net margin method) determines the profitability indicator compared — operating margin, mark-up on costs, return on assets. It must fit the functional profile.
A commercial database of financial statements is queried, starting from a broad set filtered automatically. The database must be appropriate to the type of transaction.
The activity code, the independence criterion, minimum activity thresholds and the treatment of loss-making companies. The loss filter is delicate: neither automatic exclusion nor inclusion without analysis is correct.
Each potential comparable is reviewed individually and accepted or rejected, with a documented reason. This is where the quality of the study is actually decided.
The most common is the working-capital adjustment, which neutralises differences in how the activity is financed. Adjustments must operate in both directions and be documented — not applied selectively.
The profitability indicators of the comparables produce a series of values. The full range is not used; instead, the interquartile range is applied — the portion between the lower quartile (Q1) and the upper quartile (Q3), with the median in the middle. In effect, the first and last quarter of values are removed, because the extremes often reflect atypical situations that would distort the result.
The essential nuance in disputes: if the tested party's margin lies anywhere between Q1 and Q3, the price is treated as arm's length — it need not be at the median. A company with its margin at Q1 is, technically, just as much within the range as one at the median. Positioning becomes decisive only when the tested party falls below Q1 (or above Q3) — that is when an adjustment arises, and the question becomes how far the price is adjusted.
This last question is among the most disputed in Romanian practice. ANAF typically positions the result at the median. In litigation, however, it has been argued that where the taxpayer was already close to the range, adjustment to the median is excessive, and the correct point would rather be the lower quartile. The difference between Q1 and the median, applied across the whole flow of the audited years, can on its own be worth as much as the entire adjustment.
At audit, the tax authority does not re-read the study as a text; it tests it at its most fragile links.
Are the companies in the set genuinely independent, or are some themselves part of groups? A related comparable is not a comparable.
Do the comparables perform the same functions, with the same risks, as the tested party? A routine distributor is not compared with a manufacturer.
Does the financial data relate to the correct period? A study built on old data, never refreshed, is vulnerable.
Were the necessary comparability adjustments made — or omitted because they widened the range against the authority's interest?
OPANAF 828/2026 imposes a hierarchy for widening the search: first the tested party's jurisdiction, then a region and, only if there are not enough comparables, the international level. Skipping the hierarchy is challengeable.
When it rejects part of the comparables, ANAF rebuilds the range on the remaining set (often with higher margins) and, if the result falls below the new range, brings it to the median.
Caution. A study filed for form's sake but built superficially is more dangerous than having none: it creates the appearance of compliance while handing the audit exactly the weak points to attack. Under OPANAF 828/2026, the study is also presented in a machine-readable electronic format, with the search strategy, the selection formulas, the set of comparables and the list of rejected companies with the reasons for rejection, plus the calculation of the range. The taxpayer is effectively required to put on the table the very instrument by which its own study can be verified — and, if weak, dismantled.
The most common and most serious: companies with functions, risks or assets different from the tested party, included only to enlarge the set.
An un-refreshed study, with outdated financial information that no longer reflects the market of the audited period.
Both automatic inclusion and blanket exclusion of loss-making companies are wrong; each case calls for an analysis of the cause of the loss.
Omitting the adjustments that would have brought the comparables closer to the tested party, especially the working-capital one.
A source unsuited to the type of transaction, or a query that misses relevant comparables.
Accepting the automatically generated set without individually reviewing each company.
In many audits, the dispute reduces to two studies: the one in the file and the one rebuilt by the tax authority, each rejecting the other's comparables. Here the taxpayer's position is not defended by assertions, but by technical demonstration — why the comparables rejected by ANAF were in fact valid, and why those added by the audit are not comparable. When the dispute reaches litigation, the ground shifts to the transfer-pricing expert report and to the court-appointed expert report as evidence, and a well-documented study becomes reconstructable and therefore defensible. An opaque study cannot be defended even by the best expert.
A second front concerns the data the authority's own study relies on. Sometimes ANAF uses information the taxpayer cannot access, invoking tax or commercial secrecy. An adjustment founded on "secret" comparables, which the party cannot rebut, is vulnerable from the standpoint of the right to a defence — a procedural argument that adds to the technical one, useful also in defending the group's position at audit.
Updating and simplifications. A benchmarking study is not a document "done once": the financial data is refreshed annually, and the search is fully re-run, as OECD good practice, at a periodic interval. For one special category — low value-adding intra-group services — the OECD Guidelines provide a simplified approach, with a standard, reduced mark-up applied on costs, without a full benchmarking exercise, provided the services fall within the definition. The benefit is not automatic: eligibility must be justified, and services with real added value remain outside the simplification.
A defensible study is measured not in page count but in transparency and traceability: functional analysis, justification of the method and the indicator, the full search strategy, the accepted set, the list of rejections with reasons, the adjustments and the calculation of the range — in machine-readable electronic format.
If the margin falls between Q1 and Q3 you are, technically, within the range, and the price is treated as arm's length — you need not be at the median. An adjustment arises when the margin falls below Q1 (or above Q3). Only then does the debate move to where the price is adjusted, and automatic positioning at the median is not always correct.
With caution. OPANAF 828/2026 requires a geographic hierarchy — local comparables first, then regional, then international. A regional study that does not test the availability of local comparables can be rejected. In addition, the tested party and the functional profile must match the reality of the Romanian subsidiary.
Usually on grounds of independence, functional relevance, the year of the data, or missing adjustments. Sometimes the rejection is justified, sometimes not — and the difference is demonstrated technically, comparable by comparable. This is why the study must be transparent and reconstructable.
The financial data is refreshed annually, and the search is fully re-run, as OECD good practice, at a periodic interval. For multi-year analyses the file requires data across several fiscal years. An un-refreshed study is one of the most common vulnerabilities at audit.
Informational material, updated on 18 July 2026. It does not constitute legal or tax advice; individual situations must be analysed on their specific facts.
Time limits run from the moment of communication. An initial discussion clarifies what is being alleged, what you must justify and how the defence is built — before an estimate becomes a tax assessment.