A maximum drawdown is the largest fall a series makes from a running peak to a later low, written as a percentage of that peak. We measured it on URA's private residential price index, as published on data.gov.sg. The index fell 44.87% from 1996-Q2 to 1998-Q4. It did not get back to its 1996-Q2 level until 2010-Q2. Fifty-six quarters, fourteen years, passed between the peak and the quarter it was regained, and undoing the fall took a gain of 81.4%.
Recovery from a drawdown comes only when a series regains its running peak
At every reading, the drawdown is how far the value sits below the highest value seen so far: (peak − value) ÷ peak. The maximum drawdown is the deepest of those readings. Magdon-Ismail and colleagues, in a 2004 paper in the Journal of Applied Probability, describe it informally as "the largest drop from a peak to a trough". Their drop is in level; the percentage form is the convention this guide uses.
Take five readings: 100, 120, 90, 110, 125. The running peak is 120 by the second reading, so the low of 90 is a 25% drawdown, even though 90 is only 10% below where the series started. The fall ends at 125, the first reading at or above 120.
Retirement planning uses the same word for taking money out of a pot, as in our guide to the 4% rule; here it always means a fall from a peak.
Every loss is recouped only by a larger percentage gain
After a fall of d, the series needs a gain of 1 ÷ (1 − d) − 1 to get back to its peak, because the gain is earned on the smaller amount that is left. A 50% fall followed by a 50% rise leaves the series at 75% of where it began.
| Fall from the peak | Gain that gets back to the peak |
|---|---|
| 10% | 11.11% |
| 20% | 25.00% |
| 30% | 42.86% |
| 45% | 81.82% |
| 50% | 100.00% |
| 70% | 233.33% |
| 80% | 400.00% |
Rounding the loss matters. URA's measured 44.87% needs 81.4% on our arithmetic; a round 45% needs 81.82%.
URA's private home index regained its peak 46 quarters after the trough
We downloaded URA's Private Residential Property Price Index, quarterly from 1975-Q1 to 2026-Q3 for three property types, and HDB's Resale Price Index from 1990-Q1. We computed each drawdown twice, in Python and with our calculator's code, and the two agreed to the quarter.
| Index | Deepest fall | Peak | Trough | Back at the peak | Quarters down | Quarters to recover |
|---|---|---|---|---|---|---|
| URA, All Residential | 44.87% | 1996-Q2 | 1998-Q4 | 2010-Q2 | 10 | 46 |
| URA, Landed | 48.33% | 1996-Q2 | 1998-Q4 | 2010-Q3 | 10 | 47 |
| URA, Non-Landed | 46.95% | 1981-Q2 | 1986-Q2 | 1991-Q2 | 20 | 20 |
| HDB resale | 30.20% | 1996-Q4 | 2002-Q1 | 2008-Q3 | 21 | 26 |
For All Residential the index went from 129.7 to 71.5 in ten quarters, and the first reading back above 129.7 was 131.7 in 2010-Q2. The recovery took 4.6 times as long as the fall. Counted as readings, 55 quarters sat below the peak, from 1996-Q3 to 2010-Q1. The series had two other falls of 10% or more: 36.27% from 1983-Q4 to 1986-Q2, regained in 1989-Q3, and 11.64% from 2013-Q3 to 2017-Q2, regained in 2020-Q4.
HDB's resale index won back its smaller fall sooner
The resale index peaked at 99.0 in 1996-Q4 and reached 69.1 in 2002-Q1, a fall of 30.20%. It took 21 quarters to reach its low, against the private index's ten, then 26 more to pass the old peak, with 99.4 in 2008-Q3. Peak to recovery was 47 quarters, against 56 for the private index. Its only other fall of 10% or more since 1990 ran from 2013-Q2 to 2019-Q2, 12.45%, and was regained in 2021-Q3.
Non-landed private homes made good a different worst fall, from the early 1980s
The "worst" episode depends on which sub-index you're looking at. Landed homes fell furthest in 1996–98, with the All Residential series. Non-landed homes fell further in 1981–86: 46.95% over 20 quarters, and back at the 1981-Q2 level in 1991-Q2. Their 1996–98 fall was shallower in quarterly data. A headline built on "private homes" therefore picks an episode as well as a number.
Yearly samples of the same index recover as early as 2008
Sampling less frequently means the true low or high can fall between readings and go unrecorded. We kept one quarter in four from the All Residential series, starting at each quarter in turn. Our drawdown calculator reproduces every row: load the URA sample, set "Read every 4th value" and choose the starting value (value 2 is Q2, because the series begins in a Q1).
| URA All Residential, read | Deepest fall | Trough | Back at the peak |
|---|---|---|---|
| every quarter | 44.87% | 1998-Q4 | 2010-Q2 |
| once a year, Q1 | 38.35% | 1999-Q1 | 2008-Q1 |
| once a year, Q2 | 38.01% | 2004-Q2 | 2010-Q2 |
| once a year, Q3 | 38.44% | 1998-Q3 | 2010-Q3 |
| once a year, Q4 | 42.20% | 1998-Q4 | 2010-Q4 |
Read each Q2, the trough moves to 2004 and the fall lasts eight readings. HDB's index read once a year gives 27.49% to 29.92%, against 30.20% every quarter. Non-landed homes read each Q4 swap episodes outright, to 40.07% from 1996-Q4 to 1998-Q4, back in 2007-Q4. The sampling effect is well known; these figures show its size on one public series.
Before a fix dated 4 November 2026, the calculator silently dropped every value of 0 or below, so pasted returns were read as prices, and its one sample, labelled S&P 2007-2010, was a made-up 32-point list. The fixed version stops on a negative value in values mode and ships the two indices above as samples.
A price index recovers its level without saying what any owner recouped
These are price indices. They leave out rent, upkeep, stamp duty, borrowing and how any single home moved, so a drawdown in them is not what an owner gained or lost. URA's dataset page describes its index as compiled by a stratified hedonic regression, weighted by the value transacted in recent quarters. The page does not say the 1970s and 1980s quarters were built the same way, so comparing the 1981–86 fall with 1996–98 crosses whatever method changes the index has had. We used the series as published in October 2026 and did not check for back-revisions.
A past fall also says nothing about the next one. The figures in this guide are our own measurements from the published indices; URA and HDB have not reviewed or endorsed them.
Contains information from Private Residential Property Price Index (Base Quarter 2009-Q1 = 100), Quarterly accessed on 4 November 2026 from data.gov.sg which is made available under the terms of the Singapore Open Data Licence version 1.0 (https://data.gov.sg/open-data-licence).
Contains information from HDB Resale Price Index (1Q2009 = 100), Quarterly accessed on 4 November 2026 from data.gov.sg which is made available under the terms of the Singapore Open Data Licence version 1.0 (https://data.gov.sg/open-data-licence).
