How the figures are made, and how far they can be trusted

The Japanese edition is the authoritative version. The English and Korean editions are reference translations; if anything differs, the Japanese text takes precedence.

• Regional prices
• Quality adjustment
• Prices by station
• Land prices
• Price factors
• Auction discounts
• Limits

Regional prices

• From MLIT’s real estate transaction-price information (actual transactions) we use used condominiums, residential land with building, and residential land (land only). The price per ㎡ is the transaction price divided by the area (the unit floor area for condominiums, the land area for the others).
• Transactions flagged with “special circumstances” (including private roads, related parties, mediation/auction; about 5%) are excluded.
• The representative value is the median, not the mean (to limit the effect of extreme values). The interquartile range is the interval that contains the middle half of the transactions.
• The period is the latest 5 years (through 2026Q1). “Year-over-year” compares the medians of the latest four quarters and the four quarters before them, and is shown only when each period has 5 or more transactions.
• Municipality and station tables show only those with 30 or more transactions. Tables by building age, area and walking time show only bands with 10 or more.

Quality adjustment (price trends)

The median changes with the mix of building age, area and distance to the station among the properties traded. We therefore fit a statistical model called hedonic regression for each municipality and property type, and estimate each quarter the “price per ㎡ of a reference property” with the effects of building age, area, distance to the station, zoning and renovation removed. The reference property is: for used condominiums, 60㎡ floor area, 20 years old, 8-minute walk; for detached houses (land and building), 100㎡ land, 20 years old, 10-minute walk; for land, 100㎡ and a 10-minute walk. The prefecture lines combine the municipal estimates weighted by the number of transactions.

• The model is a statistical estimate, not the actual transaction price itself. We have confirmed that errors remain even when the model predicts the future from a past point in time.
• The latest quarter may be revised later because the publication of transactions is delayed.

Prices by station

For each “nearest station” in the transaction-price information, we aggregate the median price per ㎡ over the latest 3 years (used condominiums and detached houses only; land is omitted because its relation to station distance is weak). The distance from the station and the mix of building age and area differ by station, so the ranking also reflects differences in such conditions. Station names are shown in Japanese.

Land prices (official land prices)

We use the standard parcels of the official land price data in National Land Numerical Information (as of January 1 each year). The municipal figures are the median of the latest year’s standard parcels, and only municipalities with 5 or more standard parcels are shown. The index chains the yearly median of point-by-point year-over-year changes; it is less affected by changes in the set of standard parcels, but is not suited to comparing levels.

Price factors (curves)

The coefficients (effects of building age, area and walking distance) of each municipality’s hedonic model are summarized by prefecture as medians weighted by the models’ sample sizes, and drawn as curves of multipliers with the reference property set to 1.0. A curve is drawn only when there are 3 or more separately fitted models for the same prefecture and type. They show the average tendency with other conditions held equal, not the price of an individual property.

Auction discount rates

• Winning bid ÷ market price: the winning bid divided by the estimated market price (central estimate) of a property of the same area and building age in the same municipality and type.
• Winning bid ÷ reference sale price: the winning bid divided by the reference sale price set by the court. The observed values are divided directly; no model estimate is involved.
• Both values are computed from the same sample (6,844 cases). They use only the winning bid, reference sale price, area, construction year and municipality from the past sale results on the courts’ real estate auction information site (BIT), for winning bids in fiscal years 2023–2026; no documents, photos, addresses or case information are used.
• The reference time of the market price is the middle of the fiscal year of the winning bid, and the latest quarter of the model is 2026Q1.
• The tallies by number of bidders are built from the winning-bid results collected by this site (note the small sample).

The following cannot have a market price computed and are therefore excluded from the tally: winning bids on land (vacant lots), cases where the municipality cannot be identified, cases with no market model for that municipality and type, cases without area or construction year, cases outside the range the model covers, and cases with an abnormally small reference sale price (values that look like placeholders). The counts are in the table below.

For the auction tally we publish only bands (prefecture × type) with 10 or more samples by building-age band, and municipalities with 30 or more (condominiums, all building ages). This is because a small sample is strongly affected by chance; when we randomly split the sample of the same area in half, the medians differed considerably.

Limits

• Market prices are estimates. Factors the model cannot explain (occupancy, management condition, floor and orientation, need for repairs) are not reflected. Auction properties often have such unfavorable conditions, so a low “winning bid ÷ market price” is not a pure discount.
• The auction tally for detached houses is for reference. The past winning-bid data have no building area, so the estimate uses only land area and building age. When checked against winning bids whose building area is known, the medians did not shift much, but the error for each case is large.
• The time resolution of auctions is low. The source data have only the fiscal year, not the winning date. The mix of building age and area among the properties sold also changes from year to year, so connecting values by fiscal year can show a trend that does not exist. We therefore do not publish a yearly trend for auctions.
• The winning bid is not the total cost of acquisition. Acquisition costs, costs of vacating, repairs and taxes are not included.
• The source data may be changed or corrected. This site aggregates public information as of the time it was obtained.

Updates

The figures are re-aggregated automatically once a week. The reference dates are shown at the bottom of each page (market 2026Q1 / land prices 2026 / auction tally 2026-10-10).