Every few months a client asks me some version of the same question: can HDB prices actually fall? And the honest answer is that they can, they have, and the last time they did it properly it took eleven years and nine months before the index got back to where it started.

That is the part missing from almost everything written about an HDB price crash. The current coverage is all about the last three quarters. Nobody goes back and looks at the record, which is public, complete, and sitting on a government open-data portal in 146 quarterly numbers running from the start of 1990 to the middle of this year.

So I pulled all 146, read them off two separate government datasets to make sure I had them right, and worked out every single decline: how deep, how long, and how long until it was over. Then I pulled real four-room transactions from August 1996, August 1999, August 2003 and August 2008 in four towns, so the index could be checked against actual dollars changing hands.

Two things came out of it that I did not expect. One of them means correcting a figure I have published myself.

Has there ever actually been an HDB price crash?

Yes. Once, unambiguously. Here is the entire record of declines in HDB’s Resale Price Index, which starts at 24.3 in the first quarter of 1990 and stands at 202.8 in the second quarter of 2026.

Two episodes above 3%. In thirty-six years. That is the whole list, and it is worth sitting with for a moment, because the popular framing — that HDB prices are permanently and gently upward — is wrong in a specific and expensive way. They are mostly upward. But when they turn, they can stay turned for a decade.

1996: down 30.2%, and eleven years underwater

Here is where I have to correct my own work.

With that fixed, the shape of it:

Forty-six consecutive quarters below the old high. If you bought at the top in late 1996 and simply held, you were underwater on paper from the first quarter of 1997 until the middle of 2008. A child born the quarter you bought would have been in secondary school before your flat was worth what you paid for it.

And that is nominal. In real terms, adjusting for eleven years of inflation, the recovery point is considerably later than 2008.

It was a triple bottom, not a V

This is the structural point I think matters most for anyone reading today’s numbers, and I have not seen it laid out anywhere.

The 1996–2002 decline was not one event. It was three, with two rallies in between that each looked exactly like a recovery at the time:

Look at leg two. Someone who bought in the first quarter of 2000, after the index had recovered nine points and the newspapers were full of the word “rebound”, lost another 13.95% and waited seven and a half years to get back to par. They did not buy at the top of the bubble. They bought after the crash, in the recovery, doing what everyone advises — and it still took until late 2007.

That is the lesson I would take from this history over any other. In a long decline, the dangerous moment is not the peak. It is the first convincing rally.

The recession ended in 1999. The housing fall didn’t.

Singapore’s economy went from around 8% growth in 1997 to 1.5% in 1998, and fell into recession in the second half of that year. Unemployment hit 3.2%, with an average of 62,800 people out of work in 1998 against 34,800 in 1997, and 28,300 retrenched. Then it recovered fast: 1999 delivered 7.2% GDP growth.

The flats did not care. The index bounced with the economy — 71.2 to 80.3 across 1999 — and then fell for eight more quarters to a lower low than the recession itself produced. The macro recovery was real and it did not rescue housing.

Why not? The cleanest explanation is supply, and here I have to be careful about sourcing. Singapore’s Minister for National Development has said in Parliament that HDB was left holding roughly 31,000 unsold flats when the crisis struck in 1997, with home owners paying “a heavy price” as resale values stayed depressed and some buyers ending up in negative equity. I have that from reporting of his 2023 remarks rather than from the parliamentary record itself, which I could not retrieve, so treat the exact figure as attributed rather than verified. The direction of it, though, is corroborated structurally: HDB was building ahead of projected demand under the old Registration for Flats System, and the shift to Build-To-Order — where construction only starts once enough buyers commit — came in the early 2000s precisely because the old model had produced a glut.

So the mechanism was: an enormous overhang of new public housing competing with the resale market, for years, regardless of what GDP was doing. If that sounds familiar, it should. The single biggest competitor to HDB resale prices today is not other resale flats — it is the BTO pipeline. That is a very different situation from 1998 in scale, but it is the same mechanism.

What it looked like in dollars, in four towns

An index is an abstraction. So I pulled every four-room transaction in four towns for four separate Augusts, spanning the peak, the first-leg low, the real trough and the recovery point. Sixteen cells, each reconciled against the registry’s own reported total.

