Half of store-days carry a discount — and almost none of them should

THE ELASTICITY IS -1.49 AND THAT IS NOT ELASTIC ENOUGH. Across 2,457,493 fully-stocked store-days, with store-product and day fixed effects and the chain's own marketing flag held constant, a one per cent price cut raises units sold by 1.49 per cent (95% CI [-1.51, -1.46]). For constant-elasticity demand the profit-maximising gross margin is one over that, so a markdown only improves profit on a product already carrying more than a 67.3% full-price margin.

FRESH GROCERY DOES NOT RUN 67% MARGINS, AND THE CHAIN DISCOUNTS ANYWAY. 51.4% of store-days carry a discount, at a median depth of 14.4%. The threshold the chain's own demand response sets is above any plausible fresh-grocery margin, so on this evidence the markdown programme is not a profit-optimising exercise. It is doing something else — clearing stock, defending footfall, matching a competitor — and those are legitimate aims that the elasticity framework sold as 'markdown optimisation' does not price.

11% OF THE APPARENT PRICE RESPONSE WAS THE CAMPAIGN, NOT THE PRICE. With fixed effects but no marketing control the elasticity is -1.66; adding the chain's own activity flag moves it to -1.49. A retailer does not cut a price silently — it runs a promotion — and an elasticity estimated without separating the two credits the campaign's traffic to the discount, then recommends deeper discounts on the strength of it.

NO CATEGORY CLEARS THE BAR AT A REALISTIC MARGIN. Of 23 product categories with enough data to estimate separately, 0 clear the threshold at a 25% gross margin and 1 at 30%. Even at an implausible 50% margin only 8 do. The most elastic category sits at -3.85 and the least at -0.50, so this is not an average concealing a profitable subset.

THE UNCONTROLLED NUMBER IS 1.03 AND IT IS THE ONE A SPREADSHEET PRODUCES. Regressing log units on log price with nothing held constant gives -1.03. Every difference between stores, products and seasons lands in that coefficient. It is published first because it is what the tooling gives you by default, and the distance between it and -1.49 is the cost of not asking what else was moving.

MOST STORE-DAYS COULD NOT MEASURE DEMAND AT ALL. 44.1% of store-days carry at least one hour with an empty shelf, averaging 3.2 hours out of the sixteen-hour trading day. Sales on those days measure supply, not demand, and a markdown study that includes them understates the response. Only 2,504,437 store-days were fully stocked, and the estimate uses those; this file is unusual in making that distinction possible at all.

A SENTINEL VALUE WOULD HAVE DOMINATED THE ESTIMATE. 16,039 rows carry a discount of exactly zero — free — and they average 14.5 units against roughly one for the file as a whole. They are not hundred-per-cent markdowns, they are a placeholder, and because log(0) is undefined the only way to keep them is a transformation that would have let sixteen-times-typical volume anchor the price response.