Everything needed to check, reuse, or challenge the number. The definition is versioned; changes are listed at the bottom and annotated on the chart at the month they take effect. History is restated on cutover, never silently redefined.
The Warning-Sign Index is under review and is not being published this month. The aggregates on this page are unaffected. The Warning-Sign Index history file is not currently published.
WSI = ZIP markets at WATCH or ACT ÷ all scored ZIP markets, as a percentage, computed monthly. A ZIP is scored when at least two of the four index signals are known for the month. Insufficient-data ZIPs are excluded from both numerator and denominator. Seller's-market readings — a market with no danger line crossed and every strength condition met — count in the denominator only.
These are the index's thresholds, and they are deliberately frozen. The Warning-Sign Index is a historical series, so it is computed on a constant four-signal basis over closed-sale statistics — restating it under new rules would make this month incomparable with every month before it.
The site's current per-ZIP readings are computed differently: three signals over active-listing statistics, with two lines recalibrated for that basis. Months of supply and price-cut share need a closed-sale count that active-listing data cannot see, so they are not part of a current reading. See the reading methodology for the reading engine; the table below describes this index only.
| Signal | Danger line |
|---|---|
| Months of supply | > 4 (severe > 6) |
| Median sale price, y/y | < −2% (fast < −5%) |
| Time to sell, y/y | > +40% |
| Inventory, y/y | > +50% |
A fifth signal — the share of listings cutting price (> 35%) — was used by per-ZIP readings under the previous engine but has no history before 2026, so it is deliberately excluded from this index; measured impact of the exclusion at adoption was 0.2 points (62.2% vs 62.4%). The current reading engine does not use it either: the data source behind it is no longer in use.
The continuous series begins June 2020: a prior data vendor’s sold-home statistics — the file the site refreshed from until 14 August 2026, and no longer does — ~25,000 scored ZIPs per month. The 2012–2019 tail is reconstructed from that vendor’s legacy tracker (~18,000 ZIPs, a prior universe), drawn in a lighter stroke, and excluded from every record, delta, and superlative. The two sources overlap for 72 months; per-ZIP level agreement across 1.36M shared zip-months is 72.7% — similar, not the same, which is exactly why claims never reach across the seam. The current month is always computed from the site's own published per-ZIP data, so the index and the ZIP pages a reader can check agree by construction.
"Highest since {month}" names the last month the index was at or above the current value — ties block the bigger claim, so no superlative can contradict the archive. "Record" always means "within the continuous series", never "ever". ZIP warning streaks count consecutive months at WATCH or ACT; a month without a score breaks the streak.
ZIPs map to Metropolitan Statistical Areas via the Census 2020 ZCTA↔county relationship file (largest land-overlap county per ZCTA) chained to the OMB 2023 CBSA delineation — the same definitions the press already uses. ZCTAs approximate ZIP codes; the divergence is the standard documented compromise, because USPS publishes no ZIP geography. Metro league tables require ≥ 15 scored ZIPs.
Each signal's threshold was tested against the FHFA's official ZIP-level house-price outcomes (182,644 zip-year pairs, 2012–2026, outcomes through 2025): of markets past a line at year-end, the share whose FHFA index declined the following year, versus markets clear of it.
| Signal crossed | Declined next year | Clear of the line |
|---|---|---|
| Months of supply > 4 | 36.7% | 12.4% |
| Prices falling y/y | 20.4% | 10.9% |
| Time-to-sell up > 40% y/y | 20.4% | 11.8% |
| Inventory up > 50% y/y | 20.0% | 12.2% |
Warning ≠ crash: a WATCH/ACT market is one where decline risk is elevated versus baseline, not one guaranteed to fall. The method is the paragraph above: each signal's line, applied at year-end, scored against the FHFA index the following year. The lines themselves and the calibration record are further down this page.
Markets slow before they fall. Prices are the last thing to move, because sellers hold their asking price long after buyers have stopped paying it. What changes first is behaviour, and behaviour is visible.
Three things happen, in this order, while the headline price still looks fine: homes take longer to sell, unsold homes pile up, and sellers start cutting asking prices. Each is measurable monthly from licensed market statistics, per ZIP code.
So we watch three gauges — the year-over-year price trend, how long listed homes have been on the market, and how the pool of homes for sale is changing — each with a published danger line. The price line is drawn from what happened in past downturns; the two volume lines were recalibrated when the underlying statistics changed from closed sales to active listings, because the old values fired on well under one per cent of markets on the new basis. Every line is published on this page and never moved to fit a story. A market past enough of them reads WATCH or ACT.
We also watch the speed of approach: a market can sit well inside every line and still be closing on one fast. That drift is usually visible months before the crossing itself.
This is not a prediction. It is a smoke detector. It tells you that the conditions which have preceded past declines are present in your market now. Smoke detectors go off for burnt toast, and some markets cross a line and recover — which is why one of the cases below is a market that did exactly that.
From 2026-05 onward we snapshot every month's readings as published, so they can be scored against what prices actually do next — recall on real declines, precision of the flags, and lead time. The first 12-month scoring lands in 2027-05; until then this section says "collecting" rather than showing partial numbers. Updated monthly once live. One caveat applies now and always: danger lines are calibrated on past downturns; future downturns may differ.
Index values, league tables, and release CSVs may be used, quoted, charted,
and republished in journalism, research, analysis, and commentary, with
attribution: "Source: ShouldISellYet (shouldisellyet.com)." The CSVs
carry ShouldISellYet's derived indicators only — never upstream raw metrics,
which belong to their publishers, and we grant no rights in any third party's
underlying data. Redistributing the files as a dataset, or using them to build
a competing data product or service, is not permitted; the full terms ship
with each release as LICENSE.txt. Media/data questions:
press@shouldisellyet.com.
| Version | Effective | Change |
|---|---|---|
| v1.0 | June 2026 | Index launched. Four-signal constant methodology (months of supply, price trend y/y, time-to-sell y/y, inventory y/y — the site's published danger lines, price-cuts share deliberately excluded for lack of pre-2026 history). Continuous series begins June 2020 (a prior data vendor's hub, ~25k scored ZIPs/month); 2012–2019 context tail reconstructed from the legacy market tracker (~18k ZIPs, prior universe), charted separately and excluded from records. Measured cross-source level agreement over the 72 overlapping months: 72.74% of 1.36M shared zip-months. |