The next U.S. recession — a three-step ladder, registered and running

Registered in “A GDP-Weighted Multi-Signal System for Real-Time Recession Detection,” which carries the full definitions, data, and scoring rules.

The recession-detection paper registers a standing prediction about the next U.S. recession, built as a three-step alert ladder.

Step one, WATCH: national unemployment filings run at least 10% above the prior year for four consecutive weeks, confirmed by at least 25% of the economy sitting in deteriorating states. Step two, ELEVATED: a stricter national trigger — 18% for three weeks — confirmed the same way. Step three, CONFIRMED: the Sahm Rule, the slow and widely trusted unemployment benchmark, crossing its public half-point trigger.

The three rungs do three different jobs: warning, detection, confirmation — and the official recession-dating committee’s eventual call is the validation. The Sahm Rule sits last by design, and it earned that place on the measurements in this paper. Compared head to head at the paper’s registered default settings — this system’s own detector against the Sahm Rule, each on its own — the Sahm Rule was the more accurate of the two: it detected all eight recessions to this detector’s seven, with two matured false alarms to this detector’s four. What it trades away is speed. Built from smoothed monthly survey data rather than raw weekly filings, it is a coincident instrument by construction: it speaks once a downturn is under way rather than before one arrives, its crossings falling on average about three months after the peak date the committee later assigns. That is not a shortcoming — it is what a confirmation rung is for, and it is why the faster, jumpier signals sit ahead of it rather than the other way round. Comparing the Sahm Rule against the full ladder would not be a fair test in either direction, since the ladder’s final rung is the Sahm Rule itself.

No instrument is exempt from this, including that one. The July 2024 employment report pushed the Sahm Rule across its trigger, no recession followed, and by the pre-registered scoring used here that episode matured as a false alarm. Dr. Sahm said as much herself at the time. Days before the trigger she wrote that “a recession is not imminent, even though the Sahm Rule is close to triggering,” and once it fired she stated plainly that the country was “not in a recession, despite the Sahm Rule indicator bearing my name” saying otherwise — while adding that recession risk was elevated and that the conversation the rule had started was one worth having. Her stated reason was that unusual labor-force swings were feeding into the unemployment-rate change her rule reads. We note the episode because it illustrates the point this series keeps returning to, and because the way she handled it — publishing the limits of her own instrument while it was being read too confidently in her name — is the standard this project is trying to meet.

The registered claim: ahead of the next officially declared U.S. recession, the ladder escalates in that order.

The paper carries one further layer that sits outside the ladder, because it answers a different question: not whether a downturn has begun, but where it began. Working only from data available before the recession’s peak, it asks which industry was already declining across the largest share of the deteriorating map. For 2007 it names real estate, with a footprint of 0.87 across 46 declining states, cross-checked against a separate government employment dataset with the one disagreeing state printed. Economists identified housing as the origin of that recession long ago, and this layer does not claim to have discovered it. What is unusual is the manner of arriving there: a rule fixed in advance, run on pre-peak data, with a scoring bar that lets the method decline to answer. For 2001 the strongest candidate scores 0.196 — noise — and no epicenter is named; for 2020 the leader falls below the bar as well, and is left there rather than rounded up into a story.

A note on settings, because the distinction matters when reading the paper alongside this board. The ladder above is not the paper’s default configuration. The default was re-tested on a freshly sealed data vintage, missed the 2001 recession by a knife edge — a government revision to old data moved a borderline confirmation just outside its window — and the paper records that result as a refutation. The ladder’s settings come from the pre-registered sensitivity grid that followed, and this warning → detection → confirmation combination detects all eight recessions since 1967. The paper carries both results in full: the default’s failure and the grid that produced these settings.

The failure conditions are registered too. A WATCH that ELEVATED never joins counts, by the written rule, as a self-cancelling false alarm against the system. A recession arriving in the fast, concentrated shape of 2001 is named in advance as the design’s known weak spot.

The starting position, frozen at the paper’s data vintage:

Anyone can keep score. The data is public, the rules are published, and the official recession-dating committee’s eventual call settles it.

This page reports a registered research prediction. It is education, not advice: nothing here is a forecast service or an investment signal.