An economic indicator is a statistical measure used to assess some aspect of economic conditions, such as output, employment, prices or expectations. For a quant, its usefulness depends on what it measures, when it becomes available and whether it adds information beyond what is already known. A good description of last quarter can be a poor input to a decision made before its release.
Indicators can inform business-cycle analysis, forecasts and market scenarios. They are measurements with scope and uncertainty, rather than direct observations of a single hidden variable called “the economy.”
Three dates belong to one observation
Consider a fictional monthly real-activity index, seasonally adjusted, with its 2025 average set to 100. It is a volume measure, not a currency amount or an annualised growth rate. Its January 2026 observation is first published as 100 on 12 February, then revised to 103 on 12 March.
| Field | Meaning in this example |
|---|---|
| Observation period | January 2026: the activity being measured |
| Release date | 12 February 2026: the first estimate becomes public |
| Vintage | The version of the series available as of a specified date; the 12 March vintage includes a revised January value |
A decision on 20 February can use 100. It cannot use 103, even if a download made today places 103 next to the January observation date. The revision is +3 index points, not new activity occurring in March. ALFRED's real-time-period documentation makes the distinction between today's information about the past and information known at an earlier date.
Fictional monthly real-activity volume index, 2025 average = 100; seasonally adjusted, not annualised. Calendar positions show publication order. Both reported values describe January, and the March revision is unavailable at the February decision.
This is a look-ahead bias problem in a historical model test. Keep observation period and availability time separately, and preserve revisions rather than overwriting history. For an intraday strategy, release date alone is insufficient: check the release timestamp, time zone and the time your system actually received the data. A missing release is missing information, not a zero observation.
A small offline example selects the latest eligible release for this one observation:
from datetime import date
# Fictional releases for the same January 2026 observation.
releases = [
(date(2026, 2, 12), 100),
(date(2026, 3, 12), 103),
]
decision_date = date(2026, 2, 20)
available = [row for row in releases if row[0] <= decision_date]
latest = max(available, key=lambda row: row[0]) if available else None
value = latest[1] if latest else None
print(value)
# 100The date-only convention assumes the decision occurs after publication on an eligible date. A production dataset also needs series identifiers, observation periods, missing-value handling and timestamp precision.
Leading relative to what?
Leading, coincident and lagging describe the usual timing of an indicator relative to a specified reference measure or cycle: it tends to turn before, around the same time as, or after that reference. These are empirical relationships conditional on country, sample, transformation and horizon. They do not guarantee the next turning point or establish causation.
For example, the OECD Composite Leading Indicators target fluctuations around trend. Their signals are qualitative; a fall is not a forecast of an equally sized fall in GDP, and a growth-cycle slowdown is not automatically an economy-wide recession.
Economic timing and publication timing also differ. An indicator can track current activity closely yet arrive late. Another may provide an early signal but be noisy. Evaluate predictive usefulness with the vintages and release schedule that a forecaster could actually have used.
Hard data, surveys and the information in a release
Hard data usually refers to quantitative measures of realised activity, such as payroll employment or sales. Soft data usually refers to qualitative assessments, intentions or expectations, including consumer confidence. The New York Fed explains how its regional surveys complement later activity measurements with timely reports of conditions and expectations.
The distinction is about what is measured, not simply how it is collected. Hard data can come from surveys: the US Current Employment Statistics survey produces monthly employment, hours and earnings estimates. Neither label removes sampling error, nonresponse, coverage gaps or revision risk. Survey sentiment is not realised spending; an index of respondents reporting improvement is not automatically an output growth rate.
For market research, distinguish the level from the surprise. A release surprise compares the announced value with an expectation recorded before the announcement, using the same series definition and units. If the fictional first release of 100 had been expected to be 102, the surprise would be −2 index points. Its later revision to 103 does not retroactively turn that first announcement into a positive surprise.
The New York Fed's research on economic news studies this unexpected component. A negative surprise need not produce a particular asset return: revisions, other simultaneous news, policy expectations and the prevailing regime can all affect interpretation. A consensus forecast is also only a proxy for the expectations reflected in prices.
Read the series definition before the headline
Before combining indicators, record the jurisdiction and population covered, frequency, units, reference period, seasonal-adjustment status and release vintage. For monetary amounts, distinguish nominal values from inflation-adjusted volumes. For growth, distinguish month-on-month, year-on-year and annualised rates. For indexes, check the base and construction; index points and percentage changes answer different questions.
Seasonal adjustment removes estimated recurring seasonal influences so that adjacent periods are easier to compare. It does not remove all noise or isolate a causal effect. Adjustment can itself change historical observations: US seasonally adjusted CPI indexes can be revised when seasonal factors are recalculated, as the BLS methodology page explains.
Use several economically distinct indicators to test a story, while checking whether they repeat the same underlying information. A correlation or an isolated threshold crossing does not prove a mechanism, a recession or a profitable trading rule. Evaluate forecasts on later observations, respect information availability and report the failures alongside the useful signals.
For an advanced application of the timing discipline, the Real-time macro forecasting capstone compares forecasts of US industrial production using historical vintages and separates first-release from revised targets. Membership and sign-in are required. Its teaching concerns a particular forecasting study; it does not establish a universal ranking of indicators.
Sources
- The Conference Board: Description of Components — historical timing classification; no current component weights or values used.
- St. Louis Fed: FRED API Real-Time Periods — information known at historical dates.
- OECD: Composite Leading Indicators — reference cycle, qualitative signals and interpretation.
- New York Fed: Tracking the Economy in Real-Time Through Regional Business Surveys — hard and soft data and survey timeliness.
- BLS: CES Frequently Asked Questions — survey-based measurement of employment, hours and earnings.
- New York Fed: How Economic News Moves Markets — announcement surprises; a historical study, not a current trading rule.
- BLS: Seasonal Adjustment and Seasonal Adjustment in the CPI — adjustment and revision conventions.