Abstract:
Central banks emphasize expectations as an important channel of monetary transmission. Yet when households actually update their inflation beliefs in response to policy news--and which households do so--has been hard to pin down empirically. This paper studies when households "listen" to the Fed. Our central claim is that attention to macroeconomic conditions is a key, heterogeneous, and time-varying determinant of the pass-through from conventional monetary policy (MP) surprises to household inflation expectations. We combine a simple model of endogenous attention with new micro and time-series evidence from a long-running U.S. household survey and externally identified policy shocks. Four key results emerge: attention gates the individual-level impact of MP on beliefs; aggregate pass-through scales with the economy's average attentiveness; the effect strengthens in periods of elevated uncertainty; and the response is largest for households with higher payoffs to being informed.
We begin with a minimal behavioral framework, following Gabaix, 2020, in which each household chooses an attention level prior to the arrival of shocks and forms expectations as an attention-weighted combination of a long-run anchor and the fully informed forecast. Attention balances forecast-loss reductions against mental costs and is increasing in the payoff-relevant news variance--the volatility of monetary and non-monetary disturbances that would move the fully informed forecast. The model delivers four testable implications: (i) only the attentive component of beliefs loads on policy news (attention gates pass-through); (ii) aggregate pass-through in time series is proportional to average attentiveness; (iii) higher uncertainty raises attention and therefore amplifies belief responses to policy; and (iv) pass-through is larger for households with higher payoffs to information (e.g., stockholders and homeowners), consistent with a higher benefit parameter in the model.
We then take these predictions to the data using the Michigan Survey of Consumers (MSC). Exploiting its rotating panel, we construct a predetermined attentiveness indicator by contrasting respondents' assessments of recent business conditions with an external benchmark. Monetary policy surprises are identified with high-frequency methods. Our empirical strategy tests each of the model's predictions: we begin with a micro event-study of the effect of conventional MP surprises on revisions in one-year-ahead inflation expectations, followed by a time-series regression that tests the scaling with aggregate attentiveness. We then analyze state dependence by interacting shocks with macro uncertainty and, finally, test the payoff-heterogeneity predictions using household characteristics including stockholding, homeownership, age, and income.
Four sets of findings align closely with the model's predictions. First, in the micro data, a contractionary shock reduces one-year-ahead inflation expectations only among respondents classified as attentive; the estimate for inattentive respondents is small and statistically indistinguishable from zero. This individual-level pattern is the attention-gated pass-through predicted by the model and directly links policy surprises to belief updates when attention is high. Second, in a time-series design that splits months by ex ante economy-wide attentiveness, the pass-through of a contractionary monetary policy shock is large and negative in high-attentiveness months and near zero otherwise, consistent with aggregate pass-through being proportional to average attention. Third, pass-through strengthens in more uncertain periods--during recessions and when real or financial uncertainty is elevated--and this amplification is concentrated among the attentive households. These facts match the comparative statics that optimal attention rises with payoff-relevant news variance and help reconcile why measured effects of MP on the economy could vary across environments [Vavra, 2014,Tenreyro and Thwaites, 2016,Alpanda et al., 2021]. Fourth, consistent with the model's payoff logic, we find systematic heterogeneity in the response. Among attentive respondents, stockholders and homeowners exhibit an especially large pass-through, while younger and middle-aged individuals react more strongly than older ones. These patterns confirm the prediction that groups with a higher stake in the economy are endogenously more responsive to policy news. They also complement a growing body of evidence on firm attention heterogeneity and the efficacy of MP (e.g. [Afrouzi and Yang, 2021,Yang, 2022,Afrouzi, 2024,Wu,2024]). Our findings provide a household-level analogue: just as more complex firms pay closer attention, households with greater financial stakes are more attuned to policy news. For both firms and households, higher attention leads to expectations that align more tightly with fundamentals and react more to policy news.
Our contribution is to show, in a single framework and dataset, that households' attention mediates how conventional monetary policy shocks pass through to inflation expectations, that average attentiveness organizes the strength of the expectations channel over time, and that the effect becomes stronger in more uncertain periods and for households with higher payoffs to information. Conceptually, the results underscore that the expectations channel is attention sensitive: the same policy action can have sharply different effects on beliefs depending on how much attention the audience endogenously devotes to macroeconomic news. In practice, they suggest that communication strategies and policy evaluations should account for variation in attentiveness across groups and over time.
This paper bridges theories of inattentive expectations with empirics on the monetary transmission of beliefs. On the theory side, our setup nests classic information frictions--sticky information and rational inattention [Mankiw and Reis, 2002,Sims, 2003,Ma$$\large{\'{c\/}}$$kowiak and Wiederholt, 2009]--within the behavioral expectations operator of Gabaix, 2020, and relates to broader bounded-rationality approaches [Angeletos and Lian, 2018,Bordalo et al., 2018]. On the empirical side, we connect to work on limited information and learning among households and firms [Coibionand Gorodnichenko, 2015a,Candia et al., 2024], the effects of central-bank communications on household beliefs [Carvalho and Nechio, 2014,Lamla and Vinogradov, 2019,Claus and Nguyen, 2020,Kryvtsov and Petersen, 2021,Coibion et al., 2022,Bauer et al., 2024], and experience/salience in expectation formation [Malmendier and Nagel, 2016,Cavallo et al., 2017,DAcunto et al., 2021]. Our contribution is to fuse these strands by embedding classic information frictions within a behavioral expectations model that delivers sharp, state-contingent predictions for belief updating after externally identified MP shocks, and testing these predictions using a widely used household survey by measuring attentiveness prior to policy news and showing that it governs who updates, by how much, and when.
Household and firm attentiveness to inflation has been measured in several complementary ways. One strand uses "revealed attention" from search behavior and news supply, such as internet search for inflation-related queries and counts of inflation articles in major outlets [Kumar et al., 2015,Marcellino and Stevanovic, 2022,Korenok et al., 2023]. Pfäuti, 2024 infers attention from updating behavior, estimating a time-varying attention parameter from how strongly short-run inflation expectations load on recent inflation and classifying "high-attention" regimes when this responsiveness exceeds an estimated threshold. Kroner, 2025 introduces a complementary pre-announcement index of investor attention around CPI releases aggregates news coverage, mainstream media mentions, and Google search intensity for inflation into a CPI-attention measure used to predict market reactions. Micro-based approaches complement these aggregates by inferring attentiveness directly from survey behavior (e.g., [Braitsch and Mitchell, 2022,Song and Stern, 2024]). In particular, Bracha and Tang, 2024 proxy inattention from the MSC's two-step inflation module: among respondents who first say prices will "stay the same," low attention is flagged if they answer "don't know" at the numeric follow-up or, if they give a number, when it departs substantially from contemporaneous inflation. Relative to these papers, our contribution is to measure attentiveness at the respondent level before policy news and connect it to externally identified monetary policy shocks, showing that attention governs who updates, how much, and when--and that aggregate pass-through scales with independently measured attentiveness over time. This bridges aggregate search and news-based indicators and micro consistency-based measures by providing a direct, policy-linked mapping from attention to belief updating.
Recent evidence indicates that inattention itself is endogenous and varies with the environment: when inflation or macro risk is high, agents acquire more information and align beliefs more closely with fundamentals [Flynn and Sastry, 2024,Weberet al., 2025]. We build on these insights to provide a unified, micro-founded explanation of how attention shapes the MP expectations channel when policy shocks are identified externally and attentiveness is measured before the shock realizes. We also speak to state dependence in monetary policy. While prior explanations emphasize non-linear pricing Vavra, 2014, and broader nonlinear propagation Tenreyro and Thwaites, 2016, we highlight an informational channel: in more volatile or uncertain environments, agentsendogenously raise attention, which amplifies the beliefs response to policy. This mechanism complements recent evidence on time-varying firm inattention and MP efficacy Song and Stern, 2024.
The paper is organized as follows. Section 2 presents the behavioral expectations model and testable implications. Section 3 describes the data and the construction of the attentiveness proxy. Section 4reports the main empirical results, and Section 5 provides robustness checks. Section 6 concludes.
This section develops a minimal behavioral framework in which households choose how much attention to devote to inflation-relevant news. Building on the bounded-rational expectations operator of Gabaix, 2020 and the endogenous-attention logic used in Dietrich, 2024, we derive four testable implications that guide our empirical work in Sections 3 and 4: (i) attention gates the pass-through of monetary policy (MP) shocks to household inflation expectations; (ii) aggregate MP pass-through in time series scales with the economy's average attentiveness; (iii) state dependence is stronger for already-attentive agents, as higher payoff-relevant uncertainty raises attention and amplifies responses; and (iv) payoff heterogeneity: groups with a higher benefit of being informed (larger $$\omega_i$$) or lower attention costs (smaller $$\kappa_i$$) choose more attention, are more likely to be classified as attentive, and exhibit larger pass-through. Section 3 introduces our empirical proxy for attentiveness; Section 4 implements the corresponding tests.
