Published June 20, 2025Updated September 7, 202618 min read
Quantitative Screening Filters for Transfer Pricing Benchmarking: Complete GuideQuantitative Screening Filters for Transfer Pricing Benchmarking: Complete Guide
Learn how to apply quantitative screening filters systematically in transfer pricing benchmarking. This guide covers industry codes, independence indicators, size thresholds, and a practical filter sequence for narrowing a broad company universe to a defensible candidate set.
Borys UlanenkoCEO of ArmsLength AI
18min read
Contents↓
Contents
TL;DR key takeaways
A commonly efficient filter sequence is: industry → geography → independence → status → data availability → size → profitability. Other sequences can work if documented consistently.
Use 4-digit NACE/SIC codes as a starting point for precision. Validate with keywords and qualitative review—codes alone can be outdated or misassigned.
Independence is a core screening criterion: typically exclude companies with >50% ownership (BvD indicators C, D). Some jurisdictions apply stricter thresholds.
Document every filter, threshold, and the count of companies remaining at each step. This audit trail demonstrates transparent, systematic, traceable screening (per EU JTPF recommendations).
Many practitioners target a few dozen companies after quantitative screening for manual review, often expecting to finalize a smaller, higher-quality set after manual review. There is no required minimum or maximum number of comparables.
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 Are Quantitative Screening Filters?
Before launching an external-company database search, check whether reliable internal comparables exist. OECD's typical process reviews internal comparables before external sources (), and EU JTPF guidance states that internal comparables should be preferred where they exist and meet the comparability standard. The search for comparables follows the delineation of the transaction and does not drive it.
Quantitative screening filters narrow a database of thousands of companies to a manageable set of potential comparables before manual review. The key filters are: industry codes (NACE/SIC/NAICS), geographic location, independence indicators, size thresholds, financial data availability, and profitability requirements.
Apply filters in a logical sequence from broad to narrow—starting with industry and geography, then independence and status, then size and profitability. Document all criteria and the number of companies remaining at each step (per EU JTPF transparency recommendations). A well-executed quantitative screen yields a manageable set of candidates for qualitative review—often a few dozen companies—from which practitioners finalize a smaller, higher-quality set after manual review.
Quantitative Filter Catalog (At a Glance)
Use this as a checklist of the most common quantitative filters. Exact thresholds depend on the tested party and data universe—what matters is that each filter is economically motivated and traceable.
Filter Category
What it screens
Common patterns (illustrative)
What tax authorities look for
Industry classification
Wrong line of business
Start with specific codes (e.g., 4-digit NACE), broaden only if needed
Evidence codes match real activities (not just labels)
Geography
Different market conditions
Start local; broaden to region if local data is sparse
Quantitative screening is the first step in a comparables search. Its purpose is to filter a database universe—often containing millions of companies—down to a few dozen or hundred candidates that warrant individual examination.
Note
Why It Matters: Without systematic quantitative filters, you'd either (a) review thousands of companies manually, which is impractical, or (b) apply filters arbitrarily, which is indefensible. Quantitative screening ensures efficiency and consistency.
The process uses objective, measurable criteria that databases can apply automatically:
Industry classification — selecting companies in the same line of business
Geography — limiting to relevant markets
Independence — excluding subsidiaries and affiliates
Size — filtering by revenue, assets, or employees
Data availability — requiring minimum years of financial data
Profitability — screening out persistent loss-makers or distressed companies
Operational status — focusing on active, going-concern businesses
After quantitative filters, the remaining companies undergo manual (qualitative) screening—reviewing business descriptions, functional profiles, and financial details to confirm true comparability.
Industry Classification Filters
Industry filters ensure only companies engaged in comparable economic activities are considered. They rely on standardized classification codes.
Common Code Systems
System
Origin
Digits
Coverage
NACE Rev.2
European Union
4
EU-standard; used in Amadeus, Orbis
SIC
United States
4
U.S. Standard Industrial Classification
NAICS
North America
6
U.S./Canada; more modern than SIC
ISIC
United Nations
4
International standard; less common in TP databases
Most databases support multiple systems. Orbis, for example, allows searching by NACE Rev.2, NAICS, US SIC, and proprietary BvD sectors.
