Multi-Year Trend Analysis: Reading Filings as One Series, Not a Pile of PDFs

Ask for five years of revenue and you’ll usually get five PDFs, not five consistent numbers. A company restates a prior year after an accounting change. A line item that used to sit under “other operating expense” moves under cost of sales. The fiscal year-end shifts. None of that is unusual, and none of it means the underlying business is any less comparable year to year, but it does mean that stitching the filings together by hand is where trend analysis quietly goes wrong.

The newest filing wins, but only for the year it actually covers

The basic rule sounds simple: when two filings disagree on the same fiscal year, the newer one wins. A company that restates FY2023 in its FY2025 annual report is telling you the FY2023 figures printed back in 2023 are no longer the right ones to use. A trend line that still shows the original, superseded FY2023 number next to the current FY2024 and FY2025 figures isn’t being conservative: it’s comparing a corrected year to two uncorrected ones, which produces a YoY swing and a CAGR that don’t describe anything real.

Getting this right means pivoting actual cells across every filing an entity has published, not just the latest one, and letting a restated year quietly take its value from whichever filing most recently covered it, while every year in the series still drills back to the specific source atom it came from, so nobody has to take the composite trend on faith. That’s the mechanic behind restatement detection and corroboration: it’s not enough to notice that a number changed, the trend has to know which version of that year is current and keep the ability to show its work for both.

A trend line only means something if the periods actually match

The other way multi-year analysis goes wrong is quieter: comparing periods that aren’t really comparable. A mid-year balance-sheet instant sitting next to year-end instants, a discrete quarter compared against a cumulative interim, or a fiscal year with a shifted year-end lined up against a calendar one: each of these produces a chart that looks fine and means nothing. Year-over-year growth should only be computed for consecutive, like-for-like years, and a CAGR needs a real elapsed span with two positive endpoints, or it should say so rather than print a number that implies more precision than the inputs support.

The same logic scales down to a single year’s interim data, where H1, 9M, and the full year have to compose into consistent discrete quarters rather than being read as independent figures, the same discipline covered in quarterly and interim TTM views. A five-year trend is really a chain of correctly-composed periods, and if one link in that chain silently mixes bases, the whole series inherits the distortion.

Trend and variance are the same question asked at different distances

A multi-year trend and a single-period budget-vs-actual variance are answering the same underlying question (did this move, and does it matter), just at different zoom levels. The same KPI variance analysis that flags a margin compression against this year’s budget is what should also be driving which multi-year swings get surfaced instead of buried in a wall of consistent-looking figures: a metric that’s flat for four years and jumps in the fifth is exactly the kind of thing a trend view should make impossible to miss, tied back to the actual filed numbers behind it rather than a smoothed line that hides where the jump happened.

None of this requires treating every year the same way. It requires knowing when a year has been superseded, when a period is genuinely comparable to the one next to it, and being willing to show “not computable” instead of a plausible-looking number built on mismatched bases. A multi-year trend that survives a restatement and a presentation change is worth more than one that was never tested by either. For more on how finance teams turn raw filings into decisions, see the finance team use cases hub.