Timing. At the start of month $$t$$, household $$i$$ chooses attention $$m_{i,t}\in[0,1]$$. Then the period-$$t$$ shocks are realized, and the household forms a one-year-ahead inflation expectation using a behavioral operator. We study the impact change in expectations around the shock arrival (holding $$\pi_t$$ fixed and varying only the news realized within $$t$$).
Inflation fundamentals. The fully informed (rational) forecast of next-period inflation is
| $$\displaystyle \pi^{*}_{t+1} \;=\; \bar{\pi} + \rho\,(\pi_t-\bar{\pi}) + \theta\,\varepsilon^{mp}_t + \Gamma'\varepsilon^{o}_t,$$ | (2.1) |
Shock normalization and covariance. We normalize the shocks to be mean-zero Gaussian:
Behavioral expectations and attention choice. Household $$i$$ forms a behavioral expectation by blending a coarse anchor with the fully informed forecast:
| $$\displaystyle \mathbb{E}^B_{i,t}\!\big[\pi_{t+1}\big] \;=\; (1-m_{i,t})\,\bar{\pi} \;+\; m_{i,t}\,\mathbb{E}_t\!\big[\pi^{*}_{t+1}\big],$$ | (2.2) |
| $$\displaystyle m_{i,t} \;=\; \arg\min_{m\in[0,1]} \frac{1}{2}\,\omega_i\,U_t\,(1-m)^2 + \frac{\kappa_i}{2}\,m^2,$$ | (2.3) |
| $$\displaystyle m^{*}_{i,t}(U_t) \;=\; \frac{\omega_i\,U_t}{\omega_i\,U_t+\kappa_i} \;\in\;[0,1].$$ | (2.4) |
| $$\displaystyle U_t \;\equiv\; \mathrm{Var}_t\!\big(\pi^{*}_{t+1}\big) \;=\; \theta^2\,\mathrm{Var}_t(\varepsilon^{mp}_t) \;+\; \Gamma'\Sigma_{o,t}\Gamma \;+\; 2\,\theta\,\mathrm{Cov}_t\!\big(\varepsilon^{mp}_t,\,\Gamma'\varepsilon^{o}_t\big),$$ | (2.5) |
| $$\displaystyle \mathrm{Var}_t(\varepsilon^{mp}_t)=1,\qquad \mathrm{Cov}_t\!\big(\varepsilon^{mp}_t,\varepsilon^{o}_t\big)=0 \;\Rightarrow\; U_t \;=\; \theta^2+\Gamma'\Sigma_{o,t}\Gamma.$$ | (2.6) |
We now characterize individual and aggregate responses to a contractionary MP surprise ( $$\varepsilon^{mp}_t>0$$ with $$\theta<0$$). Proofs are deferred to Appendix A.
| $$\displaystyle \Delta\pi_{i,t+1} \;\equiv\; \mathbb{E}^B_{i,t}\!\big[\pi_{t+1}\big]-\pi_{i,t} \;=\; \theta m^{*}_{i,t}(U_t)\varepsilon^{mp}_t.$$ | (2.7) |
| $$\displaystyle \Delta\pi^{e}_{t+1} \;=\; \underbrace{\Lambda_t}_{\in[0,1]}\,\theta\,\varepsilon^{mp}_t \;+\; \upsilon_t, \qquad \Lambda_t\;\equiv\; \int m^{*}_{i,t} di,$$ | (2.8) |
Endogenous attention creates state dependence: when the environment is more uncertain (larger $$U_t$$), attentive agents reduce their inflation expectations by more after a contractionary MP shock, and the sensitivity to $$U_t$$ is itself stronger for the already-attentive group.
In sum, the simple behavior expectations model delivers four testable implications: (i) attention gates the impact of MP shocks on individual expectations; (ii) aggregate MP pass-through scales with the economy's average attentiveness; (iii) higher payoff-relevant uncertainty strengthens pass-through--especially for already-attentive agents; and (iv) groups with higher payoff from information (larger $$\omega_i$$) or lower attention costs (smaller $$\kappa_i$$) choose more attention and exhibit larger pass-through.5 InSection 3, we define the empirical attentiveness proxy and construct the aggregate attentiveness index used to verify these predictions. Section 4 then implements the corresponding micro and time-series tests.
This section describes the datasets and the construction of our empirical attentiveness proxy, which we will take to the tests implied by Section 2. We first outline sources and sample definitions, then construct an individual-levelaccuracy indicator (our proxy for attention in the model), and finally define an aggregate attentiveness index used in our time-series exercises. Section 4 will bring these measures to the micro and aggregate regressions implied by Propositions 1-4.
Micro survey and demographics. Our micro data come from the Michigan Survey of Consumers (MSC), which interviews a nationally representative sample monthly and re-interviews a rotating panel of respondents roughly six months later. We use the rotating-panel structure to construct revisions in expectations at the individual level and to control for observed heterogeneity (age, income, education, homeownership, stock ownership, gender, region). The MSC provides one-year-ahead inflation expectations and a rich set of qualitative questions on recent business conditions. We focus on the one-year horizon because it is standard for near-term transmission, aligns with our six-month panel and identification window, and is the measure most responsive to contemporaneous macro and policy news in household data (e.g., [Cavallo et al., 2017,Coibion et al., 2022,DAcunto et al., 2023]). Our baseline micro sample spans September 1998 to March 2020, which is the intersection of MSC availability for the necessary items and the availability of our high-frequency monetary policy shocks.6
Monetary policy shocks. Our baseline measure of monetary policy (MP) surprises uses the high-frequency series from Nakamura and Steinsson, 2018, as extended by Bauer et al., 2022. These surprises are identified from changes in federal funds futures prices in a narrow window around FOMC announcements and are standard in the literature. A potential concern with this approach is that it may capture not just pure policy actions but also a Fed "information effect." We retain this series as our baseline because its narrow identification window is crucial for precisely timing policy news relative to our survey's interview dates. To ensure our results are not driven by information effects, we confirm our findings using alternative shocks from Bu et al.,2021 that are designed to purge such effects. For our time-series analysis of the Great Moderation (Section 4.2)), we also use the narrative-based shocks from Romer and Romer, 2004.
To ensure consistent interpretation across all specifications, we normalize the shock series. First, we set the sign so that a positive value always represents a contractionary surprise (an unexpected policy tightening). Second, we scale the series so that a one-unit change corresponds to a one-percentage-point (100 basis point) tightening. This normalization allows our reported regression coefficients to be interpreted directly as the percentage-point response of inflation expectations to a one-percentage-point policy shock.7 Our analysis uses all identified surprises, both contractionary and expansionary. For expositional clarity, we discuss the effects of a "contractionary" shock in the text, as the model's predictions are symmetric.
Other macro series. We obtain our macroeconomic data from the St. Louis Federal Reserve's FRED database. We use the unemployment rate, Industrial Production (IP), inflation, and the National Financial Conditions Index (NFCI) as either benchmarks for our attentiveness proxy or as contemporaneous controls. For our state-dependence analysis, we use the NBER-dated recession indicator and the CBOE Volatility Index (VIX).
Section 2 formalizes attention as a latent weight $$m^{*}_{i,t}\in[0,1]$$. In the data, we proxy attentiveness with a pre-determined indicator based on each respondent's qualitative assessment of recent business conditions, recorded at the first interview.
