When Food Becomes a Luxury

Food bought for consumption at home cost roughly a third more in the middle of 2026 than it had in 2019. NielsenIQ scanner data put the cumulative increase at 33 percent, the Department of Agriculture’s own series at a little under 30, and neither has retreated from that plateau. What moved instead, beginning late last year and speeding up through the spring, was the rate of climb, which decelerated far enough that the official forecasting machinery now treats grocery inflation as ordinary. The Economic Research Service’s October 23 release projected a 2.3 percent rise in food-at-home prices for the calendar year, marginally below the twenty-year average and the third downward revision in three consecutive monthly updates.

Read alone, that forecast is sound and reassuring, and a good deal of commentary has read it that way. Put it beside the physical quantity of food moving through American stores and the reassurance shrinks. Grocery units sold were 1.8 percent lower year over year in June, the fifth consecutive month of contraction, following four months of annual declines near two percent spread fairly evenly across regions and store formats. Dollar sales held up longer than units did, because price increases were wide enough to cover for the difference, and somewhere around February they stopped being wide enough.

Both series are accurate. Neither one on its own describes what is happening. Prices can stop rising while the quantity a household can buy at those prices keeps falling, and it is the second of those that reaches a plate. What the country entered in 2026 is not an inflation emergency in any technical sense, since the rate is mild against the 2022 peak and several categories are genuinely cheaper; it is a problem of level and of distribution, and the two have settled into an arrangement stable enough that every improving month of headline data makes the underlying condition harder to see.

Groceries make an unusually good instrument for watching household finances, and mostly for reasons that have nothing to do with groceries. A car purchase can be postponed two years, a roof one, a dental appointment months, and none of those deferrals becomes visible to a statistician until it turns into a default. Eating cannot be postponed at all. Food is therefore the budget line that absorbs pressure generated elsewhere, and the point at which a household starts trimming is the point at which its other adjustments have run out. When the quantity of food crossing a till falls in a country with low measured unemployment, the likeliest explanation is that the money available for food has fallen, though other explanations exist and get a hearing below.

The rest of this article follows that question through several registers: the measurement problem that makes the official picture look better than the lived one, the volume data and the competing accounts of it, the pricing behavior of manufacturers who depend on exactly these households, the way a cart changes composition long before it changes size, the categories where supply cannot answer a price signal, the migration of the weekly shop into short-term credit, the fixed costs that crowd food out of the budget, the withdrawal of public provision, and the way the burden is spread across a country whose average conceals it. It closes with the observations that would overturn the reading.

Two boundaries belong at the start. Nothing below rests on a named household, and the evidence is administrative, corporate and survey data throughout. Nor is the claim that Americans are going hungry in large numbers, though a substantial minority are, or that grocery prices cause the country’s budget stress. Prices are mostly a transmission channel. The cause sits in the ratio between a cost level that ratcheted upward across seven years and a wage distribution that did not follow it, and the channel is where the ratio becomes legible.

Two official series that should move together, and the reasons they have come apart

The revisions make a useful starting point, because they show a competent statistical machine producing a misleading impression without malfunctioning. Between May and October the ERS cut its 2026 grocery estimate three times, landing at a midpoint of 2.3 percent inside an interval of 1.9 to 2.7, tracking real softening in the September consumer price index, where food at home rose a tenth of a percent on the month and 1.8 percent on the year against 3.7 percent for all items and 3.5 percent for restaurant food. Grocery inflation now trails general inflation, which for a household that spent four years watching food outrun everything else is a genuine reversal, and exactly the sort of fact that gets generalized past its warrant.

Three properties of the index undercut the generalization.

Weighting is the least intuitive of them. The basket is built around average consumption, and averages are dominated by the households spending most in absolute terms, so a professional couple with $400 of weekly groceries and a family of four with $180 do not inhabit the same index even when buying identical items, since the same percentage change is a different quantity of forgone food. They do not buy identical items either. National weights carry specialty coffee, prepared foods and premium cuts at frequencies that lower-income carts do not reproduce, so inflation in the categories a poor household actually buys diverges from inflation in the categories the index weights.

Substitution does more damage. The CPI handles it by assuming households shift toward cheaper alternatives as relative prices move, which is defensible for a cost-of-living measure and useless for describing a fixed basket. Once a family has traded ribeye for ground chuck and ground chuck for leg quarters, the index records the shift as mitigation and the family records it as a loss. Each round of trading down resets the baseline, so an improvement in the index and a deterioration in the household turn out to be the same event seen from two places.

