The effects of interest rate increases on consumers' inflation expectations: The roles of informedness and compliance

Knotek, Edward S.; Mitchell, James; Pedemonte, Mathieu; Shiroff, Taylor Working Paper The effects of interest rate increases on consumers' inflation expectations: The roles of informedness and compliance IDB Working Paper Series, No. IDB-WP-1641 Provided in Cooperation with: Inter-American Development Bank (IDB), Washington, DC Suggested Citation: Knotek, Edward S.; Mitchell, James; Pedemonte, Mathieu; Shiroff, Taylor (2024) : The effects of interest rate increases on consumers' inflation expectations: The roles of informedness and compliance, IDB Working Paper Series, No. IDB-WP-1641, Inter-American Development Bank (IDB), Washington, DC, https://doi.org/10.18235/0013184 This Version is available at: https://hdl.handle.net/10419/309115 Standard-Nutzungsbedingungen: Die Dokumente auf EconStor dürfen zu eigenen wissenschaftlichen Zwecken und zum Privatgebrauch gespeichert und kopiert werden. Sie dürfen die Dokumente nicht für öffentliche oder kommerzielle Zwecke vervielfältigen, öffentlich ausstellen, öffentlich zugänglich machen, vertreiben oder anderweitig nutzen. Sofern die Verfasser die Dokumente unter Open-Content-Lizenzen (insbesondere CC-Lizenzen) zur Verfügung gestellt haben sollten, gelten abweichend von diesen Nutzungsbedingungen die in der dort genannten Lizenz gewährten Nutzungsrechte. Terms of use: Documents in EconStor may be saved and copied for your personal and scholarly purposes. You are not to copy documents for public or commercial purposes, to exhibit the documents publicly, to make them publicly available on the internet, or to distribute or otherwise use the documents in public. If the documents have been made available under an Open Content Licence (especially Creative Commons Licences), you may exercise further usage rights as specified in the indicated licence. https://creativecommons.org/licenses/by/3.0/igo/ The Effects of Interest Rate Increases on Consumers’ Inflation Expectations: The Roles of Informedness and Compliance Edward S. Knotek II James Mitchell Mathieu Pedemonte Taylor Shiroff WORKING PAPER No IDB-WP-1641 InterA merican Development Bank Department of Research and Chief Economist September 2024 * Federal Reserve Bank of Cleveland ** Inter-American Development Bank The Effects of Interest Rate Increases on Consumers’ Inflation Expectations: The Roles of Informedness and Compliance Edward S. Knotek II* James Mitchell* Mathieu Pedemonte** Taylor Shiroff* InterA merican Development Bank Department of Research and Chief Economist September 2024 Cataloging-in-Publication data provided by the Inter-American Development Bank Felipe Herrera Library The effects of interest rate increases on consumers’ inflation expectations: the roles of informedness and compliance / Edward S. Knotek II, James Mitchell, Mathieu Pedemonte, Taylor Shiroff. p. cm. — (IDB Working Paper Series; 1641) Includes bibliographical references. 1. Monetary policy-Mathematical models-United States. 2. Inflation (Finance)- Mathematical models-United States. 3. Communication policy-Mathematical models-United States. 4. Surveys-United States. I. Knotek, Edward S. II. Mitchell, James. III. Pedemonte, Mathieu. IV. Shiroff, Taylor. V. Inter-American Development Bank. Department of Research and Chief Economist. VI. Series. IDB-WP-1641 http://www.iadb.org Copyright © 2024 Inter-American Development Bank ("IDB"). This work is subject to a Creative Commons license CC BY 3.0 IGO (https://creativecommons.org/licenses/by/3.0/igo/legalcode). The terms and conditions indicated in the URL link must be met and the respective recognition must be granted to the IDB. Further to section 8 of the above license, any mediation relating to disputes arising under such license shall be conducted in accordance with the WIPO Mediation Rules. Any dispute related to the use of the works of the IDB that cannot be settled amicably shall be submitted to arbitration pursuant to the United Nations Commission on International Trade Law (UNCITRAL) rules. The use of the IDB's name for any purpose other than for attribution, and the use of IDB's logo shall be subject to a separate written license agreement between the IDB and the user and is not authorized as part of this license. Note that the URL link includes terms and conditions that are an integral part of this license. The opinions expressed in this work are those of the authors and do not necessarily reflect the views of the Inter-American Development Bank, its Board of Directors, or the countries they represent. Abstract∗ We study how monetary policy communications associated with increasing the federal funds rate causally affect consumers’ inflation expectations in real time. In a large-scale, multi-wave randomized controlled trial (RCT), we find weak evidence that communicating these policy changes lowers consumers’ mediumterm inflation expectations on average. However, information differs systematically across demographic groups in terms of ex ante informedness about monetary policy and ex post compliance with the information treatment. Monetary policy communications have a much stronger effect on the subset of consumers who had not previously heard news about monetary policy and who take sufficient time to read the treatment. Our findings show that, in an inflationary environment, these consumers expect that raising interest rates will lower inflation. More generally, our results emphasize the importance of measuring both respondents’ information sets and their compliance with treatment when using RCTs in empirical macroeconomics to better understand the real-world implications of monetary policy communications. JEL classifications: E31, E52, E58 Keywords: Expectations formation, Policy communication, Monetary policy, Inflation, Surveys ∗ Knotek: Federal Reserve Bank of Cleveland, Email: [email protected]; Mitchell: Federal Reserve Bank of Cleveland, Email: jam[email protected]rg; Pedemonte: Inter-American Development Bank, Email: [email protected]; Shiroff: Federal Reserve Bank of Cleveland, Email: [email protected]. We thank Oli Coibion, Francesco D’Acunto, Ricardo Reis, Michael Weber, and Mirko Wiederholt for helpful comments, as well as conference and seminar participants at the Federal Reserve Bank of Cleveland, the Oxford Communication Workshop, the Bank of Canada, the Cleveland Fed’s Central Bank Communications: Theory and Practice conference, the Second Paris Conference on the Macroeconomics of Expectations, ISF 2024, Venice Summer Institute, and the CEBRA Annual Meeting 2024. The RCT was registered as AER RCT ID AEARCTR-0009172 (Knotek 2022). The views expressed here are solely those of the authors and do not necessarily reflect the views of the Federal Reserve Bank of Cleveland or the Board of Governors of the Federal Reserve System. 2 1. Introduction Monetary policymakers on the Federal Open Market Committee (FOMC) raised the federal funds rate at a relatively rapid pace during 2022 via a sequence of rate increases that began on March 16, 2022. The primary reason for this rapid tightening of policy was mounting concern that inflation was proving long-lived, with the risk of inflation expectations becoming unanchored. This led FOMC participants to talk openly and in advance of FOMC meetings about the pivot to a tighter monetary policy stance and the need for higher interest rates. While these communications were reflected in rising market-based expectations of the future federal funds rate, it is not clear whether the consumers comprising the “general public” were paying attention or, if they were, how they were reacting to these communications.1 This paper estimates the causal effects of communicating interest rate increases on consumers’ inflation expectations using five waves of a randomized controlled trial (RCT) conducted via an online survey. We focus on communicating to consumers in very simple terms the federal funds rate increases of 2022 and assessing the impact on their inflation expectations. We do so by conducting in real time a specially designed set of RCTs immediately following the March, May, June, July, and September 2022 FOMC meetings. Each of these meetings resulted in increases to the federal funds rate target. Our RCT-based estimates of the average causal effect of communicating these hikes on households’ inflation expectations have a wide range. We find that providing consumers with information about the latest interest rate hike reduced their expectations for inflation over the next five years, on average, between 0.17 and 2.18 percentage points, depending on the precise information transmitted. While these treatment effects are statistically significant on average, we raise two difficulties about their interpretation that, as explained below, much of the wider RCT-based literature in empirical macroeconomics also faces. First, our RCTs treat consumers with information that is already publicly available and, moreover, they do so in an environment where monetary policy news was salient given high inflation readings. We provide novel evidence that consumers were indeed more informed about monetary policy in 2022 than they had been in prior years.2 As a result, some treated individuals likely already had the treatment in their ex ante 1 See https://www.cmegroup.com/insights/economic-research/2022/fed-rate-hikes-expectations-and-reality.html. 2 This finding that consumers were more aware of monetary policy in 2022 than in prior years complements other work that has documented rising awareness of and attention paid to inflation in particular during this episode, such as Pfäuti (2024) and Bracha and Tang (2024). 