Step 1: Perceived business conditions (favorable / unfavorable / no news). At the first interview in month $$t$$, each respondent reports whether they have heard favorable or unfavorable changes in business conditions in "the last few months," or have not heard of changes. We code a three-way categorical variable
Step 2: Benchmark for business conditions. To construct our accuracy benchmark, we seek a macroeconomic indicator that is canonical, widely reported, and maps closely to the survey's phrasing of "changes in business conditions." The unemployment rate is arguably the most salient and easily understood measure of real economic health for the general public. Specifically, we compare perceived favorability with the three-month change in the unemployment rate to smooth out high-frequency noise while still capturing the recent economic developments respondents were asked about:
Accuracy classification. We define three mutually exclusive groups at the first interview date $$t$$:
Timing and identification. Crucially, the attentiveness indicators, $${\textbf{A}}_{i,t}$$, are measured at the first interview in month $$t$$, prior to the FOMC announcement window that defines the monetary policy surprise $$\varepsilon^{mp}_t$$. Hence they are pre-determined with respect to the shock. Under our high-frequency identification,
Table 1: Demographic and Socioeconomic Characteristics by Attentiveness Group
Panel A: Homeownership
| Accurate | Inaccurate | Haven't Heard | |
|---|---|---|---|
| (1) Homeowner (%) | 83.2 | 81.9 | 76.9 |
| (2) Renter (%) | 16.8 | 18.1 | 23.1 |
Panel B: Stockownership
| Accurate | Inaccurate | Haven't Heard | |
|---|---|---|---|
| (3) Stockholder (%) | 76.4 | 75.6 | 60.8 |
| (4) Non-stockholder (%) | 23.6 | 24.4 | 39.2 |
Panel C: Education level
| Accurate | Inaccurate | Haven't Heard | |
|---|---|---|---|
| (4) Grade 0-8 no hs diploma (%) | 0.5 | 0.6 | 1.5 |
| (5) Grade 9-12 no hs diploma (%) | 1.6 | 1.3 | 4.1 |
| (6) Grade 0-12 w/ hs diploma (%) | 16.1 | 16.3 | 28.0 |
| (7) Grade 13-17 no col degree (%) | 25.7 | 26.7 | 29.8 |
| (8) Grade 13-16 w/col degree (%) | 30.7 | 29.3 | 22.8 |
| (9) Grade 17 w/ col degree (%) | 25.2 | 25.5 | 13.6 |
Panel D: Age
| Accurate | Inaccurate | Haven't Heard | |
|---|---|---|---|
| (10) 18-34 (%) | 14.4 | 14.8 | 22.8 |
| (11) 35-64 (%) | 63.4 | 61.7 | 53.3 |
| (12) 65+ (%) | 22.0 | 23.3 | 23.7 |
Panel E: Gender
| Accurate | Inaccurate | Haven't Heard | |
|---|---|---|---|
| (13) Male (%) | 56.3 | 56.2 | 53.1 |
| (14) Female (%) | 43.6 | 43.7 | 46.8 |
Panel F: Region
| Accurate | Inaccurate | Haven't Heard | |
|---|---|---|---|
| (15) West (%) | 22.3 | 22.2 | 20.5 |
| (16) North Central (%) | 27.0 | 27.0 | 27.8 |
| (17) Northeast (%) | 17.4 | 17.4 | 16.3 |
| (18) South (%) | 33.2 | 33.2 | 35.2 |
Panel G: Marital status
| Accurate | Inaccurate | Haven't Heard | |
|---|---|---|---|
| (19) Married/partner (%) | 67.2 | 67.0 | 60.0 |
| (20) Divorced (%) | 13.5 | 13.5 | 13.9 |
| (21) Widowed (%) | 6.3 | 6.3 | 8.4 |
| (22) Never married (%) | 12.8 | 13.0 | 17.4 |
Panel H: Average income
| Accurate | Inaccurate | Haven't Heard | |
|---|---|---|---|
| (23) Average income | 93,911.9 | 93,886.0 | 71177.6 |
| Accurate | Inaccurate | Haven't Heard | |
|---|---|---|---|
| Total (%) | 37.2 | 29.7 | 33.1 |
Descriptive statistics by accuracy group. Table 1 reports respondent characteristics across the three groups. The groups are balanced in the sample (Accurate: 37.2%, Inaccurate: 29.7%, Haven't heard: 33.1%). Accurate and Inaccurate respondents look strikingly similar on observables: homeownership (83.2% vs. 81.9%), stockholding (76.4% vs. 75.6%), education (about 56% vs. 55% with a college degree), age (35-64: 63.4% vs. 61.7%; 65+: 22.0% vs. 23.3%), gender, region, marital status, and average income (both $$\sim\$94k$$). By contrast, the Haven't heard group differs systematically: lower homeownership (76.9%), lower stockholding (60.8%), lower educational attainment (36.4% with a college degree; 5.6% less than high school), younger on average (18-34: 22.8%), less likely to be married/partnered (60.0%), and lower average income ($71.2k). These patterns are consistent with interpreting our accuracy indicator $$\boldsymbol{\textbf{A}}_{i,t}$$ as an attentiveness proxy rather than a proxy for fixed traits; observable composition differences are concentrated in the Haven't heard category, while Accurate and Inaccurate respondents are similar on observables. Section 4 will control for the full set of demographics in all specifications.
Discussion of the accuracy proxy. Our accuracy-based indicator is a noisy measure of the latent attention variable, $$m_{i,t}$$, in our model. To strengthen the theoretical justification for this proxy, we ground it directly through the lens of our model. Proposition 4 (part 3) provides a direct motivation to our empirical classification: $$A_{i,t}=\mathbf{1}\{m^{*}_{i,t}\ge \tau\}$$, where an agent is classified as "attentive" if their chosen attention level $$m^{*}_{i,t}$$ surpasses a certain threshold $$\tau$$ required for accurate perception.
From this perspective, our "Accuracy" indicator is not just a proxy for the continuous latent variable $$m^{*}_{i,t}$$, but an empirical implementation of this theoretical classification rule. An agent is classified as "Accurate" because their attention level was sufficiently high to correctly parse the direction of recent economic news. This approach still allows for misclassification--an attentive agent ( $$m^{*}_{i,t}\ge\tau$$) might misread a specific signal, or an inattentive one ( $$m^{*}_{i,t}<\tau$$) might guess correctly--which would induce attenuation bias and work against us finding significant results. Nonetheless, the robust alignment of our empirical results with all four of the model's predictions, as shown in Section 4, suggests that this proxy successfully captures this salient theoretical dimension of household attentiveness.
One might be concerned that our "Accuracy" proxy captures factors other than attention, such as cognitive ability or political bias. While we cannot rule these channels out entirely, three pieces of evidence point toward an attention interpretation. First, our framework provides a unified explanation for the full set of our findings: the scaling of aggregate pass-through, the amplification during uncertain times, and the stronger response among high-payoff groups like stockholders and homeowners, and it is less clear how these other factors would jointly explain this specific constellation of results. Second, the robustness of our main empirical findings to using different macro indicators, as shown in Section 5, mitigates concerns that the results are driven by a specific political narrative tied to unemployment. Third, politicalcomposition is balanced across attentiveness groups suggesting that partisan bias is not the primary driver of the classification.8 Taken together, these facts make it difficult for non-attention explanations to jointly account for the constellation of patterns we document.
To test Proposition 2 in time series, we construct an aggregate attentiveness measure as the cross-sectional share of attentive respondents at the first interview date $$t$$:
Expectation revisions. For individual $$i$$, we compute the revision in one-year-ahead inflation expectations over the six-month panel window, aligning the timing so that the first interview (where $$A_{i,t}$$ is measured) precedes the MP shock and the second interview falls at $$t+h$$ (typically $$h=6$$ months). Aggregate revisions $$\Delta\pi^{e}_{t+h}$$ (e.g., median) are computed analogously across individuals interviewed in month $$t$$ and re-interviewed in $$t+h$$.
Controls and scaling. When used, contemporaneous macro controls are measured between the first and second interviews (e.g., $$\Delta$$ IP and $$\Delta\pi$$ from $$t$$ to $$t+h$$). Monetary shocks are cumulated from $$t$$ through $$t+h-1$$ to match the survey horizon when appropriate; Section 4 reports the exact horizon choice and robustness to alternatives.
This section brings the model's predictions to the data. We test four implications from Section 2 using the measures defined in Section 3. First, at the micro level, attention gates the pass-through of contractionary monetary policy surprises to one-year-ahead inflation expectations: only attentive (accurate) respondents revise down on impact. Second, in time series, the aggregate pass-through scales with the economy's average attentiveness. Third, pass-through is state dependent and strengthens when uncertainty is high, especially among the attentive. Fourth, cross-sectional heterogeneity lines up with payoff differences: groups for whom information is more valuable--homeowners, stockholders, prime-age, and higher-income households--display larger responses when they are accurate. Throughout, identification exploits exogenously identified monetary policy shocks and the fact that accuracy is measured before the shock window; we report robustness to alternative shock measures, controls, and samples.
We begin by testing Proposition 1 in the micro data: only attentive (accurate) respondents should load on contractionary monetary policy (MP) news on impact. Identification rests on two timing features. First, attentiveness is measured at the first interview in month $$t$$ and is therefore predetermined with respect to the FOMC announcement window that generates the MP surprise in month $$t$$. Second, the MP shock is measured in high frequency around the announcement and then cumulated from $$t$$ to $$t+5$$ so that the information set between the two interviews (typically six months apart) aligns with the survey horizon. Under this timing, and conditional on observables, the surprise component of $$MPS_t$$ is orthogonal to respondents' pre-shock attentiveness and demographics, so the interaction coefficients below identify differential pass-through rather than reverse causality or within-month information acquisition.
Our baseline specification, adapted from Coibion and Gorodnichenko, 2015b, tests this attention-gating mechanism by interacting the policy shock with our attentiveness indicators:
| $$\displaystyle \Delta\pi_{i,t+6}^{e} = \alpha + \beta_{M,A}'\big(MPS_{t}\times \boldsymbol{\text{\textbf{A}}}_{i,t}\big) + \beta_{Z,A}'\big(\mathbf{Z}_{t}\times \boldsymbol{\text{\textbf{A}}}_{i,t}\big) + \Gamma'\mathbf{X}_{i,t} + \varepsilon_{i,t},$$ | (4.1) |
Table 2: Attention Shapes Monetary Policy Effects on Inflation Expectations
| (1) Accurate | (2) Inaccurate | (3) Haven't Heard | |
|---|---|---|---|
| (1) $$MPS_{t}$$ | -0.360*** (-4.56) | 0.088 (0.81) | -0.155* (-1.66) |
| (2) $$\Delta$$IP$$_{t}$$ | 0.060*** (3.60) | -0.008 (-0.49) | 0.013 (0.70) |
| (3) $$\Delta\pi_{t}$$ | 0.370*** (9.81) | 0.272*** (6.41) | 0.325* (7.21) |
| Controls | Yes | Yes | Yes |
| Observations | 37,445 | 37,445 | 37,445 |
| $$R^{2}$$ | 0.0138 | 0.0138 | 0.0138 |
Table 2 reports the estimates. The results line up closely with the gating prediction. For the Accurate group, a 1 pp tightening in the shadow policy rate lowers one-year-ahead expected inflation by $$-0.359$$ percentage points $$(t=-4.56)$$. For the Inaccurate group, the slope is small and statistically indistinguishable from zero $$(0.088,\, t=0.81)$$. The Haven't heard group shows a modest negative and only marginally significant coefficient $$(-0.155,\, t=-1.66)$$, an effect much smaller in magnitude than that of the Accurate group.10 Quantitatively, the Accurate-Inaccurate wedge is large: accurate respondents revise down by roughly one third of a percentage point per 1 pp tightening, while inaccurate respondents do not react on impact. This pattern is exactly what Proposition 1 implies when attentive agents have $$m_{i,t}^*>0$$ and inattentive agents have $$m_{i,t}^*\approx 0$$.
