Method & limits

Everything on the market page is measured from property adverts we collect and re-check. This page says exactly what that lets us claim, and — at greater length — what it does not.

Where the numbers come from

We collect public listings from Japanese property portals, translate them, geocode them from the coordinates the listing itself publishes, and re-check each live listing on a rolling schedule. The market figures are computed from that collection on a nightly job — the page itself computes nothing at request time, which is why it can carry a single “computed at” timestamp for the whole page.

As of this page load, the collection holds 204,872 active listings, last computed 2026-08-28 11:30 UTC.

Deduplication

The same property is frequently advertised more than once — by different agents, under different listing IDs, sometimes at slightly different times. We collapse listings that share the same coordinates, the same building and land area, and the same asking price. That removed 16,299 rows from the counts on the market page.

This rule is conservative for detached houses, where identical coordinates and identical areas make a genuine coincidence very unlikely. It is less safe for apartments, where many units share one building footprint and two units in the same block can legitimately share a floor area and an asking price. Apartment counts should therefore be read as a lower bound.

Price per m²

Asking price divided by the stated building area. Listings with no stated floor area are excluded from per-m² figures rather than estimated, so the n on a per-m² figure is always smaller than the n on a headline count. House and apartment areas are not comparable — an apartment's stated area excludes shared structure — which is why no figure on the page pools property types.

Median, not mean

Every headline is a median. The mean sits near the 60th percentile of this distribution, pulled up by a long tail of very expensive listings, so a page that leads with the mean overstates what a buyer is looking at. Where the mean appears at all — as a hollow marker on the distribution chart — it is labelled as the mean and is drawn so it cannot be mistaken for the median.

What we deliberately do not publish

Each of these is something a property-data page is normally expected to have. We have the raw material for all of them and publish none of them, because in each case the measurement we could make is not the measurement a reader would assume they were getting.

“Time to sell” and sell-through — still not published

We have never observed a sale, so nothing on this site measures how long a property takes to sell or what share of listings sell at all. What the Market Speed view publishes instead is defined next — read the difference before using either.

Days until removal — what the Market Speed view measures

The clock starts when a listing first entered our index and stops when our nightly re-check confirms it is gone. A removal is a sale or a withdrawal or an expiry or a relisting by another agent — indistinguishable events to us, and the figure is named accordingly: days until removal, never days to sell.

Two more limits are part of the definition. First, the cohort only starts on 2026-08-12: 88% of the removals on file before that date were stamped inside one 13-day operational cleanup, and a figure computed across it would measure our backlog, not the market — so everything earlier is excluded permanently. Second, the cohort is young and censored: only listings already removed are in it, so listings that take months to leave have not had time to enter, and early readings skew fast while the window widens. The clock deliberately does not use the portal's own listing date — agents re-register it mid-listing (16.6% of the cohort carries a listing date newer than our first sighting), which would silently restart the clock and understate time on market.

Sold prices, and anything derived from them

We have never observed a sale. Listings that disappear may have sold, been withdrawn, expired, or been relisted by a different agent — to us these are one indistinguishable event. No price on this site is a transaction price, and no figure describes what buyers paid.

An asking-price index

We can match a listing's price against its own earlier price, which is the right way to build an index. We are not publishing one yet because the price-change record is still catching up on a backlog: three consecutive days of it annualise to −50.4%, −6.2% and −1.5%. Until that settles, any trend line would be measuring the backlog draining rather than the market moving.

The negotiation figures survive this problem because they are levels, not trends: the share of price moves that are cuts, and the typical depth of a cut, come out the same in every month measured on its own.

A composite hazard score against price

The seismic figure uses the raw 30-year probability per listing. It does not use our composite hazard score, which takes only 15 distinct values across 11,104 Fukuoka listings — that column is close to a per-prefecture constant, so charting price against it would only be charting price against prefecture with extra steps.

How to read the figures that are here

Sample sizes

Every figure carries its own n, because the n varies enormously between them. A prefecture median rests on thousands of listings; a station-distance band inside one prefecture may rest on a hundred and fifty. Prefectures with fewer than 30 priced listings are not published in the league table at all. Where a scope has no data for a section, the section says so rather than showing an estimate.

Correlation, in a market with only one price

The station-distance, age and seismic figures each compare medians across bands of a single variable. Nothing controls for anything else. Near-station stock is also more urban; new stock is also more likely to be in a growing municipality; high-seismic zones are the Pacific-side conurbations. Read each as a description of what listings in that band ask, not as an estimate of what that attribute is worth.

Price cuts

A cut is a change to an advertised price, recorded when we re-check a live listing. It is not evidence that a seller is motivated, that the property is overpriced, or that it will sell at the new number. It is evidence of exactly one thing: the advert changed.

Update cadence

Listings are re-checked on a rolling schedule; the market aggregates are recomputed nightly. The bar at the top of every market page shows when the figures on screen were computed, and turns amber if a nightly run was missed and red if the job has not run for four days.

That indicator exists because the previous version of this page carried a computed-at timestamp in its data and displayed it nowhere. It went 104 days stale and looked exactly the same as it had on day one.

Corrections

If a figure here looks wrong, it may be. Email manekihomes@proton.me with the scope and the figure and we will re-check it. Listing removal requests go to the takedown page.