3 information sets, a phenomenon also described in Weber et al. (2024) and Mackowiak and Wiederholt (2024). Second, while our RCTs explicitly communicate information to the treatment groups, it is hard to know how much of this information is really read and/or absorbed by respondents within the survey. We make the analogy to medicine and a physician prescribing a pill to a patient: the patient is only truly treated after swallowing the pill, not by having been prescribed or even given the pill.3 Voluntary compliance with the treatment can only be confirmed ex post.4 Methodologically, we introduce to empirical macroeconomic studies that use RCTs the importance of ex ante informedness and ex post compliance. We propose a novel, easy, and accessible way to control for both the ex ante information set of respondents and how compliant they are with the information treatment ex post in a manner that is feasible in most online surveys. In our application, we capture informedness by asking households whether they had recently heard monetary news, and we measure voluntary compliance—rather than attempting to enforce compliance—through the time that respondents choose to spend reading the information treatment in the online survey. In this way, we proverbially do not force patients to take the pill, but we observe through a two-way mirror whether they choose to take the pill. Using econometric methods familiar to microeconomists when undertaking causal inference with non-experimental data, we obtain complier average treatment effects by upweighting (downweighting) control group respondents based on their predicted probability of complying (not complying) with the treatment and excluding treatment group non-compliers. After accounting for compliance with the treatment and controlling for consumers’ ex ante informedness in our regressions, we find that monetary policy communications about increases in interest rates have a statistically significant and economically meaningful negative impact on medium-term inflation expectations for the consumers who were previously uninformed about recent policy actions and compliant with the treatments. To the extent that the informed individuals proverbially swallowed the real-world monetary policy pill upon hearing monetary policy news prior to taking our survey, then they too may have reduced their inflation expectations at that time. However, 3 In the 1999 movie “The Matrix,” when Morpheus gives Neo the choice between taking the blue pill and staying blissfully unaware or taking the red pill and joining the movement to undermine the matrix, he watches as Neo takes the red pill and washes it down with a glass of water. 4 An alternative interpretation is that there is a key intensive margin to treatments that can vary across respondents and may not be completely random. Fuster et al. (2022) find that people with less uncertain prior beliefs are more likely to spend time reading the information treatment. 4 RCTs are unable to adequately capture their behaviors, because the information treatments are already part of their information set, as noted in Weber et al. (2024).5 Our empirical results contrast with the findings in Andre et al. (2022), who reported strong disagreement among consumers’ responses to hypothetical situations involving monetary policy communications based on survey responses from a low-inflation period in 2019. Our findings, coming from a high-inflation period when consumers may respond differently to information shocks, suggest that consumers had a better understanding of the objectives of monetary policy when given the right information—they understood the basic mechanism that higher interest rates would reduce inflation. In a theoretical rational inattention model, Mackowiak and Wiederholt (2024) show that individuals have more incentive to pay attention to the macroeconomy and to comply with information treatments when inflation is high. We explicitly measure both margins and show empirically that the treatment effect, consistent with their model, is higher for the uninformed and for those consumers who pay attention during the survey. By observing rather than enforcing compliance, we can analyze consumers’ choice to comply with the treatment. We document systematic demographic differences across individuals who are both less informed about monetary policy and more compliant with the treatment, and we show that these differences are relevant when interpreting average and heterogeneous effects of information treatments in the macroeconomics RCT literature. For example, Coibion, Gorodnichenko, and Weber (2022) find that communication directly by the FOMC is more effective in moving household expectations than indirect communication via the media. They also find considerable heterogeneity across respondents, with female respondents’ inflation expectations reacting more strongly to monetary policy information treatments. In our RCT, we find that women are more likely to spend a longer time reading the information treatments than men and to thus be considered as having complied with the treatment. This finding means that the gender differential in Coibion, Gorodnichenko, and Weber (2022) may simply reflect women paying more attention to the RCT treatment than men. We find support for this conjecture in our 5 At the same time, we do not believe that an event study around the FOMC meetings in our sample is appropriate. Increases in the federal funds rate during our sample were telegraphed in advance of the meetings—subject to some uncertainty about their size—and hence it is unclear when the “events” transpired. Furthermore, we provide ample evidence that there are systematic differences across individuals who are and who are not informed about monetary policy. Thus, comparing individuals who had not heard about monetary policy before an FOMC meeting and those who had heard about monetary policy after the meeting likely confounds a number of factors beyond the actual information provided by the meeting—i.e., selection into informedness is not random. 5 study: after controlling for informedness and compliance, the gender gap is no longer statistically significant. Our analysis also allows us to consider issues related to monetary policy awareness more broadly. Our findings suggest that central bank communications could be augmented to reach consumers who typically do not hear much monetary policy news. At the same time, our compliance results show that even when portions of the public are presented directly with monetary policy information, there is no guarantee that the information will be processed. Our focus on monetary policy awareness and compliance thus dovetails with the work of D’Acunto, Fuster, and Weber (2021), who find that the salience of female and minority representation on the FOMC affects how Fed information influences consumers’ expectations, particularly for selected demographic groups, potentially offering a pathway to enhance real-world compliance to engage with monetary policy news. Related Literature RCTs have gained prominence in empirical macroeconomics to understand expectations formation; e.g., see Armantier et al. (2016), Binder and Rodrigue (2018), and Coibion, Gorodnichenko, and Weber (2022). Using RCTs, Haldane and McMahon (2018) and Bholat et al. (2019) find simple relatable communications by the central bank to be more effective in influencing households’ expectations, thus motivating our focus on information treatments that are “short and sweet.” To test whether the information treatment has additional power to affect expectations if the rationale for the policy change is also communicated, additional treatment groups in our survey are given some narrative or “vignette” (see Andre et al., 2022, around the rate increase; e.g., by explaining that the FOMC is raising rates to reduce inflation. Our paper thus revisits the question of whether it is best to communicate targets or instruments, but it does so for the United States, orienting the information treatments around the actual federal funds rate decisions made by the FOMC through 2022 rather than hypotheticals. In our case, we provide simple treatments