How we work out every HDB figure

Where the data comes from

• HDB resale transactions — data.gov.sg, “Resale flat prices”. We hold 984,879 transactions from Jan 1990 to Aug 2026.
• HDB Resale Price Index (RPI) — data.gov.sg, quarterly, 1990-Q1 onwards. This is the index used to restate HDB prices, and only HDB prices.
• HDB rental transactions — data.gov.sg, approved rents by block, 2021 onwards.
• HDB Property Information — data.gov.sg, block completion year and sold units by flat type.
• OneMap, © Singapore Land Authority — block coordinates, and the MRT stations, schools, parks and hawker centres shown on each block page's map. Distances are straight-line, never walking routes.

We mirror the published data. When HDB restates or withdraws a record, our copy changes with it — a month whose count shrinks after a sync is the source correcting itself, not a fault at our end.

Restating a price to today's market level

A price paid in 1994 and a price paid in 2026 are not comparable sums. Pooling them into one median — as a “median price since 2017” does — quietly understates any block whose sales skew early. So wherever we compare prices across years, we restate the older ones:

real_price = nominal_price × (RPI_latest ÷ RPI_at_transaction_quarter)

• The quarter comes from the transaction's own month: months 1–3 are Q1, 4–6 Q2, and so on.
• RPI_latest is the most recently published quarter of the HDB Resale Price Index — Q2 2026 (202.8) at the time this page was built. It is read from the data, never hardcoded.
• HDB publishes transactions monthly and the index quarterly, so the newest month or two always sit past the last published quarter. Those are treated as already being at today's market level rather than being dropped.
• Medians are taken over restated transactions, never by restating a median: applying one quarter's factor to an aggregate would assume every sale in it happened in the same quarter.
• HDB prices are restated with the HDB Resale Price Index; private property is restated with the matching URA Property Price Index. One index is never used across both.
• Lease decay is not corrected for. The index moves a past price to today's market level; it does not account for the lease the flat has used up in between. Two sales thirty years apart are two different assets — the later one has thirty fewer years left to run — so a long-run gain read off a restated figure is measured against a younger asset, and overstates how much of that value the lease itself preserved. We say so rather than adjust for it: the only public schedule of leasehold value is Bala's Table, which is a valuation model, not a record of what buyers paid.
• The index tracks HDB resale prices, not the cost of living, and the two are far apart: resale prices have risen several times faster than consumer prices since 1990, so a restated figure is much larger than the same sum adjusted for purchasing power. It answers whether a block beat the property market. It never says what that money would have bought.

Worked example

Take a real sale from the record-price ranking — 1G CANTONMENT RD (CENTRAL AREA), 5 ROOM, which transacted at S$1,600,000 in Aug 2025 (2025-Q3, index 203.7).

S$1,600,000 × (202.8 ÷ 203.7) = S$1,592,931

Both figures appear on the ranking together, the transacted one first. The restated one is our calculation and is always marked †, set in italics, and shown beside the sum that was actually paid — never on its own.

Over a single year the correction is small enough to look like rounding. Over three decades it is the whole story: 7 TELOK BLANGAH CRES has a 1990–94 1 ROOM median of S$10,300 as transacted, which is S$80,788 at Q2 2026 market level. Measuring today's price against the first number rather than the second is what produces the impossible-looking gains quoted elsewhere.

Why appreciation is always per flat type

Ranking a block on its blended median across decades measures the block's changing unit mix as much as it measures its prices. The worst case in our data runs to a gain of more than 2,400%: a block whose early-1990s sales were tiny rental-conversion flats going for under S$10,000, set against a completely different mix of flats today. The arithmetic is correct and the number means nothing.

Computed per flat type it becomes a like-for-like comparison, and that is the one we publish. The top of our since-1990 ranking is 7 TELOK BLANGAH CRES, 1 ROOM: a 1990–94 median of S$80,788 at today's market level against S$268,000 now — a gain of 234.0% beyond the market. Every metric on our pages that compares a price across time is computed per flat type for this reason.

Every ranking, its window and its threshold

A median over one or two sales is one household's price wearing a statistic's clothes. Each ranking states the minimum sample a block must reach before it is eligible, and blocks below it are left out rather than shown with a caveat.

Record HDB resale prices

The highest price ever paid for an HDB flat, one row per block so a single block cannot fill the list.

• Window: Every resale transaction since January 1990.
• Minimum sample: None — every row is a single named transaction or a published unit count, not an average.
• Rows shown: 10. These are individual transactions rather than aggregates, so the list is deliberately short and has no sorting or paging.
• Read with: These are individual transactions, not block averages.

Most expensive HDB blocks

Blocks with the highest median price per square foot.

• Window: Resale transactions over Sep 2024 – Aug 2026.
• Minimum sample: At least 3 transactions in the window.
• Rows shown: Top 100.

Most affordable HDB blocks

Blocks with the lowest median price per square foot.

• Window: Resale transactions over Sep 2024 – Aug 2026.
• Minimum sample: At least 5 transactions in the window — a stricter floor than the expensive end, because a cheap block is far more often a thin one.
• Rows shown: Top 100.

Fastest-appreciating blocks, last 5 years

Median price per square foot now against the same block and the same flat type five years ago.