Beyond statistical significance, our estimates imply that attention has an economically meaningful impact on the monetary transmission mechanism. Our baseline micro-level estimate indicates that for attentive ("Accurate") individuals, a standard 25-basis-point contractionary policy surprise lowers one-year-ahead inflation expectations by approximately 9 basis points. For a given path of the nominal interest rate, this revision directly amplifies the intended policy tightening by raising the perceived short-term real interest rate for this group.
The controls behave sensibly. IP growth between interviews is positively associated with revisions only for the Accurate group (consistent with real-side news being processed by attentive respondents), while contemporaneous inflation changes load positively and significantly for all groups, reflecting the salience of price changes in household belief formation. Crucially, the primary empirical support for our mechanism comes from the sharp contrast between the "Accurate" and "Inaccurate" groups. As shown in Table 1, these two groups are nearly identical across a wide range of demographic and socioeconomic characteristics. Their divergent responses to monetary policy shocks therefore cannot be easily attributed to observable heterogeneity, lending strong support to our interpretation that pre-shock accuracy--our proxy for attention--is the key mediating factor. Taken together, the specification, timing, and magnitudes support a "attention gates pass-through" interpretation at the micro level: contractionary MP news lowers expected inflation primarily among respondents who accurately perceived recent business conditions before the policy news arrived.
Figure 1: Aggregate Attentiveness: Share of Accurate Respondents (1985-2007)
This figure represents the monthly aggregate attentiveness (accuracy) rate from January 1985 to December 2007, defined as the share of respondents at the first interview
in month $$t$$ whose assessment of recent business conditions aligns with the sign of the three-month change in the unemployment rate (see Section 3 for construction). We use data through
2007m6 to define the "high-attentive" regime as the top $$30\%$$ of the distribution employed in the time-series analysis.
We now test Proposition 2 in aggregate time series: the impact of a conventional monetary policy (MP) surprise on revisions in inflation expectations should be proportional to the economy's average attentiveness $$\Lambda_t$$. To leverage a longer time series and focus squarely on conventional policy actions prior to the zero lower bound, we focus on the Great Moderation (1985m1-2007m12) and use the narrative-based shocks Romer-Romer shock series, $$RRshock_t$$ Romer and Romer, 2004. We construct an aggregate attentiveness index $$\textbf{A}^{\text{agg}}_t$$ as the cross-sectional share classified Accurate at the first interview (Section 3.3). To define our policy regimes, we classifymonths as "high-attentive" if the aggregate attention index falls in the top 30% of its historical distribution. We measure this distribution using data only through June 2007 to ensure the classification is pre-determined relative to our full sample (Figure 1). This top-30% cutoff is a standard approach for regime analysis, and our qualitative findings are robust to using alternative thresholds, such as the top quartile or tercile. Proposition 2 implies a larger (more negative) policy slope in these months: $$\beta^{H}=\theta\,\mathbb{E}[\Lambda_t\,\vert\,$$High$$]$$ vs. $$\beta^{L}=\theta\,\mathbb{E}[\Lambda_t\,\vert\,$$Low$$]$$ with $$\vert\beta^{H}\vert>\vert\beta^{L}\vert$$ for contractionary MP shocks ($$\theta<0$$).
Table 3: Aggregate Pass-Through Scales with Attentiveness
| Accuracy Regime | (1) High | (2) Low |
|---|---|---|
| (1) $$RRshock_{t}$$ | -0.620*** (-3.17) | -0.009 (-0.09) |
| (2) $$\Delta$$IP$$_{t}$$ | 0.183*** (2.82) | -0.032 (-1.54) |
| (3) $$\Delta\pi_{t}$$ | 0.333*** (3.01) | 0.223*** (4.66) |
| Observations | 269 | 269 |
| $$R^{2}$$ | 0.394 | 0.394 |
Our time-series regression mirrors the micro design but aggregates the dependent variable to the monthly median revision, and splits months by $$I^{A}_{t-1}=\mathbf{1}\{\textbf{A}^{\text{agg}}_{t-1}\text{ in top 30\%}\}$$:
| $$\displaystyle \Delta\pi^{e}_{t+6} =\alpha +\beta_{M}\,\big(RRshock_t\times I^{A}_{t-1}\big) +\beta'_{Z,A}\,\big(\mathbf{Z}_t\times I^{A}_{t-1}\big) +\varepsilon_t,$$ | (4.2) |
Table 3 shows that the results align tightly with Proposition 2. In high-attentive months, a 1 pp conventional tightening reduces one-year-ahead expected inflation by about $$-0.62$$ pp (significant), whereas in low-attentive months the slope is small and statistically indistinguishable from zero. Controls also behave sensibly: real activity and inflation changes load positively in the high-attentive regime and are muted otherwise. The difference in slopes is consistent with a higher average attentiveness $$\Lambda_t$$ in high-attentive months: $$\widehat{\beta}^{H}\approx \theta\,\widehat{\Lambda}^{H}$$ vs. $$\widehat{\beta}^{L}\approx \theta\,\widehat{\Lambda}^{L}\approx 0$$. Quantitatively, in high-attentive months, a 25-basis-point tightening reduces median inflation expectations by a substantial 16 basis points. This suggests that during such periods, the expectations channel can amplify the effect of a policy surprise on ex-ante real rates by more than 60%. Conversely, the absence of this effect in low-attentive periods demonstrates how a crucial channel of monetary transmission can become dormant, highlighting that the state of household attentiveness is a key determinant of the overall potency of monetary policy.
Our results imply that belief pass-through is state-dependent and scales with an independently measured attentiveness index. This complements micro evidence on information frictions in expectations formation (e.g., Coibion and Gorodnichenko, 2015a; Gabaix, 2020) by providing a clean time-series counterpart: when more households are attentive, aggregate expectations respond strongly to policy news; when fewer are attentive, pass-through is weak.
Proposition 3 predicts that when payoff-relevant uncertainty $$U_t$$ is higher, optimal attention $$m^{*}_{i,t}$$ rises and the impact of a contractionary MP shock on expectations becomes more negative, with a stronger sensitivity among already-attentive agents. We bring this to the data by interacting MP surprises with (i) our accuracy indicators and (ii) proxies for $$U_t$$ measured at $$t\!-\!1$$: NBER recessions, the Ludvigson et al., 2021 real-uncertainty index (LMN), and financial-market volatility (VIX). We select these three measures to span canonical business cycle, real, and financial uncertainty, ensuring our findings are not specific to one domain. For the LMN and VIX indices, our definition of a high-uncertainty state is based on their cyclical component to isolate deviations from the recent trend in uncertainty, which may be more salient to households than the absolute level. The estimating equation extends Equation eq1 with a triple interaction,
| $$\displaystyle \Delta\pi_{i,t+6}^{e} = \alpha + \beta_{M,A,C}'\big(MPS_t \times \boldsymbol{\text{\textbf{A}}}_{i,t} \times \mathsf{State}_{t-1}\big) + \beta_{Z,A,C}'\big(\mathbf{Z}_t \times \boldsymbol{\text{\textbf{A}}}_{i,t} \times \mathsf{State}_{t-1}\big) + \Gamma'\mathbf{X}_{i,t} + \varepsilon_{i,t},$$ | (4.3) |
Table 4: Uncertainty Raises Attention and Amplifies Expectation Responses
Panel A: NBER
| (1) Accurate | (2) Inaccurate | |
|---|---|---|
| (1) $$Recession\times MPS_{t}$$ | -1.730* (-4.01) | -1.125 (-1.00) |
| (2) $$Normal\times MPS_{t}$$ | -0.039 (-0.49) | 0.115 (1.12) |
Panel B: LMN Real Uncertainty
| (3) Accurate | (4) Inaccurate | |
|---|---|---|
| (3) $$High\times MPS_{t}$$ | -0.539* (-5.51) | 0.048 (0.35) |
| (4) $$Low\times MPS_{t}$$ | -0.269* (-1.77) | 0.250 (1.33) |
Panel C: VIX
| (5) Accurate | (6) Inaccurate | |
|---|---|---|
| (5) $$High\times MPS_{t}$$ | -0.456* (-4.06) | 0.040 (0.22) |
| (6) $$Low\times MPS_{t}$$ | -0.007 (-0.07) | 0.100 (0.79) |
| Panel A | Panel B | Panel C | |
|---|---|---|---|
| Controls | Yes | Yes | Yes |
| Observations | 37,445 | 37,445 | 37,445 |
| $$R^{2}$$ | 0.0170 | 0.0146 | 0.0182 |
The results, reported in Table 4, closely match the theory.11 During recessions, Accurate respondents revise down strongly on impact ($$-1.73$$ pp per 1 pp tightening; significant), while Inaccurate respondents do not respond.12 In High-LMN and High-VIX months, the same qualitative pattern holds: Accurate households reduce expected inflation by $$\approx\!-0.5$$ pp; Inaccurate households again show no significant reaction.13 In Low-uncertainty or Normal states, policy slopes are small and statistically indistinguishable from zero or weakly significant for all groups. This cross-state contrast is the empirical counterpart of
These findings complement existing state-dependence evidence obtained from prices and quantities. Vavra, 2014 shows that time-varying volatility changes firms' adjustment behavior and thereby alters aggregate inflation dynamics; our mechanism works on the expectations margin, with uncertainty inducing greater household attention and sharper belief updates to policy news. Relatedly, our findings can be reconciled with macro studies documenting weaker ultimate effects of policy on real activity in certain states (e.g., deep recessions or high volatility). Our evidence points to a stronger initial impact through the expectations channel: in high-uncertainty states, attentive households align their inflation expectations more sharply with policy news. This leads to a larger adjustment in their perceived ex-ante real interest rates. This very alignment, however, can explain why the ultimate real effects on spending might be muted. If expectations adjust swiftly, there is less scope for policy surprises to generate real effects through informational frictions or misperceptions. In this view, a more potent expectations channel could lead to a more muted response in real activity, as well-informed agents have already incorporated the policy stance into their decisions.
