By Gary Lim · Senior Division Director, ERA Realty Network · CEA Reg. No. R009877B · Published 22 September 2026 · Figures computed from HDB’s resale registry, June–September 2026

Two records, three days apart, both set from the lower floor

The HDB high floor vs low floor question has never had a proper answer in Singapore. It has folklore, it has one very large number that turns out to be mostly an illusion, and it has a lot of confident agents. So let me settle it with HDB’s own transaction data. And let me start with the two sales that made me go and check.

On 19 September 2026 two HDB towns set new records on the same day, and both of them broke the rule every agent in Singapore repeats.

In Queenstown, a four-room at Blk 94 Dawson Road sold for $1,388,000 from the 19th to 21st floor. The record it beat was a Blk 90 Dawson Road unit that sold for $1,380,000 back in August 2024 — from the 34th to 36th floor. Fifteen floors higher, and it lost.

In Sengkang, a five-room at Blk 279B Sengkang East Avenue sold for $1,180,000 from the 10th to 12th floor. It beat a Blk 216C Compassvale Drive unit that went for $1,128,888 in March 2026 from the 16th to 18th floor — a flat that was not only higher but newer, with a 2017 lease against the winner’s 2015.

I pulled all four transactions out of HDB’s own resale registry rather than taking them from the news. They check out to the dollar. So that is the question this article answers, properly and with data: in the HDB high floor vs low floor argument, what is a storey actually worth?

The short version: about 0.6% of your price per floor. Which is real — and roughly fifteen times smaller than the market thinks.

The answer at a glance

What I actually did, and why it’s different

Here is the problem with every “high floor premium” article you have read. They compare the average price of low-floor flats to the average price of high-floor flats, and they report the gap.

That gap is enormous — about 96% on current data — and it is almost entirely rubbish. High floors are concentrated in newer blocks, with longer leases, in redeveloped estates, near newer malls and newer stations. When you compare a 40th-floor Dawson flat to a 3rd-floor flat in an old slab block, you are measuring forty years of lease, a precinct redevelopment and a change of neighbourhood. The floor is the smallest thing in the comparison.

So I did it the other way. I built matched pairs: two sales in the same block, of the same flat type, with identical floor area to within half a square metre, on different storey bands, within two to three months of each other.

Same building. Same lease. Same model. Same lift lobby, same MRT, same school catchment, same market conditions. The only thing that differs is how high up you are. That is as close to a controlled experiment as the public data allows.

I transcribed 713 transactions across six town-and-flat-type samples — Queenstown, Sengkang, Punggol and Ang Mo Kio, across three-, four- and five-room flats — and that produced 184 usable matched pairs. Every figure below is computed from those rows, not from a summary.

HDB high floor vs low floor: one storey is worth 0.6%

Here is the whole thing in one table. The estimator is a gap-weighted regression through the origin on log psf, which is the matched-pair equivalent of block fixed effects.

Robustness check: if I throw away every short-gap pair and keep only pairs at least nine storeys apart — the least noisy ones, n = 88 — the estimate moves to 0.593% per storey. Essentially the same answer. The unweighted median of the individual pair slopes is higher at 0.737%, but short-gap pairs bias that upward, so I don’t lead with it.

By sample, the spread looks like this:

And the distribution of individual pair slopes, which tells you how noisy this is: 10th percentile −0.13%, 25th +0.32%, median +0.77%, 75th +1.29%, 90th +2.03%. Read that bottom decile again. In one pair out of ten, going up costs you money.

If you want to put your own numbers through this, the affordability and stamp-duty calculators at listings.sg/tools will tell you what an extra $40,000 of purchase price actually does to your monthly commitment. In my experience that is the number that changes people’s minds, not the percentage.

The raw market gap says 96%. Almost all of it is a lie.

Take every HDB resale in a recent window, split it into the lowest storey band and the highest, and the psf gap comes out near 96%. I have seen that number, or something like it, used in listing presentations.

It is a measurement of Singapore’s building programme, not of height. A 46th-floor flat in Singapore is, by construction, in a block finished in the last fifteen years, in a precinct the government chose to redevelop, on a lease with eighty-plus years to run. A 3rd-floor flat is very often in a 1970s slab with forty years left.

Strip all that out with matched pairs and 96% collapses to about 6% per ten floors. The raw number overstates the floor effect by roughly fifteen times. That, not the 0.6%, is the finding I would want a client to walk away with.

The noise floor: two identical neighbours trade 3.7% apart

This is the part I had not seen anyone publish, and it is the reason I think most of this debate is misdirected.

I took pairs of sales in the same block, in the same storey band, with the same floor area, in the same two-to-three month window. Flats that are, on paper, identical in every respect — including height. Then I measured how far apart they traded.

