Costs did not fall. They split.

The decision this is for

A brand manager or senior brand manager writing the 2027 plan, with the director above them approving it. The question in the room is whether the margin target can be met without another price increase.

The two answers lead to opposite moves. If the recovery came from input costs falling, the room is structural and the plan can spend price back into pack architecture to buy units. If it came from prices rising, the room is borrowed from the shopper, it is already spent, and the plan has to take price again or cut cost.

On this evidence it is neither, for most of the category — and which one it is for your business depends on a commodity exposure you can look up. That is the usable part.

The measurement

The category and the control moved in opposite directions. That is the whole reason the control is here: without it, a fall of under a point looks like the economy. With it, it is a category.

The inputs did not move together

The oils and the packaging metal came down. The tree crops went up, and cocoa doubled. That is why the category median barely moved while individual companies moved enormously — the relief was real, and it was commodity-specific.

By industry group, exposure shows

Sugar and confectionery is the worst group in the set at −5.96 points, over a period when cocoa bean import values more than doubled. Bottled and canned soft drinks is among the best at +2.13, over a period when aluminium fell 16%. Grain mill products, poultry and cosmetics all held or gained.

Every filer, worst move to best

All 42. No filer was added or dropped by hand after the pull — the set is every food and household filer above $1bn in CY2024 revenue that tagged enough data, and the ones that did not are listed further down with the reason.

The two worst moves in the set are the two largest chocolate companies. The six best are an energy drink, a bottler, a corn processor, a cereal company, a poultry processor and a cosmetics house — none materially exposed to cocoa or coffee.

The control, published in full

Drawn by a declared rule with a fixed seed from every filer above $1bn outside food and household, excluding finance and insurance where gross margin is not a meaningful construct. Nothing hand-picked.

What we said beforehand, and what it cost

Four explanations with probabilities, written down before any data was pulled. Two came out wrong, including the one we led with.

We expected a sector that took heavy price and then watched commodities fall to be sitting on widened margins. The opposite happened, and the control rules out blaming the economy. The candidate that won was the one written down as the boring one.

Four ways this measurement can go wrong, and what we found

Anyone computing margins from SEC data will hit these. They are published because the figures above are only worth what the method is.

Usable filers by version of the instrument: 34, then 48, then 50 of 59. The finding did not change sign across those versions, but the filer set did — and three of the largest filers in the category were in the recovered group.

What was dropped, and why

Zero filer-years were rejected by the plausibility guard — every year had at least one revenue candidate inside the band once the candidates were tested rather than ranked by name.

What this study cannot tell you

• Nothing here reads a filing’s prose. The SEC’s document host is unreachable from the workspace that ran this, so there is no price-versus-volume split and no management explanation of any margin move. Every figure is a tagged fact or computed from two of them.
• Causation is not established. The commodity split and the filer split are consistent with exposure driving the outcome. No company’s cost line was decomposed by input, because that decomposition is prose.
• The consumer side is missing entirely. Shelf prices and retail dollars need two sources this pass could not reach. Nothing here observes what a shopper paid.
• The qualitative half is missing, so this read is not finished by the standard this estate holds itself to. It is a set of numbers with nobody in it, and saying so is better than staying silent about it.
• An import unit value is not a company’s input cost. It is a national average across all importers, in dollars per kilogram, before contracts, hedging, grade and freight.
• Five of 66 commodity requests failed with a rate limit and are named rather than reported as absence: cocoa beans 2019, soybean oil 2020, cane sugar 2022, rice 2019, paper cartons 2023. The 2019 cocoa gap is why that row starts at 2022.
• Quantity is flagged estimated on 30 of 66 commodity rows, including every aluminium year and five of six coffee years. A unit value with an estimated denominator is partly derived.
• One pass, one day, no re-run, per the stopping rule set before retrieval. The control is a seeded sample of 15, not a census.

How this page changes

A standing study, not an article. Corrections are published in place and dated, never applied quietly.

Related reading, free and in full

• All standing studies — the index and the method each one follows
• The same spread result in a different measure — 1,633 filers, and the category median predicts nothing
• Two AI engines, two different security vendors, 72.7% of the time
• A premium-format allocation, and the number both answer engines stated that none of their sources contain