Depth of Codes: 2-Digit vs. 4-Digit
Best practice: Use the most specific code that captures the tested party's activity—typically 4-digit level.
4-digit NACE 46.52 = "Wholesale of electronic and telecommunications equipment" — specific, targeted
2-digit NACE 46 = "Wholesale trade" — includes wholesalers of food, textiles, machinery, etc.
Using 4-digit codes yields a homogeneous set performing very similar functions. Broader 2-digit codes introduce "false positives"—companies in the same sector but with different economics.
Warning
When to Broaden: Only expand to 3-digit or 2-digit codes if the specific code yields insufficient results. If NACE 46.52 returns only 3 companies in your target geography, step back to 46.5 before abandoning the search.
Primary vs. Secondary Codes
Companies have one primary code (main business) and may have secondary codes (other activities). Filter by primary code initially—this ensures the company's core business matches your tested party. If results are too narrow, you can include secondary codes, but expect more manual vetting.
Code Accuracy Caveat
Industry codes aren't perfectly precise. They're often self-reported or assigned by analysts and may be outdated. A company's registered code doesn't always reflect its actual current business. The EU JTPF notes industry classification inconsistencies and recommends combining codes with other elements.
Best practice: Codes are a starting point; validate with keywords and qualitative review of business descriptions. Don't rely on code granularity alone.
Geographic Filters
Geographic filters limit the search to companies operating in relevant markets. Market differences—competition, cost levels, consumer behavior—can significantly affect profitability.
Local vs. Regional Approach
Start with the tested party's country. Many jurisdictions prefer local comparables where available, though explicit legal requirements vary:
Jurisdiction
Preference
Poland
Domestic comparables preferred where reliable; regional/EU benchmarks commonly used when domestic data is insufficient (no explicit legal "local-first" rule)
India
India has no explicit legal preference for domestic over foreign comparables; choice depends on the tested party and facts. In practice, local comparables often require stronger justification to bypass
Japan
Domestic comparables are preferred in Japan's framework; foreign comparables may still be used where they are more reliable or where domestic data is insufficient
EU general
Pan-European acceptable, especially if local data is sparse; EU JTPF notes pan-European searches shouldn't be rejected solely for being non-domestic
If a single-country search yields insufficient comparables (a very small set may justify broadening if it is not sufficiently reliable for the facts and method), expand to the region:
DACH region (Germany, Austria, Switzerland) for German tested parties
Nordic countries for Scandinavian transactions
Pan-European when single-country data is insufficient
Documenting Geographic Rationale
Whatever scope you choose, document why:
"A local search in [Country X] yielded only 3 independent companies after all filters. Therefore, the search was broadened to include [Country Y and Z], which share similar economic conditions and wage structures with [Country X]."
Independence filters exclude companies that are part of multinational groups. This is critical—comparables analysis relies on observing uncontrolled results. A subsidiary's financials might be influenced by transfer pricing, making them unsuitable comparables.
BvD Independence Indicator
Bureau van Dijk databases (Orbis, Amadeus) use a standardized indicator:
Code
Meaning
Include?
A
No shareholder owns >25%
✅ Yes
B
No shareholder >50%, but at least one 25-50%
✅ Yes
C
Majority owned (>50%)
❌ No
D
Direct subsidiary (>50% direct ownership)
❌ No
U
Unknown ownership
⚠️ Case-by-case
Standard Practice: Include A and B, Exclude C and D
The 50% threshold balances data availability with comparability. Applying stricter independence criteria (e.g., requiring no shareholder >25%) can materially reduce the available sample—sometimes to the point where a meaningful analysis becomes difficult.
Note
Why 50%? A company with a 40% minority shareholder may remain in the initial pool under a >50% starting-point test, subject to further review of governance structure and related-party influence. Some jurisdictions apply stricter thresholds (e.g., less than 25% ownership).
Handling Unknown (U) Companies
Companies with "U" status have no ownership information available. A common conservative approach is to exclude them by default—they could be undisclosed subsidiaries. If your sample is critically small, you might include U-rated companies but investigate each manually. Document your reasoning.
