Does the Messenger of Central Bank Communication Matter? Effects on Belief Updating.

At the core of this study is an information experiment (RCT) with participants from different EA nationalities (DE, FR, IT, ES) who also reside in the respective countries. After being presented with historic 10 periods of annual inflation, participants forecast the next period of Euro area inflation in multiple different forecasting tasks. After submitting a first forecast, participants are given professional inflation forecasts and additional information that helps participants with the forecasting task. The treatment consists of a randomly varying source (i.e. the messenger) of additional information. Messenger treatments vary along nationality and institutional dimensions of the information source. After this, participants submit an updated inflation forecast. The outcome of interest is the extent to which participants update their inflation expectations in a Bayesian manner, particularly the extent to which they update towards the signal of the provided information/forecast (representing central bank communication), and whether/how this varies across messengers.

2023-02-01

2024-03-31

Signal uptake;
Similarly: Extent of Bayesian Updating

Signal uptake: measured as the extent to which the signal is used participants' inflation forecast revisions (i.e. in their posteriors, after treatment);
Extent of Bayesian Updating: i.e., how rational/close to the rational posterior the updated inflation forecasts are

1) Attention to provided analyses (i.e. central bank communication) of the messengers.
2) Exposure and trust of institutions and (examples of) policymakers that are similar to the generic treatments in the trial.
3) Other differences: across nationalities, levels of monetary policy expertise
4) Effects on uncertainty

1)+2) To better understand potential mechanisms to heterogeneous belief updating, participants' attention to information, as well as trust and exposure to similar policymakers are being elicited. Here, exposure is how well institutions and policymakers are known, and how regularly news about them are followed.
3) Sample splits by participant nationality (requires final sample to be large enough to ensure enough power), and level of self-reported extent of following key monetary policy announcements (to compare across financial expertise levels, and compare to some observational results)
4) Both, within-subject change in forecast uncertainty before and after information provision, and between-subject "disagreement" before and after information provision.

An RCT run via Prolific to test how individuals update inflation expectations, given information by different messengers.

An RCT run via Prolific, where the information treatment is within-subjects, but there is a nationality component (of participants) that is by definition between-subjects (4 participant nationalities: French, German, Spanish and Italian). At the core of the RCT is that participants from these different nationalities are incentivised to accurately perform forecasting tasks. Participants make point and range forecasts and will be rewarded for the accuracy of those (incentivised).
In each of 6 forecasting task, participants are first presented with historic 10 periods of annual inflation. Participants then forecast the next period of Euro area inflation. After submitting a first forecast, participants are given professional inflation forecasts and additional information (the treatment). There are a total of six within-subject treatments regarding the messengers that every participant faces:

1. Nationality ingroup (messenger is expert of same nationality as participant)
2. Nationality outgroup (messenger is expert of random other nationality as participant; note that other nationality is always one of the big 4 nationalities)
3. ECB (messenger is expert from ECB)
4. NCB (messenger is expert from NCB)
5. ECB ingroup (messenger is expert from ECB who is of same nationality as participant)
6. ECB outgroup (messenger is expert from ECB who is of random other nationality as participant; same as in 2.)

After the information treatment, participants get the opportunity to revise their point and range forecasts.

The idea here is to see how Bayesian participants are, when they update their forecasts after the information treatment. This design allows to test, whether there are differences in signal uptake when updating expectations, depending on who the messengers are and/or what the participant nationality is. In particular, it can be assessed whether there exists an "ingroup effect", where participants who share their nationality with the messenger update differently to the information than participants in the "outgroup". In addition, this tests differences in updating of ECB and NCB (national central bank) information. As well as the interacted version of ingroup vs. outgroup representatives of the ECB, holding the institution constant.
Thereby, insights for 2 potential policy recommendations can be derived: First, should the existing network of NCBs be used more for central bank communication? And second, should board member of other nationalities in the ECB be used more for central bank communication?

In addition to the experimental intervention, this survey will also collect information on trust and knowledge about the monetary policy institutions and some policymakers to make inferences about potential mechanisms.

Randomisation done by the computer (via a randomisation algorithm)

Individual: Subjects participated individually. We deployed our experiment online via Prolific.

No

n/a (4 nationalities)

at least 100 participants per nationality (400 in total), ideally at least 200 per nationality given extra funding availability (800 in total)

at least 400 participants per (nationality-pooled) treatments (within-subject design);
splitting by nationality (participant or outgroup nationality) this will fall accordingly