and compare that treatment to others that provide additional information. Our results build on a body of literature that studies communication as a central policy or lever of modern central banking. Much of this work has focused on whether and how central banks can communicate to consumers and firms (e.g., Blinder et al., 2008, Blinder et al., 2024, and Binder, 2017). Effective communication is important when central banks want to shape the 12 analysis, omitting them has little impact on our results.8 We do not use demographic controls in any of our regression models unless otherwise specified. Our sample is well-randomized and balanced, with only minor differences between the demographic make-up of any of the treatment groups and the control groups (see Appendix Table B2). A daily quota is used to ensure that the appropriate quantity of responses is collected even if respondents drop out of the survey before completing it. Respondents who fail the survey’s ReCAPTCHA check or who are otherwise flagged by Qualtrics as likely bots or spammers do not count toward the quota, nor are they included in our data set. We drop or otherwise alter as few responses from our sample as possible. The only respondents who are outright removed from our sample are those whose total survey completion times are either too short or too long. We drop respondents with total survey completion times in the 1st percentile (N=512 consumers) as well as those who took more than an hour to complete the survey (N=220), since the quality of responses from these respondents is typically poor, owing to rushing through the survey or simply forgetting about it. The median respondent took 14.5 minutes to complete the survey, and just under 90 percent completed it in less than 25 minutes. For the July and September survey waves, a timer was enabled to capture the time that respondents spent reading the screen with an information treatment, if relevant. Importantly, respondents choose how long to spend on the information treatment page in our survey; we do not force them to spend a specific amount of time on this page. To deal with outliers, we winsorize responses at the 2nd and 98th percentiles for all point expectations and use Huber-robust regressions. Twelve percent of our sample reported that they were expecting deflation over the next year; among these respondents, 59 percent anticipated deflation between zero and 10 percent. The raw median (across consumers and over time) response to the question on prior (that is, pre-treatment) year-ahead inflation expectations was 8 percent, with the 25th percentile at 4 percent and 75th percentile at 20 percent. The median posterior (that is, post-treatment) five-year average inflation expectation was 5 percent, with the 25th percentile at 2 percent and 75th percentile at 15 percent. After applying Huber weights, the median, 25th percentile, and 75th percentile prior 8 These weights are not used in the propensity score weighted regressions in Section 5. Rather than reweighting to match population distributions, these regressions reweight the control group to match distributions within the treatment group compliers/non-compliers. 13 expectation are 7 percent, 4 percent, and 11 percent, respectively, and 5 percent, 2 percent, and 10 percent for the posterior. 3. Treatment Effects of Different Communication Tools To estimate the average treatment effect, 𝛽𝛽𝑗𝑗, for treatments j=1,2,3,4, we run the following regression: 𝜋𝜋𝑖𝑖,𝑡𝑡 5𝑦𝑦− 𝜋𝜋𝑖𝑖,𝑡𝑡 1𝑦𝑦=𝛼𝛼+∑𝛽𝛽𝑗𝑗×𝐼𝐼�1 𝑖𝑖𝑖𝑖 𝑡𝑡𝑡𝑡𝑡𝑡𝑡𝑡𝑡𝑡𝑡𝑡𝑡𝑡𝑡𝑡𝑡𝑡𝑗𝑗= 1� 4 𝑗𝑗=1 +𝜀𝜀𝑖𝑖𝑡𝑡, (1) where 𝜋𝜋𝑖𝑖,𝑡𝑡 5𝑦𝑦 is the posterior, five-year inflation expectation for individual 𝑖𝑖 in wave 𝑡𝑡, 𝜋𝜋𝑖𝑖,𝑡𝑡 1𝑦𝑦 is the prior, 12-month inflation expectation, and 𝐼𝐼�1 𝑖𝑖𝑖𝑖 𝑡𝑡𝑡𝑡𝑡𝑡𝑡𝑡𝑡𝑡𝑡𝑡𝑡𝑡𝑡𝑡𝑡𝑡𝑗𝑗= 1� is a 0-1 dummy that takes a value of 1 if respondent 𝑖𝑖 received treatment 𝑗𝑗. If treatment 𝑗𝑗 is effective in changing the posterior inflation expectation relative to the prior inflation expectation, then 𝛽𝛽𝑗𝑗 will be different from zero, implying that the treatment induces a different response on average for individuals who receive the treatment relative to the control group. Note that 𝛽𝛽𝑗𝑗 measures the average treatment effect; changes to the distribution of responses that do not affect the average will not be captured in our regression. Consequently, a value of 𝛽𝛽𝑗𝑗< 0 signifies that treatment 𝑗𝑗 lowers respondents’ inflation expectations relative to their prior on average. To help filter outlier responses, we apply Huber weights obtained from a similar regression of five-year inflation expectations on one-year expectations, treatment indicators, and their interactions. Table 1 shows the estimation results for each wave separately, as well as a pooled version that includes a wave fixed effect. The time fixed effects are important, since they control for the common information that the treated and control groups had at the time. Thanks to the different waves, in that specification we can control and talk about a general effect over the full-sample period. Table 1 reports a negative average treatment effect for each treatment, implying that the average respondent reacts to the information treatment by reducing their inflation expectations. This negative effect is present even when consumers only receive information about the new federal funds rate (Treatment 1). This suggests that consumers may have some understanding of the mechanism behind monetary policy actions. This is particularly relevant for Treatment 1, as it does not include information about either the policy objective or inflation. The effect for Treatment 2, however, seems to be smaller in magnitude. This could be explained by an “information effect” 14 (see Nakamura and Steinsson, 2018) that can confound communications about interest rates: when consumers are told that the FOMC wants to reduce inflation with their actions, there is implicitly an acknowledgment of an inflationary problem, which might reduce the size of the treatment effect.9 Table 1. Posterior Minus Prior on Treatments (1) (2) (3) (4) (5) (6) March May June July Sept Pooled Treatment 1 -1.93*** -0.17 -0.38* -1.18*** -0.36 -1.57*** (0.33) (0.19) (0.21) (0.28) (0.32) (0.13) Treatment 2 -0.24 -0.18 -0.45** -0.84*** -1.59*** -0.35*** (0.20) (0.18) (0.21) (0.29) (0.36) (0.09) Treatment 3 -2.18*** -2.14*** (0.32) (0.31) Treatment 4 -0.72*** -0.66*** (0.21) (0.19) Placebo 1.09*** 0.36* (0.27) (0.18) Observations 7879 5204 4899 5994 5872 29833 Notes: Columns 1-5 contain no controls or fixed effects. Column 6 shows a pooled regression with treatment period fixed effects. * p < 0.10, ** p < 0.05, *** p < 0.01 There are some effects in Table 1 that are hard to rationalize. For example, the placebo seems to increase inflation expectations, depending on the specification. We re-evaluate these results below, once we have accounted for consumers’ ex ante informedness about monetary policy and their ex post compliance with the information treatment. 4. Informed and Uninformed Respondents: Expanding the Reach of Monetary Policy Table 1 showed that the information treatment was effective in reducing the inflation expectations of the average respondent. But since this information treatment is conveying public information at the time of the experiment, in principle our treatment should only affect respondents who are uninformed about recent monetary policy decisions. Informed respondents should already know the information communicated in the treatment and hence should be unaffected by it. In this 9 In Appendix C, we complement Table 1 by showing that the treatment effects are larger for those consumers with higher prior expectations for inflation. 15 section, we explore whether there are differences between the respondents who receive the treatment and are likely uninformed (the “local average treatment effect”) relative to the effect on all of the treated (the average treatment effect). We start by identifying the uninformed group and exploring its demographic, behavioral, and socio-economic characteristics. We do so by exploiting the fact that our survey includes a pretreatment question that asks respondents if mortgage rates had changed recently (and, if so, how) and whether they had heard news about monetary policy. As our treatments provide information about monetary policy decisions and changes in interest rates, these questions help identify which respondents were likely to have already known the information in the treatment. We find that 82.1 percent of respondents who had heard news about monetary policy were also aware that interest rates had risen recently, while only 45.7 percent of respondents who had not heard monetary policy news were aware of such an increase, and only 29.7 percent correctly described the change. Since the question on monetary policy news is indeed correlated with informedness about interest rates, we use as our indicator of informedness whether respondents had heard news.10 Figure 1 shows the evolution of the answers to this question over time. Figure 1 reveals that the share of respondents who indicate that they have heard news about monetary policy has been rising since 2020. While only between 25 and 30 percent of respondents heard news about monetary policy in October 2020, around 45 percent of respondents had heard news of monetary policy in October 2022. This is consistent with the rise in inflation and the consequent increased public discussion of monetary policy through 2022. In addition, we also see from Figure 1 that respondents are more likely to have heard news about monetary policy immediately after FOMC meetings, as represented by the vertical dashed lines in Figure 1.11 10 Using the question on interest rates also introduces additional complications. For example, since the possible initial responses were “No,” “Not Sure,” and “Yes,” we would need to consider the cases of truly uninformed (i.e., “No” respondents) and little-informed (i.e., “Not sure” respondents), under the tenuous assumption that these labels are accurate. Using these as alternative indicators of informedness does not significantly change our results, however. 