Ang Mo Kio’s −30.2% and Bedok’s −28.1% bracket the index’s own −30.20% almost exactly. That is a real corroboration: two entirely different methods — a national hedonic index built by HDB, and my own medians off raw registrations in two towns — landing on the same number. The dollar reality matches the index.

But the column that stops me is the last one. Twelve years after the peak, Tampines was still 11.0% below its August 1996 median and Bedok 1.8% below — in nominal dollars, before any adjustment for inflation. Only Woodlands and Ang Mo Kio were meaningfully above water, and Woodlands’ figure is the least trustworthy of the four for a reason I should be straight about: its median floor area shifts across the four samples, from 92 sqm in 1996 to 103, 101 and 100 sqm later, so part of Woodlands’ apparently milder fall is a mix effect rather than a price change. The 1999 Woodlands sample is visibly split into two clusters of different flat sizes. I am reporting it, and I am telling you not to lean on it.

The other honest caveat: sample sizes collapse between the late 1990s and the 2000s — Tampines goes from 258 transactions in August 1999 to 55 in August 2003. Ang Mo Kio’s two later cells rest on 27 sales each, which is thin. That collapse is itself the subject of the next section.

The part nobody mentions: volume rose as prices fell

I expected transaction volume to dry up as prices fell. It did the opposite, and then did what I expected two years later.

July 1999 recorded nearly twice the volume of July 1997, while the index was 21 points below its peak. Then volume fell by 59% from that high, bottoming near the price trough in 2003 — and was still only 2,458 in the month the market finally got back to even.

The reading I would offer is that these were two different phases wearing one label. 1997–99 was a high-volume repricing — sellers capitulating into a market that was still liquid, lots of transactions at rapidly falling prices. 2001–2005 was a low-volume stagnation — the sellers who had to sell had sold, everyone else simply stopped, and the market went quiet for four years at the bottom.

Which matters practically, because high volume is routinely read as market health. In 1999 it was the opposite: it was the sound of a market clearing at much lower prices. If you want to know how healthy a market is, volume alone will not tell you — you need volume against listings, which is what I have written about elsewhere as months of inventory.

2013 to 2019: shallower, longer, and almost monotonic

The second real decline looks nothing like the first, and it is arguably the more instructive one because it happened for an entirely domestic, policy-driven reason.

The index peaked at 149.4 in 2Q2013 and fell for twenty-four consecutive quarters to 130.8 in 2Q2019 — down 12.45%. It did not regain 149.4 until 3Q2021, which is 33 quarters, or 8.25 years, underwater.

What makes it unusual is how featureless it was. Between 3Q2013 and 3Q2019 no quarterly rise exceeded two-tenths of an index point. 2016 read 134.7, 134.7, 134.7, 134.6 — four quarters, one-tenth of a point of total movement. There was no crash, no panic, no month where anything obviously happened. Prices simply declined by about half a per cent a quarter for six years, which is the kind of decline that is very easy to talk yourself out of noticing while it is happening to you.

Less than half the depth of the 1996 crash, and it kept people underwater for only three fewer years.

What the Government did, and the one thing it refused to do

Both declines have clear policy fingerprints, and the contrast between the two responses is genuinely interesting.

1996–98: loosen taxes and supply timing, never lending

In May 1996 the Government introduced an anti-speculation package: a stamp duty surcharge plus income tax on gains from property sold within three years of purchase, restrictions on purchases by non-Singaporeans and companies, a loan-to-value cap of 80%, and a bar on foreigners taking Singapore-dollar housing loans. Private property was already about 16% off its peak before the Asian Financial Crisis even arrived. The HDB index, notably, kept climbing for two more quarters after the measures and only turned in 1Q1997.

Then, on 18 November 1997, the first reversal. The May 1996 stamp duty surcharge was suspended for contracts made on or after 19 November; eight residential government land sale sites due that November and December were postponed; the 1998 land sales programme was deferred; and project completion periods were extended from four or five years to a maximum of eight.

