← Daybreak
RECESSION, HONESTLY · A DATA-DRIVEN OVERVIEW

Can you see a recession coming?

0 years·0 recessions

1 question: can you see one coming?

The yield curve has a perfectrecord of preceding recessions — but it says if, not when, and its deepest-ever warning just misfired. We test the three best-known alarms — the curve, a machine reading the present, and credit spreads — and ask what each can honestly tell you.

Built from public US data across 64 years and eight recessions (1962 → mid-2026), tested only on data the models had never seen. We sell no predictions. The job is to separate what the record supports from the headlines. Observation, not advice.

THE WHOLE PICTURE

Eight recessions, each a grey band.

Here is the S&P 500 across the modern era, shown on a log scale — the only honest way to compare percentage moves across decades. Each grey band is one official US recession: the start-to-end dates the National Bureau of Economic Research later marked as a contraction.

Eight recessions since 1962, eight grey bands. Every one of them shows up as a stumble in the index. The question this whole report asks is simple: does anything warn before the grey arrives?

The three parts that follow each test one well-known alarm — the yield curve, a machine reading the present, and credit spreads — against this same record.

overview

loading overview
The index climbs roughly a hundredfold across the era, but every grey band marks a real drawdown. Nothing on this chart warns you the grey is coming — that is what the alarms are for.
PART 1 — THE CURVE: A WARNING WITH NO CLOCK

A perfect record — that just misfired.

Normally longer loans pay more than shorter ones. When that flips — when the 1-year Treasury rate climbs above the 10-year, an “inverted” curve — it has historically been the single most reliable warning that a recession is coming. Every one of the eight recessions since 1962 was preceded by one.

But a perfect record hides two problems. The warning says a recession is likely — it cannot tell you when. And the deepest, longest inversion ever recorded, in 2022–24, has so far been followed by no recession at all.

curve_spread

loading curve_spread
Every recession (grey) sits just to the right of the curve dipping below zero (amber) — zero misses in 64 years. But the giant 2022–24 amber block has no grey band after it: the loudest alarm on record was, so far, a false one.

curve_lead_times

loading curve_lead_times
Across the recessions it correctly warned about, the curve led by anywhere from 9 to 19 months — a typical lead of about 13, but with a wide spread. It reliably says “a recession is coming.” It cannot say when.

curve_base_rate

loading curve_base_rate
The deeper the inversion, the worse the odds. With a steeply positive curve a recession almost never followed within a year; deeply inverted, it followed about 60% of the time — over four times the all-history average. Depth matters, not just the bare fact of an inversion.
PART 2 — THE MACHINE THAT READS THE PRESENT

It spots a recession nine months before the official word.

The curve forecasts the future and pays for it in vagueness. A different approach gives that up: instead of guessing months ahead, a machine reads the presentmonth of the economy — unemployment momentum, payroll growth, the curve, the credit spread, consumer sentiment — and outputs one number: the chance a recession is underway right now.

We tested it the honest way: each year it was retrained only on the past and judged only on data it had never seen. It tracks recessions almost perfectly — and because the official body that dates recessions publishes with a long delay, the machine’s read of “now” lands about nine months before the press release.

nowcast_prob

loading nowcast_prob
The probability jumps through 50% right at the start of each grey recession band — while the official announcement (amber) lands far to the right. The machine reads the present; its “lead” is over a slow committee, not over the economy.
WHAT THE NOWCAST DELIVERS, OUT OF SAMPLE
0 motypical lead over the official recession announcement
0recessions in the test era (1980 onward) it caught
0recessions it missed at the time they began
0.0%its recession reading for May 2026

nowcast_lead

loading nowcast_lead
The machine (blue) beat the official announcement by 3 to 17 months, clustered around nine, edging out a simpler unemployment-only rule (grey). Both work because the official call is slow — but the machine’s lead is steadier across very different recessions.

nowcast_drivers

loading nowcast_drivers
Payroll growth and unemployment momentum carry the call, with sentiment and credit close behind. The famous yield curve barely registers here: it warns a year ahead but reads the present poorly, so the machine leans on the labor market instead.
PART 3 — CREDIT: SMART MONEY, BUT LATE

Credit prices stress — usually after it has already broken.

The bond market is supposed to be the smart money: the extra yield investors demand to hold riskier corporate bonds over safe Treasuries — the credit spread — should widen before trouble, a warning the stock market hasn’t caught yet. That is the story. The record is more sobering.

Most of the time the spread blows out duringa sell-off, not before it. Of every sharp spread spike in a century, nearly two-thirds happened when the S&P was already down 10% or more. Credit is smart money only in the rare case when it moves first; usually it is just confirming what stocks already know.

credit_spread

loading credit_spread
The spread spikes into and during almost every recession (grey), then settles. Today it sits near the 35th percentile of its whole history and below the long-run average (amber line) — by this gauge, a low-stress moment.

credit_event_study

loading credit_event_study
There is no average crash after a credit-spread blowout. The S&P drifts up into the spike, goes flat for a month or two, then resumes climbing — up about 7% a year later. Because most spikes happen mid-decline, the spike marks a pause near the bottom, not a top.
PUTTING IT TOGETHER · JUNE 2026

Where we are, honestly.

This is not advice to position for, or against, a recession. It is a map of what each well-known alarm can honestly tell you — and what all three are saying right now. Observation, not advice.

The curve

A perfect record with no clock. Every recession since 1962 was preceded by an inverted yield curve — zero misses — but the lead ranges 9 to 19 months, so it says if, not when. And the deepest, longest inversion ever (2022–24) has so far produced no recession at all.

The machine

Reads the present, not the future. A model of the labor market, the curve, credit and sentiment tracks recessions almost perfectly and crosses 50% about nine months before the official announcement — but that lead is over a slow committee, not over the economy. It nowcasts; it does not forecast quarters ahead.

Credit

Smart money, but usually late. The credit spread widens during stress at least as much as before it — nearly two-thirds of spikes fire when stocks are already down 10%+. The rare spike from a calm market is a genuine warning; most are just confirmation.

Where we are

June 2026: nothing is flashing. The curve has un-inverted back to a normal positive slope; the machine's recession reading is about 0.8%; credit sits calm near its 35th percentile. The honest catch — the loudest alarm just failed, so 'all clear' deserves a discount.

LIMITATIONS — READ THESE
  • Only eight recessions since 1962 (six in the machine's test era). Every statistic here rests on a handful of events — treat the rates as directional, not precise.
  • The curve's 2022–24 misfire is real and unresolved. The deepest inversion on record has produced no recession yet, so the historical mapping may be weakening — the report leads with that tension, it doesn't bury it.
  • The machine is a nowcast, not a forecast. It reads the present month well; its lead is over the official body's deliberate slowness, not over the economy. Pushed even a quarter ahead, its skill drops sharply.
  • These use final, revised data. A true real-time test would use the messier first-release numbers, which would add noise and trim the realized lead — so the skill shown is an upper bound.
  • Credit is mostly coincident. Its only honest edge is flagging a higher chance of a deep drawdown when the spread is extreme — a risk gauge, not a timing signal. Calm credit reassures about now, not next year.
  • Descriptive, not predictive. This is a record of what has happened around past recessions; it is not a validated forecast and it sells no predictions.

Descriptive research, not financial advice. Recessions and bear markets can arrive with little warning — and the signals that usually warn have failed before, as the 2022–24 curve inversion shows.

Appendix — every chart & data inventory →