How we measure, and why every figure can be checked

Every number we publish can be produced again. Not looked up in a note, not re-derived by hand from memory of what we did, but recomputed from the original records by a process we built and committed to. If a figure changes, we find out. If it was wrong, we say so in public and leave the correction where the mistake was.

That is unusual in property commentary, and this page explains what it means and why we do it.

Why this matters

Most property figures in circulation cannot be checked. A number appears in an article, gets quoted in a second article, and by the third nobody can say what it measured, over what period, or whether it is still true. The number outlives the evidence for it.

That has a cost, and the cost falls on people making the largest financial decision of their lives. A buyer told that small units command a price premium may be working from a rule that was true a decade ago and is not true now. An owner told their unit has underperformed may be reading a comparison that mixed different buildings in different districts of different ages. The advice is not dishonest. It is just unverifiable, and unverifiable advice cannot be corrected when it goes wrong.

We take a different position: a number we cannot reproduce is a number we should not publish.

What we guarantee

Every published figure is reproducible. Each one is tied to the exact question that produced it, and can be recomputed from source records at any time. Not approximately, not by someone repeating the analysis from a description, but by running the same defined query again.

Every figure is rechecked on a schedule. Figures decay at different rates. A number covering the current quarter can move within weeks as late records arrive; a median for a year that closed five years ago barely moves at all. Each figure carries its own recheck interval accordingly, and the checking is automatic rather than something we remember to do.

When a figure moves, we know. A recheck that returns a materially different number raises a flag. So does a change in the underlying population, even when the headline figure looks unchanged, because a stable number over a shifted base is a different claim.

We publish corrections in place. When we get something wrong, the correction sits in the article that carried the error, dated, saying what was wrong and what replaced it. We do not quietly edit and move on. Several of our articles carry corrections; you can read them.

We publish results that do not help us. If a test we ran to explain something fails to explain it, that is a finding and it publishes. If a cut of the data shows nothing, the nothing is reported. We have withdrawn a published claim when better measurement showed it could not be supported.

What this means for you

If you are buying or selling, you can check the basis of anything we tell you. Every article states what it measured, over what period, on how many properties, and what it cannot tell you. If a figure looks wrong against what you are seeing in the market, the disagreement is a real one that can be resolved, not a matter of whose impression is better.

If you are a professional, you can use our figures with your clients knowing they will not quietly become false. When something changes, it changes visibly.

If you are writing about the market, our numbers are citable in the sense that matters: we can tell you exactly what any of them measured, and we will tell you if one of them turns out to be wrong.

And more broadly: property is the largest asset most households will ever own, and the quality of public information about it is poor. Not through bad faith, mostly, but because there is no cost to publishing a number nobody can check. We think there should be. This is our attempt to hold ourselves to that, and we would rather more people did the same than have it be a competitive advantage.

How we measure

The rest of this page is the technical basis. It is here so that anyone who wants to interrogate our work can.

Sources

Private residential transactions come from URA caveat data. Rental contracts come from URA's non-landed rental records. HDB resale transactions come from HDB. Every article states its data boundary, the latest date on which a record could have entered the analysis, because caveats arrive with a lag and a boundary date is part of what a figure means.

Comparing like with like

Most misleading property statistics come from comparing different things. A market-wide median can fall while every individual property rises, simply because the mix of what sold shifted toward cheaper stock. This is not a subtle effect; it is often the largest thing in the data.

We handle it in two ways.

Within the same development wherever possible. When we compare small units against large ones, we compare them inside the same building in the same year, then take the middle result across buildings. Location, age, tenure and building quality are held constant because they are literally the same building.

Medians across properties, not pooled across transactions. Pooling every transaction lets a development that happened to sell heavily dominate the result. Taking each development's own median first, then the median of those, gives each development one vote.

Where the stricter comparison is impossible we say so and explain what we used instead. Some questions cannot be answered inside a single building: prime-district blocks rarely contain both a 500 sqft unit and a 1,400 sqft one, so requiring both would make the prime market unmeasurable.

Definitions

Categories are defined in print, with their source. A shoebox is 538 sqft or less, matching the regulator's own 50 sqm threshold rather than a convenient round number. Where a definition changes, the change is dated and the affected figures are re-run.

We learned this the hard way. We published four articles using a shoebox band that did not match the regulator's, while citing the regulator's threshold in the same articles. Correcting it meant re-deriving every published figure. Defining the term in the first article would have cost a paragraph.

Testing explanations rather than asserting them

When we find a pattern, the interesting question is what caused it, and the honest way to answer is to try to rule things out.

Our work on small-unit prices found the premium had collapsed. Five explanations were available: the mix of buildings had shifted, larger units sat on higher floors, buyers wanted more space after the pandemic, supply restrictions had changed the market, or the tenure mix had moved. Each was tested against the data. Each failed. What survived was the explanation we could not rule out, and we said plainly that it survived by elimination rather than by direct observation.

That distinction matters. An explanation reached by elimination is weaker than one directly measured, and an article that does not tell you which it has is hiding something.

What we will not do

We do not fill gaps. Where a number is not available, we say it is not available. We do not estimate and present the estimate as a measurement.

We do not delete outliers. Unusual transactions are disclosed and discussed, not quietly removed to make a line look cleaner.

We do not give verdicts. We will tell you that a small unit yields more and has grown less, and what each of those means over different holding periods. We will not tell you whether to buy it. That depends on circumstances no transaction record contains.

We do not hide the limits. Every article ends with what it cannot tell you. Those sections are not disclaimers; they are the boundary of the claim.

When we get it wrong

We will. The published record includes a wrong instruction to sellers, a definition that did not match the regulator's, and a stated cause that better measurement contradicted. Each was corrected in the article that carried it, dated, with the reasoning.

Two things follow from that. The corrections are the evidence that the checking works, so we would rather have them visible than tidy. And if you find something wrong in our work, tell us, and we will check it and say what we find.

Questions

If you want to know exactly what a figure measured, ask. We can tell you the question that produced it, the properties it covered, and the date it was last verified.