11 Using a stratified random sample of the US public two days before and two days after the FOMC press conference, Lamla and Vinogradov (2019) also find that monetary policy announcements lead to an increase in the proportion of people who have heard monetary policy news. 16 Figure 1. Percentage of Respondents Hearing News about Monetary Policy We should expect smaller treatment effects for respondents who have already heard news about monetary policy. To test this hypothesis, we re-run the regressions in Table 1 but only on that subset of respondents who reported that they had not heard news about monetary policy. Note that by design (random) assignment to a treatment group is not correlated with having heard news about monetary policy (see Appendix Table B2).12 Table 2 confirms that the estimated treatment effects for respondents who have not heard news are indeed often larger than those reported in Table 1, likely because for these respondents the information contained in the treatment is more informative than it is for those who have already heard some monetary policy news. In particular, in July and September we observe stronger treatment effects in Table 2 than in Table 1. 12 Appendix Table E2 presents estimates showing how the probability that respondents had heard news about monetary policy varies with demographic factors and as a function of the time since the most recent FOMC meeting. 17 Table 2. Posterior Minus Prior on Treatments, Heard News = No (1) (2) (3) (4) (5) (6) March May June July Sept Pooled Treatment 1 -1.24** -0.20 -1.34* -2.03*** -0.93** -1.34*** (0.62) (0.59) (0.70) (0.45) (0.47) (0.26) Treatment 2 0.26 -0.01 0.73 -1.30*** -2.81*** -0.70*** (0.46) (0.56) (0.58) (0.43) (0.61) (0.23) Treatment 3 -1.26** -1.40*** (0.55) (0.50) Treatment 4 -0.75 -0.91** (0.49) (0.44) Placebo -0.61 0.01 (0.45) (0.41) Observations 4508 3075 2669 3503 3337 17091 Notes: Columns 1-5 contain no controls or fixed effects. Column 6 shows a pooled regression with treatment period fixed effects. * p < 0.10, ** p < 0.05, *** p < 0.01 This differential effect speaks to a general limitation of this type of RCT. The treatment effect of communicating information depends on how “informative” the information provided actually is. Respondents who are already fully informed have no need to update their priors posttreatment. In the next section, we propose a way of distinguishing between the informational content of the (specific) treatment using a measure of whether the respondent complied with the treatment. 5. Who Reads the Treatment? Measuring Compliance Assigning a respondent to a treatment group does not guarantee that the respondent complies with—and thus actually receives—the full extent of the information contained within the treatment. A parallel can be made with medical RCTs, in which patients may be randomly assigned the treatment, but they may not comply, for example, by not swallowing the pill prescribed by the physician. In the context of our survey of consumers, there could be many reasons for noncompliance. Consumers might be inattentive within the survey itself: they could be distracted, for example, and continue to the next question without having processed the treatment. Alternatively, they might not be willing to pay attention to the treatment at all, and they simply skip to the next question as fast as possible without digesting the information. While demographic characteristics and question-based assessments of reading and/or numerical literacy can help control for varying 18 levels of attentiveness and understanding of the treatment, they cannot assess who reads—and hence complies with—the treatment and who does not. Because non-compliance may be selfselected, it is “nonignorable,” since it undermines the random allocation into the treatment group required for unbiased estimation of the local average treatment effect (the complier average causal effect); see Imbens and Angrist (1994) and Angrist, Imbens, and Rubin (1996). Therefore, the estimates in Tables 1 and 2 offer unbiased estimates of the effect of assignment, the so-called “intention-to-treat,” not of the treatment itself. They reflect the efficacy of both the treatment and the compliance. In this section, we propose a novel tool to measure whether respondents read (that is, comply with) the treatment, which we then use to estimate the average treatment effect free from any confounding effects of non-compliance. This is facilitated by the fact that in the July and September waves we measured how much time respondents chose to spend on the treatment page.13 We use this measure as a proxy for whether respondents read and processed the treatment; we then separate the sample into attentive and inattentive respondents. Because the treatments vary in length and content, and across waves, separate rules are calculated for each treatment in each wave. We use a rule whereby a respondent is considered to have “read the treatment” if they took at least half of the average amount of time spent by respondents assigned to their treatment in their wave to read the treatment. This cutoff is 4.1 and 5.4 seconds for Treatments 1 and 2 in July, respectively, and 4.3, 5.9, and 2.5 seconds for Treatment 1, Treatment 2, and the placebo in September, respectively.14 Table 3 shows how demographic and other characteristics predict attentiveness to the treatment. While we cannot know from reading time alone whether a respondent truly read, processed, and correctly understood the information in the treatment, we can make reasonable assumptions that those with treatment reading times below these cutoffs are not likely to have done so. Even though our treatments provide succinct, “tweet-style” snippets of information, it is unreasonable to imagine that a respondent could read the entirety of the treatment in fewer than 4 or 5 seconds. 13 This type of information is easy to implement in online surveys. Fuster et al. (2022) use a similar counter to track the relevance of information. They measure time spent reporting the posterior. They find a positive relationship between uncertain priors and the time spent reporting the posterior. 14 Our results are not sensitive to using different cutoffs. Using an alternative standard of the 25th, 50th, or 75th percentile reading time produces similar results, with stronger treatment effects associated with longer reading time requirements. In general, our “half-the-average” rule requires a treatment reading time slightly below the median. For a plot of the empirical distribution of reading times, see Appendix Figure D1. For treatment effects by the reading time percentile used as the cutoff point, see Appendix Figure D3. 