Radar Perene / Archive / science

Business cycles: describing the past is not predicting

◦ Index methodology v2.2 (working papers with DOI). See the methodology.

Science

At every quarterly GDP release, the same scene: if the number comes in negative for the second time, headlines declare a recession; if it comes in positive, they declare it over. Months later, a committee of academics publishes a sober note saying the recession began in a quarter nobody reported and ended in another nobody celebrated. The two versions use the same word for different objects — and the distance between them is an entire lesson on what research can and cannot do with cycles.

A business cycle is the alternation, documented in virtually every market economy, between periods of expansion and contraction of activity. The scientific dating of a cycle is done after it, by committees that examine multiple series — output, employment, income — and not by an automatic rule. Describing the past cycle is measurement work; predicting the next one is a statistically different problem, and a far more fragile one.

The tradition runs deep: Arthur Burns and Wesley Mitchell systematized cycle measurement in Measuring Business Cycles, in 1946, and the American NBER has maintained since the 1970s the dating committee that became the world's model. Brazil has its own: CODACE, the committee housed at FGV, which maintains the reference chronology of Brazilian recessions.

How a cycle earns its name

The committees' method has three traits that distinguish it from the headline. First, multiplicity of series: no single number — not even GDP — decides; a set of activity, employment and income indicators is examined. Second, the tradition's three D's: depth, diffusion and duration — a contraction must be deep, spread across sectors and persistent to deserve the name. Third, and most counterintuitive: deliberate patience. The committee announces with months or years of delay, on purpose, because activity data gets revised and a peak is only confirmed once the subsequent decline consolidates.

That delay is usually read as academic slowness. It is the opposite: it is the formal admission that, in real time, not even the best measurers know with confidence which phase of the cycle the economy is in. The newsroom's "technical recession" — two consecutive quarters of falling GDP — is a shortcut useful for its speed, but it is a mechanical rule over a single, revisable series, and it diverges from the official chronology with documented frequency.

The Brazilian chronology, in open data

CODACE's chronology is public and rewards reading. In numbers: the recession that began in the second quarter of 2014 extended to the fourth quarter of 2016 — eleven quarters, one of the longest contractions in the Brazilian chronology; the 2020 recession, by contrast, lasted two quarters, short and violent. Two recessions adjacent on the calendar and opposite in anatomy: one slow and dragging, the other sudden and deep. Any theory of "the Brazilian cycle" has to accommodate both — which is why research speaks of cycles in the plural, dated case by case, and not of a regular wave with a fixed period.

The house archive crossed the first of those recessions with the record open: the essay the 2014-2016 cycle, from top to bottom rereads that arc with the data the archive kept — and the rereading confirms what late dating suggests: lived from inside, the cycle does not announce its turns; they only gain contour in the rearview mirror.

Why predicting is another problem

The temptation is immediate: if cycles exist and repeat, why not anticipate them? Research's answer is uncomfortable and has three layers. The sample is small — an economy accumulates only a few dozen dated cycles in its entire statistical history, and small samples support little generalization. Real-time data is provisional — the GDP known today is not the GDP that will stand in the revised series, and a model fed provisional numbers inherits the provisionality. And the object changes — the economy's structure between two cycles is not the same, so a regularity learned in one cycle arrives partially expired at the next.

None of this forbids forecasting research — leading indicators are a respectable literature. But it explains why the house keeps the boundary drawn in the entry changing level is not changing regime: describing rigorously what the archive records is one activity; announcing the next turn is another, and methodological honesty lies in not dressing the second in the first one's clothes.

Frequently asked questions

What is a technical recession?

Two consecutive quarters of seasonally adjusted GDP decline. It is a useful journalistic shortcut, not the scientific definition: the official chronology uses multiple series and criteria of depth, diffusion and duration — and the two classifications frequently diverge.

Who dates cycles in Brazil?

CODACE — the Business Cycle Dating Committee, housed at FGV — publishes the reference chronology of Brazilian recessions, modeled on the American NBER's committee.

Why does the announcement take so long?

Because activity data gets revised and a turning point is only confirmed once the subsequent trajectory consolidates. The delay is the price of reliability — the alternative would be dating fast and correcting forever.

Does a recession require negative GDP?

Not necessarily throughout its whole extent. The committees' criterion is the ensemble: a deep, widespread and persistent contraction of activity. GDP is the most visible indicator, not the only judge.

From the cycle that crosses the whole economy, the trail descends to a price that crosses it alongside — one Brazil pays like few others: credit and bank spread →

House readings: each day's state is in today's note, in the Daily; whole cycles, reread with consolidated data, in the precedents, in the Atlas.

Placing a specific period within the cycle chronology, with the series declared, is an exercise the house conducts on request.