Property Price Indices, Explained

In the second quarter of 2026, URA reported that private residential prices rose. Our own transaction record, over the same quarter and the same market, shows the median price falling by more than ten percent. Both statements are true. Understanding why is the most useful hour you can spend on Singapore property data.

Two numbers, one quarter

Start with the contradiction, because it is not a small one.

URA's private residential property price index rose 0.5% in Q2 2026, easing from a 0.9% rise the quarter before.

Across the same period, the median transacted price for private homes excluding ECs fell from S$2,136,000 to S$1,907,500 — a drop of 10.7%.

If you have ever wondered why property headlines seem to disagree with each other, this is the reason, and it is not that someone is lying.

Slicing finer makes it worse, not better

The instinctive fix is to look closer. If the aggregate number is confusing, break it into segments.

Segment	Q1 2026 median psf	Q2 2026 median psf	Change
OCR	S$1,952	S$2,024	+3.7%
RCR	S$2,111	S$2,189	+3.7%
CCR	S$2,986	S$2,365	−20.8%

Two of the three segments went up. The overall median went down by more than ten percent. Already the picture has inverted.

So slice again, into the CCR where the fall lives.

CCR, by sale type	Q1 count	Q1 median psf	Q2 count	Q2 median psf
New sale	721	S$3,163	152	S$3,017
Resale	524	S$2,255	613	S$2,205

Within each sale type, the change is modest: new sales down 4.6%, resales down 2.2%. The twenty-percent fall has almost entirely evaporated.

What actually happened is visible in the count column. CCR new sales collapsed from 721 to 152 between quarters while resales rose, and new sales there transact around nine hundred psf above resales. The share of expensive transactions fell, and the median followed.

Nothing in this tells you that any home became cheaper. And on the question of whether they did, there is an instrument built precisely to answer it: over the same quarter, in the same segment, URA's index recorded non-landed CCR prices rising 1.8%.

One market, one quarter, two numbers twenty-two points apart. Only one of them is about price.

The thing worth understanding

Notice the pattern. At every level of aggregation, slicing finer made the apparent fall smaller — because at every level, what we were measuring was who bought, not what things cost.

This is why you cannot slice your way out of the problem. Every slice, however fine, still has a composition. Segment down to a single project and you have the mix of stacks, floors and unit sizes that happened to transact that quarter. There is no level at which the composition effect disappears.

A median tells you about the middle transaction in a group. Change the group and you change the middle, and no amount of subdividing repairs that.

An index exists to answer the other question: what happened to prices, holding the property constant?

Two markets, two indices

Singapore publishes two residential price indices, and they are not interchangeable.

HDB's Resale Price Index (RPI) covers public housing resale transactions.

URA's Private Residential Property Price Index (PPI) covers the private market, with separate sub-indices for landed and non-landed homes.

What is striking is how similar they are by construction. Both moved to a stratified hedonic regression within two quarters of each other — HDB from Q4 2014, URA from Q1 2015 — replacing a simpler stratification method in each case. Both use five-quarter fixed weights, revised every three years. Both are set to 100 in the first quarter of 2009.

That last coincidence is the trap, and it is worth naming early: because both read 100 at 1Q2009, they look directly comparable. They are not. Each measures its own market's distance from its own 2009 baseline. An RPI of 200 and a PPI of 200 do not mean HDB flats and private condos cost the same, or even that they have travelled the same road — only that each has doubled relative to where its own market stood in early 2009.

How the method works

Rather than grouping similar homes and comparing group medians, both indices estimate the relationship between price and a property's characteristics, then report how prices moved once those characteristics are held fixed.

Both halves of "stratified hedonic regression" are doing work.

Hedonic describes a statistical model that breaks a price into the contributions of the thing's individual characteristics: how much of this home's price is attributable to its floor level, how much to its age, how much to sitting near an MRT station. Once you can estimate those contributions, you can ask what a property with identical characteristics would have cost last quarter versus this one. That comparison is the price movement, and it is what makes the number a price index rather than a price average.

Stratified means the market is first divided into segments and the model is fitted within each, rather than one model being asked to describe the whole island at once.

The attributes each index controls for differ with the market. HDB's model works with age, floor level and proximity to amenities, among others. URA's works with age, unit size and micro-location such as MRT proximity, alongside tenure, property type and completion status.

The segment-level movements are then combined using capital-value weights: each segment counts in proportion to its share of the market's total value rather than its share of transactions. Those weights are averaged over five quarters and held fixed, so one unusual quarter cannot reshape the aggregate, and they are refreshed once every three years. URA's switch to this approach in 2015 replaced a twelve-quarter moving average, which had made the index slower to reflect a turning market.

Base period and rescaling

An index set to 100 in the first quarter of 2009 is using that quarter as an arbitrary reference point, not a valuation. An index of 200 means prices have doubled since 1Q2009, and nothing more.