And here is the bit I find most striking. The same announcement explicitly declined to stimulate demand by relaxing credit. The 80% mortgage loan limit stayed exactly where it was. In the middle of the worst property downturn in the country’s history, the one lever the Government would not pull was the lending lever.

June 1998 brought a S$2 billion off-budget package, which for property meant deferring stamp duty for buyers of uncompleted homes until completion and allowing developers to offer deferred payment schemes. November 1998 brought a second package of S$10.5 billion aimed at cutting business costs. Land sales were suspended.

2013: the lending lever, pulled hard

The second decline started the other way round. On 28 June 2013 the Monetary Authority of Singapore introduced the Total Debt Servicing Ratio framework, effective the next day. In its own words, MAS expected “any property loan extended by the FI to not exceed a TDSR threshold of 60%“, covering “loans for the purchase of all types of property, loans secured on property, and the re-financing of all such loans” — you can read the original MAS press release in full.

Two months later, reportedly on 27 August 2013 and effective the same afternoon, came the HDB-specific package: the Mortgage Servicing Ratio cut from 35% to 30% of gross monthly income, the maximum HDB loan tenure cut from 30 years to 25, and a new three-year wait for permanent residents before buying an HDB resale flat. I could not retrieve the original MND or HDB release for that one, so I am describing it as reported rather than quoting it. (A word of caution if you go looking: MAS’s own macroprudential summary page appears to date the 30% public-housing MSR to August 2015, which conflicts with all the contemporaneous reporting. I would not rely on that page for the date.)

The index peaked at 149.4 in 2Q2013 — the last quarter before TDSR — and fell in every one of the twenty-four quarters that followed. Whatever else you conclude, the timing is not subtle.

The takeaway for today: the 1996 crash was caused by a shock plus a supply glut, and the 2013 decline was caused by credit rules. Those are the two mechanisms that have ever produced a sustained HDB decline. If you want to know whether one is coming, those are the two things to watch — not sentiment, and not a single quarter’s print. Credit rules, incidentally, have moved the other way recently; the income ceiling changes of August 2026 widened the eligible buyer pool rather than narrowing it.

Private property crashed harder and recovered faster

Since most people asking about an HDB crash are also weighing up private property, it is worth putting the two side by side. I pulled URA’s private residential price index from the same portal, on the same 1Q2009 = 100 basis, so the comparison is apples to apples.

In the crisis, private housing fell half again as hard as HDB and did it in less than half the time — −44.87% in ten quarters, bottoming at the end of 1998. That figure is independently corroborated: a Bank for International Settlements working paper on the Singapore property market puts the private index trough at “45% below its 1996 peak”. Two routes, same answer.

And then private rebounded hard, up 34.1% off the bottom within four quarters. HDB, meanwhile, was still thirteen quarters away from its own low. So the two markets behaved completely differently in the same crisis: private was violent and quick, HDB was gentler and interminable.

The 2013–19 cycle is the opposite story — near-identical depth (−12.45% against −11.64%) but HDB’s decline ran nine quarters longer and it regained its peak four quarters later. Both times, whatever the depth, HDB stayed down longer. If you hold public housing, that is the risk you are carrying: not a violent drop, but a very long flat patch. It is also part of why the gap between the two markets is worth understanding structurally rather than emotionally, which I have covered in the HDB versus condo price gap.

So is 2026 an HDB price crash?

No. And the arithmetic is not close enough for it to be a matter of opinion.

Peak to date: −0.44% over three quarters. Set against the two real declines:

• 68 times shallower than 1996–2002 (−0.44% against −30.20%).
• 28 times shallower than 2013–2019 (−0.44% against −12.45%).
• Shallower than the Global Financial Crisis, which took 0.79% off in a single quarter — 1.8 times the total decline of the last three quarters put together.

There is one nuance in that table worth pointing out, because it is a small original finding and it slightly corrects the standard account. You will read everywhere that 1Q2026 was the “first quarterly decline in nearly seven years”. On the published figures rounded to one decimal, that is true — 4Q2025 reads as flat. But the underlying arithmetic gives 4Q2025 as −0.049%, which is already negative. The turn happened one quarter earlier than the headlines say; it was simply too small to survive rounding. That does not change anything material. It does tell you how small these numbers are: we are discussing a turning point that is invisible at one decimal place.

