Multi-Year Data: How Averaging Is Applied

Multi-year data is the case where one fiscal year is not a representative sample of the tested party or the pool — the start-up ramp, the cyclic industry, the restructuring year. The product handles it in three moves: it detects the years in the file, it computes the PLI per year for every company, and it consolidates the years into the single value the range is built on. This page is how those moves work, what is stored, and the consistency discipline that keeps a multi-year study defensible.

How the years are detected

Multi-year files carry the year in the column name — one row per company, with a column per line item per year:

Company | Revenue_2022 | Revenue_2023 | Revenue_2024 | Cost_2022 | ...

The detection recognizes the common dump conventions — Revenue_2022, Revenue 2022, Revenue (2022), Revenue FY22, Revenue-FY2022 — for each year-aware line item, and records the detected years alongside the column mapping on the study’s dump record. A file with no year-suffixed columns is a single-year file: the detected years are one, and the run behaves accordingly. The file layout and the detection itself are in Uploading Your Data File.

The three computations

For each company, the engine works the years in order:

1. Per-year PLI. For every detected year, the PLI on the study’s definitions — the numerator over the denominator for that year, per the chosen margin type. A year where the denominator is missing or zero is not a ratio; it is recorded as such.
2. The consolidated PLI. Across the detected years, the consolidation follows the Golden Rule: sum the numerators across the years, sum the denominators across the years, and divide. The per-year margins are never averaged against each other — averaging ratios is a different number than the ratio of the sums, and the engine does not do it.
3. The per-period record. The per-year PLIs are kept on the study’s data, alongside the consolidated value and the count of years used. That is the per-period view: the trend the file can show, year by year, without re-running anything.

The consolidation runs over the comparable pool. The tested party’s own position is not derived from the file: it is the margin you configure in the parameters step, and it is the single number the range is measured against. That asymmetry is a discipline you carry, not something the tool enforces — see the consistency section below.

The choices, and what they do

Choice	Where it is set	What the engine does
Single Year (Latest)	The Multi-Year PLI Averaging selector in the parameters step	The latest detected year is the study year: one PLI per company, no consolidation
Multi-Year (Golden Rule — Aggregate Raw Financials)	The parameters step — this is the default	The numerators are summed across the detected years, the denominators are summed, and the ratio of the sums is the company’s PLI
simple (legacy stored value)	Not offered in the wizard	Older studies may carry it; the engine treats it exactly as the Golden Rule option, so nothing about the computation differs

Two consequences of the design:

• The label never changes the arithmetic. Because every multi-year setting aggregates the raw financials (sums over sums) rather than averaging per-year ratios, a three-year pool is the same pool whichever legacy label sits on the study; the stored method is a declaration, not a switch between computations.
• The per-period values survive. The per-year PLIs and the years-used count are on the study’s data for every comparable, so the trend read — the direction and the cycle, the distribution per period — is available from the study itself, and the workbook lays the ratios out year by year (up to three fiscal years displayed).

The method theory — when a single year is not enough, the Golden Rule arithmetic as a matter of technique, and the loss-year treatment — is in the multi-year averaging guide; the multi-year average entry has the one-paragraph definition including the per-period method.

Consistency: the tested party and the pool on the same clock

The multi-year choice binds the pool. What it does not do is derive the tested party’s side, and the honesty of a multi-year study rests on understanding that split:

• The pool is consolidated on the detected years. Every comparable’s PLI is the ratio of summed numerators to summed denominators across the years the file carries — or, under Single Year (Latest), the most recent year alone. One rule, applied to the whole population.
• The tested party’s position is a parameter. The margin the conclusion is measured against is the value you enter in the parameters step. The product does not compute a multi-year tested-party PLI from a second data set, and it cannot check that your entered figure was itself struck on the same years and the same Golden Rule as the pool. That check is yours, and it is the first thing an examiner runs when the pool is multi-year.
• The method is stated. The choice is on the study’s parameters block and is printed in the report — the examiner reads the declared method from the file, not from an inference.

The TPO’s expectation is the consistency itself: same periods, same method, stated — applied to the tested party and the comparables without distinction. A multi-year file whose tested-party side cannot show all three is a file the examination opens on that point.

What screening does with the years

The multi-year checks run before the financial filters, so a rejected company carries the specific reason:

Check	The company is rejected when
Incomplete multi-year financials	Fewer than two detected years carry both operating profit and revenue — the reason names the years that are missing
Persistent operating loss	Every year with data is loss-making at the operating line
High intangible intensity	Intangible assets exceed 10% of total assets, where both figures are present

A company that is loss-making in one of the years but positive in another survives the loss check and screens on the period numbers — the multi-year design absorbing exactly the single-year noise it exists for. Note what the first two checks read: the universal operating-profit and revenue columns, not the selected margin’s own numerator and denominator, so the year-level completeness bar is the same whichever PLI the study runs on. Note also the floor the first check sets: a multi-year file needs at least two usable years per company, so a comparables dump with one populated year of a three-year header row is an incomplete record, not a single-year observation. The loss-year edge cases, as a matter of technique, are in the multi-year averaging guide.

When averaging masks a trend

The consolidated value is a smoothing instrument, and the product keeps the per-period record precisely because the smooth line can hide what the examination cares about: the pool drifting across the period, a one-off year averaged in silently, a set that only looks stable because two opposite years cancel. The defence is the trend itself — the per-year PLIs across the pool, the distribution struck for each displayed year, and your explanation of the movement. The tested party’s side of that comparison is the single configured margin, so the per-year narrative for the tested party is yours to evidence from its own accounts. The per-year values are stored on the study rather than discarded at consolidation, which is what makes the pool’s trend readable; the full treatment of the trend risk is in the multi-year averaging guide.

FAQ

Does the product recompute the per-year PLIs at review time? No — the per-year values, the consolidated value, and the years used are all on the study’s analysis data after the run. The review reads the record; it does not re-derive it.

Can a multi-year file have missing years for one company? It can, within a limit. A gap in one year of a three-year header is consolidated over the years that do carry both operating profit and revenue; the incomplete- multi-year check rejects a company only when fewer than two detected years carry that pair, and the rejection reason names the years it looked for. A dump where a company has a single populated year is therefore an incomplete record rather than a one-year observation.

Where is the averaging choice set? In the parameters step of the wizard, on the Multi-Year PLI Averaging selector beside the PLI and the filters — Single Year (Latest) or Multi-Year (Golden Rule — Aggregate Raw Financials), the latter by default. See Choosing PLI, Years and Benchmarking Parameters. It is recorded on the study, part of the run’s record.