How to Read a URA Caveat Yourself

Almost every price claim made about Singapore private property traces back to one free public dataset. URA's transaction e-Service holds five years of private residential sales, updates twice a week, and costs nothing. Anyone can open it. Very few people read it correctly.

The errors are not subtle, and they are not the reader's fault. The dataset is honest about its own limits, in fine print that nobody clicks. This is that fine print, translated.

What a caveat actually is

URA's glossary is precise: a caveat is "a legal document lodged with the Singapore Land Authority by a purchaser to protect his/her interests after an option to purchase is exercised or a Sales & Purchase Agreement is signed" (URA).

It is a defensive filing, not a record of sale. Under the Land Titles Act, any person claiming an interest in a property may lodge one, and in a typical purchase three parties do: the buyer, the mortgagee bank, and CPF Board where CPF monies are released (SLA). A caveat lapses five years after lodgement, or on registration of the dealing it protects, and can be withdrawn at any time by the caveator (Land Titles Act, Part XII).

That legal purpose is why the data behaves the way it does. Nobody lodges a caveat to inform the market. The market is reading somebody else's insurance policy.

Where to get it, free

The Private Residential Property Transactions e-Service is the main door. Filter by project or postal district, property type, sale date and type of sale. URA SPACE puts the same records on a map. The URA Data Service API returns them as JSON for anyone who wants to do this in a spreadsheet, and data.gov.sg carries the quarterly aggregates. HDB resale is a separate, non-caveat collection owned by HDB and published on data.gov.sg by registration date.

Two constraints before you start. The window is 60 months — a ten-year project price history cannot be built from the free tools, only from the paid REALIS subscription. And the refresh is not what most people assume: resale and sub-sale update every Tuesday and Friday; new sale updates every Friday (URA data release calendar).

Misreading one: treating it as a complete record

It is not, and URA says so twice.

On the search page itself: "It is not mandatory to lodge a caveat. As such, some transacted properties may not have caveats lodged against them." On data.gov.sg, with a number attached: "As the latter (i.e. lodgment of caveats) is voluntary, caveats account for about 80 to 90% of all sub-sale and resale transactions" (URA via data.gov.sg).

One transaction in eight, give or take, is simply absent from the free search.

There is a second filter you cannot see. URA's REALIS methodology states that "a small number of outlier records (i.e. transactions with exceptionally high or low prices) may also be excluded as they are not a good reflection of general market prices" (URA). The public dataset is already lightly curated, and the curation is not itemised.

The important corollary is that URA's own price index is not subject to either limitation. Since 1Q2015 the index draws on caveats "supplemented with Stamp Duty data from IRAS, as well as data provided by developers for new sales," and URA is explicit that "transactions which are not caveated will still be included as the contracts are submitted to IRAS for stamp duty payments." The index sees the market. The free search sees most of it.

Misreading two: mixing three different sales

The Type of Sale field carries three values, and they are not variants of one thing (REALIS data dictionary):

• New Sale — a developer selling before the Certificate of Statutory Completion and titles are issued. Uncompleted, progressive payment, developer pricing.
• Sub Sale — one buyer assigning an uncompleted unit to another before completion.
• Resale — a completed unit with a title, sold in the secondary market.

They come from different pipelines. Since 25 May 2015, New Sale records are not caveats at all — they are weekly submissions of Options to Purchase by licensed developers. The common phrase "the caveat data shows the launch sold at S$X psf" has therefore been technically wrong for eleven years.

They also arrive in wildly different volumes. In 2Q2026 there were 3,813 resale transactions — 62.0% of all sale transactions — 2,141 developer sales, and just 194 sub-sales, or 3.2% (URA, 24 July 2026). A project-level sub-sale median is usually a median of three or four prints. That is an anecdote with a decimal place.

A related trap: a cluster of same-month caveats across an entire project at a uniform premium is an en-bloc signature, not a price surge. URA identifies these separately and strips them from the index and from the quarterly unit counts. The raw search does not label them for you.

Misreading three: comparing PSF across two different rulers

This is the expensive one, because it looks like arithmetic.

URA circular URA/PB/2022/09-DCG, issued 1 September 2022, harmonised how URA, SLA, BCA and SCDF measure floor area. The problem it fixed, in URA's words: "URA's GFA is measured to include the full thickness of external walls but excludes voids, whereas SLA's strata area is only measured up to the middle of the wall and may include voids."

