Prepare a transfer pricing benchmarking study: choose databases and comparables, calculate profit indicators, and document the results with worked examples.
Borys UlanenkoCEO of ArmsLength AI
22min read
Contents↓
Contents
TL;DR key takeaways
A database is a source of potential comparables, not proof of comparability. Document its coverage, fields, version, search date, and limitations.
Define the transaction, tested party, method, PLI, period, and geography before running a search; otherwise the filters can end up driving the conclusion.
Retain the search population, filter sequence, raw export, manual-review evidence, reject reasons, financial mapping, and calculations so another reviewer can reproduce the study.
Losses or extreme results trigger investigation rather than automatic exclusion. The facts and comparability analysis decide whether a candidate remains.
The OECD does not prescribe a fixed number of years or make the interquartile range mandatory in every case. Use multiple-year data and statistical tools only when they improve reliability.
We tested three AI configurations on 10 transfer pricing questions. ChatGPT with web search scored 83%, but models without web access fabricated most quotes. Here's what we learned.
The Berry Ratio (gross profit ÷ operating expenses) is a specialized PLI for low-risk distributors and agents. Learn when it's appropriate, when to avoid it, and how to defend your analysis—with calculation examples and jurisdictional guidance.
On November 10, 2025, HMRC updated INTM485120 with critical details on searching for comparables. The key takeaway? The Interquartile Range is not a default right - and relying on it blindly carries a median-sized risk.
More resources for your next workflow stepMore resources for your next workflow step
Browse the full resource library or contact us if you want recommendations for your specific use case.
Quick answer: what makes a defensible benchmarking study?
A transfer pricing benchmarking study compares a transaction between related parties with evidence from comparable transactions or businesses operating independently. It helps assess whether the price or profit earned is consistent with the arm's length principle.
This guide explains how to prepare the study, from choosing a database and search criteria to reviewing comparables, calculating profit indicators, and documenting the results. The worked search funnel and accept/reject log show how to record the decisions behind the final comparable set.
For a one-sided profit method such as TNMM or the US Comparable Profits Method, the study usually identifies independent companies with sufficiently comparable activities and calculates a profit level indicator for the tested party and the accepted comparables.
The conclusion is only as sound as the route used to reach it. A reviewer should be able to see:
what transaction was tested and why the selected method fits;
why the tested party and profit level indicator were selected;
which data sources and search criteria were used;
why each candidate was accepted or rejected;
how financial data, adjustments, and the arm's length range were calculated; and
how the tested result ties back to the financial records.
Note
OECD Guidelines paragraphs 3.30-3.34 describe commercial databases as useful sources with real limitations. They should be used objectively, and a database-only search may need to be refined with annual reports, company websites, registries, and other public information. Quantity does not replace comparability.
A search is not the first step. Start with an accurately delineated transaction and a functional analysis of all parties. Record the following decisions in a short search-planning memo:
Decision
Question to answer
Evidence to retain
Controlled transaction
What goods, services, financing, or rights are being tested?
Which denominator has a sound economic relationship to the tested activity?
Segmented accounts and PLI calculation policy
Market
What geography and economic conditions matter?
Market analysis and rationale for local or regional scope
Data timing
Is the study being used to set prices, test outcomes, or both?
Pricing policy, true-up process, and filing calendar
Under , the tested party is generally the party to which the method can be applied most reliably and for which the most reliable comparables can be found. It will often, but not automatically, be the party with the less complex functional profile.
Warning
Do not label an entity "routine" simply because the group policy gives it a fixed return. Confirm what its people actually do, which assets it uses, which risks it controls, and whether it makes any unique and valuable contribution.
2. Check internal comparables first
An internal comparable is an uncontrolled transaction between one party to the controlled transaction and an independent party. OECD paragraphs 3.27-3.28 note that internal comparables can have a closer relationship to the tested transaction and may use the same accounting practices. They are not automatically reliable: volume, contractual terms, geography, product features, risk allocation, and other comparability factors still need to be tested.
Ask these questions before commissioning an external company search:
Does either party conduct the same or a similar transaction with independent customers or suppliers?
Can the uncontrolled transaction be isolated in the accounts?
Are the commercial terms, volumes, markets, assets, and risks sufficiently comparable?
Can material differences be adjusted with reasonable accuracy?
