Measuring inflation expectations: How stable are density forecasts?

Intervention: Eliciting inflation density forecasts using different scales.

2022-12-16

2022-12-23

(i) Individual density forecasts of the 12-month ahead inflation rate [Q1]
(ii) Individual estimate if 12 months from now the inflation rate will be positive or negative (deflation) [Q3]
(iii) Individual point forecast of the 12-month ahead inflation rate [Q5]

(i) Participants distribute probabilities on a scale of possible inflation rates with 8-10 bins (depending on treatment). Probabilities need to add up to 100.
(ii) Participants are asked to state whether they think inflation or deflation is more likely occur 12 months from now.
(iii) Participants are asked about a point estimation depending on their answer in (ii). If they answered “deflation”, they will be asked for a point forecast for deflation and similarly for inflation.

(i) Participants certainty regarding their answers to [Q1, Q3, Q5]. [Q2, Q4, Q6]
(ii) Demographics (age, gender, education, US state of residence) [Questionnaire]
(iii) Financial literacy [Questionnaire]
(iv) Knowledge of inflation target [Questionnaire]

(i) Ordinal measurement of certainty (six-point Likert scale).
(ii) Answers will be used as control variables.
(iii) We use the three multiple-choice questions from Lusardi and Mitchell (2011).
(iv) Multiple-choice question that tests if participants know the inflation target of the Fed.

Participants in the survey will be asked about their density forecast for the inflation 12 months from now [Q1]. The question for this density forecast is directly taken from the NY Fed. After the density forecast, participants will state whether they expect inflation or deflation 12 months from now [Q3] and then give their inflation expectations again in a point forecast [Q5]. Both questions are also taken from the NY Fed SCE. This will allow us to check if the predictions of these questions align. After each of these questions, participants state how certain they are regarding their answer on a six-item Likert scale [Q2, Q4, Q6]. After this, the participants will answer a questionnaire including control and demographic questions [Questionnaire].

Participants in the main part of the experiment will take part in either the Baseline or one of eight possible treatments. In the Baseline participants will answer the density forecast as it is currently used by the NY Fed and the Bundesbank. The Baseline will serve as a benchmark for our other treatments.

Participants in the survey will be asked about their density forecast for the inflation 12 months from now [Q1]. The question for this density forecast is directly taken from the NY Fed. After the density forecast, participants will state whether they expect inflation or deflation 12 months from now [Q3] and then give their inflation expectations again in a point forecast [Q5]. Both questions are also taken from the NY Fed SCE. This will allow us to check if the predictions of these questions align. After each of these questions, participants state how certain they are regarding their answer on a six-item Likert scale [Q2, Q4, Q6]. After this, the participants will answer a questionnaire including control and demographic questions [Questionnaire].

Participants in the main part of the experiment will take part in either the Baseline or one of eight possible treatments. In the Baseline participants will answer the density forecast as it is currently used by the NY Fed and the Bundesbank. The Baseline will serve as a benchmark for our other treatments. Our treatment interventions fall into two broad categories: Shifting and Centralization.
In the six Shifting treatments, the scale will be shifted to either the right or the left. Both shifts are implemented in different proportions, to test how larger shifts affect the answers. Specifically, these six treatments include the four Shifting treatments from 2021, which had encompassed to treatments shifting the scale towards inflation and two that shifted the scale towards deflation. As the current US inflation rate has steadily increased since 2021, we include two additional Shifting treatments that further shift the scale towards inflation.
The second set of treatments is designated Centralization. A feature of the NY Fed density forecast is that bins closer to the midpoint are smaller compared to others. This can complicate some computations, such as deriving a continuous belief distribution. Our 2021 Centralization treatments vary the number of bins around the midpoint. In the 2022 data collection we include one of these treatments, which has completely even bin sizes. Additionally, we include the same even-width Centralization treatment but shifted towards higher inflation rates. This allows us to further test how well the even-width Centralization treatment performs for the currently higher inflation rates.

In order to recruit a representative sample of the US population, we will run the survey on the online research platform Prolific. We will recruit two samples: First, a sample of 100 new participants for each of the nine treatments. Second, we will try to recruit as many participants as possible from the 2021 data collection. These participants will be assigned to the same treatment as in 2021. Data for both samples will be collected in parallel, to ensure a sufficient time window to resample participants from 2021. We aim to gather data for at least half of the participants from 2021. New participants will be assigned one of the nine treatments iteratively when starting the study. This way, we make sure that any differences we obtain are not driven by timing but instead reflect a genuine effect of our treatment intervention. Participants will earn a fixed fee of £1 for their participation.

Randomization of treatments done by the experimental software.

Individual

No

1200 Prolific participants

1200 Prolific participants

100 new participants per one of the 9 treatments (900 total) plus 50+ participants per one of the 6 old treatments (up to 300 total) from 2021

Test if the mean between any two treatments differs. Test used is two-sample t-test, assuming equal variances in both treatment groups. For a group size of 100, power calculation (alpha = 0.05, beta = 0.8) informs us that the minimum detectable effect size would be a difference in means of 0.398.