Bottom line: Independence is a core screening criterion for external company comparables. In BvD terms, analysts typically include A/B and exclude C/D, subject to local rules and the specific facts of the analysis.
Size and Revenue Filters
Size filters exclude companies that are extremely larger or smaller than the tested entity. Scale differences affect profitability—larger companies may have economies of scale; very small companies may be volatile or unsustainable.
Common Thresholds
These ranges are frequently used in practice; adjust based on tested party characteristics and local norms:
Filter
Common Ranges
Purpose
Minimum revenue
€1M - €10M
Exclude micro-enterprises
Maximum revenue
5-10x tested party's revenue
Exclude companies with different scale economics
Employees
>10 employees (optional)
Exclude shell companies or dormant entities
Setting Thresholds Relative to Tested Party
The tested party's size should guide your thresholds. For a €25M distributor:
Minimum: €5M (excludes one-person operations)
Maximum: €100-250M (excludes multinational distributors with different scale)
If the tested party is €500M, you might set minimum €50M and no maximum.
Watch the Sample Count
Size filters significantly reduce candidate numbers. If moving from €1M to €5M minimum drops your pool from 200 to 30, that's fine. If it drops from 30 to 3, reconsider.
Financial Data Availability Filters
These filters ensure companies have sufficient data for analysis. There's no point including a company if it lacks the financial metrics you need.
Years of Data
Common practice: Require at least 3 years of financial data where available. Transfer pricing analysis often uses multi-year data to understand business cycles and improve comparability—though this is context-dependent, not a universal minimum. Hungary's new transfer pricing decree (45/2025 NGM), published on 23 December 2025, applies mandatorily for tax years starting in 2026 and may be applied early for 2025. As a main rule, the review period covers the three years preceding the year under review, with data available for all three years.
Required Fields
For TNMM/CPM analysis, you need:
Revenue (operating revenue/sales)
Operating profit (EBIT or similar)
For ROA: Total assets
For WCA: Accounts receivable, payable, inventory
Set database criteria to exclude companies with null values in critical fields for your analysis years.
Consolidated vs. Unconsolidated
Entity-level (unconsolidated) financials are often preferable where they better match the tested party, but consolidated accounts should not be automatically excluded at the quantitative stage. If consolidated comparables remain in the pool, accept them only where comparability can be demonstrated; consolidated accounts involving more than 10 subsidiaries should generally be excluded.
Profitability and Loss Exclusion Filters
Profitability filters screen out companies with problematic financial patterns. However, they're contentious—excluding all loss-makers can introduce upward bias.
Common Approaches
Filter
Description
Risk
All-year loss-makers
Exclude if losses in every year of analysis period
Reasonable
Any-year loss-makers
Exclude if loss in any single year
Too strict—may bias results
Multi-year average negative
Exclude if 3-year average profit < 0
Reasonable
At most 1 loss year in 3
Allow companies with one bad year
Balanced approach
Warning
Bias Risk: Excluding all loss-makers can create upward bias—you're essentially selecting only "winners." The Italian Supreme Court (Decision No. 19512, 2024) ruled that potentially comparable entities cannot be excluded from the comparability analysis solely because they have low profits or losses—the exclusion must be based on substantive economic analysis, not automatic rules.
Practical Guidelines
These are policy choices, not universal rules—adjust based on case-specific analysis consistent with OECD guidance:
Exclude companies with losses in all analysis years (likely non-viable or facing extraordinary circumstances)
Allow companies with 1 loss year out of 3 (may reflect normal business cycles)
Consider excluding companies with negative equity (signal of financial distress)
Apply filters in a logical order—from broad to narrow. This commonly efficient sequence maximizes efficiency and ensures the most relevant cuts happen first. Other sequences can be equally defensible if documented and consistently applied.