If a reliable internal comparable exists, an external database search may add little. Document the decision either way.
3. Choose the data source that fits the transaction
There is no universally best transfer pricing database. The right source depends on the type of evidence required, the market, the reporting period, and whether the data can be reviewed and reproduced.
Database-selection matrix
Evidence needed
Typical source category
Useful fields
Main limitations to test
Company profitability for TNMM/CPM
Company accounts and ownership database
Income statement, balance sheet, ownership, activity codes, business descriptions
Uneven private-company disclosure, company-level rather than transactional data, inconsistent accounting classifications
Comparable prices or terms
Internal transaction systems, public filings, contract and agreement databases
Different credit risk, implicit support, market timing, covenants, and liquidity
Qualitative screening
Company websites, annual reports, registries, regulatory filings
Products, functions, business model, group membership, intangibles
Pages change, descriptions may be promotional, and historic evidence may disappear
The commercial product name matters less than whether the retained evidence answers the comparability questions. Before choosing a source, test:
relevant country, entity-type, and industry coverage;
financial detail needed to calculate the selected PLI consistently;
ownership and related-party indicators;
access to original accounts and historic releases;
export fields, record identifiers, and search reproducibility;
qualitative information available for manual review; and
licence terms governing retention and disclosure of extracts.
Record the database name, product or module, release or version, access date, and field definitions in the report. If more than one source is used, explain which source controls when the records conflict.
4. Build a search strategy with a reason for every filter
A search strategy translates the tested party's functional profile into observable criteria. Industry codes are useful for generating a population, but they are not a substitute for manual review.
Common quantitative filters
Filter
Purpose
Question for the file
Activity codes and keywords
Find candidates performing potentially similar activities
Why do these codes and terms represent the tested function?
Geography
Reflect relevant economic circumstances and available data
Why is the selected market sufficiently comparable?
Independence
Remove entities whose results may be affected by controlled dealings
Which ownership fields, thresholds, and manual checks were applied?
Operating status
Remove inactive or unsuitable legal entities
What status date was used?
Financial-data availability
Ensure the PLI can be calculated consistently
Which years and fields are mandatory, and how is missing data treated?
Size
Remove candidates whose scale creates a material comparability difference
What fact supports the chosen measure and threshold?
Loss or extreme-result screen
Flag candidates for further analysis
Is the rule investigative, or does local law require a particular treatment?
Keep a count after each filter. A reviewer should be able to reconstruct the funnel from the initial universe to the manual-review population.
Note
Amount B checkpoint: Before running an ordinary company search for a baseline marketing and distribution transaction, confirm whether the relevant jurisdiction and period apply the OECD's Amount B simplified and streamlined approach, and whether the transaction meets its scope conditions. Domestic implementation and options control. Where Amount B does not apply, use the ordinary method-selection and benchmarking analysis required by the relevant law.
Illustrative search funnel
Assume the tested party is a routine distributor of industrial components in a regional market:
Stage
Candidates
What happened
Relevant activity codes and keywords
1,842
Initial population generated
Active entities in the selected geography
1,106
Dormant and out-of-scope locations removed
Documented independence screen
284
Controlled or insufficiently independent entities removed
Required financial data available
96
Candidates without the fields needed for the PLI removed
Quantitative review completed
47
Scale and other documented screens applied
Qualitative review completed
11
Functions, products, risks, and intangibles reviewed manually
These figures are illustrative. A small accepted set is not inherently weak, and a large set is not inherently strong. The question is whether the retained companies are sufficiently comparable and the exclusions follow the stated criteria.
Worked accept/reject log
Candidate
Evidence reviewed
Decision
Reason recorded in the study
Alpha Distribution
Database description, website, annual report, ownership record
Accept
Distributes similar industrial components, no material manufacturing identified, and passes the documented independence screen
Beta Industrial
Website and annual report show manufacturing and distribution; no segmental accounts
Reject
Manufacturing results cannot be separated reliably from distribution results
Gamma Components
Comparable activity; losses in two reviewed years
Investigate
Losses are a signal for further analysis, not an automatic rejection; determine whether they reflect normal market conditions and comparable risks
Delta Systems
Database code matches, but annual report describes proprietary technology and strategic product development
Reject
Owns and develops valuable technology and controls functions not present in the tested activity
Epsilon Trading
Similar products but majority ownership by a larger group
Reject
Does not satisfy the documented independence criterion
This treatment follows the logic in OECD paragraphs 3.63-3.66. Extreme results and losses call for investigation. A candidate should not be excluded only because its result differs from the rest of the set; the decision should follow the underlying facts.