Base effects are arithmetic and nothing else. A 2.3 percent increase is small against a year that opened at a level a third above 2019, because the comparison point is no longer the pre-pandemic world but the elevated plateau. Every modest month gets measured against a month that was already expensive, and the resulting calm belongs to the denominator rather than to the price.

None of this makes the forecast wrong. It makes it a statement about velocity in a system whose position is the trouble, and the confusion between the two is not innocent. Slowing inflation and falling prices get explained constantly and lost constantly, partly because the two generate similar-looking charts and partly because the political value of the first vastly exceeds that of the second. Nobody campaigns on the observation that the level will not come down.

What a wage series beside a price series does to the reassuring version

Real average hourly earnings fell 0.6 percent in the twelve months to September, according to the Bureau of Labor Statistics, after flat readings through the spring and early summer. Nominal earnings are still climbing and prices are climbing faster, which in an ordinary year would be a footnote on the fourth page of a release nobody reads; against a price level a third higher than seven years ago it means the catch-up households were implicitly promised after the pandemic has not arrived and shows no sign of arriving on any schedule that matters to somebody buying food this week.

Why the cooling framing covers for this is worth stating plainly, since the coverage is structural rather than conspiratorial. The CPI is a rate of change while a household’s problem is a ratio between the cost of what it needs and the money it holds. Cooling inflation means the ratio stops deteriorating as quickly and does nothing to improve it, yet a great deal of public communication treats the first as though it were the second. When General Mills chief executive Jeff Harmening told the Consumer Analyst Group of New York in February that cost of living and housing pressure were reshaping spending patterns and that value had become a core expectation which was here to stay, he was describing a ratio to an audience of investors who needed to hear that the old pricing model was finished.

Manufacturers reached that conclusion at different speeds and in opposite directions, and the divergence is more informative than any single announcement. Through the first half of 2026 the industry said aloud what the volume data had been saying quietly. General Mills cut prices across roughly two-thirds of its North American grocery portfolio. PepsiCo announced reductions of up to 15 percent on many snack lines and moved to eliminate a fifth of its product range. J.M. Smucker trimmed its own. Circana cut its 2026 growth projection for retail food and beverage from a range of three to five percent down to two to four.

Behind the announcements sat a sequence of executive admissions with an unusually confessional tone. Conagra’s Sean Connolly told Food Dive in February that his company had swallowed the final inflation wave rather than pass it on, after four straight years of increases, because the consumer had reached a breaking point, and said of competitors who took one more round that they had simply pushed the shopper too far. Mondelēz chief executive Dirk Van de Put, at the same conference, was blunter: “I am worried about the U.S. consumer. I don’t see how anything will change until the disposable income of the consumer goes up or cost starts to go down in a big way.”

By August the record had become more interesting than the announcements. Conagra reversed itself and raised price, surrendering 1.4 points of volume. General Mills held or gained share in roughly two-thirds of its ten largest categories, the same proportion of the portfolio it had cut, and still lost volume overall, with North America Retail operating profit down a fifth. Two companies moving in opposite directions on price both lost units, which is not what elasticity theory predicts and not what either company’s guidance assumed.

The most plausible inference, and it stays an inference, is that the binding constraint is not the price of any individual item but the total size of the budget items compete for. A household holding $140 for the week does not buy more cereal because cereal got cheaper; it buys cereal instead of something else, and the something else leaves the cart. Category-level losses get redistributed rather than recovered, so a discount buys share of a pot that keeps shrinking and the manufacturer pays margin for a prize that is smaller each quarter.

Something similar happens in the delivery of the relief. Much of the discounting arrives as promotional depth, digital coupons and loyalty pricing rather than as a lower shelf price, and promotions are conditional and expiring while harvesting them costs time, transport and usually a smartphone. A household with two stores within walking distance and an hour to plan gets a materially lower bill than a household with one store and no hour, so discount depth distributes relief regressively while presenting itself as a universal benefit. The shopper who most needs the markdown is the least equipped to catch it.

Range reduction follows margin compression as a matter of course. PepsiCo’s SKU cull and Smucker’s are the visible edge of a broader thinning, and fewer products on shelf means less variety to trade down into. The long tail of niche items, often where the cheaper and nutritionally different options live, goes first, which narrows the ladder at the moment more people are stepping onto it.

Read the Whole Article