19 Our measure of compliance, therefore, is a conservative one. There may very well be respondents we consider compliant who in fact were not compliant, but there could only be very few respondents who we could be incorrectly considering non-compliant, given our reading cutoff times. Table 3. Likelihood of Reading the Treatment Full Sample No News Only (1) (2) (3) (4) Logit Odds Ratio OLS Logit Odds Ratio OLS Male 0.78*** (0.05) -0.05*** (0.01) 0.80*** (0.07) -0.05*** (0.02) Nonwhite 0.67*** (0.05) -0.08*** (0.02) 0.64*** (0.07) -0.09*** (0.02) Hispanic 0.63*** (0.07) -0.09*** (0.02) 0.64*** (0.09) -0.09*** (0.03) Primary Shopper 0.89 (0.10) -0.02 (0.02) 1.09 (0.14) 0.02 (0.02) Numerical Literacy 1.83*** (0.16) 0.11*** (0.02) 1.50*** (0.17) 0.08*** (0.02) Heard News 0.94 (0.06) -0.01 (0.01) Age: 36-50 1.89*** (0.17) 0.13*** (0.02) 2.17*** (0.24) 0.16*** (0.02) 51-65 4.71*** (0.45) 0.35*** (0.02) 4.48*** (0.53) 0.34*** (0.02) 66+ 13.38*** (1.56) 0.53*** (0.02) 10.18*** (1.59) 0.49*** (0.03) Income: $35,000-$49,999 1.16 (0.11) 0.03 (0.02) 1.23* (0.13) 0.04* (0.02) $50,000-$99,999 0.92 (0.07) -0.02 (0.02) 0.94 (0.09) -0.01 (0.02) $100,000 or more 0.77** (0.08) -0.05** (0.02) 0.65*** (0.09) -0.09*** (0.03) Education: Some College 1.63*** (0.13) 0.10*** (0.02) 1.46*** (0.14) 0.08*** (0.02) Bachelor's Degree 1.57*** (0.14) 0.09*** (0.02) 1.73*** (0.20) 0.11*** (0.02) Advanced Degree 1.47*** (0.17) 0.07*** (0.02) 1.83*** (0.30) 0.12*** (0.03) Political Party: Democrat 0.89 (0.07) -0.02 (0.01) 0.86 (0.08) -0.03 (0.02) Republican 1.08 (0.09) 0.02 (0.02) 1.05 (0.11) 0.01 (0.02) Constant 0.63** (0.11 0.39*** (0.04) 0.56*** (0.12) 0.37*** (0.04) Observations 9421 9421 5347 5347 Notes: Standard errors are reported in parentheses. Columns 1 and 3 report results from a logit model predicting compliance as a function of the listed variables as odds ratios, while columns 2 and 4 report results from OLS regressions of an otherwise identical model. * p < 0.10, ** p < 0.05, *** p < 0.01 We see from Table 3 that women tend to pay more attention to the treatment, even conditional on not having already heard news. In addition, white, non-Hispanic respondents are more likely to read the treatment. Older and more educated respondents tend to pay more attention, but income plays a negative role, if any. Political affiliation does not seem to affect compliance. 20 In total, we can explain a good proportion of compliance that does not come from idiosyncratic characteristics (for example, how tired the respondents are). This result is useful because it will allow us to characterize potential candidates who are likely to be attentive in the control group, and therefore to see if the effect of the treatment is stronger for consumers who pay attention. We find a close relationship between the predicted probabilities of compliance and the Huber weights used in Tables 1 and 2: respondents with Huber weights close to 1 are predicted to be much more likely to read the treatment than those downweighted in the Huber-robust regressions (see Appendix Figure D2). In other words, those who are predicted to be more likely to comply with their assigned treatment are far less likely to provide outlier responses, and viceversa. Accounting for compliance therefore has an added benefit in that doing so downweights outlier responses on the basis of the respondents’ behavior instead of using measures derived from the statistical properties of all (or a subset of) responses, as in the case of Huber weights. While having heard news does not predict assignment to a treatment group, reading the treatment does; respondents could only pass the reading time cutoff if they had a treatment to read in the first place. This problem is compounded by the predictability of compliance demonstrated in Table 3: selection into compliance or non-compliance is predictable by respondent demographics. Therefore, simply excluding non-compliers from the sample and rerunning the regressions in Tables 1 and 2 would no longer leave us with a randomly assigned treatment group. This implies that the average treatment effect estimates provided in Tables 1 and 2 are likely to understate the local average treatment effect (for compliers), because those estimates do not account for the partial endogeneity of treatment selection; that is, respondents must be randomly assigned a treatment group to be treated, but they may or may not comply with being treated in a non-random fashion. This underestimation is understood by noting (see Imbens and Angrist, 1994, and Angrist, Imbens, and Rubin, 1996) that, under the “exclusion restriction” that treatment does not affect compliance, the local average treatment effect is the ratio of the estimated intent-to-treat effect (as shown in Tables 1 and 2) and the estimated proportion of compliers (as modeled in Table 3). Given the evidence from Table 3 that we have covariates that explain compliance, we use these to calculate propensity scores that, in turn, are used to re-estimate the treatment effect. Following Jo and Stuart (2009), we use a two-step process to estimate complier and non-complier treatment effects. In the first step, similar to Follmann (2000), we estimate propensity scores (𝑝𝑝) 21 for respondents in the control group of a (given) treatment period by fitting a logit model of compliance within the treatment group using the covariates seen in Table 3. In the case of additional sample restrictions (that is, excluding respondents who heard news about monetary policy), the logit model is estimated on a subsample with the same restrictions, and fitted values are only calculated for respondents in the control group belonging to the same group (that is, who also did not hear news about monetary policy). Because of randomization, the covariates used in the treatment group should also explain compliance in the control group, for whom we cannot measure compliance directly, of course, as they did not receive a treatment and thus were not timed. In the second step, we return to the regression model in Tables 1 and 2, but now use the estimated propensity scores to reweight respondents. To estimate the average causal effect among compliers (CACE), respondents in the control group are assigned weights 𝑝𝑝𝑖𝑖/(1 − 𝑝𝑝𝑖𝑖), while treatment group compliers and non-compliers are assigned weights equal to one and zero, respectively (thus equivalently excluding treatment group non-compliers). To estimate the average causal effect among non-compliers (NACE), respondents in the control group are assigned weights (1 − 𝑝𝑝𝑖𝑖)/𝑝𝑝𝑖𝑖, while treatment group compliers and non-compliers are assigned weights of zero and one, respectively (the inverse of the CACE weighting regime).15 Intuitively, these regressions estimate complier (non-complier) treatment effects by giving the most likely and least likely compliers (non-compliers) within the control group the largest and smallest weights, respectively, effectively reweighting the control group as a whole to match the characteristics of the treatment group compliers (non-compliers). Table 4 reports CACE and NACE estimates obtained via the two-step procedure outlined above, for the July and September waves and broken down by prior news exposure. To account for the estimation uncertainty from using estimated propensity scores in the second-stage weighted regression, we bootstrap each step of the two-step process. Table 4 reports the means and standard deviations of the distribution of each coefficient’s draws, with p-scores calculated using these means and standard deviations. 15 The reason that treatment group compliers and non-compliers get weights of 1 and 0, respectively (and vice-versa when estimating the NACE), rather than predicted scores, is that we know with certainty whether they passed the reading cutoff time or not. Using predicted scores over the observed compliance would discard this information. 28 While our treatments and analysis focus on the communication of monetary policy decisions, consumers may be more aware of other, longer-term interest rates—which potentially incorporate forward-looking expectations of monetary policy—because these rates are more relevant for the real-world monetary transmission mechanism for borrowers and savers. In Appendix G, we re-estimate the treatment effects in Table 4 while controlling for two additional measures of broader awareness about interest rates that our survey asks about pre-treatment: whether the respondent reported that the interest rates that people pay to borrow money in general had changed recently; and whether the respondent indicated that borrowing rates in general had gone up by more than 1 percentage point recently.20 As in Table 4, we continue to find negative and statistically significant treatment effects for all respondents who complied with the treatment and for those respondents who complied and were uninformed about monetary policy. We find no evidence that prior awareness about borrowing rates attenuates the treatment effects, as seen by the statistically insignificant interaction terms. Thus, our results are not dependent on providing information to consumers who were completely inattentive to the economic environment; the monetary policy treatments also moved the inflation expectations of those who already had some sense of what was happening to interest rates. These results provide suggestive evidence that it is the combination of communicating the actor (that is, the FOMC) and the action (that is, the decision to raise the policy rate) in our treatments that has delivered the large negative impact on inflation expectations, and not a change in interest rates alone. Future work can refine the role of these two margins further. Our results highlight the importance of controlling for both informedness and compliance so that RCTs have a higher degree of external validity and can better help economists understand the real-world effects of monetary policy communications. Informedness is not the same for everybody (see Appendix Table E2), with some agents accruing information about policy ahead of others in the “real-world” rather than in the “laboratory” of the RCT. Informed consumers update their inflation expectations outside of the RCT. This means that the small and statistically insignificant treatments effects we find for them (Table 4, Columns (2) and (5)) do not mean that the monetary policy communication has been ineffective, just that this group of consumers may 20 Appendix G documents that there is a high correlation between informedness about monetary policy and awareness of increases in borrowing rates more generally, but not a perfect correlation between the two different concepts, which forms the basis for these regression results. 