Quantitative Screening: Filters, Order and Discipline (2026)

Quantitative screening is the part of the study that a reviewer can verify: every filter has a number, and every exclusion has a cause. That verifiability is its value — and its limit. Passing the quantitative screen proves nothing about actual comparability (a company can clear every ratio test and still make a different product for different customers). It only proves that the company is plausibly comparable on the dimensions the data covers. The qualitative review is what decides; the quantitative screen is what makes the decision reproducible.

The filter sequence

Apply the filters in this order, and no other:

Order	Filter	Typical criterion	Why this order
1	Industry / sector classification	NIC/NACE code or a defined search string over product lines	Sets the population; everything else runs inside it
2	Size	Revenue (or asset) band around the tested party, e.g. 0.5x–5x	Scale drives cost structure; a 40x-larger “comparable” is not one
3	PLI denominator availability	The tested party’s denominator (revenue, operating costs) present, non-zero, consistent across the data years	A company you cannot compute the PLI for cannot enter the pool, however good it otherwise looks
4	Profitability sanity	Loss-making across all data years, or extreme outliers vs the cohort	Structural losses signal a different business reality; flag rather than auto-drop where temporary
5	Geography	The countries in the search	Applied here, not first, so the size/sector population is visible before it is cut

The order matters for two reasons. First, each filter narrows the population, so a filter’s effect (how many companies it drops) is only interpretable against the population it receives. Second, reviewers re-run the screen in this order; a screen that cannot be re-run in sequence is a screen that cannot be defended.

Size filters in practice

The size band is expressed as a ratio to the tested party, applied to revenue or total assets depending on the PLI (revenue-based PLIs → revenue band; cost-based PLIs → cost or asset bands are better proxies).

• Wider is not safer. A band of 0.25x–10x captures companies whose cost structures are not comparable to the tested party’s; the filter stops doing work and the qualitative screen inherits the mess.
• Narrower is not better. A band so tight it returns four companies produces a range with no statistical meaning.
• The band is a fact, not a dial. The same band, every year, unless the tested party’s size moved. Changing the band year-on-year without a documented reason reads as range management — and it is.

Loss-making companies

A loss-making candidate needs a classification, not an automatic drop:

Situation	Treatment
Loss in one of the data years, otherwise healthy	Keep; the multi-year average already dilutes it. Note it.
Loss in all data years, with a documented turnaround in progress	Keep or reject — document the reasoning either way
Loss in all data years, no turnaround evidence	Reject: the PLI is not measuring the same thing for this company as for the tested party

The reason string in the log is the point: “loss-making FY 2023-24 to 2025-26, no documented turnaround” is a screen; “unprofitable” is noise.

The exclusion log

Every company the screen drops — and, for the audit trail, every company it keeps — gets a row:

Company	Data year	Sector	Size	Denominator	Profitability	Geography	Disposition	Reason
J	2023-25	✓	✗ (₹38 cr)	✓	✓	✓	Reject	Below 0.5x size band
K	2023-25	✓	✓	✓	✗	✓	Reject	Loss-making all three years

The log is the exhibit. When the TPO asks “why is company J not in your pool?”, the answer is a row, with the threshold and the value in the same line. A screen without a log is a range without a defence.

Common mistakes

1. Dropping companies before the log exists — the deletion happens in the spreadsheet, the reason gets written (or not) later. The log must be written as the screen runs.
2. Thresholds that flatter the range — the size band quietly tightened after a first run produced an uncomfortable range. The criteria are frozen before the first run.
3. A profitability filter doing a qualitative job — dropping a company because “its margins look different” is a qualitative decision wearing a quantitative costume; it belongs in the next stage, with a qualitative reason.
4. Forgetting the denominator filter — the pool includes companies whose operating-cost line is a mix of functions, so the PLI values are computed on an incommensurate base. The PLI is only comparable if the denominator is.

FAQ

What size band should I use? Start from the tested party’s position: a 0.5x–5x revenue band (or the cost/asset equivalent for cost-based PLIs) is the common Indian practice range. Widen or narrow only with a documented, transaction-based reason — and keep that reason with the criteria.

How many companies should survive the quantitative screen? Enough for the qualitative stage to be meaningful — typically ten to twenty, so that qualitative rejection of a few does not leave a pool with no statistical weight. Under five survivors means the search was over-narrow; go back to the search design, not to looser filters.

Do I re-run the whole screen every year? Yes — same criteria, fresh data. The prior year’s log is kept as the trend line. Re-running is cheap; explaining a rolled-forward screen is not.