Two additional patterns are worth noting. First, the state dependence we uncover does not require time variation in the volatility of the MP shock itself; increases in $$\Gamma'\Sigma_{o,t}\Gamma$$ (e.g., energy or markup volatility) suffice to raise $$U_t$$ and, therefore, attention. Second, controls behave sensibly across states: real-side changes (IP) load more in high-uncertainty states for the Accurate group, while contemporaneous inflation changes remain salient across groups. Together, the micro evidence supports a simple message: the expectations pass-through of monetary policy shocks is attention weighted and therefore state dependent.
Figure: Uncertainty Regime from Real Sector
Figure: Volatility Regime from Financial Sector
Guided by Proposition 4, in this section, we ask whether groups for whom being informed is more valuable (higher $$\omega$$) or less costly (lower $$\kappa$$) display larger monetary policy pass-through when they are accurate. We proxy these higher payoff groups withr three charecteristics. First, we use asset exposure (stockholding and homeownership), as policy moves directly affect portfolio values and mortgage financing. Second, we examine age, where different life-cycle stages present distinct incentives: younger households' lifetime earnings are highly sensitive to the business cycle, while prime-age households (35-64) typically have the largest balance sheet exposure through assets and mortgages. Third, we use higher income, which correlates with both asset ownership and information use.
Empirically, we extend Equation eq1 by interacting $$MPS_t$$ with the accuracy indicators and each demographic partition, controlling for group means and the full set of covariates. Let $$\boldsymbol{D}_{i,t}$$ be a mutually exclusive demographic partition (e.g., Stockholder/Non-stockholder; Homeowner/Renter; Young/Middle/Old; Income quartiles), with one category omitted in estimation. Our general specification replaces the demographic block as needed:
| $$\displaystyle \Delta\pi^{e}_{i,t+6} = \alpha + \underbrace{\beta_{M,A,D}'\!\big(MPS_t \times \boldsymbol{\text{\textbf{A}}}_{i,t} \times \boldsymbol{D}_{i,t}\big)}_{\text{group- and accuracy-specific MP pass-through}} + \beta_{Z,A,D}'\!\big(\mathbf{Z}_t \times \boldsymbol{\text{\textbf{A}}}_{i,t} \times \boldsymbol{D}_{i,t}\big) + \Gamma'\mathbf{X}_{i,t} + \varepsilon_{i,t},$$ | (4.4) |
Table 5: Attention and Demographic Heterogeneity in Monetary Policy Pass-Through
Panel A: Stockholding
| (1) Accurate | (2) Inaccurate | |
|---|---|---|
| (1) $$Stock\times MPS_{t}$$ | -0.410* (-4.57) | 0.150 (1.20) |
| (2) $$NonStock\times MPS_{t}$$ | -0.228 (-1.42) | -0.047 (-0.21) |
Panel B: Homeownership
| (3) Accurate | (4) Inaccurate | |
|---|---|---|
| (1) $$Homeowner\times MPS_{t}$$ | -0.436* (-5.08) | 0.063 (0.54) |
| (2) $$Renter\times MPS_{t}$$ | 0.026 (0.13) | 0.214 (0.76) |
Panel C: Age
| (5) Accurate | (6) Inaccurate | |
|---|---|---|
| (1) $$Young\times MPS_{t}$$ | -0.613*** (-3.22) | 0.260 (0.82) |
| (2) $$Middle\times MPS_{t}$$ | -0.350*** (-3.68) | 0.140 (1.12) |
| (3) $$Old\times MPS_{t}$$ | -0.234 (-1.22) | -0.264 (-0.98) |
| Panel A | Panel B | Panel C | |
|---|---|---|---|
| Interaction | Stockownership | Homeownership | Age Group |
| Controls | Yes | Yes | Yes |
| Observations | 37,445 | 37,445 | 37,445 |
| $$R^{2}$$ | 0.0142 | 0.0144 | 0.0150 |
Stockholding
Proposition 4 predicts stronger monetary-policy (MP) pass-through among households for whom the payoff to paying attention is higher (larger $$\omega_i$$). Stockholders are a natural candidate: the value of their portfolios is more exposed to macro and policy news, which raises the marginal benefit of tracking and interpreting such news.
Panel A of Table 5 estimates Equation eq:hetero_general with interactions between MP shocks and (i) our pre-determined attentiveness proxy and (ii) stockholding status. We find a large and statistically significant response only for accurate stockholders: a 1 pp contractionary MP surprise lowers their one-year-ahead inflation expectations on impact by about $$-0.41$$ pp $$(t=-4.57)$$. In contrast, the coefficient is smaller and statistically indistinguishable from zero for accurate non-stockholders, and all coefficients are near zero for the inaccurate groups. The absence of any response among inaccurate stockholders, alongside the strong effect for accurate ones, points to attention--rather than simple selection on unobservable traits--as the operative channel. This pattern maps tightly to Proposition 1 (attention gates pass-through) andProposition 4 (higher-$$\omega$$ types exhibit stronger pass-through).
Ahn and Xie, 2024 independently document that stock-market participation is associated with greater household attentiveness and more accurate inflation beliefs. Using MSC micro data, they show that stockholders are more attentive and hold more accurate inflation beliefs; they update more to macro news than non-holders, and the attention gap widens when uncertainty is high (consistent with a risk-hedging motive). Our finding is complementary along the MP margin: conditioning on a pre-determined attentiveness proxy, the impact pass-through of conventional MP surprises is concentrated among accurate stockholders, whereas inaccurate stockholders do not react--exactly the attention-gating logic of Proposition 1. Quantitatively,this delivers a larger (more negative) slope for stockholders within the Accurate group, an empirical counterpart to Proposition 4 (higher $$\omega$$).
Homeownership
For homeowners, interest-rate movements are directly salient via mortgage payments, refinancing options, and housing wealth, raising the marginal benefit of tracking policy news and plausibly increasing optimal attention $$m_i^{*}$$. This mechanism complements evidence that homeowners are especially sensitive to rate changes through refinancing/payment channels (e.g., Ahn et al., 2024).
Estimating Equation eq:hetero_general with interactions between MP surprises, our pre-determined accuracy indicators, and homeownership status supports these predictions. Panel B of Table 5 shows that, among accurate respondents, a 1 pp contractionary MP surprise reduces one-year-ahead expected inflation by about $$-0.434$$ pp for homeowners ($$t=-5.08$$), whereas renters exhibit no detectable impact response; for the inaccurate groups, coefficients are small and statistically indistinguishable from zero. This sharp contrast provides evidence that the results are driven by the proposed attention channel, rather than by selection on unobservable characteristics correlated with homeownership. The pattern mirrors Proposition 1--attention drives pass-through--and aligns with Proposition 4: conditional on being attentive, the homeowner group (a high-payoff-to-information margin) transmits policy news more strongly into expectations.
In magnitude, the homeowner effect is comparable to the stockholder effect in Panel A, suggesting two complementary margins--portfolio exposure and mortgage-linked exposure--through which higher $$\omega_i$$ amplifies expectation responses when attention is present. Crucially, the prerequisite of accuracy remains central: absent pre-shock attentiveness, neither homeowners nor renters transmit policy news into expected inflation on impact.