29 have already updated their expectations. The uninformed do have an understanding of the main mechanism of monetary policy, so new, targeted communications to reach this group in the realworld should also be expected to weaken their responsiveness to treatments within the RCT. 7. Conclusion This paper reports and analyzes results from a specially designed multi-wave RCT to test whether and how communications about actual increases in the federal funds rate in 2022 causally affected consumers’ inflation expectations. We find that simple communications about increases in the federal funds rate reduced consumers’ medium-term inflation expectations, but most notably only for those who were previously unaware of but willing to pay attention to (that is, in our language, “compliant” with) the information communicated in the RCT. Our results thus highlight the importance of measuring the ex ante informedness and the ex post compliance of respondents in applied macroeconomics RCTs. Our results therefore suggest that the FOMC’s policy actions of 2022 likely helped to directly (re)anchor medium-term inflation expectations and contribute to the disinflation process for some consumers. In turn, the fact that real-time and real-life monetary policy communications of the policy action alone (even absent any information on its intent) are found to lower longerterm inflation expectations suggests that consumers did have some common understanding of the monetary policy transmission mechanism and of the FOMC’s intention that the rate hikes should drive inflation down. Our finding that communicating the monetary policy changes of 2022 had small effects on inflation expectations on average, but larger effects on the previously uninformed and compliant, reinforces the growing consensus in the literature (for example, Coibion and Gorodnichenko, 2015, and Andre et al., 2022) that there is considerable heterogeneity across consumers in terms of how they form and update their inflation expectations. During the high inflation of 2022, we find that not everyone was paying attention to monetary policy nor was everyone equally attentive to the information treatment administered in our five RCTs, even when presented with very short, tweetlike monetary policy communications designed to be easy to read. By identifying groups of consumers who tend to report being less informed about monetary policy news and to be more likely to pay attention to news when it is shared with them, and by controlling for those groups in our regressions, we find evidence that there is scope to increase the impact of monetary policy 30 communications by targeting specific groups of the general public, notably women. More generally, our results suggest that it is important when interpreting the heterogeneous treatment effects commonly found in RCTs in macroeconomics to unpack both the compliant from the noncompliant and those for whom the informational treatment is news from those for whom the information is already known. Otherwise, as we illustrate, it is possible to misclassify the drivers of the heterogeneity and misdiagnose the best ways of increasing the effectiveness of monetary policy communications. 31 References Andre, Peter, Carlo Pizzinelli, Christopher Roth, and Johannes Wohlfart. 2022. “Subjective Models of the Macroeconomy: Evidence from Experts and Representative Samples.” Review of Economic Studies 89 (6): 2958-91. https://doi.org/10.1093/restud/rdac008. Angrist, Joshua D., Guido W. Imbens, and Donald B. Rubin. 1996. “Identification of Causal Effects Using Instrumental Variables.” Journal of the American Statistical Association 91: 444–55. https://doi.org/10.1080/01621459.1996.10476902. Armantier, Olivier, Scott Nelson, Giorgio Topa, Wilbert van der Klaauw, and Basit Zafar. 2016. “The Price Is Right: Updating Inflation Expectations in a Randomized Price Information Experiment.” Review of Economics and Statistics 98 (3): 503–23. https://doi.org/10.1162/REST_a_00499. Bholat, David, Nida Broughton, Janna Ter Meer, and Eryk Walczak. 2019. “Enhancing Central Bank Communications Using Simple and Relatable Information.” Journal of Monetary Economics 108: 1-15. https://doi.org/10.1016/j.jmoneco.2019.08.007. Binder, Carola. 2017. “Fed Speak on Main Street: Central Bank Communication and Household Expectations.” Journal of Macroeconomics 52 (C): 238-51. https://doi.org/10.1016/j.jmacro.2017.05.003. Binder, Carola and Alex Rodrigue. 2018. “Household Informedness and Long-Run Inflation Expectations: Experimental Evidence.” Southern Economic Journal 85: 580-98. https://doi.org/10.1002/soej.12306. Blinder, Alan. S., Michael Ehrmann, Marcel Fratzscher, Jakob De Haan, and David-Jan Jansen. 2008. “Central Bank Communication and Monetary Policy: A Survey of Theory and Evidence.” Journal of Economic Literature 46 (4): 910–45. https://doi.org/10.1257/jel.46.4.910. Blinder, Alan S., Michael Ehrmann, Jakob de Haan, and David-Jan Jansen. 2024. “Central Bank Communication with the General Public: Promise or False Hope?” Journal of Economic Literature 62(2): 425-57. https://doi.org/10.1257/jel.20231683. Bracha, Anat and Jenny Tang. 2024. “Inflation Levels and (In)Attention.” Review of Economic Studies. https:// doi.org/10.1093/restud/rdae063. 32 Coibion, Olivier, and Yuriy Gorodnichenko. 2015. “Information Rigidity and the Expectations Formation Process: A Simple Framework and New Facts.” American Economic Review 105 (8): 2644-78. https://doi.org/10.1257/aer.20110306. Coibion, Olivier, Yuriy Gorodnichenko, and Saten Kumar. 2018. “How Do Firms Form Their Expectations? New Survey Evidence.” American Economic Review 108 (9): 2671–2713. https://doi.org/10.1257/aer.20151299. Coibion, Olivier, Yuriy Gorodnichenko, Saten Kumar, and Mathieu Pedemonte. 2020. “Inflation expectations as a policy tool?” Journal of International Economics, 124: 103297. https://doi.org/10.1016/j.jinteco.2020.103297. Coibion, Olivier, Dimitris Georgarakos, Yuriy Gorodnichenko, and Michael Weber. 2023a. “Forward Guidance and Household Expectations”. Journal of the European Economic Association. 21 (5): 2131-71. https://doi.org/10.1093/jeea/jvad003. Coibion, Olivier, Yuriy Gorodnichenko, Edward S. Knotek II, and Raphael Schoenle. 2023b. “Average Inflation Targeting and Household Expectations.” Journal of Political Economy Macroeconomics 1 (2): 403-46. https://doi.org/10.1086/722962. Coibion, Olivier, Yuriy Gorodnichenko, and Michael Weber. 2022. “Monetary Policy Communications and Their Effects on Household Inflation Expectations.” Journal of Political Economy, 130 (6): 1537-84. https://doi.org/10.1086/718982. D’Acunto, Francesco, Daniel Hoang, Maritta Paloviita, and Michael Weber. 2020. “Effective Policy Communication: Targets versus Instruments.” Working Paper 2020-148. University of Chicago Becker Friedman Institute for Economics. https://doi.org/10.2139/ssrn.3712658. D’Acunto, Francesco, Andreas Fuster, and Michael Weber. 2021. “Diverse Policy Committees Can Reach Underrepresented Groups.” Working Paper 29275. National Bureau of Economic Research. https://doi.org/10.3386/w29275. Dietrich, Alexander M., Edward S. Knotek II, Kristian Ove R. Myrseth, Robert W. Rich, Raphael S. Schoenle, and Michael Weber. 2022. “Greater Than the Sum of the Parts: Aggregate vs. Aggregated Inflation Expectations.” Federal Reserve Bank of Cleveland, Working Paper No. 22-20. https://doi.org/10.26509/frbc-wp-202220. 33 Follmann, Dean A. 2000. “On the Effect of Treatment Among Would-Be Treatment Compliers: An Analysis of the Multiple Risk Factor Intervention Trial.” Journal of the American Statistical Association 95: 1101–09. https://doi.org/10.1080/01621459.2000.10474306. Fuster, Andreas, Ricardo Perez-Truglia, Mirko Wiederholt, and Basit Zafar. 2022. “Expectations with Endogenous Information Acquisition: An Experimental Investigation.” The Review of Economics and Statistics 104 (5): 1059–1078. https://doi.org/10.1162/rest_a_00994 Gorodnichenko, Yuriy, Tho Pham, and Oleksandr Talavera. 2023. “The Voice of Monetary Policy.” American Economic Review 113 (2): 548-84. https://doi.org/10.1257/aer.20220129. Haaland, Ingar, Christopher Roth, and Johannes Wohlfart. 