Median spread: 3.7%. Mean: 4.3%. Across 21 such groups.

Here are the widest same-band spreads I found, so you can see the scale of it:

Honest caveat on that first row: same block, same band, same size, same month — but almost certainly a different stack and a very different renovation state. I show it as how wide the spread can get, not as a controlled estimate. The 3.7% median is the number to use.

The Dawson lofts: nine sales, the whole population

There is a unit type in Queenstown called the Premium Apartment Loft — double-height duplexes at Blks 92 and 93 Dawson Road, all 122 sqm, all on a 2016 lease. Since they first hit the resale market, exactly nine have ever sold. That is the entire population, not a sample, which makes it an unusually clean natural experiment.

Look at the last two rows. A unit on the 19th to 21st floor sold for $1,700,000 in February 2026. Four months later, a unit on the 4th to 6th floor — same size, same model, adjacent block, same lease year — sold for $1,650,000.

Fifteen floors of height bought $50,000. That is 3.0%, or 0.20% per storey.

And that June 2026 fourth-floor sale is, as far as the registry goes back, the seventh-highest price ever paid for an HDB flat in Singapore. A fourth-floor flat. If you want a single fact to keep in your head from this whole article, that is the one.

When the higher floor sells for less — 25 times out of 184

In 25 of my 184 matched pairs — 14% — the higher unit sold for less money than the lower one, in the same block, at the same size. Here are the biggest gaps:

Blk 88 Dawson is the one that makes agents uncomfortable. Same block, same 83 sqm, same month — and the 40th-to-42nd floor beat the 46th-to-48th floor by $128,000.

There is a reason the very top can go backwards, and it is not just noise. Published work on top-floor units finds a genuine discount, which I come back to below.

“New town means tall” is wrong

I believed this one myself until I counted it. Here is the highest storey band that actually appears in recent four-room transactions, town by town:

Punggol and Sengkang — the two newest towns in Singapore, the ones everybody pictures as forests of towers — have the lowest ceilings of the four. Punggol tops out at 22 to 24 storeys. Ang Mo Kio, an old town, out-tops both. And Queenstown’s Dawson precinct reaches 46 to 48.

The tallest HDB stock in Singapore is not in the new towns. It is in redeveloped mature estates, where the land is expensive enough to justify building up. If a listing presentation tells you Punggol is where the high floors are, the person showing it has not checked.

The premium is steepest where the blocks are shortest

Now put the two tables together, because they interact in a way I did not expect.

Punggol, ceiling 22 to 24 storeys, has the steepest floor gradient in my sample at +1.089% per storey. Queenstown, ceiling 46 to 48, has the flattest at +0.461% — less than half.

The reading I find most plausible: height is priced like any other scarce thing. In a 22-storey town, a 20th-floor flat is genuinely near the top of what exists, and buyers pay for that. In Dawson, where a large share of the stock is already above the 20th floor, height stops being scarce and the premium flattens out. What you are paying for is not altitude. It is relative position in the local stack.

Practically: if you are buying in Punggol or Sengkang, a high floor is worth more of a stretch than it is in Queenstown or Bishan. That is the opposite of how most people assume it works.

78.6% of the market lives on floors 1 to 12

I split every single HDB resale registered in August 2026 by storey band. The seventeen band counts sum to exactly 2,521, which matches the month’s independently-queried total — so every integer in this table reconciles.

Three numbers worth carrying: 78.6% of all HDB resales happen on floors 1 to 12. Only 11.0% happen at storey 16 or above. And just 3.1% at storey 25 or above.

In an entire month of 2,521 national resales, four flats changed hands above the 39th floor. Four. The “high floor” market that so much listing copy is written for is a rounding error.

Which has a direct consequence for sellers. If you own on the 30th floor and you are waiting for the buyer who specifically wants a 30th floor, you are fishing in a pond with about twenty fish a month in it nationally. That queue is a big part of why liquidity and price are different problems — something I get into in more detail in my 2026 HDB resale price outlook.

The twenty highest HDB prices ever recorded

If height really drove price, the top of the market would be uniformly in the sky. Here is the actual list, sorted by price, straight out of the registry and read twice.

Three things fall out of that list.

One: yes, the top of the market does skew high. The median storey of a top-20 sale is around 30. I am not going to pretend otherwise.

Two: seven of the twenty are below storey 25, and number seven on the all-time list is on the fourth-to-sixth floor.

Three, and this is the big one: every single entry is a five-room, an Executive or a loft. There is not one four-room and not one three-room anywhere in the top twenty. At the top of the HDB market, size and layout beat height, and it is not close. If you want to understand how the biggest HDB prices get made, you would learn far more from my write-ups on jumbo flats and million-dollar HDB flats than from any amount of thinking about floors.