Age Group
Age offers another natural partition for the attention sensitivity. Younger and prime-age households have greater labor-market exposure and more high-frequency economic decisions, which plausibly raises $$\omega_i$$; they may also face lower information costs (lower $$\kappa_i$$). Moreover, the personal-experience framework of Malmendier and Nagel, 2016 implies that younger individuals place more weight on recent macro information and thus update beliefs more strongly, whereas older individuals rely more on longer-horizon experience and update less on impact.
Estimating Equation eq:hetero_general with interactions between monetary policy (MP) surprises, our pre-determined accuracy indicators, and age-group status (Young 18-34, Middle 35-64, Old 65+) yields a clear gradient within the accurate group (Panel C of Table 5). Accurate young respondents revise one-year-ahead inflation expectations the most after a 1 pp contractionary MP surprise ($$-0.611$$, $$t=-3.23$$), accurate middle-aged respondents respond less but still significantly ($$-0.349$$, $$t=-3.68$$), and accurate older respondents show a smaller, statistically insignificant coefficient ($$-0.234$$, $$t=-1.22$$). For inaccurate respondents, coefficients are small and indistinguishable from zero across all age groups.
This pattern mirrors Proposition 1: attention gates pass-through, with virtually no impact among the inaccurate. Conditional on being attentive, the magnitude ordering (Young > Middle > Old) is consistent with higher $$\omega_i$$ and/or lower $$\kappa_i$$ for younger/prime-age households, and with the experience-based updating of Malmendier and Nagel,2016, whereby younger individuals place greater weight on recent policy-relevant information. In sum, the age gradient in impact responses provides an additional cross-sectional validation of the model's payoff-based heterogeneity.
Table 6: Attention and Income Quartile in Monetary Policy Pass-Through
| (1) Accurate | (2) Inaccurate | |
|---|---|---|
| (1) $$YTL1\times MPS_{t}$$ | 0.048 (0.18) | -0.192 (-0.58) |
| (2) $$YTL2\times MPS_{t}$$ | -0.669*** (-3.90) | 0.046 (0.19) |
| (3) $$YTL3\times MPS_{t}$$ | -0.361*** (-2.58) | 0.201 (1.04) |
| (4) $$YTL4\times MPS_{t}$$ | -0.298 (-2.47) | 0.132 (0.77) |
| Interaction | Income Quartile | Income Quartile |
| Controls | Yes | Yes |
| Observations | 37,445 | 37,445 |
| $$R^{2}$$ | 0.0153 | 0.0153 |
Income Quartile
Lastly, income offers another natural partition: relative to the bottom quartile, middle- and higher-income households typically have more policy-exposed stakes (labor-market risk, asset portfolios, mortgage/credit margins), which raises $$\omega_i$$ and, in turn, the attention-scaled response $$\vert\theta\,m_i^{*}\vert$$.
Using the MSC income quartiles, we estimate Equation eq:hetero_general with the demographic partition $$\boldsymbol{D}_{i,t}=\{$$YTL1$$,\ldots,$$YTL4$$\}$$ and report results in Table 6.15 The accuracy prerequisite remains first-order: across all quartiles, inaccurate respondents do not react on impact. Within the accurate group, we find a clear gradient: middle-income households (YTL2, YTL3) display the largest and most precisely estimated declines in 1-year-ahead expectations after a 1 pp contractionary MP surprise (-0.667 and -0.360, respectively), high-income households (YTL4) react moderately (-0.297), and the lowest-income quartile (YTL1) shows no detectable impact response. This pattern is consistent with our payoff-based mechanism (higher $$\omega_i$$ outside the bottom quartile) and with the idea that groups whose expenditure baskets load more on energy and other policy-sensitive categories anticipate larger near-term disinflation following a tightening.16
Table 7: Alternative Monetary Shock Measure
Panel A: Bu et al., 2021
| (1) Accurate | (2) Inaccurate | (3) Haven't Heard | |
|---|---|---|---|
| (1) $$MPS_{t}$$ | -1.411*** (-6.66) | -0.256 (-1.19) | -0.505** (-2.14) |
| (2) $$\Delta$$IP$$_{t}$$ | 0.043*** (2.77) | -0.001 (-0.10) | 0.004 (0.24) |
| (3) $$\Delta\pi_{t}$$ | 0.349*** (9.29) | 0.268*** (6.33) | 0.319*** (7.04) |
Panel B: Bauer and Swanson, 2023
| (4) Accurate | (5) Inaccurate | (6) Haven't Heard | |
|---|---|---|---|
| (1) $$MPS_{t}$$ | -1.343*** (-4.46) | -0.535 (-1.57) | -0.898*** (-2.79) |
| (2) $$\Delta$$IP$$_{t}$$ | 0.079*** (3.97) | 0.056** (1.97) | 0.056*** (2.16) |
| (3) $$\Delta\pi_{t}$$ | 0.275*** (7.45) | 0.268*** (6.32) | 0.275*** (6.11) |
| Panel A | Panel B | |
|---|---|---|
| Controls | Yes | Yes |
| Observations | 37,445 | 35,592 |
| $$R^{2}$$ | 0.0148 | 0.0168 |
We assess the robustness of our findings along five dimensions and report full details and tables in Appendix Section C. First, we address concerns that high-frequency (HF) monetary policy surprises may bundle a Fed information-effect component. We thereforere-estimate our baseline specifications using the Bu et al., 2021 "BRW" shocks (Panel A of Table 7). The signs, cross-group ordering, and significance mirror the HF results; magnitudes are somewhat larger under BRW, which is consistent with differences in the mapping from shocks to rates (BRW innovations move 2-year yields nearly one-for-one, whereas HF factors need not). We also verified robustness to the reassessed HF series in Bauer and Swanson, 2023; because that sample endsin 2019:M7, we do not tabulate it, but the core patterns persist (Panel B of Table 7). Second, we vary the construction of Accuracy. Our benchmark measure uses recent changes in the unemployment rate; to check that results do not hinge on this choice, we reclassify Accuracy using two alternative aggregate signals that proxy the real and financial sides of the macro environment: Industrial Production (IP) and the National Financial Conditions Index (NFCI). The baseline gating and heterogeneity patterns are unchanged when we use IP (Appendix Tables in Appendix C.3) or NFCI (Appendix Tables in Appendix C.4) instead of unemployment to define Accuracy.17
Third, we evaluate representativeness by reweighting the micro regressions with household-head weights (wt). Because the recontacted MSC panel in a given month contains at most about 250 respondents, weighting is a natural correction. Weighted regressions (Appendix Table C.18) yield coefficients that are statistically and economically indistinguishable from our baseline, suggesting that small-sample composition does not drive our results.
Fourth, we augment the macro controls to account for the salience of gasoline prices in household belief formation. We add the log change in U.S. Regular All Formulations Gas Price between the two interviews (from FRED) to the baseline controls (replacing crude oil prices used elsewhere). The gating and heterogeneity results are robust to this addition (Appendix Table C.19), indicating that our findings are not an artifact of omitted gasoline-price movements.
Finally, we revisit the state-dependence analysis using an alternative uncertainty proxy. We construct the volatility state from the Economic Policy Uncertainty (EPU) index Baker et al., 2016a--a text-based measure that captures policy-relevant uncertainty spanning both real and financial sources. Defining high-uncertainty months by the cyclical component of EPU and re-estimating the triple-interaction design reproduces our baseline pattern: Accurate respondents load more strongly on contractionary policy news in high-uncertainty states, while Inaccurate respondents do not (Appendix Table C.20).
Across all checks--alternative Accuracy definitions (IP, NFCI), alternative shock measures (BRW, reassessed HF), population weighting, richer price controls, and alternative uncertainty splits (EPU)--the core results remain: attention (Accuracy) mediates pass-through on impact, aggregate pass-through scales with attentiveness, state dependence is stronger for the attentive, and high-payoff groups (stock-holders, homeowners, prime-age, higher-income) display larger effects when accurate.
We develop a minimal behavioral framework in which households optimally choose attention to inflation-relevant news and derive four predictions: attention drives the pass-through of monetary policy to inflation expectations; aggregate pass-through scales with the economy's average attentiveness; pass-through is state dependent and rises with payoff-relevant uncertainty; and, conditional on being attentive, groups with a higher payoff from being informed display stronger effects. Using pre-determined Accuracy, high-frequency identified MP surprises, and both micro and aggregate designs, the data align closely with these predictions. On impact, attentive households revise down expected inflation after contractionary shocks, the aggregate response is larger in high-attentive months, state dependence is concentrated among the attentive, and stockholders, homeowners, prime-age, and higher-income households react more when accurate.
These findings have clear policy and macro implications. Attention acts as an expectations multiplier: when attention is low, policy news barely reaches household beliefs; when high, the same news moves expectations strongly. This provides a microfoundation for why broad-based communications can have limited effects, as a large share of the audience may be in a low-attention state. Our results suggest that the expectations channel is most potent when communications are timed to coincide with periods of high uncertainty or targeted toward high-payoff groups--like homeowners and stockholders--who are endogenously more attentive. The effectiveness of tools like forward guidance is therefore not constant but is likely amplified during turbulent economic times. This uneven transmission, while useful for fast-acting policy, means central banks may confront distributional asymmetries in how expectations are updated. From a macro lens, stronger belief pass-through amplifies the short-run real-rate effect of a given nominal tightening, potentially making conventional MP more powerful in disinflating while sharpening near-term trade-offs.
