STEP 01
Define the information set
Specify survey dates, economies, forecast horizons and indicator conventions before testing.
A CONSENSUS USE CASE
Test the expectation, not the benefit of hindsight.
Quantitative research needs the information set that existed at each decision date. Consensus survey vintages preserve the expectations of professional private sector forecasters as they were recorded at the time, rather than retrofitting today’s knowledge into yesterday’s models.
THE QUESTION
“Can we test a signal using only what forecasters knew at the time?”
Quants & Data TeamsCONSENSUS ECONOMICS
01 / THE SITUATION
Here’s the decision.
The research question
A systematic team is testing whether changing inflation expectations contain useful information.
The signal looks promising on a modern dataset, but the researchers need to know whether it would have been observable in real time. They want the timestamped survey sequence, the dispersion around the consensus and a defensible treatment of publication timing.
THE OBJECTIVE
Build a reproducible research panel of forecast vintages and test the hypothesis without look-ahead bias.
02 / THE WORKFLOWFrom raw question
From raw question
to useful evidence.
Build features from forecasts available at each release date; test revisions and dispersion without introducing information from later surveys.
STEP 02
Load historical vintages
Use the preserved forecast observations available in each survey round, not final or reconstructed values.
STEP 03
Engineer expectation features
Calculate changes in consensus, distribution measures and cross-market comparisons where coverage permits.
STEP 04
Validate the backtest
Apply publication lags, out-of-sample splits and robustness checks before drawing a conclusion.
03 / A PRACTICAL EXAMPLEA worked example.
A worked example.
The numbers in context.
Turn successive survey releases into a testable inflation-revision feature.
A systematic researcher holds the target period fixed and records only the forecasts released before each hypothetical trading decision.
WORKING EXAMPLEIllustrative example
Swipe sideways to view the complete example table
| Data available after | Forecast for same target | Revision vs prior release |
|---|---|---|
| January publication | 3.21% | — |
| February publication | 3.54% | +0.33 percentage points |
| March publication | 3.36% | −0.18 percentage points |
THE PRACTICAL DELIVERABLE
Research-ready panel: release timestamp, target period, consensus, revision, coverage flag and test window.
SUITABLE STARTING POINTExplore historical Excel data
Preserved survey vintages, consistent spreadsheet structure and individual observations where covered.
GO FURTHER / APPLY THE FULL DISTRIBUTION
The range behind
The range behind
the average.
Create point-in-time factors from revisions, disagreement and individual forecast behaviour. Research the relationship; do not imply that every revision is a tradable signal.
FORECAST DISTRIBUTIONLatest fixed-target inflation forecast after March release
Consensus mean3.36%
Low2.57%
High4.18%
Standard deviation0.41 pp
BEHIND THE AVERAGE
Selected individual views: 3.12% and 3.73%. The corresponding cross-forecaster variance is about 0.17 pp² (standard deviation squared). Match contributor identities across releases and handle missing observations.
LOW · 2.57%HIGH · 4.18%
Consensus mean Selected individual forecasts
A PRACTICAL SCENARIO OR RESEARCH TEST
Illustrative exampleDesign a signal with a dispersion-aware control
Swipe sideways to review the scenario table
| Research feature | Value | Possible use in a model |
|---|---|---|
| Latest mean revision | −0.18 pp | Test revisions from 3.54% to 3.36% against subsequent asset returns. |
| High-forecast sensitivity | 4.18% | Stress-test a model with the upper submitted inflation view. |
| Disagreement filter | 0.41 pp | Flag survey releases where cross-sectional disagreement is elevated. |
Use each release only after its actual availability time, keep the forecast target fixed and separate forecaster disagreement from statistical forecast uncertainty.
OTHER APPLICATIONS FOR THIS AUDIENCE
Further applications.
Forecast-surprise backtest
Test whether expectation revisions anticipate asset-price moves.
USE THE DATA
Join historical survey vintages to returns only after each release was available; estimate out of sample.
OUTPUT
A vintage-safe factor panel with lagged signals and test results.
Forecaster breadth indicator
Ask whether changes are concentrated among outliers or shared across institutions.
USE THE DATA
Calculate the share of matched forecasters revising upward, cross-sectional SD and mean revision.
OUTPUT
A revision breadth and dispersion monitoring feed.
04 / WHAT YOU TAKE AWAYA practical result.
A practical result.
Ready to review.
Build an output that can be explained, challenged and updated as forecasts change.
How the consensus is calculatedEXAMPLE DELIVERABLE↗
Illustrative research deliverableA point-in-time factor research dataset
Versioned forecast features linked to their available-at dates, accompanied by a methodology log and reproducible tests.
WHAT IT CONTAINS
Adapt the example to your own data and decision process.✓Survey-vintage feature table
✓Release timing and data lineage
✓Out-of-sample results and limitations
05 / THE CONSENSUS ADVANTAGEWhy the data
Why the data
matter here.
Match the forecast inputs to the decision. The average, the distribution and the historical record serve different purposes.
RELEVANT FORECAST INPUTS
01
Historical survey databases
02
Consensus revisions and forecast dispersion
03
Individual panel estimates where supplied
WHY CONSENSUS ECONOMICS
Real-time expectations, not reconstructed history.
Survey-vintage records support point-in-time research; revisions to the mean and disagreement across individual forecasters provide separate testable features. Stable survey dates and release timing matter as much as the figures.
1,000+professional forecasters1989first survey