Outlier: The Comparables Value Outside the Distribution’s Logic
The outlier defined: the comparable whose PLI sits outside the pool’s distribution logic — the detection, the treatment and the documentation of the exclusion or the retention.
Definition
The outlier is the comparable whose PLI sits outside the pool’s distribution logic — the value so far from the pool’s centre that it is the member’s difference, not the distribution’s tail: the one high-margin member in the low-margin pool, the one loss-making member in the profitable pool. The outlier is a comparability question before it is a statistical one — the detection (the statistical signal, the distance from the distribution) raises the question; the treatment (the exclusion, with the reason, or the retention, with the support) is the comparability decision the documentation carries.
| The stage | The content |
|---|---|
| The detection | The statistical signal — the value outside the pool’s distribution (the distance from the IQR, the gap to the nearest member, the dispersion read) — the outlier candidate, flagged |
| The investigation | The member’s facts — the products, the revenue mix, the assets, the one-time items, the data quality — why the value is where it is (the qualitative screen read on the flagged member) |
| The treatment | The exclusion (the member is not comparable on the facts — the reason stated, the record kept) or the retention (the member is comparable, the value is the member’s real position — the support stated, the extraordinary event adjustment where the one-time item is the cause) |
| The documentation | The outlier’s treatment in the benchmarking annex — the detection, the investigation, the treatment, the reason — the accept/reject record extended to the statistical flag |
The working read (the IQR vs full range guide and the loss-making comparables guide): the IQR’s construction (the p25–p75) is robust to the outlier by design (the middle 50%, the tails outside the range) — but the outlier is not ignored: the member outside the distribution logic is investigated and treated (the exclusion with the reason, or the retention with the support), because the pool’s comparability — not just its range — is what the file defends. The loss-making member is the outlier’s edge case: the loss-making comparables guide has the keep / adjust / exclude framework on the loss values.
Example
The pool (12 comparables, the adjusted PLIs): the IQR is 2.1%–3.4%, the median 2.8% — and one member at 9.6%. The outlier flag: 9.6% is outside the distribution logic (the gap to the p75, 3.4%, is 6.2 points — the distance the pool’s logic does not explain). The investigation: the member’s revenue mix (60% a high-margin product line the tested party does not carry), the asset structure (the owned brand, the tested party licenses it in) — the member is not comparable on the product mix and the brand ownership. The treatment: excluded, the reason stated (the product mix, the brand ownership — the comparability facts), the record kept in the accept/reject matrix. The pool is 11 members; the IQR recomputed on the 11 (the bounds, the median) — the outlier’s treatment, documented.
See also
FAQ
Is the outlier excluded automatically, by the statistic? No — the statistic flags (the distance, the gap, the dispersion read); the comparability decision treats (the investigation of the member’s facts, the exclusion with the reason, or the retention with the support). The automatic exclusion (the “drop the p95+” rule, applied without the investigation) is the benchmarking mistake — the statistical flag is the question, the comparability facts are the answer, and the answer is documented either way (the excluded member’s reason, the retained member’s support).
Outlier or loss-making member — how are the two treated differently? The outlier is the value outside the distribution logic (the high or the low, the distance the logic does not explain) — investigated on the comparability facts, excluded or retained with the reason. The loss-making member is the edge case where the value is negative (the loss) — the loss-making comparables framework (the temporary vs the structural loss, the data-missing treatment, the exclusion’s documentation). The loss-making member is often the outlier (the negative value is outside the profitable pool’s logic) — the framework’s keep / adjust / exclude applies, on the loss’s character (temporary, structural, the data quality).
Where does the outlier’s treatment appear in the documentation? In the benchmarking annex’s screening record: the detection (the statistical flag, the distance), the investigation (the member’s facts — the products, the mix, the assets, the one-time items), the treatment (the exclusion with the comparability reason, or the retention with the support), and the range’s recomputation (the IQR on the treated pool). The accept/reject matrix carries it — the outlier is the matrix’s statistical-flag row, treated like the qualitative-screen row: the decision, the reason, the record.
Run the screens as a study, not a spreadsheet
Quartyl applies the method, PLI and screening steps above as a pipeline — and keeps a documented reason for every exclusion.
Related docs
Interquartile Range (IQR): The OECD Arm's Length Range
The interquartile range defined: the band between the first and third quartiles of the comparable pool — the OECD\'s preferred construction of the arm's length range.
Read docLoss-Making Comparables: Keep, Adjust or Exclude?
The framework for loss-making companies in the comparable pool: temporary versus structural losses, the matching rule, missing data, and the documentation the TPO expects.
Read doc