A business cycle is the sequence of expansions and contractions in broad economic activity. For a quant researcher, the useful question is both what state the economy is in and what evidence of that state was available when a forecast or portfolio decision was made. The word cycle does not imply a fixed period or a clockwork trading rule.
Direction, level and trend answer different questions
The classical business cycle concerns rises and falls in the level of aggregate real activity. A peak precedes a contraction; a trough precedes an expansion. Real means adjusted for price changes: rising nominal spending alone could reflect inflation, rather than increased production. The RBA's growth explainer develops this distinction.
Three readings of economic news can therefore disagree without any arithmetic error:
| Reading | Question it answers | What it does not establish |
|---|---|---|
| Classical cycle | Is broad real activity expanding or contracting? | Whether its level has regained the previous peak |
| Growth cycle | Is activity above or below an estimated trend, and is that deviation rising or falling? | An outright decline in the level of activity |
| Two falling quarters of real GDP | Has this particular quarterly output measure declined twice in succession? | An NBER recession determination |
The OECD's composite leading indicators target growth-cycle turning points around trend and provide qualitative signals. Its methodology explicitly distinguishes a slowdown from a recession. A falling deviation from trend can accompany an activity level that is still rising.
For the United States, the NBER dates peaks and troughs using broad evidence, including employment, real income and industrial production. Depth, breadth and duration matter; there is no fixed indicator weighting or GDP-only rule. The NBER FAQ explains why two negative quarters are an imperfect shortcut. Other jurisdictions need their own definitions and dating sources.
An upturn can leave the economy below its old peak
Take a fictional monthly index of real activity, assumed seasonally adjusted, with Month 1 set to 100. The observations are 100, 110, 105, 108. They are index levels, not annualised growth rates or actual GDP releases.
Fictional monthly real-activity index, Month 1 = 100; assumed seasonally adjusted. Straight segments join observations. Local turning points illustrate direction versus level; they do not establish an economy-wide recession.
Month 2 is a local peak and Month 3 a local trough within this short sample. The subsequent increase is
Yet 108 is still 2 index points, or approximately 1.82%, below 110. Positive growth and an unrecovered level can coexist. Calling the move a partial recovery describes the example; four readings of one fictional index do not establish sustained, widespread expansion or an official recession. Nor do they tell us when the next turning point will occur.
A historical label has an availability date
NBER dates are retrospective: the committee waits for enough evidence and revisions. An eventual trough month and the later announcement date are different dates. The Philadelphia Fed's real-time dataset preserves successive data vintages precisely so researchers can examine what was known at earlier dates.
In a backtest, joining today's recession labels to historical returns is useful for describing outcomes. Feeding those labels to a strategy as though they were known contemporaneously creates look-ahead bias. The same problem arises if later-revised observations enter an earlier signal.
An implementable cycle signal needs an as-of record: jurisdiction, indicator definitions, observation periods, release dates and the values published by the decision time. Record frequency, seasonal adjustment, real or nominal basis, and whether growth rates are annualised. Evaluate that signal separately from the eventual chronology. A perfect explanation of past phases is not evidence that a model could identify them live.
For an advanced application, the MV real-time macro forecasting capstone examines historical vintages and release timing for US industrial production. Membership and sign-in are required. Its target is a production forecast, not an official recession classifier.
A timing check
Suppose a committee later assigns a trough to an earlier month. A researcher uses that final label to switch a simulated portfolio in the trough month. Does waiting until the next trading day remove the information problem?
Answer. No. The label was still unavailable on that next day. The simulation must use the information and any cycle estimate available at its actual decision time; a one-day execution delay cannot repair a retrospective label.
Sources
- NBER — Business Cycle Dating Procedure: Frequently Asked Questions, especially indicators, GDP shortcuts, direction versus level, and announcement timing.
- OECD — Composite Leading Indicators, About and Methodology, including its growth-cycle clarification.
- Reserve Bank of Australia — Economic Growth, real versus nominal growth.
- Federal Reserve Bank of Philadelphia — Real-Time Data Set for Macroeconomists, vintage snapshots and first-, second- and third-release values.