• Window: Sep 2024 – Aug 2026 against Sep 2019 – Aug 2021.
• Minimum sample: At least 5 transactions of that flat type on each side.
• Rows shown: Top 100.
• Read with: Nominal change, the way the market quotes it. Over five years the market’s own drift is a small part of the move; over thirty-six it is most of it, which is why the since-1990 ranking is measured against the index instead.

Biggest gainers since 1990

What the same block and the same flat type sold for in the early 1990s against today, both sides put on the same footing with the HDB Resale Price Index.

• Window: January 1990 – December 1994 against Sep 2024 – Aug 2026.
• Minimum sample: At least 5 transactions in the early window and 3 in the recent one.
• Rows shown: Top 100.
• Read with: Per flat type, never a blended block median: the early-1990s mix included 1-ROOM rental-conversion stock that no longer transacts, and blending it against today produces gains of several thousand percent that mean nothing. The comparison also runs against the HDB Resale Price Index, not the cost of living — a block that simply kept pace with the national market shows close to zero here, however many times its price multiplied in dollars.

Most liquid HDB blocks

Share of the block’s sellable flats that changed hands over the window, as a yearly rate.

• Window: Sales over Sep 2024 – Aug 2026, halved to a yearly rate.
• Minimum sample: At least 20 sellable units in the block. Rental flats are excluded from the denominator: they never reach the resale market.
• Rows shown: Top 100.
• Read with: A rate, not a count — ranking by raw transaction count just ranks big blocks. Read the top of this list as a first-resale wave, not a standing rate: a block reaching its 5-year MOP sells a burst of flats that have never been resaleable before, which is why recently completed blocks dominate the top of this list. That burst does not repeat, so it is a poor guide to how long a sale would take in a mature block.

Highest gross rental yield

A year of rent as a percentage of the price, for the same block and flat type.

• Window: Rentals and resales over the last 24 months of each dataset.
• Minimum sample: At least 3 resales and 3 rentals of that flat type in the block.
• Rows shown: Top 100.
• Read with: Gross means before costs — and before lease decay. A short lease cuts the price but not the rent, which is why the top of this list is all short leases. Part of a high yield on a short lease is simply your own capital coming back as the flat runs down. Read the lease column alongside the yield, not after it.

Blocks finishing MOP soon

Flats reaching the end of their 5-year Minimum Occupation Period, and therefore becoming sellable.

• Window: Estimated MOP year of 2026 or 2027.
• Minimum sample: None — every row is a single named transaction or a published unit count, not an average.
• Rows shown: Every block in the window. This list enumerates a cohort rather than ranking a continuum, so it is not cut at 100.
• Read with: Estimated to the YEAR only. MOP runs five years from each household’s key collection, which is in no public dataset, so this is the block’s completion year plus five.

Shortest remaining leases

Blocks with 50 years or less left on the lease, and what that block’s flats actually fetch against the same flat type across the town.

• Window: Median psf over Sep 2024 – Aug 2026.
• Minimum sample: At least 3 transactions of that flat type in the block.
• Rows shown: Top 100.
• Read with: Both psf figures are medians of recorded HDB resale prices, not Bala’s Table. Bala’s Table is a valuation model; this is what buyers paid.

Ties are broken on street and block so the same dataset always produces the same order. HDB resale records carry a month, not a date, so within a month there is no true ordering to preserve.

Known limitations

• Rebuilt and converted stock. A block redeveloped under SERS, or converted from rental, carries transactions from two different buildings under one address. Computing per flat type removes most of this; it does not remove all of it.
• Short leases inflate rental yield. A short remaining lease depresses the price but not the rent, so gross yield rises as a flat runs down. The yield ranking prints remaining lease in the same row for that reason. We do not cap or hide the figure.
• Block records are not complete. About 270 of roughly 10,000 transacting blocks — mostly pre-1970 SIT walk-ups and older Kallang stock — are absent from HDB Property Information. They have no unit count, so they are left out of the turnover and MOP rankings rather than shown with a guessed denominator.
• MOP dates are estimated to the year. The Minimum Occupation Period runs five years from each household's key collection, which is in no public dataset. We use the block's completion year plus five.
• Per square foot, with per-square-metre alongside it. HDB publishes floor area in square metres. We convert once, at 10.7639 square feet to the square metre, and every ranking is ordered on psf. Prices are shown in both units on purpose — every psf figure in the ranking tables and on a block page carries S$/m² beneath it, and the HDB market report quotes S$/sqm with S$/sqft beneath that — because buyers here read prices one way or the other. Where both appear they are the same figure converted, never two figures computed on different bases or over different windows.
• Lease figures are measured, not modelled. The shortest-lease ranking prints a block's median psf beside the median for the same flat type across its town, and both are medians of recorded HDB resale transactions. Neither is Bala's Table, which is a valuation model rather than a record of prices — and the gap between the two is not a pure lease effect: location, storey and flat condition inside the town land in the same number.

What these numbers are not

They are not a valuation. A median tells you what a group of flats sold for; it does not tell you what yours is worth, because it knows nothing about your floor, your outlook, your renovation or your lease. They are not advice, and they are not a forecast — every figure here is a record of what has already happened.

If you want a figure for your own flat rather than for its block, start here.