2023. “Designing Information Provision Experiments.” Journal of Economic Literature 61 (1): 3-40. https://doi.org/10.1257/jel.20211658. Haldane, Andrew and Michael McMahon. 2018. “Central Bank Communications and the General Public.” AEA Papers and Proceedings 108: 578-83. https://doi.org/10.1257/pandp.20181082. Hazell, Jonathon, Juan Herreño, Emi Nakamura, and Jón Steinsson. 2022. “The Slope of the Phillips Curve: Evidence from U.S. States.” The Quarterly Journal of Economics 137 (3): 1299–1344. https://doi.org/10.1093/qje/qjac010. Hoffmann, Mathias, Emanuel Moench, and Guido Schultefrankenfeld. 2023. “A KISS for Central Bank Communication in Times of High Inflation.” mimeo Imbens, Guido W. and Joshua D. Angrist. 1994. “Identification and Estimation of Local Average Treatment Effects.” Econometrica 62 (2): 467–75. https://doi.org/10.2307/2951620. Jo, Booil and Elizabeth A. Stuart. 2009. “On the Use of Propensity Scores in Principal Causal Effect Estimation.” Statistics in Medicine 28 (23): 2857-75. https://doi.org/10.1002/sim.3669. Knotek, Edward S., II, Raphael S. Schoenle, Alexander M. Dietrich, Keith Kuester, Gernot J. Müller, Kristian Ove R. Myrseth, and Michael Weber. 2020. “Consumers and COVID-19: A Real-Time Survey.” Economic Commentary 2020-08. Federal Reserve Bank of Cleveland. https://doi.org/10.26509/frbc-ec-202008. Lamla, Michael J. and Dmitri V. Vinogradov. 2019. “Central Bank Announcements: Big News for Little People?” Journal of Monetary Economics 108: 21-38. https://doi.org/10.1016/j.jmoneco.2019.08.014. 34 Maćkowiak, Bartosz and Mirko Wiederholt. 2024. “Rational Inattention during an RCT.” https://cepr.org/publications/dp18875. Maćkowiak, Bartosz, Filip Matějka, and Mirko Wiederholt. 2023. “Rational Inattention: A Review.” Journal of Economic Literature 61 (1): 226-73. https://doi.org/10.1257/jel.20211524. Nakamura, Emi and Jón Steinsson. 2018. “High-Frequency Identification of Monetary NonNeutrality: The Information Effect.” Quarterly Journal of Economics 133 (3): 1283– 1330. https://doi.org/10.1093/qje/qjy004. Pedemonte, Mathieu. 2023. “Fireside Chats: Communication and Consumers’ Expectations in the Great Depression.” Review of Economics and Statistics. https://doi.org/10.1162/rest_a_01200. Pfäuti, Oliver. 2024. “The Inflation Attention Threshold and Inflation Surges.” Unpublished manuscript. Qualtrics. 2019. “ESOMAR 28: 28 Questions to Help Buyers of Online Samples.” April 2019, Technical Report. https://www.iup.edu/arl/files/qualtrics/esomar.pdf. Roth, Christopher and Johannes Wohlfart. 2020. “How Do Expectations about the Macroeconomy Affect Personal Expectations and Behavior?” Review of Economics and Statistics 102 (4): 731-48. https://doi.org/10.1162/rest_a_00867. Sims, Christopher A. 2003. “Implications of Rational Inattention.” Journal of Monetary Economics 50 (3): 665-90. https://doi.org/10.1016/S0304-3932(03)00029-1. Weber, Michael, Bernardo Candia, Tiziano Ropele, Rodrigo Lluberas, Serafin Frache, Brent H. Meyer, Saten Kumar, Yuriy Gorodnichenko, Dimitris Georgarakos, Olivier Coibion, Geoff Kenny, and Jorge Ponce. 2024. “Tell Me Something I Don’t Already Know: Learning in Low and High-Inflation Settings.” Working Paper 31485. National Bureau of Economic Research. https://doi.org/10.3386/w31485. 35 Online Appendix A. The 5 FOMC Statements Exhibit A1. March 16, 2022, FOMC Statement 36 Exhibit A2. May 4, 2022, FOMC Statement 37 Exhibit A3. June 15, 2022, FOMC Statement 44 Table C1. Posterior on Prior-x-Treatment (1) March (2) May (3) June (4) July (5) Sept (6) Pooled Prior 0.96*** 0.92*** 0.93*** 0.65*** 0.69*** 0.91*** (0.01) (0.01) (0.01) (0.01) (0.01) (0.01) Treatment 1 3.26*** -0.54** -0.30 0.49** 0.29 2.07*** (0.23) (0.22) (0.23) (0.20) (0.25) (0.10) Treatment 2 0.16 -0.57*** -0.50** 0.55*** 0.46* -0.45*** (0.23) (0.22) (0.23) (0.20) (0.25) (0.10) Treatment 3 3.22*** 2.66*** (0.24) (0.20) Treatment 4 0.03 -0.54*** (0.23) (0.20) Placebo -1.38*** 0.10 (0.24) (0.19) Treatment 1 x Prior -0.54*** 0.04*** -0.01 -0.18*** -0.10*** -0.39*** (0.02) (0.01) (0.01) (0.02) (0.02) (0.01) Treatment 2 x Prior -0.04*** 0.03** 0.01 -0.11*** -0.20*** 0.01 (0.01) (0.02) (0.01) (0.02) (0.02) (0.01) Treatment 3 x Prior -0.57*** -0.52*** (0.02) (0.02) Treatment 4 x Prior -0.08*** -0.02* (0.02) (0.01) Placebo x Prior 0.25*** 0.03** (0.02) (0.01) Observations 7879 5204 4899 5994 5872 29833 Adjusted R2 0.83 0.89 0.89 0.64 0.77 0.83 Notes: Columns 1-5 contain no controls or fixed effects. Column 6 shows a pooled (over waves) regression with treatment period fixed effects. * p < 0.10, ** p < 0.05, *** p < 0.01 The abundance of statistically significant coefficients in Table C1 suggests that the treatments did affect respondents’ posterior beliefs. We find that the coefficient on the prior (𝜃𝜃 in Eq. C1, the weight that respondents in the control group assigns to their prior in forming their posterior expectations) is generally high, suggesting a high correlation between those two variables for the control group. This suggests that one-year inflation expectations are a good measure for a hypothetical prior five-year inflation expectation, which we do not directly observe. Table C1 also shows that, in general, consumers put a lower weight on their prior after the treatment, evidence that they are learning from the information treatment. We observe bigger updates for those with higher ex ante inflation expectations. In the case of the pooled regression (column 6), Treatment 1 45 has a significant effect when interacted with the prior. There is heterogeneity in the different waves, but, in general, we find negative coefficients, indicating that the treatment did cause respondents to revise down their prior. Online Appendix D. Treatment Reading Times and Compliance Figure D1. Treatment Reading Times (Pooled Across the July and September Surveys) 46 Figure D2. Compliance Probabilities vs. Huber Weights 47 Figure D3. Treatment Effects on Compliant and Uninformed Respondents by Reading Time Percentile 48 Online Appendix E. Inflation Expectations by Informedness and Informedness by Demographics Figure E1: Inflation Expectations Based on Whether Consumers Had Heard News about Monetary Policy 49 Table E2. Probability of Having Heard News about Monetary Policy: Logit and Linear Probability Models (1) (2) (3) (4) (5) (6) Male 2.22*** (0.08) 2.22*** (0.08) 2.22*** (0.08) 0.17*** (0.01) 0.17*** (0.01) 0.17*** (0.01) Nonwhite 0.97 (0.05) 0.97 (0.05) 0.97 (0.05) -0.01 (0.01) -0.01 (0.01) -0.01 (0.01) Hispanic 1.55*** (0.10) 1.56*** (0.10) 1.55*** (0.10) 0.09*** (0.01) 0.09*** (0.01) 0.09*** (0.01) Primary Shopper 1.71*** (0.11) 1.72*** (0.11) 1.75*** (0.11) 0.11*** (0.01) 0.11*** (0.01) 0.11*** (0.01) Numerical Literacy 1.10* (0.05) 1.10* (0.05) 1.10** (0.05) 0.02* (0.01) 0.02* (0.01) 0.02* (0.01) Age: 36-50 0.92 (0.05) 0.92 (0.05) 0.93 (0.05) -0.02 (0.01) -0.02 (0.01) -0.02 (0.01) 51-65 0.78*** (0.04) 0.78*** (0.04) 0.78*** (0.04) -0.05*** (0.01) -0.05*** (0.01) -0.05*** (0.01) 66+ 1.76*** (0.11) 1.76*** (0.11) 1.76*** (0.11) 0.12*** (0.01) 0.12*** (0.01) 0.12*** (0.01) Income: $35,000-$49,999 1.27*** (0.06) 1.27*** (0.06) 1.27*** (0.06) 0.05*** (0.01) 0.05*** (0.01) 0.05*** (0.01) $50,000-$99,999 1.43*** (0.06) 1.43*** (0.06) 1.44*** (0.06) 0.07*** (0.01) 0.07*** (0.01) 0.07*** (0.01) $100,000 or more 1.92*** (0.12) 1.92*** (0.12) 1.93*** (0.12) 0.14*** (0.01) 0.14*** (0.01) 0.14*** (0.01) Education: Some College 1.25*** (0.06) 1.25*** (0.06) 1.26*** (0.06) 0.05*** (0.01) 0.05*** (0.01) 0.05*** (0.01) Bachelor's Degree 2.16*** (0.11) 2.16*** (0.11) 2.18*** (0.11) 0.17*** (0.01) 0.17*** (0.01) 0.18*** (0.01) Advanced Degree 3.20*** (0.21) 3.20*** (0.21) 3.21*** (0.21) 0.26*** (0.01) 0.26*** (0.01) 0.26*** (0.01) Political Party: Democrat 1.34*** (0.06) 1.33*** (0.06) 1.34*** (0.06) 0.06*** (0.01) 0.06*** (0.01) 0.06*** (0.01) Republican 1.25*** (0.06) 1.25*** (0.06) 1.25*** (0.06) 0.04*** (0.01) 0.04*** (0.01) 0.04*** (0.01) Days Since FOMC 0.97*** (0.01) -0.01*** (0.00) Days Since FOMC, Sq. 1.00*** (0.00) 0.00*** (0.00) Wave Fixed Effect No Yes Yes No Yes Yes Observations 33728 33728 33728 33728 33728 33728 Notes: * p < 0.10, ** p < 0.05, *** p < 0.01. Columns 1-3 show results as odds ratios from a logit model, while columns 4-6 show results from OLS regression. 50 Online Appendix F. Treatment Effects on Additional Variables Table F1. Treatment Effects on GDP Expectations via Propensity Score Weighted Regressions Compliers Non-compliers (1) (2) (3) (4) (5) (6) Heard News = All Yes No All Yes No July Treatment 1 -0.55 0.53 -1.66 2.32* -0.02 3.73** (0.84) (1.15) (1.19) (1.40) (2.30) (1.78) Treatment 2 -0.49 -1.25 -0.17 3.34** 5.21** 1.78 (0.85) (1.14) (1.24) (1.42) (2.30) (1.83) Observations 4672 1992 2680 4283 1832 2451 September Treatment 1 0.82 -0.80 1.79 1.26 -0.40 3.04 (1.10) (1.47) (1.54) (1.60) (2.47) (2.16) Treatment 2 0.44 -3.23** 2.67* 1.81 0.26 3.57* (1.12) (1.57) (1.53) (1.54) (2.42) (2.04) Placebo 0.38 -2.51* 2.31 0.75 -1.73 3.45 (1.05) (1.42) (1.49) (1.68) (2.54) (2.30) Observations 4338 1900 2438 3978 1832 2146 Notes: * p < 0.10, ** p < 0.05, *** p < 0.01. The first-stage logit regressions are exactly as in Table 3. The dependent variable in the second-stage regressions is the post-treatment five-year average GDP growth expectation minus the pre-treatment year-ahead expectation, each winsorized at the 2nd and 98th percentiles. 