Our analysis focuses on impact revisions and leaves longer-horizon dynamics and general-equilibrium propagation to future work. Natural next steps include causal manipulation of attention (e.g., information treatments), linking belief updates to spending/refinancing/portfolio behavior, and integrating household and firm attention in a structural model, and studying optimal communication under attention constraints. While our work focuses on monetary policy, the model implies that attention gates responses to any inflation-relevant news; investigating this mechanism for other disturbances, like fiscal or energy shocks, is a fruitful avenue for future research. A companion agenda is to connect time variation in attention inequality to the changing effectiveness of policy over the business cycle.
Afrouzi, H. (2024).
Strategic inattention, inflation dynamics, and the nonneutrality of money.
Journal of Political Economy, 132(10):3378-3420.
Afrouzi, H. and Yang, C. (2021).
Dynamic rational inattention and the phillips curve.
Working Paper 8840, CESifo.
Ahn, H. J. and Xie, S. (2024).
Stock-driven household attention.
Ahn, H. J., Xie, S., and Yang, C. (2024).
Effects of monetary policy on household expectations: The role of homeownership.
Journal of Monetary Economics, 147:103599.
Alpanda, S., Granziera, E., and Zubairy, S. (2021).
State dependence of monetary policy across business, credit and interest rate cycles.
European Economic Review, 140:103936.
Angeletos, G.-M. and Lian, C. (2018).
Forward guidance without common knowledge.
American Economic Review, 108(9):2477-2512.
Bachmann, R., Berg, T. O., and Sims, E. R. (2015).
Inflation expectations and readiness to spend: Cross-sectional evidence.
American Economic Journal: Economic Policy, 7(1):1-35.
Baker, S. R., Bloom, N., and Davis, S. J. (2016a).
Measuring economic policy uncertainty.
Quarterly Journal of Economics, 131(4):1593-1636.
Baker, S. R., Bloom, N., and Davis, S. J. (2016b).
Measuring economic policy uncertainty.
The quarterly journal of economics, 131(4):1593-1636.
Bauer, M. D., Lakdawala, A., and Mueller, P. (2022).
Market-based monetary policy uncertainty.
The Economic Journal, 132(644):1290-1308.
Bauer, M. D., Pflueger, C. E., and Sunderam, A. (2024).
Perceptions about monetary policy.
The Quarterly Journal of Economics, 139(4):2227-2278.
Bauer, M. D. and Swanson, E. T. (2023).
A reassessment of monetary policy surprises and high-frequency identification.
NBER Macroeconomics Annual, 37(1):87-155.
Bordalo, P., Gennaioli, N., and Shleifer, A. (2018).
Diagnostic expectations and credit cycles.
Journal of Finance, 73(1):199-227.
Bracha, A. and Tang, J. (2024).
Inflation levels and (in)attention.
The Review of Economic Studies, 92(3):1564-1594.
Braitsch, H. and Mitchell, J. (2022).
A new measure of consumers' (in)attention to inflation.
Economic Commentary, 2022(14):1-7.
Bu, C., Rogers, J., and Wu, W. (2021).
A unified measure of fed monetary policy shocks.
Journal of Monetary Economics, 118:331-349.
Candia, B., Coibion, O., and Gorodnichenko, Y. (2024).
The inflation expectations of u.s. firms: Evidence from a new survey.
Journal of Monetary Economics, 145.
Supplement issue.
Carvalho, C. and Nechio, F. (2014).
Do people understand monetary policy?
Journal of Monetary Economics, 66:108-123.
Cavallo, A., Cruces, G., and Perez-Truglia, R. (2017).
Inflation expectations, learning, and supermarket prices: Evidence from survey experiments.
American Economic Journal: Macroeconomics, 9(3):1-35.
Claus, E. and Nguyen, V. H. (2020).
Monetary policy shocks from the consumer perspective.
Journal of Monetary Economics, 114:159-173.
Coibion, O. and Gorodnichenko, Y. (2015a).
Information rigidity and the expectations formation process: A simple framework and new facts.
American Economic Review, 105(8):2644-2678.
Coibion, O. and Gorodnichenko, Y. (2015b).
Is the phillips curve alive and well after all? inflation expectations and the missing disinflation.
American Economic Journal: Macroeconomics, 7(1):197-232.
Coibion, O., Gorodnichenko, Y., and Weber, M. (2022).
Monetary policy communications and their effects on household inflation expectations.
Journal of Political Economy, 130(6):1537-1584.
DAcunto, F., Malmendier, U., Ospina, J., and Weber, M. (2021).
Exposure to grocery prices and inflation expectations.
Journal of Political Economy, 129(5):1615-1639.
DAcunto, F., Malmendier, U., and Weber, M. (2023).
What do the data tell us about inflation expectations?
In Handbook of economic expectations, pages 133-161. Elsevier.
Dietrich, A. M. (2024).
Consumption categories, household attention, and inflation expectations: Implications for optimal monetary policy.
Journal of Monetary Economics, 147:103594.
Flynn, J. P. and Sastry, K. (2024).
Attention cycles.
Working Paper 32553, National Bureau of Economic Research.
Gabaix, X. (2020).
A behavioral new keynesian model.
American Economic Review, 110(8):2271-2327.
Jaravel, X. (2019).
The unequal gains from product innovations: Evidence from the us retail sector.
The Quarterly Journal of Economics, 134(2):715-783.
Korenok, O., Munro, D., and Chen, J. (2023).
Inflation and attention thresholds.
The Review of Economics and Statistics, pages 1-28.
Kroner, N. (2025).
How markets process macro news: The importance of investor attention.
Finance and Economics Discussion Series 2025-022, Board of Governors of the Federal Reserve System.
Kryvtsov, O. and Petersen, L. (2021).
Central bank communication that works: Lessons from lab experiments.
Journal of Monetary Economics, 117:760-780.
Kumar, S., Coibion, O., Afrouzi, H., and Gorodnichenko, Y. (2015).
Inflation targeting does not anchor inflation expectations: Evidence from firms in new zealand.
Brookings Papers on Economic Activity, pages 151-208.
Lamla, M. J. and Vinogradov, D. V. (2019).
Central bank announcements: Big news for little people?
Journal of Monetary Economics, 108:21-38.
Ludvigson, S. C., Ma, S., and Ng, S. (2021).
Uncertainty and business cycles: exogenous impulse or endogenous response?
American Economic Journal: Macroeconomics, 13(4):369-410.
kowiak and Wiederholt, 2009] --> Ma$$\large{\'{c\/}}$$kowiak, B. and Wiederholt, M. (2009).
Optimal sticky prices under rational inattention.
American Economic Review, 99(3):769-803.
Malmendier, U. and Nagel, S. (2016).
Learning from inflation experiences.
The Quarterly Journal of Economics, 131(1):53-87.
Mangiante, G. and Lauper, C. (Forthcoming).
Monetary policy shocks and inflation inequality.
International Journal of Central Banking.
Mankiw, N. G. and Reis, R. (2002).
Sticky information versus sticky prices: A proposal to replace the new keynesian phillips curve.
Quarterly Journal of Economics, 117(4):1295-1328.
Marcellino, M. and Stevanovic, D. (2022).
The demand and supply of information about inflation.
Working Papers 22-06, University of Quebec in Montreal's School of Management.
Nakamura, E. and Steinsson, J. (2018).
High-frequency identification of monetary non-neutrality: the information effect.
The Quarterly Journal of Economics, 133(3):1283-1330.
Pfäuti, O. (2024).
The inflation attention threshold and inflation surgers.
Working Papers 2024-02, EMPCT Working Paper Series.
Romer, C. D. and Romer, D. H. (2004).
A new measure of monetary shocks: Derivation and implications.
American economic review, 94(4):1055-1084.
Sims, C. A. (2003).
Implications of rational inattention.
Journal of Monetary Economics, 50(3):665-690.
Song, W. and Stern, S. (2024).
Firm inattention and the efficacy of monetary policy: A text-based approach.
The Review of Economic Studies, page rdae102.
Tenreyro, S. and Thwaites, G. (2016).
Pushing on a string: Us monetary policy is less powerful in recessions.
American Economic Journal: Macroeconomics, 8(4):43-74.
Vavra, J. (2014).
Inflation dynamics and time-varying volatility: New evidence and an ss interpretation.
The Quarterly Journal of Economics, 129(1):215-258.
Weber, M., Candia, B., Afrouzi, H., Ropele, T., Lluberas, R., Frache, S., Meyer, B., Kumar, S., Gorodnichenko, Y., Georgarakos, D., Coibion, O., Kenny, G., and Ponce, J. (2025).
Tell me something i don't already know: Learning in low- and high-inflation settings.
Econometrica, 93(1):229-264.
Wu, E. (2024).
Rationally inattentive seller: The geographic dispersion of u.s. retail chains and its aggregate implications.