What HDB itself charges for a better unit

HDB does not publish a floor-price schedule. But it does publish, for every BTO exercise, the range of prices for each flat type in each project — the cheapest unit and the dearest unit of the same type in the same block set. That spread is the closest thing to an official number we have.

From the February 2026 BTO price annex published by HDB:

Median spread within a project and flat type: 31.3%, ranging from 22.1% to 39.5%.

Now be careful with that, because I nearly overclaimed it myself. HDB’s annex says only that “actual prices may vary based on the actual attributes of the flats at the time of selection”. It never says the spread is about floor. It bundles floor, stack, orientation, and small area differences together.

But the magnitude is instructive. A 20-to-35-storey spread at 0.6% per storey would give you 13% to 23%. HDB’s observed spread is 31.3%. So the arithmetic is consistent with roughly two-thirds of that spread being height and the remaining third being stack, orientation and aspect — which is, more or less, exactly what the resale matched pairs say too. Two completely independent datasets, pointing the same way.

What the published research says

I went looking for peer-reviewed work rather than agency blog posts, and verified the citations at the publishers rather than at whatever page happened to surface them. That matters: I have been burned before by a figure that turned out to be attributed to the wrong authors.

Higher floors gained more from noise barriers

Lee, Kwan Ok & Pang, Alvan Hui Kiat (2022), “Railway sound barriers and housing prices,” Real Estate Economics 50(5), 1282–1306. Studying Singapore’s railway noise barrier programme, they find public-housing prices within 300 m of tracks rose about 0.6% on announcement and a further 1.2% after installation — and that units on the seventh floor and above saw the larger increase, because they were the ones more exposed to noise beforehand.

Two honest caveats. First, the wording I have is from the working-paper abstract, not the published journal text, because the publisher’s site is behind a paywall — so I am not presenting it as a verbatim journal quote. Second, and more importantly: that study measures barrier-induced appreciation near railway lines. It is not a general floor premium, and it should not be read as one.

The top floor carries a discount

Hwang, Sam Il Myoung & Ma, Leo (2023), “Top-floor discounts in residential buildings: Evidence from South Korea,” Real Estate Economics 51(2), 441–469. Using 1.9 million resale transactions, they find top-floor discounts of 1.3% to 8.3% of resale price, which they attribute to heating and cooling costs from larger indoor temperature swings directly under the roof.

That is Korea, not Singapore, and I am not going to pretend the climates are comparable. But it is a documented, large-sample finding that the very top of a block is not the best place to be — and it lines up uncomfortably well with Blk 88 Dawson Road, where the 46th-to-48th floor lost $128,000 to the 40th-to-42nd.

There is also a 2019 paper on floor-level premiums in Singapore condominiums (Khiew & Lee, Applied Economics Letters 26(6)). I could confirm that the paper exists but could not access its findings, so I am not quoting a number from it. I mention it so you know it is there, not so you think I have read it.

Nobody can separate the view from the height

Here is a gap in the entire literature, commercial and academic, and I think it is worth saying out loud.

No published Singapore analysis separates the view from the height. Not one. TRIBE says openly that it cannot control for view obstruction. Stacked says the same. And there is a structural reason: HDB’s transaction data has no view field, no orientation field and no aspect field. It is not that researchers have been lazy. It cannot be done from the public data at all.

So when someone tells you the 25th floor is worth a premium, what they are really measuring is a bundle: height, plus whatever view tends to come with height in that block, plus the better light, plus the quieter air, plus the fact that high-floor units in many blocks are the corner stacks. My 0.6% is that whole bundle. It is not the price of altitude.

And it cuts the other way too. A 12th-floor flat looking out over a park is not the same product as a 12th-floor flat looking into the next block’s kitchen, but the registry records them identically. That is a large part of the 3.7% noise floor.

HDB high floor vs low floor: who should pay up

What I actually tell clients

Three things, every time.

Buy the stack, not the storey. Which direction you face, what you look at, and whether the afternoon sun cooks your living room will matter more to you in ten years than six floors will. Go and stand in the unit at 3pm, not at 10am.

Six floors is inside the noise. If you are agonising between the 8th and the 14th, you are agonising over a difference smaller than the spread between two identical flats in the same block. Take whichever one has the better layout, the better renovation, or the more motivated seller.

Never let the floor cost you lease. I have watched buyers pay 6% more for ten floors and accept a flat with twelve fewer years of lease to do it. That is a terrible trade, and it is the single most expensive floor-related mistake I see. If you are pricing a specific unit and want to know whether the number stacks up, my note on HDB valuation and COV covers how to frame it, and you can run the financing side yourself at listings.sg/tools.

Frequently asked questions