51 Table F2. Treatment Effects on Average Income Growth via Propensity Score Weighted Regressions Compliers Non-compliers (1) (2) (3) (4) (5) (6) Heard News = All Yes No All Yes No July Treatment 1 0.53 0.60 -0.18 2.32* 1.22 3.26* (0.77) (1.12) (1.07) (1.33) (2.22) (1.68) Treatment 2 -0.13 -0.22 -0.64 3.17** 3.55 3.13* (0.76) (1.00) (1.13) (1.37) (2.26) (1.75) Observations 4672 1992 2680 4283 1832 2451 September Treatment 1 -0.81 -0.24 -1.70 -1.18 -1.26 -0.90 (0.85) (1.16) (1.18) (1.61) (2.55) (2.10) Treatment 2 -0.05 -1.82 0.63 -0.48 -0.38 -0.31 (0.97) (1.53) (1.29) (1.58) (2.45) (2.11) Placebo -0.27 -2.11 0.75 -1.54 -2.96 0.05 (0.90) (1.29) (1.24) (1.71) (2.70) (2.25) Observations 4338 1900 2438 3978 1832 2146 Notes: * p < 0.10, ** p < 0.05, *** p < 0.01. The first-stage logit regressions are exactly as in Table 3. The dependent variable in the second-stage regressions is the post-treatment five-year average income growth expectation minus the pre-treatment year-ahead expectation, each winsorized at the 2 nd and 98 th percentiles. 52 Online Appendix G. Treatment Effects Conditional on Interest Rate Priors Table G1. Treatment Effects with Qualitative Prior Interest Rate Change Interaction Compliers Non-compliers (1) (2) (3) (4) (5) (6) Heard News = All No Yes All No Yes Treatment 1 -2.93** -3.05** -1.41 0.33 1.02 -2.60 (1.21) (1.26) (3.08) (1.35) (1.49) (3.32) Treatment 2 -3.35*** -3.24** -3.23 0.39 1.34 -3.65 (1.29) (1.41) (2.54) (1.38) (1.51) (3.36) Placebo 0.60 0.42 2.93 -0.57 2.69 -14.19** (1.73) (1.88) (3.91) (2.11) (2.01) (6.98) Rates Changed -1.88** -1.17 -2.99* 1.97 0.45 0.01 (0.90) (1.02) (1.63) (1.33) (1.56) (2.50) Changed x Treatment 1 1.39 -0.49 1.72 -0.83 -1.30 1.99 (1.34) (1.53) (3.17) (1.81) (2.20) (3.74) Changed x Treatment 2 1.33 0.11 2.26 -1.07 -2.06 2.99 (1.41) (1.63) (2.65) (1.85) (2.22) (3.80) Changed x Placebo -1.09 -1.98 -2.54 -0.43 -2.34 12.29* (1.92) (2.32) (4.03) (2.60) (2.81) (7.30) Observations 9010 5118 3892 8261 4597 3664 Notes: * p < 0.10, ** p < 0.05, *** p < 0.01. Pooled across the July and September waves. Treatment effects estimated using the propensity score methodology described in Section 5. The “Rates changed” dummy =1 if respondents answered “Yes” to the question, “Thinking about the interest rates that people pay to borrow money, such as mortgage rates, would you say that those interest rates in general have changed recently?” and =0 if they answered “No” or “Not sure.” 53 Table G2. Treatment Effects with Quantitative Prior Interest Rate Change Interaction Compliers Non-compliers (1) (2) (3) (4) (5) (6) Heard News = All No Yes All No Yes Treatment 1 -2.57** -3.27*** 0.59 0.01 1.42 -3.03 (1.01) (1.14) (2.09) (1.28) (1.42) (2.67) Treatment 2 -3.23*** -3.47*** -1.81 -0.39 1.19 -3.88 (1.08) (1.27) (1.81) (1.32) (1.42) (2.90) Placebo 0.84 0.28 3.44 -1.24 2.58 -9.65** (1.48) (1.70) (2.76) (1.83) (1.87) (4.19) Rates Up ≥ 1pp -1.65** -1.03 -1.89* 1.28 0.77 -0.69 (0.75) (0.97) (1.08) (1.40) (1.58) (2.60) Rates Up x Treatment 1 1.01 -0.12 -0.52 -0.50 -2.74 3.02 (1.17) (1.47) (2.21) (1.84) (2.23) (3.32) Rates Up x Treatment 2 1.30 0.53 0.76 0.20 -1.99 3.99 (1.23) (1.57) (1.96) (1.86) (2.30) (3.44) Rates Up x Placebo -1.49 -1.86 -3.41 0.82 -2.43 8.69* (1.70) (2.21) (2.95) (2.48) (2.90) (4.81) Observations 9010 5118 3892 8261 4597 3664 Notes: * p < 0.10, ** p < 0.05, *** p < 0.01. Pooled across the July and September waves. Treatment effects estimated using the propensity score methodology described in Section 5. The “Rates Up ≥1 pp” dummy =1 if respondents said that general borrowing rates had recently increased by more than 1 percentage point for the question, “Thinking about the interest rates that people pay to borrow money, such as mortgage rates, how much would you say that interest rates in general have changed recently?” and =0 otherwise. 60 would continue to reduce the size of its balance sheet. These actions were part of an effort to help bring inflation back down toward its objective. July Wave TLOv2.1 [Control Group] Please proceed to the next question. TLOv2.2 [Treatment 1] On July 27, 2022, the Federal Open Market Committee (FOMC) raised its primary policy interest rate (the federal funds rate) by three-quarters of a percentage point, to a target range of 2-1/4 to 2-1/2 percent. The FOMC also said that it would continue to reduce the size of its balance sheet. TLOv2.3 [Treatment 2] On July 27, 2022, the Federal Open Market Committee (FOMC) raised its primary policy interest rate (the federal funds rate) by three-quarters of a percentage point, to a target range of 2-1/4 to 2-1/2 percent, and anticipated that ongoing increases in the target range will be appropriate. The FOMC also said that it would continue to reduce the size of its balance sheet. These actions were part of an effort to help bring inflation back down toward its objective. September Wave TLOv4.1 [Control Group] Please proceed to the next question. TLOv4.2 [Treatment 1] On September 21, 2022, the Federal Open Market Committee (FOMC) raised its primary policy interest rate (the federal funds rate) by three-quarters of a percentage point, to a target range of 3 to 3-1/4 percent. The FOMC also said that it would continue to reduce the size of its balance sheet. TLOv4.3 [Treatment 2] On September 21, 2022, the Federal Open Market Committee (FOMC) raised its primary policy interest rate (the federal funds rate) by three-quarters of a percentage point, to a target range of 3 to 3-1/4 percent, and anticipated that ongoing increases in the target range will be appropriate. The FOMC also said that it would continue to reduce the size of its balance sheet. Federal Reserve Chair Jerome Powell said, “The FOMC is strongly resolved to bring inflation down to 2 percent and we will keep at it until the job is done.” TLOv4.4 [Placebo] From 2015 to 2021, the population in the United States grew in a range of 3 to 3-1/4 percent. QJH13 Over the next 5 years, do you think that there will be inflation or deflation on average? o Inflation o Deflation (opposite of inflation) 61 QJH13a [If QJH13 = Inflation] What do you expect the average annual rate of inflation to be over the next 5 years? Please give your best guess. I expect the average annual rate of inflation to be __ percent per year over the next 5 years. QJH13b [If QJH13 = Deflation (opposite of inflation)] What do you expect the average annual rate of deflation to be over the next 5 years? Please give your best guess. I expect the average annual rate of deflation to be __ percent per year over the next 5 years. QJH14 Over the next 5 years, do you think that there will be an increase or decrease in GDP on average? o Increase o Decrease QJH14a [If QJH14 = Increase] What do you expect the average annual rate of increase in GDP will be over the next 5 years? Please give your best guess. I expect the average annual rate of increase to be __ percent per year over the next 5 years. QJH14b [If QJH14 = Decrease] What do you expect the average annual rate of decrease in GDP will be over the next 5 years? Please give your best guess. I expect the average annual rate of decrease to be __ percent per year over the next 5 years. QJH15 In your view, will the total income of all members of your household (including you), after taxes and deductions, increase or decrease over the next 5 years on average? o Increase o Decrease QJH15a [If QJH15 = Increase] What do you expect the average annual rate of increase in the total income of all members of your household will be over the next 5 years? Please give your best guess. I expect the average annual rate of increase in the total income of all members of my household to be __ percent per year over the next 5 years. QJH15b [If QJH15 = Decrease] What do you expect the average annual rate of decrease in GDP will be over the next 5 years? Please give your best guess. I expect the average annual rate of decrease in the total income of all members of my household to be __ percent per year over the next 5 years. QJH18a What do you think is the chance that inflation will be more than 4% in the next 12 months? 62 Q53. What is your civil status? o Single o Partner (not co-habiting) o Partner (co-habiting) o Married o Divorced o Widowed Q121 What would you say is your political affiliation? o Democrat o Independent o Republican o Other Q54 How many children do you have? ___ QX2 Imagine there are white and black balls in a ballot box. You draw a ball 70 times. 56 times, you have drawn a white ball, 14 times a black ball. Given this record, what would you say is the probability of drawing a black ball the next time? The probability is ___ percent.