By Gary Lim · ERA Senior Division Director · CEA R009877B · Published 06 November 2026

I have spent seventeen years telling people that a flat is a home first. This week I finally went and measured the second part properly, and the answer to the HDB flat appreciation question is lower than almost anyone in my industry says out loud.

Take the same blocks — literally the same buildings, on the same streets — that sold a 4-room flat in August 1996 and sold one again in August 2026. There are thirty-two of them across Ang Mo Kio, Bedok, Tampines and Woodlands. Median price then: the low three hundred thousands. Median price now: the low five hundred thousands. Over thirty years that is a multiple of 1.74 times, which compounds out at 1.87% a year.

And before you go further, the caveat that makes this honest: August 1996 was the top of the 1996 HDB bubble. The index peaked that December and did not see that level again for roughly twelve years. Measuring from a peak makes any thirty-year return look worse than measuring from a trough would. I have kept that starting point because it is the one I set before I saw the answer, and I show you every decade separately below so you can see exactly where the damage was done. If I had started in 1999 instead, the number would be considerably better.

The numbers at a glance

Three ways to measure HDB flat appreciation, three answers

Here is why this question gets answered badly. There are three legitimate ways to measure it, they give three different answers, and almost everybody quotes the biggest one.

Method one — compare town medians. What did a 4-room flat in Bedok cost in 1996? What does one cost now? Divide. This is what most articles do, and it gives 2.02× over thirty years, or 2.36% a year.

Method two — HDB’s Resale Price Index. The official measure, published quarterly, which controls for flat type and model mix. The index read 96.4 in the third quarter of 1996 and 202.8 in the second quarter of 2026. That is 2.10×, or 2.51% a year. (I re-read both of those values directly at the source, digit by digit, rather than taking them from anyone’s summary — they anchor everything else here.)

Method three — hold the building constant. Find blocks that sold a 4-room flat in both August 1996 and August 2026. Compare those, and only those. This gives 1.74×, or 1.87% a year.

The third number is the one that describes what happened to an owner. The first two describe what happened to the market, and those are not the same thing — because the market keeps replacing its buildings and an owner does not.

The caveat I have to give you first

I said it at the top and I will put a number on it here, because it would be dishonest to bury it.

The HDB Resale Price Index ran like this either side of my starting point: 96.4 in the third quarter of 1996, peaking at 99.0 in the fourth quarter, then 98.6, 97.5, 93.6, 88.0, 81.7, 78.3, 74.9, 72.3, and bottoming at 71.2 in the first quarter of 1999. That is a fall of about 28% from peak to trough. The index did not regain its 1996 level until somewhere around 2008.

So a buyer who bought in August 1996 bought at close to the worst moment available in the modern history of the market, and then waited roughly twelve years just to break even on paper. My 1.87% figure includes that entire episode.

That cuts both ways, and I want to be even-handed about it. It means the thirty-year number understates what a typical holding period produced — most people did not buy at the 1996 peak. It also means the number is real: somebody did buy at that peak, and this is what happened to them. The decade table below lets you pick your own starting point and see the difference for yourself.

The thirty-two blocks that sold in both 1996 and 2026

This is the heart of it. Thirty-two blocks recorded a 4-room resale in August 1996 and another one in August 2026 — same block number, same street name, thirty years apart. Here is a representative slice of them.

Notice the floor areas. Twenty-five of the thirty-two blocks show an identical median floor area in both years, and twenty-seven agree within two square metres. That is the check that tells me these are genuinely comparable homes and not a 1996 three-bedroom being compared against a 2026 something-else.

Notice also how tight the range is once you strip out two obvious oddities. Two 1996 Ang Mo Kio records — $130,000 and $220,000 for flats that should have fetched well over $300,000 that month — look like family transfers rather than open-market sales. They barely move the median, but they inflate the average. Excluding them, every one of the remaining thirty blocks lands between 1.52× and 2.16×. Thirty years, four towns, and essentially nothing outside that band.

I ran four different screens on this — all blocks, excluding the two outliers, only blocks with matching floor areas, and both screens together — and the answer moved between 1.84% and 1.88%. It is about as stable a result as this kind of measurement gets.

Where the missing 27% went

The naive town-median method said +101.6% over thirty years. The same-block method says +74.2%. So roughly 27% of the headline gain is not appreciation at all. Where did it go?

It went into buildings that did not exist in 1996.

Two-thirds of the 4-room flats that sold in those five towns in August 2026 are in blocks that did not exist when our 1996 buyer bought. And they trade at a $175,000 median premium over the older stock — same towns, same month, same flat type.

