What Actually Defines a Recession?

Why the two-consecutive-quarters rule misses the point — and what serious economists look at instead.

Martin Harrison · Archimedes Research Group · March 2026

In the summer of 2022, the Bureau of Economic Analysis released a number that set off a firestorm. Real GDP had contracted for the second consecutive quarter — down 1.6% in Q1 and 0.6% in Q2. By the most commonly cited definition of a recession, the United States had just entered one. Media outlets ran with the story. Politicians argued over the semantics. And economists found themselves in the uncomfortable position of explaining why the answer was, in fact, considerably more complicated than a single number.

The debate that followed was not merely political noise. It exposed a genuine flaw in how most people — and many institutions — understand what a recession actually is. The two-consecutive-quarters rule is intuitive and easy to communicate, but it is not how recessions are officially determined in the United States, and history shows it can be both too aggressive and too slow depending on the circumstances.

The 2022 episode was not ultimately declared a recession. Understanding why requires a deeper look at what economic data actually measures — and what it misses.

The Official Arbiter: What NBER Actually Does

In the United States, recessions are officially dated by the National Bureau of Economic Research (NBER) Business Cycle Dating Committee — a group of academic economists who review a broad array of economic indicators before designating a peak or trough in the business cycle. They are deliberately not bound by the two-quarter GDP rule.

The NBER defines a recession as "a significant decline in economic activity that is spread across the economy and lasts more than a few months." The key word is "spread." A recession, by this definition, must show up not just in output but across employment, income, industrial production, and sales. GDP captures the value of final goods and services produced — it is an essential measure, but it is a quarterly estimate subject to revision and it reflects only one dimension of economic health.

The NBER's business cycle chronology identifies six periods of economic contraction since 1980: 1980, 1982, 1990-91, 2001, 2008-09, and 2020. Notably, the 2001 recession — despite widespread job loss, a collapse in manufacturing employment, and unemployment rising from 4% to nearly 6% — did not register as two consecutive quarters of negative GDP growth. Under the simple rule, 2001 would not be a recession. Under NBER's broader framework, it clearly was.

And in 2022? The NBER did not call it a recession. The labor market remained exceptionally strong throughout both negative-growth quarters — payrolls were expanding, unemployment was near historic lows, and consumer spending held up. The GDP contraction was real but narrow, concentrated in trade and inventory adjustments rather than reflecting genuine broad-based economic deterioration. The two-quarter rule fired. The economy, by most other measures, was not in recession.

The Coincident Index: A Better Real-Time Gauge

One of the most useful alternatives to GDP for real-time recession monitoring is the Conference Board's Composite Index of Coincident Indicators (CEI). Published monthly, it aggregates four components that collectively describe the current state of the economy: nonfarm payroll employment, personal income less transfer payments, industrial production, and manufacturing and trade sales.

Because it updates monthly rather than quarterly, the CEI provides a more granular picture of turning points. And because it combines labor, income, output, and sales into a single index, it is far less susceptible to the kinds of one-time distortions — a surge in imports, a draw-down in inventories — that can temporarily drag GDP into negative territory without signaling genuine economic contraction.

Comparing the CEI to Real GDP across the past four decades reveals a consistent pattern: every NBER-designated recession shows up in both measures, but the CEI tends to provide an earlier signal at turning points.

The 2008 Global Financial Crisis illustrates this well. The CEI began declining in Q2 of 2008, while Real GDP did not post a negative quarter until Q4 of 2008. Conversely, Real GDP turned positive in Q3 of 2009 while the CEI did not bottom out until Q4 of 2009 — suggesting the CEI tracked the actual recovery more accurately than the headline GDP figure.

Leading vs. Coincident vs. Lagging: Why the Distinction Matters

Economic indicators are typically classified into three categories that reflect their relationship to the business cycle, and understanding that classification is essential to using them correctly.

Leading indicators change before the economy as a whole changes. The yield curve — specifically the spread between the 10-year and 3-month Treasury yield — is among the most powerful. When short-term rates exceed long-term rates (an inverted yield curve), it has historically preceded recessions by six to eighteen months in every cycle since the 1960s. Initial unemployment claims, building permits, and the money supply are other examples. These are the variables that give you advance warning.

Coincident indicators move with the economy in real time — payrolls, personal income, industrial production. They tell you where you are now. GDP is largely a coincident measure, though its quarterly frequency and tendency toward revision makes it less nimble than monthly coincident series.

Lagging indicators confirm what has already happened. Unemployment is the classic example: firms do not immediately lay off workers when activity slows, and they do not immediately rehire when it recovers. The unemployment rate tends to peak well after a recession has officially ended. Using it as a leading signal — as many commentators do — is a persistent analytical error.

Confusing lagging indicators for leading ones is one of the most common errors in public economic commentary. Unemployment rising is confirmation of a downturn already underway, not a warning of one to come.

The Limits of Any Single Measure

The 2022 debate ultimately illustrated something more fundamental than a disagreement over semantics. It revealed how poorly equipped most public discourse is to handle economic complexity. The desire for a single, definitive number — a binary yes/no on whether a recession has occurred — is understandable, but it misrepresents how business cycles actually work.

Recessions are not uniform events. The 2020 contraction was the deepest on record by percentage decline in GDP but the shortest in duration — two months by NBER dating. The 2008-09 recession was longer and broader, with cascading effects across credit markets, housing, employment, and global trade that took years to fully unwind. The 1990-91 recession was mild by historical standards but marked a turning point in how the labor market functioned, with a "jobless recovery" that reshaped the relationship between output growth and hiring.

Each cycle has its own character — its own leading indicators that fired most reliably, its own sectors that led the decline, its own policy response that shaped the recovery. Reducing all of this to a single GDP figure, or to a yes/no recession classification, discards information that is essential for making good decisions.

A Multi-Indicator Approach: The ARG Framework

At Archimedes Research Group, our approach to business cycle monitoring is built on the same principle the NBER has applied for decades: recessions are multidimensional events that require multidimensional measurement.

Our ARG Recession Probability Model synthesizes the most historically reliable leading and coincident indicators into a single probability estimate — expressed as a percentage — updated monthly. Rather than asking "are we in a recession?" it asks the more useful question: "what is the probability that the economy will be in recession twelve months from now?" That forward-looking framing is what institutional investors, corporate planners, and policy analysts actually need to make decisions.

The model draws on the yield curve spread, initial unemployment claims trends, payroll growth, and financial conditions — each selected for their demonstrated predictive power across multiple business cycles back to 1975. Validated against six recessions, the model has produced elevated probability readings ahead of each. It is not infallible — no model is — but it provides a rigorous, transparent, and regularly updated signal that the two-quarter GDP rule simply cannot match.

Conclusion

The 2022 recession debate will not be the last time the public is misled by an oversimplified economic heuristic. Every cycle produces its own version of this confusion — a headline number that tells a partial story, commentators who mistake lagging indicators for leading ones, and policy decisions made on incomplete information.

The solution is not more data — it is better frameworks for interpreting data. Understanding the difference between leading, coincident, and lagging indicators. Knowing what GDP measures and what it doesn't. And building research infrastructure that synthesizes across indicators rather than deferring to whichever number is easiest to communicate.

That is the work Archimedes Research Group exists to do.