Frictions in Forecasting: Experimental Evidence from Professional Forecasters

How to cite

Pavlova, Lora; Rostam-Afschar, Davud; Hack, Lukas. 2025. "Frictions in Forecasting: Experimental Evidence from Professional Forecasters." AEA RCT Registry. AEARCTR-0017013. https://doi.org/10.1257/rct.17013-1.0

Download: BibTeX · RIS — for reference managers without a browser translator. Zotero and most others pick up this page's citation metadata directly.

Lora Pavlova* , Davud Rostam-Afschar , Lukas Hack

* Primary investigator

Registry ID
AEARCTR-0017013
DOI
https://doi.org/10.1257/rct.17013-1.0
First registered
Last updated
Status
In development
Trial dates
2025-11-03 – 2027-11-10
Countries
Private
JEL codes
C9, D83, E37
Registry record
http://www.socialscienceregistry.org/trials/17013

Abstract

We study how information about the distribution of peers’ forecasts affects professional forecasters’ own predictions. Using the Finanzmarkttest (FMT) survey of financial market experts, conducted monthly by ZEW Mannheim, we implement an information provision experiment. Participants provide quarterly GDP growth forecasts for Germany. A randomly selected subset (treatment group) receives information about the actual share of forecasters whose previous forecast was lower than their own. This allows us to causally identify whether forecasters adjust their expectations in response to information about the forecast distribution, revealing motives in professional forecasting that go beyond pure forecast accuracy considerations.

Intervention

In our experiment, we provide a random subset of all FMT professional forecasters with information about the distribution of survey forecasts. This helps us to causally identify whether the ``forecasts of others'' matter for the own prediction of a given forecaster.

Experimental design

Treatment group: Receives feedback on the true share of participants whose prior forecast (from an earlier wave) was lower than their own prior forecasts (“distribution information”). Control group and treatment group: Receives their own previous forecast and the consensus (average) forecast, but no information about the distribution. Both groups report priors and posteriors about GDP growth and beliefs about the consensus forecast.

Design details

Randomization method
Random assignment of participants to treatment and control groups within the November 2025 FMT survey wave. Randomization will be performed in survey software prior to fielding and will not be visible to participants.
Randomization unit
Individual forecaster
Clusters
Expected participants: ~120–150
Observations
Expected participants: ~120–150
Arms
Expected participants per treatment arm: ~60-75
Primary outcomes
quarter-on-quarter real GDP growth forecasts, belief about the consensus, belief about the consensus gap, belief about the relative position in the distribution, treatment intensity, forecast revision
Secondary outcomes
Responses in subsequent FMT waves (forecast persistence), Self-reported forecast methodology (model-based vs judgment-based), Comparison to expert predictions

Cited by

No citation data available for this registration yet.

Keywords