Commonly Used Sequence
INDUSTRY (NACE/SIC code)
GEOGRAPHY (country/region)
INDEPENDENCE (exclude subsidiaries)
STATUS (active only, exclude start-ups)
DATA AVAILABILITY (≥3 years)
SIZE (revenue range)
PROFITABILITY (loss filter)
MANUAL SCREENING
Why This Order Works
Industry first — Establishes the fundamental universe of comparable businesses
Geography second — Narrows to relevant markets
Independence early — Removes thousands of subsidiaries before detailed filtering
Status/Data before size — Ensures companies are usable before evaluating metrics
Size before profitability — Fundamental comparability factor comes before outcome-based filters
Profitability last — Most contentious filter; applied when you can assess impact on sample
This sequence ensures that when you apply profitability filters, you're evaluating companies of similar scale in the same industry—not mixing apples and oranges.
Note
Note on Sequence: The EU JTPF supports a step-based approach with transparent documentation of criteria and outcomes, but does not canonize any specific ordering. What matters is consistency and traceability—your sequence may vary depending on database functionality and tested party facts.
The Accept/Reject Matrix (Your Audit Trail)
An accept/reject matrix is the simplest way to make your screening defensible. It shows that companies were included or excluded based on predefined criteria, not outcome-driven judgment.
Minimum columns (recommended)
Company
Stage (Quant/Manual)
Key reason
Decision
Evidence / note
Company A
Quant
Independence
Reject
BvD indicator C (majority owned)
Company B
Quant
Data availability
Reject
Missing EBIT for an analysis year
Company C
Manual
Functional mismatch
Reject
Description indicates manufacturing, not distribution
Company D
Manual
Accept
Accept
Functionally similar; no disqualifying issues found
What makes it audit-ready
Specific reasons (avoid “not comparable” as a standalone reason)
Consistency (if one company is rejected for a threshold, others failing that threshold should be rejected too)
Traceability (link each reason to your criteria and the evidence source)
For the review that follows your database search, AI-supported benchmarking software can help organize company evidence and screening decisions. Keep the final comparability assessment with your team.
Common Screening Mistakes
Mistake 1: Over-Filtering
Symptom: Zero or near-zero companies remain after screening.
Cause: Too many restrictive criteria stacked together—narrow industry code AND single country AND strict size AND no losses ever.
Fix: Relax filters progressively. Check which filter eliminated the most companies and consider whether it's too strict.
Mistake 2: Under-Filtering
Symptom: Hundreds or thousands of companies remain, requiring impossible manual review.
Cause: Broad 2-digit industry code, no size filter, no independence filter.
Symptom: Companies in the set don't actually do what the tested party does.
Cause: Relying on codes that don't match actual activities, or codes that were accurate years ago but not now.
Fix: Verify code by reading descriptions of companies in that category. Consider multiple codes if the tested party spans activities.
Mistake 4: Ignoring Independence
Symptom: Final comparables include subsidiaries of large groups.
Cause: Database didn't have independence filter applied, or filter was set incorrectly.
Fix: Always apply independence filter. For public companies, manually check for parent ownership.
Mistake 5: Arbitrary Thresholds Without Justification
Symptom: Revenue cutoff of exactly €7.3M with no explanation.
Cause: Setting thresholds to achieve a desired outcome rather than based on comparability rationale.
Fix: Every threshold should be tied to tested party characteristics or standard practice. Document reasoning.
Practical Example: Screening a German Distributor
This illustrative example shows how filters progressively narrow the candidate pool. Actual counts vary by database version, year, and specific filter settings.
Tested Party: German limited-risk distributor of electronic components, €25M annual revenue.
Database: Orbis/Amadeus
Step-by-Step Search (Illustrative)
Step
Filter Applied
Companies Remaining (Illustrative)
0
Starting pool: NACE 46.52 worldwide
~3,500
1
Geography: Germany, Austria, Switzerland
~800
2
Independence: A and B only (no >50% owner)
~350
3
Status: Active, incorporated before 2020
~330
4
Data availability: ≥3 years (2021-2023)
~200
5
Revenue: €5M - €100M
~80
6
Profitability: Not loss-making all 3 years
~60
Output
Proceed to manual screening
~60
Threshold Rationale + Sensitivity (How to Defend Your Cutoffs)
Thresholds should be set to improve comparability, not to force an outcome. The easiest way to defend them is to document:
Why the filter matters economically (e.g., scale differences affect margins; subsidiaries may embed transfer pricing)
Why the threshold fits the tested party (anchored to tested party size/market)
What it did to the candidate pool (counts before/after each step)
Using the illustrative funnel above:
After the data availability step, the pool was ~200.
