Economic indicators can be described by the timing of their turning points relative to a chosen business-cycle reference: leading, coincident or lagging. That classification does not rank trustworthiness, publication speed or trading profitability. Examine the indicator's definition and the information available at each date before using it.

Three different clocks for data

Indicators that describe the state of the economy are grouped by how their timing falls relative to the business cycle — the successive phases of expansion and slowdown. The Conference Board, the American research body that has long published composite cycle indices, uses three categories: leading, coincident and lagging. A technically analogous trap awaits technical indicators that repaint historical bars — they appear to lead price but in reality only reveal what was knowable after the fact.

Separate cycle timing, information availability and market response. The first compares turning points with a reference cycle. The second asks which period was measured and when the value became public. The third compares a release with expectations and subsequent prices. In our fictional example, March’s new-orders survey is published in April. It might inform later production, but its March result was not already available at the start of March; assigning it to a trade then would use future information.

Smoothing alone can delay a decline

Before calling a series lagging, check what its filter does. In invented data we average only the current and two preceding values. Both series use the same signal rule: the first decline from the previous available point. Raw data meets it in period 4; smoothed data only in period 6.

Scroll sideways to see the full diagram

First decline in raw data and a three-period meanRaw values are 100,110,120,118,116,114. The mean of the current and two preceding periods is unavailable in periods 1 and 2, then 110,116,118,116. The first raw decline is in 4; the first mean decline is in 6. Both share the same period axis; the two-period difference comes from the filter in this example.A filter can delay the first declineRaw data3-period mean120110100Raw falls P4Mean falls P6123456Raw: 100, 110, 120, 118, 116, 114Mean: n/a, n/a, 110, 116, 118, 116Difference: 2 periods
The mean begins in period 3, when three observations are available. Dashes in periods 1 and 2 mean unavailable filter values, not zeroes.

Here, the same data with a different transformationshifts the first detected decline by two periods. This is not a fixed filter lag: a different sequence can give a different result. Record the window, calculation direction and signal rule before comparing indicators.

Assumptions: The series is fictional, evenly spaced, with no revisions or missing observations. The fixed three-period mean uses only values available in each period. We examine the first point-to-point decline, not the official OECD CLI method or economic forecasting ability.

Leading design is not a currency forecast

OECD’s CLI is designed to signal potential turns in activity around its long-run level; its message is qualitative rather than a forecast of the amount of growth. A falling CLI alone cannot calculate future GDP or EUR/USD pips. Nor are all leading-index inputs intentions or sentiment: financial prices and observed claims can be included.

Read the exact series. The US LEI list includes ISM New Orders, building permits, stock prices and average weekly initial jobless claims. An ISM New Orders component is not the whole manufacturing or services PMI report. The Conference Board’s current US release names payroll employment, personal income less transfers, manufacturing/trade sales and industrial production as CEI components.

A component’s cycle relationship does not fix publication speed, reliability or trading profitability. Check the country, construction, transformations and revision policy. Two related releases need not be two independent confirmations of a hypothesis.

Coincident and lagging indicators

Examples from The Conference Board’s US classification
Relative to the cycle, not minutes after an announcement
ClassSelected seriesWhat to investigate
LeadingISM New Orders; building permitsPossible later activity change
CoincidentNonagricultural payroll employment; industrial productionCurrent broad activity
LaggingAverage unemployment duration; CPI for servicesAdjustment after cycle change

The services-CPI component does not classify every inflation release as a late trading signal. Likewise, average unemployment duration is not the same series as the unemployment rate. Keep the exact definition and reference cycle.

Distinguish the period measured from the publication date. Unemployment-related series can lag a cycle reference, while GDP measures activity during its reference quarter even when an estimate is released later. A third GDP estimate is not permanently final. CPI behaviour can vary with the inflation process; publication delay alone does not make a series economically lagging.

BEA can update history after the advance, second and third GDP vintages. BLS annually recalculates seasonal factors; previously published seasonally adjusted CPI indices can be revised for up to five years. Preserve the relevant adjustment and vintage instead of mixing an adjusted monthly change with expectations for another series.

Surprise is different from improvement

A historical-period release can add information through surprises, revisions and components. Our examples below are invented, including the forecast median recorded before publication; they are not actual economic announcements.

Our comparison card
Different units cannot be added to one dollar score
MeasureInputsResult
New-orders surpriseActual 49; consensus 48 index points+1 point
New-orders period changeActual 49; prior 50 index points−1 point
Monthly CPI surpriseActual 0.4%; consensus 0.2%+0.2 percentage points
Prior employment revisionOld +180,000; revised +150,000−30,000 jobs

The same orders result beats consensus while falling from the previous period. Do not add index points, percentage points and jobs. A positive inflation surprise is also not automatically favourable economic news or a currency instruction.

Faust, Rogers, Wang and Wright’s official October 2003 research abstract examines high-frequency macro surprises and changing effects over time. It does not establish that one report class always causes the same move today. Keep “the quote rose after publication” separate from “this release alone caused the full rise”. Price measurement is also distinct from expectations; see trading around inflation releases.

Historical-period data can still contain new information through surprises, revisions and components. A release's classification alone does not determine how much is already priced or whether its reaction will be small. Compare the specific information with prior expectations and simultaneous releases.

A revision can reverse apparent lead

Today’s historical series can differ from what was reported earlier. Our model is not CPI or OECD dating: use only the school rule “first month below its previous value”. B is unchanged; A’s history is later revised.

Same decline rule, different A vintage
All three series are invented
MonthA: first releasesA: revised historyB: unchanged
January100100100
February99101101
March98102100
April979999
First declineFebruaryAprilMarch

Initial A leads B by one month; revised A lags B by one month. Only A’s past values changed, not the rule or B. The first column contains each month’s original release, not a complete January file. The later reconstruction was not knowable beforehand. This does not measure a real indicator’s effectiveness.

Predefine the country, exact series, horizon, availability dates and a condition that would refute the hypothesis. Do not change the comparison window after a contrary result. Even a correct macro hypothesis does not set an entry price, holding period or loss budget. For a strategy test, use contemporaneously available quotes, executable entries/exits and costs; cycle agreement alone does not establish profitable FX trading.

What to do at the next releases

  1. Annotate the calendar carefully. Record the observation period, publication date, revision status and any justified business-cycle classification. Do not mark a release as useless solely because its reference period is in the past.
  2. Match the references. Record the series and forecast provider before release. In the invented orders example, 49 − 48 = +1 point surprise but 49 − 50 = −1 period change. A new consensus of 49.5 changes surprise to −0.5 without changing period change or the separate −30,000-job revision.
  3. Keep the decision-time vintage. If A’s history is later revised, retain both records and the publication timestamp. A backtest of today’s reconstructed history answers a different question from a test of information available at the time.
  4. Log what was knowable at each date. Use contemporaneous data vintages and predefined windows to compare hypotheses and reactions. A few observed cycles do not establish that only leading data shifts expectations or that lagging releases merely confirm information already known.