When you compare “the Bedok median in 1996” against “the Bedok median in 2026”, you are comparing a town of 1970s flats against a town that is now more than half new-build. The difference gets recorded as price growth. For the person who owned a 1979 flat for thirty years, none of it arrived in their bank account.

I tested this three separate ways — matching on blocks, matching on lease vintage, and restricting the 2026 sample to pre-1996 buildings. They gave 1.87%, 1.83% and 1.88%. Three methodologically different controls landing within five hundredths of a percentage point of each other.

The decade of HDB flat appreciation nobody talks about

Split the thirty years into three and the shape of it changes completely.

The first decade was negative in every single town. A 4-room flat bought in August 1996 was worth roughly a quarter less in August 2006. That is not an index artefact — it is 304 real transactions in 1996 against 304 in 2006, and HDB’s own index agrees almost exactly (−2.58% a year against my −2.76%).

Everything people believe about HDB flats going up forever was earned in the two decades after 2006. Someone who bought in 2006 and sold today has had a genuinely excellent run: roughly +5.6% a year for a decade, then +4.5% for another. Someone who bought in 1996 spent ten years going backwards first.

If you want the shorter-horizon picture of where the market is right now rather than where it has been, our running HDB resale price tracker for 2026 covers the current quarter.

Checking my answer against HDB’s own index

A measurement nobody has checked is just an opinion with decimals. So I set mine against the official one.

Decade by decade, we agree to within half a percentage point and the shape matches perfectly, including the brutal first decade. That is a good sign for both of us.

The full-period gap between HDB’s 2.51% and my same-block 1.87% is 0.64 points a year, and I can account for it without either of us being wrong. HDB’s index is answering a different question from mine. It measures what the HDB resale market did. I measured what one owner holding one building earned. The index’s underlying stock is continually renewed; my thirty-two blocks are not. On top of that, my five towns are mostly mature ageing estates while the index includes Punggol, Sengkang and every other growth area, and I measured 4-room flats only.

I am not debunking HDB’s statistics. The index is doing exactly what an index should. The gap between it and the same-block number is the finding, and it is worth 0.64% a year compounded — which over thirty years is the difference between 2.10× and 1.74×.

The comparison that actually matters: your own CPF

Here is the comparison I think every HDB owner should see once, and almost none ever do.

Most people buy a flat with money from their CPF Ordinary Account. That account pays a floor rate of 2.5% a year. And when you sell, you must return to your CPF the principal you withdrew plus the interest it would have earned had you left it alone — CPF’s own wording is that you refund “the CPF principal amount you withdrew (P) and the accrued interest (I)”.

So the honest benchmark for a flat is not zero. It is 2.5%.

On the same-block measure, the flats grew more slowly than the CPF money used to buy them accrued interest. An owner who bought in August 1996, funded largely from CPF, and sold in August 2026 would find a very large share of the sale price going straight back into their own CPF account rather than into their pocket — before a single dollar of legal fees, admin fees or agent commission.

One more cost most people forget: the admin fee on a resale application is $40 for 1- and 2-room flats and $80 for 3-room and larger — payable by each side, so $160 changes hands on a typical 4-room deal. Small money, but it is one of the few costs nobody ever mentions. On agent commission I will say only this: the Council for Estate Agencies states plainly that commissions are not fixed and are negotiable, so anyone quoting you a “standard rate” as though it were regulated is mistaken.

Queenstown: how a town fakes appreciation

Queenstown was the fifth town in my sample and it produced the most quotable number in the whole exercise: 3.09× over thirty years, 3.83% a year, comfortably the best of the five.

It is also the number I trust least, and it is worth showing you why, because the same trick is running quietly in every town table you will ever read.

Queenstown recorded three 4-room resales in August 1996 and five in August 2006. Three transactions cannot carry a median. Worse, those three 1996 flats were Holland Avenue and Holland Drive stock with leases from 1974 and 1975. The 2026 Queenstown sample has a median lease commencing in 2016 — Dawson, Ghim Moh Link, the SkyTerrace-era rebuilds. Ninety-three percent of its August 2026 sales were in buildings that did not exist in 1996.

Queenstown in 1996 and Queenstown in 2026 are not the same housing stock. They are two different towns sharing a name. The 3.09× is almost entirely the town being demolished and rebuilt around its postcode — and not one Queenstown block matched between the two years, which tells you everything.

So when you see a headline saying some town appreciated 200% over twenty years, the first question is not “how much?” It is “is it still the same buildings?”

Your town explains less than your building year

A last finding, and a practical one. In August 2026, across the five towns, the 4-room flats sold at prices from $396,500 to $1,218,000 — a 3.07× spread within one flat type in one month.