Applying the revenue threshold (€5M–€100M) reduced it to ~80.
That reduction is normal. But if a single threshold wipes out almost everything, you should reconsider whether it’s too strict (or whether an earlier filter was too narrow). Your audit defense is the narrative: “we set thresholds to align scale and ensure data completeness, and we monitored the impact at each step.”
Documentation Summary
"The comparables search was conducted using Orbis (accessed December 2025). Companies were selected based on NACE Rev.2 code 46.52 (wholesale of electronic equipment), limited to DACH region to ensure market comparability. Only independent companies (BvD Independence A or B) with at least 3 years of financial data were included. Revenue thresholds of €5M-€100M were applied to align with the tested party's scale (€25M). Persistent loss-makers (losses in all three years) were excluded. The quantitative screening yielded 60 companies for manual review."
Apply filters from broad to narrow: industry → geography → independence → status → data availability → size → profitability. This sequence ensures maximum efficiency—you eliminate large swaths of irrelevant companies early (different industries, subsidiaries) before fine-tuning on size and profitability. Independence and data availability are objective, easy-to-apply filters that should come before more subjective decisions.
How many comparables should remain after quantitative screening?
Many practitioners aim for a few dozen companies for manual review, from which they finalize a smaller, higher-quality set. The reliability of the final set matters more than hitting a target count—there is no required minimum or maximum number of comparables (EU JTPF explicitly states that even one or two comparables may be acceptable in some cases). If your set after quantitative filters is very small, consider broadening geography or industry code if doing so would improve reliability. If you have more than 200, tighten size thresholds or add stricter criteria—manual review of hundreds is impractical and suggests under-filtering.
What's the difference between BvD Independence indicators A, B, C, and D?
A = No shareholder owns >25% (truly independent). B = No shareholder >50%, but one owns 25-50% (minority-held, still acceptable). C = Majority owned by a shareholder >50% (subsidiary—exclude). D = Direct subsidiary with >50% direct ownership (exclude). Standard practice: include A and B, exclude C and D. The U (Unknown) category requires case-by-case judgment.
Should I use NACE, SIC, or NAICS codes?
Use the system native to your database and jurisdiction. For European searches (Amadeus, Orbis Europe), NACE Rev.2 is standard. For U.S. searches (Compustat), SIC or NAICS work well—NAICS is more modern. Most global databases support multiple systems and can cross-reference. The key is using 4-digit specificity in whatever system you choose.
How do I set appropriate revenue thresholds?
There is no fixed OECD or EU turnover rule. Revenue or employee thresholds should be anchored to the tested party's scale and business model, and justified by how they improve comparability. For example, for a €25M tested party, a range such as €5M-€100M might be reasonable—but treat such figures as illustrations, not defaults. Always document the rationale: explain why your chosen thresholds improve comparability for your specific facts.
What if my search yields too few comparables?
First, verify your filters aren't too restrictive—check if any single filter caused a dramatic drop. Then consider: (1) expanding geography to a broader region, (2) including a related industry code, (3) lowering minimum revenue, (4) accepting companies with one loss year instead of zero. Document any broadening: "Local search yielded insufficient comparables; expanded to pan-European as permitted under OECD guidelines when local data is sparse."
Should I exclude all loss-making companies?
No. Automatically excluding all loss-makers is too aggressive and can bias results upward. explicitly states companies shouldn't be rejected solely for having losses. Exclude only persistent loss-makers (losses in all years of the analysis period) and investigate single-year losses to determine if they reflect normal business cycles or extraordinary circumstances. Document your analysis for each loss-maker considered.
How do I document my screening criteria?
Create a search strategy report listing: (1) database used and search date, (2) each filter criterion with threshold, (3) rationale for each threshold, (4) number of companies remaining after each filter, and (5) final count proceeding to manual review. This documentation should allow someone to replicate your search. Many practitioners include this as a dedicated section in the benchmarking study or as an appendix.
Related OECD Guidance
The OECD Transfer Pricing Guidelines provide guidance on comparable selection and screening:
Borys UlanenkoCEO of ArmsLength AI
