We thought hotel prices were rising. They were not — the cheap rooms were just leaving
This one is a correction. We ran the obvious calculation, got the answer everybody expects, and it was wrong in the direction most likely to be believed.
1. What we thought we saw
Prices rising as the date approached
We took every tracked stay date, computed the median nightly rate across the whole sample on the first day we measured it and again on the last, and compared.
Dates inside three months came out at a median of +1.4%. Dates further out came out at −3.3%. Individual dates swung from −17.9% to +12.8%.
Read casually, that is the familiar story: book early, prices climb as you wait. We nearly published it.
2. Why it was wrong
The cheap rooms leave the sample
A median is taken over whatever is still for sale. Cheap rooms sell first. As they disappear from the listing, the median of what remains rises — without a single hotel changing its price.
It is the same arithmetic as a room where the lightest people keep leaving. The average weight goes up. Nobody gained a pound.
The fix is to stop comparing populations and start comparing individuals: keep only hotels that returned a rate on both the first day and the last, and measure each one against itself.
3. What the matched comparison shows
Every single date fell
| Stay date | Days out | Hotels matched | Whole sample | Same hotels |
|---|---|---|---|---|
| 2026-08-18 | 1 | 248 | +5.6% | −0.8% |
| 2026-09-15 | 21 | 281 | +4.9% | −4.9% |
| 2026-09-19 | 25 | 190 | +1.1% | −3.0% |
| 2026-10-17 | 53 | 237 | +1.4% | −3.4% |
| 2026-10-20 | 56 | 268 | +4.8% | −3.7% |
| 2026-11-17 | 84 | 274 | −0.7% | −3.4% |
| 2026-11-21 | 88 | 216 | −3.4% | −3.0% |
| 2026-12-15 | 112 | 257 | −5.5% | −4.6% |
| 2026-12-19 | 116 | 235 | −17.9% | −3.8% |
| 2027-01-16 | 144 | 213 | −10.0% | −3.0% |
| 2027-01-19 | 147 | 220 | −1.1% | −3.0% |
| 2027-02-16 | 175 | 147 | +12.8% | −5.3% |
| 2027-02-20 | 179 | 142 | −1.0% | −3.6% |
Whole-sample figures scatter in both directions. Matched to the same hotels, 13 of 13 dates fell, with a median of −3.4% — and the size of the drop barely depends on how far away the date is.
4. Was it just the exchange rate?
We record two currencies so we can check
Everything above is in Korean won. A won figure can fall because hotels got cheaper or because the won got stronger. For Japanese cities we also record the yen price of a subset of the same hotels, which separates the two.
| Currency | Matched pairs | Median change | Share that fell |
|---|---|---|---|
| KRW | 2,928 | −3.09% | 80% |
| JPY | 627 | −1.95% | 68% |
Both fell. In yen — the currency the hotel actually charges — the drop is −1.95%, so roughly two thirds of the won-denominated move is a real price change and the remainder is currency. Worth knowing before quoting either number on its own.
5. How this was measured · and what it cannot tell you
- Six Japanese cities, 2026-08-08 to 2026-08-17, 2 adults, 1 night, hotels only.
- Matched comparison keeps only hotels priced on both the first and last day of the window, and takes the median of each hotel's own percentage change.
- Cannot tell you: that booking later is cheaper. A 18-day window that shows a market-wide drift is not a booking curve. Every date fell by a similar amount regardless of distance, which is what a market-wide move looks like — not what a deadline-driven one looks like.
- Cannot tell you: what happens in the last two weeks before arrival. We have not watched a date all the way in yet.