Rebasing means moving that reference point to a different period. When HDB changed its base in 2014, the historical series from 1Q1990 to 3Q2014 was rescaled — every historical value multiplied by a single constant — using a factor of 100 divided by 138.3, which was the old index's level at 1Q2009. That puts the entire history on one scale so it reads continuously across the change. Small rounding differences against the pre-rescaling series are expected.

Contract date, registration date, and where the two indices differ

A property transaction does not happen on one date. It happens over several, and this is where the two indices part company.

The contract date is when the price is agreed and the deal becomes binding — in practice, when the Option to Purchase is exercised or the sale agreement is signed. This is the date the price was actually struck.

The registration date is when the completed transaction is formally registered with the authorities. It comes later, after the legal and financial process finishes: a matter of weeks or months, not days.

A caveat sits between the two. When someone buys a private home, their lawyer lodges a caveat — a legal notice recorded against the property announcing that the buyer has an interest in it, so nobody else can deal with the property without notice. Caveats are public and they carry the transacted price, which is why a caveat database exists at all and why it is the raw material of private-market research.

Now the difference that matters:

• HDB's RPI is computed on transactions by date of registration.
• URA's PPI is compiled from caveats lodged at the option stage, supplemented with stamp duty data from IRAS and data provided by developers for new sales.

So the two indices are anchored at different points in the transaction timeline. URA's sits close to when the price was agreed; HDB's sits at the far end, after completion. In a turning market they will not turn in the same quarter, and that is a property of the instruments rather than a disagreement about the market.

It also explains something about our own figures. Our medians are keyed to contract date and drawn from caveats alone, so they run ahead of the RPI and, unlike the PPI, are not topped up with stamp duty records. A quarter's caveats keep arriving after the quarter ends — which is why a caveats-to-date median moves under you, and why URA's supplemented index does not.

What an index is good for

• Comparing one period against another within the same market, over the long run, which is what these were built for.
• Deflating nominal values — adjusting a figure for how much the market as a whole moved, so you can ask whether a particular price rose faster or slower than everything else.
• Establishing direction and magnitude, with composition removed.

Used for those, they are the most reliable instruments available.

What an index cannot tell you

It cannot value your home. An index is a market-wide movement, not a per-property estimate. A 5% rise does not mean your unit is worth 5% more; your block, floor, remaining lease and layout all sit outside what the index reports.

It hides divergence. A single aggregate figure is an average of towns and segments moving at very different speeds. In our Q2 2026 HDB review, individual towns ranged from strongly positive to clearly negative inside one quarter that the index summarised as a single number. The average is real; so is everything it conceals.

It is revised. A flash estimate — a provisional reading published within days of quarter end, computed on incomplete data — is updated when the full quarter lands. Q2 2026 shows the shape of this well. The headline held: 0.5% at flash on 1 July, 0.5% at the full release on 24 July. Underneath, every segment moved. Landed went from 2.6% to 2.5%, CCR from 2.0% to 1.8%, RCR from −1.4% to −1.2%, OCR from −0.2% to −0.1%. Small movements, but the segment numbers are the ones people quote, and they were all provisional for three weeks.

It measures quality imperfectly. A hedonic model controls for the characteristics it includes. Anything that shapes price but sits outside the model — renovation standard, precise orientation, the character of a specific block — is not held constant, only averaged over.

And it cannot be read across markets. The RPI and the PPI answer the same kind of question about two different populations. Comparing their levels is meaningless; comparing their directions is occasionally interesting, provided you remember they are clocked differently.

How to read them in practice

Level versus change. The level only means something relative to the base period. The quarter-on-quarter and year-on-year changes are what carry information.

Mind the base. Comparisons that span a rebasing need the rescaled series, which is why the agencies publish full histories on one scale rather than leaving you to splice two together.

Flash versus final. If a number is moving markets in the first fortnight after quarter end, it is a flash estimate.

And when a median disagrees with an index, ask what changed in the mix before concluding that prices moved. In our Q2 2026 example, the honest reading of a 10.7% fall is that fewer expensive homes transacted — a fact about demand composition, and a genuinely interesting one. It simply is not a fact about prices. We read that quarter in full in the Q2 2026 private residential review, and the HDB market in the Q2 2026 HDB review.

---

Data and methodology

Realila figures are computed from URA caveat data lodged as of the publication date, excluding ECs, and are keyed to contract date. Q2 2026 caveats continue to lodge after quarter end, so the medians quoted here will drift as the record completes — a further illustration of why a caveats-to-date median is a poor instrument for measuring price movement.

Index figures and index methodology are as published by HDB and URA. HDB's Resale Price Index methodology described here is effective from Q4 2014; URA's Private Residential Property Price Index methodology from Q1 2015.

Segment classification follows the market segment recorded against each transaction. Medians are unweighted within their stated population.

Source: URA, HDB & Realila Research. General market research, not financial advice.