The entire gap between the cheapest town median ($545,000, Woodlands) and the dearest ($1,080,000, Queenstown) is $535,000. The middle half of Ang Mo Kio’s own 4-room sales spans $372,500 — seventy percent of that entire between-town gap, inside one town, in one month, in one flat type.

The cause is exactly what this article has been about. Bedok sold 1970s flats at $430,000–$510,000 and 2022 Bedok South Road flats at $915,000–$1,188,000 in the same thirty days. Ang Mo Kio stacked three 2018–2021 BTO resales on top of a base of 1976–1981 stock.

Woodlands is the counter-example that proves it: a middle-half spread of only $68,500, because its stock is homogeneous, mostly built between 1996 and 2000. Homogeneous town, tight prices. Redeveloped town, enormous prices.

So telling an owner their flat is “worth about the Bedok median” is close to useless. The year on the lease tells you more than the name of the town.

What HDB flat appreciation means if you own one

Four things, plainly.

Your flat is a home that holds value, not an engine that creates it. On the honest measure it has roughly kept pace with the interest your own CPF would have earned. That is not a disaster and it is not a windfall. It is a roof you were living under for free, which is genuinely worth something — just not the number in the headlines.

Stop comparing yourself to the index. If your neighbour’s newer block down the road sold for $200,000 more, that is not your flat underperforming. That is a different building. The index includes it; your sale does not.

The lease is the variable that matters most. Every finding in this piece keeps returning to building year. It drives the town spread, it drives the mix effect, it drives the gap between my number and HDB’s. When you are valuing your own flat, start with the lease and work outwards — our guide to HDB valuation and COV walks through how that feeds into the valuation itself.

And if you are deciding between holding an HDB flat and moving to private, do it on arithmetic rather than on folklore. The relevant comparison is not “property goes up” — it is what each asset did, net of what your money would have done sitting still. We work through that gap in the HDB versus condo price gap.

Frequently asked questions

How much do HDB flats appreciate per year?

Measured on the same blocks from August 1996 to August 2026, 4-room flats in Ang Mo Kio, Bedok, Tampines and Woodlands appreciated about 1.87% a year — a multiple of 1.74 over thirty years. Comparing town medians gives 2.36% and HDB’s own Resale Price Index gives 2.51%, because both include newer buildings that entered the market after 1996.

Why is HDB’s index higher than the same-block number?

Because it answers a different question. The index measures what the resale market did, and the market continually renews its building stock. The same-block measure holds one building constant, which is what an individual owner actually experiences. The gap is about 0.64 percentage points a year.

Have HDB flats ever fallen in value?

Yes, substantially. The Resale Price Index peaked at 99.0 in the fourth quarter of 1996 and fell to 71.2 by the first quarter of 1999 — about 28%. It did not regain its 1996 level until around 2008. In our transaction data, the decade from 1996 to 2006 was negative in every one of the five towns measured, at roughly −2.8% a year.

Does HDB flat appreciation beat CPF interest?

On the same-block measure over these thirty years, no — 1.87% a year against the CPF Ordinary Account floor rate of 2.5%. This matters because CPF money used for a flat must be refunded with accrued interest on sale. The comparison is illustrative rather than a calculation of any individual’s position, since most buyers use a loan and pay instalments over time, which reduces the total accrual.

Which town had the best HDB flat appreciation?

Of our five towns, Queenstown showed the highest figure at 3.83% a year — but it is the least reliable number in the exercise. Queenstown recorded only three 4-room sales in August 1996, and 93% of its August 2026 sales were in buildings that did not exist in 1996. Of the towns with usable samples, Bedok led at 2.91% a year.

Does the age of the flat affect how much it appreciates?

Enormously, and it shows up in the price spread rather than the growth rate. In August 2026 the middle half of Ang Mo Kio’s 4-room sales spanned $372,500 — about 70% of the entire gap between Singapore’s cheapest and dearest town median — because the town sells both 1970s stock and 2018–2021 BTO resales in the same month.

How was this measured?

Every 4-room resale transaction registered in August 1996, August 2006, August 2016 and August 2026 in five towns was pulled from HDB’s resale transaction registry — 1,063 transactions in total, with every cell reconciled against the registry’s own reported count. Blocks appearing in both the 1996 and 2026 samples were matched on block number and street name.

What are the limitations of this analysis?

Three worth knowing. The registry carries no unit identifier, so blocks can be matched but individual units cannot — different floors and layouts are being compared, though 25 of the 32 matched blocks had identical median floor areas in both years. One month was sampled per decade, and August 1996 sits at a market peak. And five towns, all 4-room flats, is not the whole of Singapore.