Two changes matter to a buyer. Voids are now excluded from strata area, and AC ledges that form part of a strata unit now count as GFA — though developers may still keep ledges as common property and leave them out.

The consequence is that a pre-harmonisation unit's quoted square footage can include void space and ledges that a post-harmonisation unit's does not. Same physical apartment, larger printed area, lower printed PSF. URA's own worked examples in the circular show a penthouse of 152 sqm strata area excluding a 16 sqm void, and a strata bungalow of 1,035 sqm excluding 525 sqm of void.

The cut-off is where most write-ups get it wrong. The new definitions apply to development applications submitted on or after 1 June 2023, and to Government Land Sales sites launched from 1 September 2022. Projects that already held Provisional Permission stay on the old basis. The line is drawn by planning application date — not launch date, not TOP. Two projects launching in the same month can sit on opposite sides of it. Before comparing their PSF, a non-harmonised area needs restating downward, roughly 6% for one- and two-bedders and 8% for larger units.

Four more things the columns do not say

Floor is a band, not a storey. The published field is a five-storey range — 06-10, 11-15. Unit numbers are never released. A caveat cannot support any claim about a specific stack, facing or view premium.

The price is a gross contract price. URA defines it as the price agreed between purchaser and vendor, and it "excludes stamp duties, agency fees, legal and other professional fees." For the index, URA goes further and "collects nett prices (i.e. prices after deducting the value of the indirect discounts or benefits)." No such netting is stated for the free search, so a New Sale line may sit above the effective price once furniture vouchers and absorbed costs are counted.

"Sold" means a booking fee was paid. URA's glossary: a unit is sold "when the developer has given the purchaser the option to purchase, after the purchaser pays a booking fee." That is 5% down, not a completed transaction.

Landed and strata medians are measured differently. Landed PSF uses land area; strata PSF uses strata floor area. They are not the same number expressed two ways.

The data keeps arriving after you read it

The cleanest demonstration is URA's own 2Q2026 numbers. The flash estimate published 1 July was compiled from data up to mid-June; the full release landed 24 July.

Sources: URA flash estimate and the URA 2Q2026 release. Every sub-index moved. The headline did not. URA's own guidance on the flash figures — "the public is advised to interpret the flash estimates with caution" — applies with more force to a project-level median built from six weeks of prints.

The checklist

1. Set Type of Sale before anything else. Compare Resale to Resale. Never blend New Sale into a resale median.
2. Check which side of 1 June 2023 the project's planning application sits on before comparing PSF, and restate non-harmonised areas down by roughly 6% to 8%.
3. Count the prints. Under about ten transactions you have a range, not a median.
4. Look for same-month whole-project clusters. That is an en-bloc, not a market.
5. Compare bedroom to bedroom. Small units always print higher PSF, and project averages hide unit mix.
6. Assume the last few weeks are incomplete, and that roughly one resale in eight is missing entirely.
7. Use the index for direction and caveats for level. They are built from different inputs and answer different questions.

None of this makes the data less useful. It makes it usable. The transaction record is the best free instrument any Singapore buyer has — it simply has to be read as what it is: a partial, lagging, legally motivated register that URA publishes with the limitations printed on the box.

Methodology published. No spin.

---

Sources: Caveat definition, price definition, the definition of "sold" and the median PSF basis from URA's glossary and the REALIS data dictionary. Lodgement, lapsing and withdrawal from SLA and the Land Titles Act, Part XII. Release cadence from the URA data release calendar. Coverage, outlier exclusion, index methodology and nett-price collection from URA's REALIS coverage and methodology; the 80–90% caveat coverage figure from URA via data.gov.sg. Floor area harmonisation, void and AC-ledge treatment and the 1 June 2023 and 1 September 2022 cut-offs from URA circular URA/PB/2022/09-DCG; the 6%–8% restatement is TRIBE's working convention, matching the harmonisation factors used in our own scorecard tools, and is not a URA figure. Volumes, flash and final index readings from URA's 2Q2026 release and its flash estimate.

Silas Tan is a District Director at Huttons Asia and co-founder of TRIBE. This article is general information, not financial, tax or legal advice. CEA Registration R000303I.