Manual-review evidence
For every reviewed candidate, retain:
the legal name and stable database identifier;
the pages, filings, or registry records consulted and their access dates;
a short factual business description;
the acceptance or rejection reason using the study's stated criteria;
the reviewer and review date; and
any unresolved data conflict and how it was resolved.
Generic reasons such as "not comparable" do not allow a second reviewer to test the judgment. State the fact that caused the decision.
5. Select and calculate the PLI consistently
OECD paragraph 2.82 links PLI selection to the nature of the controlled transaction, the functional analysis, reliable information, comparability, and the reliability of any adjustments. No PLI belongs automatically to a transaction label.
PLI
Formula
Often considered when
Reliability questions
Operating margin
Operating profit / revenue
Revenue is a relevant indicator of the tested activity, often in distribution
Are revenue recognition, rebates, and operating items classified consistently?
Net cost plus
Operating profit / relevant operating costs
Costs bear a relationship to the value of the tested functions, often in services or manufacturing
Which costs belong in the base? Are pass-through items treated consistently?
Return on operating assets
Operating profit / relevant operating assets
Operating assets are a meaningful driver of the tested activity
Are asset values, leases, depreciation, and idle assets comparable?
Berry ratio
Gross profit / operating expenses
Limited intermediary facts satisfy the conditions in OECD paragraphs 2.106-2.108
Are cost classifications comparable, and is value unrelated to the value of products sold?
The Berry ratio is not a general substitute for an operating margin. The OECD states that it is sensitive to cost classification and sets specific conditions for its use, including a relationship between functions and operating expenses and no material relationship between those functions and the value of products distributed.
Segment the tested transaction
OECD paragraphs 2.83-2.85 require the net profit calculation to focus on operating items related directly or indirectly to the controlled transaction. A company-wide result can be misleading when the entity conducts different controlled transactions or material uncontrolled activities.
A financial bridge should show:
the statutory or management accounts used;
the accounts and cost centres assigned to the tested activity;
each allocation key and its source;
operating and non-operating classifications;
exceptional items and their treatment;
pass-through costs and the comparable-company treatment; and
the final numerator and denominator.
Apply the same PLI definition to the tested party and comparables. Accounting labels do not guarantee economic consistency.
6. Use multiple-year data and ranges for a stated reason
Multiple-year data
OECD paragraphs 3.75-3.79 say that multiple-year data are often useful but are not a systematic requirement. The Guidelines do not prescribe a fixed number of years and do not equate examining several years with averaging them.
Multiple-year data may help explain:
business and product cycles;
persistent or exceptional losses;
whether a candidate's result departs from its underlying functional profile; and
conditions before and during the tested period.
State which years are used, whether the analysis uses annual observations or an average, the averaging formula, and why that treatment improves reliability.
Full range, interquartile range, and point selection
An arm's length analysis may produce a single reliable figure or a range. OECD paragraphs 3.55-3.57 provide the relevant sequence:
remove observations with a lesser degree of comparability when they can be identified;
assess whether the remaining observations are relatively equally reliable; and
where unidentified or unquantified defects remain and the range has a sizeable number of observations, consider a statistical tool such as the interquartile range to improve reliability.
The interquartile range is therefore a response to residual comparability limitations, not an automatic first step in every OECD analysis. Local law may prescribe a particular range, calculation convention, or adjustment point, so document the jurisdictional rule separately.
Tip
Do not remove an observation merely to narrow the range or move the tested result inside it. Tie every exclusion and adjustment to a comparability fact that was not already considered.
7. Make the study reproducible
A reproducible study allows a qualified reviewer to repeat the search and calculations from the retained record. Use this checklist before sign-off.
