US Recession Indicators: What They Measure and What They Cannot Tell You

Dating, an indicator and a model answer different questions

A US recession date, a fall in an economic series and a model's recession probability are different kinds of information. Before comparing them, identify what each measures, which period it describes and when it became available. A signal is not an official turning-point date, and a probability is not a guaranteed outcome.

This guide explains the labels using NBER's US dating convention and a specific yield-curve model described in a 2006 New York Fed research paper. It does not report a current recession assessment, today's yields or a current model probability.

Keep retrospective US dates in their role

The National Bureau of Economic Research (NBER) maintains a chronology of US business-cycle peaks and troughs. Its recession concept considers a significant decline in activity spread across the economy and lasting more than a few months. Depth, diffusion and duration each matter to some degree; extreme conditions in one can partly offset weaker evidence in another. ^[1]

This is a retrospective dating task. A turning point describes economic activity in an earlier period, not the day on which a date is announced. It also differs from forecasting whether a recession will occur later. The US committee's convention should not be substituted for another country's definition or a legal or tax rule.

Two consecutive quarters of falling real GDP are not an automatic substitute for NBER dating. The committee considers a range of information and the depth of the decline, with monthly chronology as well as quarterly evidence. A GDP rule of thumb and the committee's chronology can therefore differ. ^[1]

Record the series, reference period and vintage

Economic series do not all count the same thing. A payroll-jobs change, a price-index change and a measure of output need their own units, populations and comparison periods. Do not combine them into an unlabeled score or assume that one favorable or unfavorable number settles the state of the economy.

For each observation, record the source, reference period, publication date, seasonal adjustment when applicable and release version. A revision to an earlier month is information about that earlier period. Comparing a first release with a later revised history without noting the difference can obscure what was available at the time.

The nonfarm payrolls guide separates jobs, people and revisions. The CPI guide separates index levels, inflation rates and comparison periods. Those reading tasks help label inputs; neither article supplies a recession-declaration rule.

The sequence labels reading tasks; it gives no real release schedule or fixed announcement lag. A historical model description is not evidence that every data input was available to a reader on the first day of its reference month.

Read the exact target and input of a yield-curve model

A yield-curve model needs more than the word inversion. Arturo Estrella and Mary Trubin's 2006 paper describes a probit model connecting a defined Treasury term spread in month t to recession in month t+12, viewed from information available in month t. The target is that particular month. ^[2]

Recession in month t+12 is not the same event as at least one recession month anywhere from t+1 through t+12. Other models can choose different targets or dating conventions. Read the definition rather than interpreting every phrase such as “one year ahead” as the same probability.

With long minus short as the stated orientation, a negative spread means the specified long yield is below the specified short yield: the curve is inverted at that pair of maturities. A negative sign alone does not determine a recession probability. The model and its definitions supply the additional relationship.

The paper examines why maturity, yield basis, averaging and persistence matter. These details belong to its historical construction. They do not establish the parameters or data processing of a current dashboard, and this guide does not reproduce its coefficients or calculate a live estimate. ^[2]

A signal can fail; one threshold is not a turning-point date

A forecasting signal can be followed by no recession at the specified horizon, while a recession can occur without that signal. Evaluating either case requires a definition of the signal and the target. Changing from a single day to a monthly average, or from a target month to an interval, changes the question being tested.

The 2006 paper discusses historical differences between short-lived daily inversions and more persistent monthly signals. Its findings belong to its sample and definitions; they are not a promise that an inversion will work without false signals today. Nor does a historical fit automatically establish the quality of a forecast made with only the information available in real time. ^[2]

Keep a model estimate separate from a policy-rate estimate too. The FedWatch guide explains a different outcome: a specified policy-rate level at a selected meeting under a futures-based model. A similar-looking percentage does not make its event or horizon interchangeable with recession risk.

Build a label-first recession record

Before drawing a conclusion, write down the source and which of the four information types it provides. Then keep the reference period, release vintage and target in separate fields. This short record makes the limits of a comparison visible.

• Source and version: which publication, series definition or model document are you using, and when was it issued?
• Input: what is measured, in which unit, for which reference period and adjustment or averaging basis?
• Availability: when was that version released, and does a later history contain revisions unavailable at the time?
• Output: is this a chronology date, an indicator, a threshold rule or a model estimate?
• Target: what exact month, interval and recession convention does the claim concern?
• Limits: what is unknown about current inputs, model changes, false signals or performance outside the estimation sample?

If a field is unavailable, leave the corresponding conclusion open. This guide provides no current model reading to fill that gap. The record organizes definitions and timing; it is not a validated forecasting method or a recommendation to buy or sell.

An event contract would also need its own recession definition, designated source, deadline and treatment of later announcements. A statistical indicator does not set those rules or establish that a product is offered or available to you. Real-money positions can lose money and incur fees; recession analysis is not a dependable income source or a route to recovering losses.

Common questions

Do two negative GDP quarters automatically define a US recession?

No. NBER considers a range of indicators and the depth, spread and duration of a decline, rather than applying only that rule. Its retrospective US chronology also differs from a forecast or another country's definition. ^[1]

Does an inverted yield curve guarantee a recession?

No. Identify the maturities, yield basis, averaging period and forecast target before assessing the signal. Historical relationships do not guarantee a future outcome or make every brief inversion equivalent.

Does twelve months ahead mean any recession during the whole year?

Not automatically. In the cited 2006 model, the target is recession in month t+12 using information at t. At least one recession month anywhere in t+1 through t+12 is a different event. Another model's documentation may specify a different target.

Why can turning-point dates be announced later?

Retrospective dating assesses the evolution of activity, whereas a real-time indicator describes the information released so far. Keep the economic turning-point month and any announcement date separate; this guide supplies no fixed announcement lag.

Can payrolls or CPI alone establish recession status?

No. Their figures describe different statistical objects. Label the series and release first, then distinguish that observation from a dating judgment, rule or probability model. An isolated headline does not substitute for that work.

Sources and scope

NBER’s US dating convention is supported by substantive official indexed FAQ text; direct access to the original page was limited. Estrella and Trubin’s 2006 paper supports the specified historical yield-curve construction and target month, not a current forecast. These sources were consulted in the 18–19 October 2026 evidence window. No current NYFed model parameters, yield values, chart or probability are presented. The timeline uses fictional labels.

1. Business Cycle Dating Procedure: Frequently Asked Questions
2. The Yield Curve as a Leading Indicator: Some Practical Issues