There is also a genuine paradox in the current numbers that is worth naming, because it confuses a lot of people. The index is drifting slightly down while record prices keep being set — a record 201 million-dollar flats in a single month, and two new all-time top-ten prices registered this September. Those are not contradictory. Record deals come from a tiny, new, well-located slice of the stock; the index measures the whole market. I have written about that split in detail in million-dollar HDB flats.

For what is actually driving the current softness — the MOP supply wave, the BTO pipeline, town-level divergence and what it means if you are selling this year — that belongs on the page I keep updated for it: HDB resale prices 2026. I am deliberately not re-litigating the present quarter here. This article is about the record.

One timing note: HDB has not yet published a 3Q2026 figure. The flash estimate is normally released on the first day of the following quarter, which means 1 October 2026 — six days after I am writing this. I will update this page when it lands.

What a real one would actually look like

Rather than leave you with a reassurance, let me be concrete about what would have to be true for me to change my mind, because a forecast you cannot check is worthless.

A supply overhang at 1998 scale. The 1996 crash needed tens of thousands of unsold new flats sitting against the resale market for years. Today’s pipeline is large and the 2026–28 MOP wave is real, but BTO only builds against committed demand, which is the specific structural fix introduced after the last glut. Watch for HDB carrying unsold completed stock, not for launch volumes.

A credit tightening on the 2013 scale. The 2013 decline came from MSR, loan tenure and the PR wait, applied at once. A cut to the mortgage servicing ratio, or a shortening of maximum tenure, is the single most reliable leading indicator there is. Nothing of the kind has happened; the recent moves have gone the other way.

An external shock plus both of the above. The GFC is the control experiment here. A genuine global financial crisis, with no domestic supply glut and no credit tightening, moved HDB resale prices by 0.79% for one quarter. On its own, a shock is not enough.

Valuations lagging transacted prices. In a genuine decline, the first thing sellers notice is not the index — it is that valuations stop supporting the prices being asked, and cash-over-valuation goes negative across whole towns rather than in isolated deals. That is a much earlier signal than a quarterly print, and it is visible at block level months before it reaches a national index. I have set out how the valuation process actually works, and what COV is really telling you, in HDB valuation 2026.

Six to eight consecutive negative quarters. This is the practical tripwire. Both real declines ran 21 and 24 quarters respectively and were unmistakably established well before the halfway mark. Three quarters totalling less than half a per cent is not a trend; it is a plateau. If we get to eight straight negatives, I will write a different article.

Frequently asked questions

Have HDB resale prices ever fallen in Singapore?

Yes, twice substantially. From a peak of 99.0 in the fourth quarter of 1996 the Resale Price Index fell 30.20% to a trough of 69.1 in the first quarter of 2002, and did not regain its 1996 level until the third quarter of 2008 — 47 quarters, or 11.75 years, underwater. From a peak of 149.4 in the second quarter of 2013 it fell 12.45% over twenty-four consecutive quarters to 130.8 in the second quarter of 2019, regaining its peak in the third quarter of 2021.

How bad was the worst HDB price crash?

Down 30.20% from peak to trough, over 21 quarters, with eleven years and nine months before the index recovered. In real transactions the fall was just as large: the median four-room resale price in Ang Mo Kio went from $326,800 in August 1996 to $228,000 in August 2003, down 30.2%, and in Bedok from $305,500 to $219,750, down 28.1%. Twelve years after the peak, in August 2008, Tampines four-room medians were still 11.0% below their 1996 level in nominal dollars.

Is there an HDB price crash happening in 2026?

No. The index peaked at 203.7 in the third quarter of 2025 and stood at 202.8 in the second quarter of 2026 — a fall of 0.44% over three quarters. That is 68 times shallower than the 1996 to 2002 decline, 28 times shallower than the 2013 to 2019 decline, and shallower than the 0.79% single-quarter dip caused by the Global Financial Crisis. It is a plateau, not a crash.

What caused the 1997 HDB property crash?

