How accurate the estimate was for HDB streets, Q2 2026: median error 3.4%

This is a backtest, not a market report. The engine's estimate for each HDB street was re-run as of Q4 2025 using only the transaction record available at that point, then compared against the resale prices actually recorded on that street afterwards. Across 514 HDB streets with at least five sales to test against, and 22,530 tested sales in total, the estimate missed the price each home then fetched by a median of 3.4%. The error is measured per sale and summarised per street, so a street's score is the typical gap on that street rather than the largest one.

The spread around that midpoint is narrow. Half the HDB streets scored between 2.7% and 4.5%, and 423 of them, 82.3%, came within 5%. A separate measure tracks the published range rather than the point estimate: on average 32.9% of tested sales fell inside the range shown for that street, which is a statement about how often the band captured the eventual price, not about how close the midpoint sat.

The chart above sorts the same 514 streets into error bands. The 2.5–5% bar is the largest at 333 streets, with 90 streets below 2.5% and 76 in the 5–7.5% band. Beyond that the counts thin quickly: 9 streets in 7.5–10%, 2 in 10–15%, 4 in 15–20%, and none above 20%. All figures are indicative and derived from recorded transactions.

Where it was closest

The ten streets in the table below all sit inside the chart's lowest band, below 2.5% median error. They are not uniformly deep samples. Tampines St 61 was tested on 162 sales, Bedok Nth St 4 on 47 and Pasir Ris St 52 on 37, while Punggol Field Walk rests on 7 and Bt Batok West Ave 7 on the smallest sample the test accepts. A low median error on a long run of sales says more about consistency than the same score on a handful.

The share of sales falling inside the published range varies across these streets even where the median error is similar, because the range width is set separately from the point estimate. Kim Keat Link, tested on 13 sales, and Bt Batok West Ave 7 captured a larger share of their tested sales within the band than Tampines Ctrl 8, tested on 30, or Woodlands Dr 71, tested on 12. Each project page carries its own score on the same basis.

Where it was furthest off

The table below lists the ten HDB streets with the highest median error, published because a score that hides its failures is not a score. All four streets in the chart's 15–20% band appear here, as do both streets in the 10–15% band. Dover Rd, tested on 6 sales, carried the largest median error of any street in the set, followed by Upp Aljunied Lane on 10 sales, C'wealth Cl on 24 and Lor 3 Toa Payoh on 14.

Sample depth is thin through most of this group, which means a small number of unusual resale prices can move a street's median. Two streets in the table, Upp Aljunied Lane and Choa Chu Kang Loop, had no tested sale fall inside the published range at all. Potong Pasir Ave 1 is the exception on depth, tested on 45 sales, and Indus Rd on 8 and Queen St on 7 sit closer to the lower end of the tested counts.

Each project's backtest re-runs the estimate as of an earlier quarter using only what was known then, and compares it with the sales recorded since. On that basis the Q2 2026 HDB run produced a median error of 3.4% across the tested streets, with the bulk of streets clustered in a band a little either side of it and a small tail where the estimate was well off. The share of sales landing inside the published range, averaging 32.9%, is measured separately. The figure is a property of the engine's past estimates, not a valuation of any home.

Each project's backtest re-runs the estimate as of an earlier quarter using only what was known then, and compares it with the sales recorded since. Every project page shows its own score. The figure is a property of the engine's past estimates, not a claim about any home.