Benchmarking reproducibility checklist
01The controlled transaction, parties, tested period, method, tested party, and PLI are defined before the search
02Internal comparables were considered and the conclusion is documented
03Database name, module, release or version, access date, coverage, and field definitions are retained
04Every search step, criterion, operator, and population count is recorded in sequence
05The unedited search export and a dated working copy are preserved
06Manual-review sources, access dates, reviewer, and company-specific accept/reject reasons are logged
07Financial fields are mapped to the PLI and checked for accounting consistency
08Adjustments include the rationale, source data, formula, and result for each comparable
09Range and quartile conventions are stated and the calculations are independently checked
10The tested result reconciles to the approved financial records and all final files are versioned
Minimum calculation checks
Recalculate every comparable's PLI from the retained source figures.
Confirm that signs, units, currencies, and decimal conventions are consistent.
Trace formula ranges to ensure no company or year is omitted or duplicated.
Check that the tested party and comparable definitions match for operating profit and the denominator.
Reperform adjustments from source data rather than accepting spreadsheet outputs at face value.
Lock the accepted set before viewing whether the tested result falls inside the range.
8. Refresh, roll forward, or run a new search
The OECD Guidelines do not create a universal three-year validity period for benchmarking studies. Local documentation rules and administrative practice may set their own expectations.
Use an annual facts review to decide whether the prior search can be rolled forward:
Question
If the answer is yes
Did the transaction, value chain, functions, assets, or controlled risks change materially?
Revisit the method, tested party, PLI, and search design
Did the tested party acquire or develop valuable intangibles?
Reassess whether a one-sided method remains reliable
Did the geography or economic conditions change materially?
Revisit market scope and comparability
Did the database change its coverage, fields, or financial restatements?
Reproduce the search using the current release and document the change
Did a comparable merge, become controlled, change activity, or lose usable data?
Re-screen the company and update the accepted set
Did local law or guidance change?
Apply the current local requirement and record its effective date
If none of these changes affects the methodology, update the financial data, repeat the qualitative checks, and explain why the original search remains suitable. A roll-forward is a documented conclusion, not simply a new spreadsheet column.
9. Structure of an audit-ready report
A concise report can still be complete. Include:
executive summary and tested result;
controlled transaction and functional analysis;
method, tested-party, and PLI selection;
internal-comparable review;
database selection and limitations;
search strategy and funnel;
manual-screening method and accept/reject log;
financial-data mapping and adjustments;
range calculation and conclusion;
reconciliation to the tested party's records; and
appendices containing the raw search, candidate log, source evidence, financial data, and calculations.
The methodology should be understandable without access to the preparer's memory. Keep source data and judgment together so the reviewer does not have to infer why a company was included.
Which transfer pricing benchmarking database is best?
No database is best for every transaction. Company accounts databases may support TNMM or CPM, agreement databases may support royalty or other CUP analyses, and market-data sources may support financial transactions. Compare coverage, field definitions, ownership data, historic releases, reproducibility, and licensing against the evidence the method requires.
Is manual screening necessary after a database search?
Usually, yes. Commercial databases often provide company-level codes, financials, and short descriptions rather than enough transactional detail to establish comparability. OECD paragraph 3.33 says database searches may need refinement with other public information. Record what was reviewed and the specific reason for every decision.
Must a benchmarking study use the interquartile range?
Not under the OECD Guidelines in every case. Paragraph 3.57 says a statistical tool such as the interquartile range may improve reliability where a sizeable set still contains unidentified or unquantified comparability defects. Local law may impose a specific approach.
How many years of data should a benchmarking study use?
The OECD does not prescribe a number. Paragraph 3.75 says multiple-year data should be used when they add value. Choose the period and any averaging convention based on the tested facts, economic cycle, data availability, and local rules.
How many comparable companies are enough?
There is no universal OECD minimum. Reliability depends on the quality of the comparables and the method, not a target count. Explain the search population, each exclusion, and why the final set supports a reliable conclusion.
Should loss-making companies be rejected?
Not automatically. OECD paragraphs 3.64-3.65 require attention to the facts. Exclude a loss-maker when its losses reflect non-comparable risks or conditions, not solely because the company has a loss.
Can industry codes establish comparability?
No. Codes are efficient population filters. Functional comparability still depends on products or services, functions, assets, risks, contractual terms, economic circumstances, and business strategies. Manual review tests those facts.
How often should a benchmark be refreshed?
Review it every year against the current transaction, comparables, data release, and local requirements. Run a new search when changes undermine the earlier methodology. If a roll-forward remains appropriate, record why and repeat the financial and qualitative checks.
Borys UlanenkoCEO of ArmsLength AI
