The Giant of the Far Market

Chapters of the quest

A giant enters a realm of guild shops

In 1994 a giant merchant sold groceries in just 21 of the realm's market towns. By 2006 it sold in 181. Its share of all grocery sales rose from just under 1% to roughly 15%.

Elsewhere in the world, modern chains brought big gains by replacing old-fashioned shops. Here the giant faced something different: established guild chains that were already modern.

Brother Knowledge

So shoppers here should have gained less than in Mexico, where modern chains replaced traditional stores. Yet the leading estimate says they gained about twice as much: 25% of their food spending. Our quest is to find out why.

Shoppers pay with coins and with miles

Ironprice believes price is all that matters. But a cheap store across town costs you a drive. Shoppers weigh both.

The giant's stores were 13.6% cheaper, but there were 45% fewer of them per resident. In the paper's estimates, one more store per 100,000 residents is worth about $9 per weekly basket, close to the giant's $12.99 price advantage.

Remove every supercenter from 1998–2001, holding prices fixed. How much do shoppers lose?

Section 3, static welfare. The price-only figure closely matches the 20.2% direct effect in Hausman and Leibtag (2007) for the same years.

Cheaper, but farther away

The giant's edge was real. Its costs of stocking each item were about 19% lower than the guild chains', and it passed much of that on as lower prices.

But it sold a different product: big stores that were farther from most homes. It grew fast without ever matching the guild chains' dense web of neighborhood stores.

Section 2 descriptives. By 2006 the giant's density inside the markets it served had doubled, yet was still 26% below the supermarket average.

A snapshot of a market the giant already shaped

The usual calculation erases the giant from the map and leaves every other store exactly where it was. Bjorn calls that “holding everything else fixed.”

Watch what that snapshot says in each era:

The loss keeps climbing as the giant spreads. But by then the guild chains' stores had been shaped by the giant: some chains had left, some new ones never came. A snapshot can't redraw the map that would have existed without it.

Static loss from removing the giant alone, by period.

Every year, every chain makes a move

Opening a grocery store is a big, lasting bet. So the paper models the market as a game played year after year. In every town, each chain picks a move, knowing its rivals are choosing too:

Each year's prices come from the stores on the map that year, and each move changes the map for next year. That's what a snapshot leaves out.

Section 4. Adjustment costs are estimated from the moves chains actually made.

Count the crowd, not each rival

To play the long game, each chain must weigh every possible map of the town. With many chains each holding any number of stores, the maps quickly outnumber the stars.

The trick: a chain cares how many rivals have how many stores, not which rival is which. Treat rivals as a crowd and most of the maps turn out to be the same map.

Counting is illustrative. The paper summarizes rivals with permutation-invariant features (Fukasawa 2025) and approximates values on grids (Adusumilli and Eckardt 2026).

Take the giant away, then let the realm respond

Now erase the giant from the start and play all thirteen years again. This time every other chain gets to react.

Snapshot: the giant vanishes and nothing else moves.

Section 5, 1994–2006. Re-solving prices alone adds only 0.90 points to the static loss; the difference is mostly about stores.

Two corrections, and the puzzle is solved

Apply the paper's two corrections one at a time:

The final answer is roughly one-fifth of the headline, and well below Mexico's gains. Adding one more modern format to an already modern market helps shoppers less than modernizing a market for the first time.

Bars are shares of shoppers' grocery welfare. The papers scale their percentages differently, so the ordering matters more than exact comparisons.

Relics claimed

Characters and art from Dungeons & Datasets. The villains are fictional; the findings are not.