Two things together. The Asian Financial Crisis, which began when Thailand floated the baht on 2 July 1997 and pushed Singapore into recession in the second half of 1998, and a very large overhang of unsold new HDB flats — reportedly around 31,000 when the crisis struck — which competed with the resale market for years. The combination is why the fall lasted so long: Singapore’s economy recovered in 1999 with 7.2% GDP growth, but the index then fell for another eight quarters to a lower low than the recession itself had produced.

Why did HDB prices fall from 2013 to 2019?

Credit rules. The Total Debt Servicing Ratio framework was introduced by MAS on 28 June 2013 with a 60% threshold, and in August 2013 an HDB-specific package reportedly cut the Mortgage Servicing Ratio from 35% to 30% of gross monthly income, cut the maximum HDB loan tenure from 30 to 25 years, and required permanent residents to wait three years before buying a resale flat. The index peaked in the quarter immediately before TDSR and fell in each of the next twenty-four quarters. Unlike 1996, there was no crash or panic — just a decline of roughly half a per cent a quarter for six years.

Did HDB prices fall during the Global Financial Crisis?

Barely. The index fell from 100.8 in the fourth quarter of 2008 to 100.0 in the first quarter of 2009 — a decline of 0.79% lasting a single quarter — and was back above its previous peak by the second quarter of 2009. This is a useful control case: a genuine global financial crisis, without a domestic supply glut or a credit tightening, moved HDB resale prices by less than one per cent.

Do private property prices fall more than HDB prices?

In a crisis, yes, and faster. URA’s private residential index fell 44.87% from its second-quarter 1996 peak to its fourth-quarter 1998 trough — half again as deep as HDB’s 30.20%, achieved in ten quarters rather than twenty-one — then rebounded 34.1% within four quarters while HDB was still thirteen quarters from its own bottom. In the 2013 to 2019 cycle the depths were almost identical, 12.45% for HDB against 11.64% for private, but HDB’s decline ran nine quarters longer and it regained its peak four quarters later. The pattern across both cycles is that private falls harder and HDB stays down longer.

How long would it take to recover if HDB prices crashed again?

On the only two precedents available, eight to twelve years from peak to full recovery. The 1996 peak was regained after 47 quarters, or 11.75 years; the 2013 peak after 33 quarters, or 8.25 years. That duration, rather than the depth of the fall, is the real risk of owning public housing through a downturn — and it is why anyone buying should be confident of holding for ten years or more rather than trying to time an entry.

How I got these numbers

The index series is HDB’s own Resale Price Index, rebased to 1Q2009 = 100, published as open data on data.gov.sg. There are 146 quarterly observations from 1Q1990 to 2Q2026, and I transcribed all of them and then read the whole series a second time from a separate, structurally different dataset published by the Department of Statistics. All 146 values agreed exactly between the two sources. The private comparison is URA’s private residential property price index, taken from the same portal on the same rebasing.

The transaction figures come from HDB’s resale registration records. Every town-and-month cell was paged to exhaustion and its transcribed row count reconciled against the registry’s own reported total for that filter. The volume figures are monthly totals read twice each under separate queries. All percentage changes, peak-to-trough calculations and recovery durations are my own arithmetic computed locally on the transcribed series — not figures taken from any commentary, and not computed by any automated summary.

Limits, stated plainly:

• Nothing before January 1990 is published in machine-readable form, so the 1985 recession is outside the record I can check.
• All figures are nominal. Adjusted for inflation, every recovery date in this article is later than stated, and the 1996 buyer’s real break-even is considerably past 2008.
• The volume figures are monthly samples, not annual totals. They show the shape of the change, not the precise annual counts.
• The Woodlands column in the dollar table is contaminated by a shift in median floor area across the four samples and should not be read as a clean price change. Ang Mo Kio’s 2003 and 2008 cells rest on 27 transactions each.
• Three claims are single-sourced and labelled as such in the text: the May 1996 measures date, the approximately 31,000 unsold HDB flats figure, and the detail of the August 2013 HDB package. I could not reach primary records for any of the three and have not upgraded them to assertions.
• The current episode is unfinished. No 3Q2026 figure exists yet; it is due on 1 October 2026.