Technical report, Kilts Center at Chicago Booth Marketing Data Center.
Yang, C. (2022).
Rational inattention, menu costs, and multi-product firms: Micro evidence and aggregate implications.
Journal of Monetary Economics, 128:105-123.
APPENDIX
This appendix provides the formal mathematical derivations for the four main propositions presented in the theoretical framework of Section 2. It details the steps for deriving the impact of monetary policy on individual expectations (Proposition 1) , the scaling of aggregate pass-through with attention (Proposition 2), the state-dependent nature of the response to uncertainty (Proposition 3), and the cross-sectionalpredictions based on payoff heterogeneity (Proposition 4) .
Combine Equation eq:fundamentals-Equation eq:behavioral_operator evaluated at the optimum $$m_{i,t}^*(U_t)$$:
Start from the individual impact change,
Let $$S_g(U)\equiv \partial[m_g(U)\theta]/\partial U = \theta \cdot \frac{m_g(U)[1-m_g(U)]}{U}$$ for group $$g$$. For any $$U>0$$, if $$m_A(1-m_A) > m_I(1-m_I)$$, then $$\vert S_A(U)\vert>\vert S_I(U)\vert$$ because $$\vert\theta\vert$$ and $$U$$ cancel in the comparison. A sufficient condition is $$m_A\in(\frac12,1)$$ and $$m_I\in(0,\frac12)$$ since $$f(m)=m(1-m)$$ is strictly increasing on $$[0,\frac12]$$ and strictly decreasing on $$[\frac12,1]$$ with maximum at $$m=\frac12$$.
Fix $$U_t>0$$. From Equation eq:mstar,
(ii) Pass-through ordering. The individual MP pass-through magnitude is $$\left\vert\partial\,\Delta\pi_{i,t+1}/\partial \varepsilon^{mp}_t\right\vert = \vert\theta\vert\,m^{*}_{i,t}(U_t)$$ by Proposition 1. Monotonicity then follows from part (i).
(iii) Selection into "attentive/accurate". For any threshold $$\tau\in(0,1)$$, $$A_{i,t}=\mathbf{1}\{m^{*}_{i,t}\ge \tau\}$$ is nondecreasing in $$\omega_i$$ and nonincreasing in $$\kappa_i$$ because $$m^{*}_{i,t}$$ is monotone in those parameters.
(iv) Conditional ordering within the attentive group. On $$\{A_{i,t}=1\}$$ we have $$m^{*}_{i,t}\ge\tau$$. Since $$m^{*}_{i,t}$$ is increasing in $$\omega_i$$ and decreasing in $$\kappa_i$$ pointwise, any upward (first-order) shift in $$\omega$$ or downward shift in $$\kappa$$ raises $$m^{*}_{i,t}$$ for every individual, and thus raises $$\mathbb{E}[m^{*}_{i,t}\mid A_{i,t}=1]$$ whenever the support above $$\tau$$ has positive measure. $$\qedsymbol$$
(v) State dependence by payoff type. Using Equation eq:mstar,
This appendix shows that our four testable implications do not rely on the baseline choice of a linear attention weight or quadratic attention costs. The first subsection establishes a general comparative-statics result (Lemma 5) for an arbitrary increasing attention mapping $$\phi(\cdot)$$ and strictly convex cost $$\psi_i(\cdot)$$: the optimal attention $$m_i^*$$ is unique, increases with payoff-relevant news variance $$U_t$$ and stakes $$\omega_i$$, and decreases with costs $$\kappa_i$$. The linear/quadratic specification follows as a corollary. The second subsection introduces a common noisy public signal observed before attention is chosen and shows that it synchronizes attention choices--micro-founding time variation in the aggregate attentiveness index--while leaving the individual gating, aggregate scaling, uncertainty amplification, and payoff-heterogeneity predictions unchanged.
We show that the main comparative statics do not rely on a linear attention weight or quadratic costs.
With mean-squared forecast loss, the per-period objective can be written (up to a positive multiplicative constant) as
| $$\displaystyle \omega_i U_t\,\big(1-\phi(m_i^*)\big)\phi'(m_i^*)\;=\;\psi_i'(m_i^*).$$ | (B.1) |
Implications. Replacing $$m_i^*$$ by $$\phi(m_i^*)$$ in the impact coefficient delivers the same four predictions: (i) individual gating (only attentive types load on policy news), (ii) aggregate scaling by $$E[\phi(m_i^*)]$$, (iii) amplification when $$U_t$$ is higher, and (iv) larger pass-through for high-$$\omega_i$$/low-$$\kappa_i$$ groups.
This appendix shows that introducing a common, noisy public signal $$s_t$$ that arrives before attention choices does not alter the four testable implications in the main text: (i) individual gating; (ii) aggregate scaling; (iii) uncertainty amplification; and (iv) payoff heterogeneity. The public signal provides a simple micro-foundation for time-variation in aggregate attentiveness by synchronizing attention choices across households.
Suppose the public signal is about the fully informed forecast of next-period inflation $$\pi^*_{t+1}$$ (e.g., a highly publicized data release or headline), observed before attention choice. Let $$s_t$$ be informative about $$\pi^*_{t+1}$$ so that the posterior variance
Given $$s_t$$ (and $$\tau_s$$), the household chooses attention to minimize
Discussion. A level signal reduces residual uncertainty and thereby lowers the marginal value of costly attention, but conditional on the chosen attention, the pass-through of monetary policy news is still multiplied by the attention weight. Since $$s_t$$ is common, its level effect on beliefs is absorbed by time variation (e.g., month fixed effects) in our empirical designs; the estimated slope with respect to policy surprises is therefore unaffected.
This appendix contains the complete regression tables that support the robustness analysis discussed in Section 5 . It includes detailed output from the state-dependence and demographic heterogeneity analyses, as well as a comprehensive set of checks using alternative definitions for the accuracy proxy (based on Industrial Production and the NFCI), alternative monetary policy shock measures, population weighting, and additional controls.
This section provides the complete regression output for the state-dependence analysis presented in Section 4.3. The tables report the full set of coefficients, including those for the "Haven't Heard" group and contemporaneous macro controls, for specifications using NBER recessions , the LMN real uncertainty index , and the VIX to define high- and low-uncertainty states.
Appendix Table C.1: Attention with NBER Business Cycle Indicator
Panel A: NBER Recession
| NBER Recession | (1) Accurate | (2) Inaccurate | (3) Haven't Heard |
|---|---|---|---|
| (1) $$Recession\times MPS_{t}$$ | -1.701*** (-4.01) | -1.125 (-1.00) | -0.988 (-1.29) |
| (2) $$Recession\times \Delta$$IP$$_{t}$$ | 0.200*** (5.19) | 0.084 (0.87) | 0.120* (1.68) |
| (3) $$Recession\times \Delta\pi_{t}$$ | 0.303* (3.29) | 0.258 (1.32) | 0.243 (1.52) |
Panel B: Normal
| NBER Recession | (1) Accurate | (2) Inaccurate | (3) Haven't Heard |
|---|---|---|---|
| (4) $$Normal\times MPS_{t}$$ | -0.039 (-0.49) | 0.115 (1.12) | -0.123 (-1.43) |
| (5) $$Normal\times\Delta$$IP$$_{t}$$ | -0.028 (-1.41) | -0.018 (-1.00) | -0.021 (-1.09) |
| (6) $$Normal\times\Delta\pi_{t}$$ | 0.332*** (8.17) | 0.269*** (6.17) | 0.275* (5.86) |
| NBER Recession | (1) Accurate | (2) Inaccurate | (3) Haven't Heard |
|---|---|---|---|
| Interaction | NBER | NBER | NBER |
| Controls | Yes | Yes | Yes |
| Observations | 37,445 | 37,445 | 37,445 |
| $$R^{2}$$ | 0.0170 | 0.0170 | 0.0170 |
Appendix Table C.2: Attention with Real Uncertainty Indicator
Panel A: High Uncertainty
| LMN Real Uncertainty | (1) Accurate | (2) Inaccurate | (3) Haven't Heard |
|---|---|---|---|
| (1) $$High\times MPS_{t}$$ | -0.539*** (-5.51) | 0.048 (0.35) | -0.137 (-1.18) |
| (2) $$High\times \Delta$$IP$$_{t}$$ | 0.130*** (4.37) | -0.003 (-0.09) | 0.022 (0.66) |
| (3) $$High\times \Delta\pi_{t}$$ | 0.304*** (5.29) | 0.137* (1.95) | 0.324*** (4.38) |
Panel B: Low Uncertainty
| LMN Real Uncertainty | (1) Accurate | (2) Inaccurate | (3) Haven't Heard |
|---|---|---|---|
| (4) $$Low\times MPS_{t}$$ | -0.269* (-1.77) | 0.250 (1.33) | -0.248 (-1.49) |
| (5) $$Low\times \Delta$$IP$$_{t}$$ | 0.034 (1.62) | -0.012 (-0.63) | 0.008 (0.37) |
| (6) $$Low\times \Delta\pi_{t}$$ | 0.381*** (7.54) | 0.397*** (7.73) | 0.310* (5.76) |
| LMN Real Uncertainty | (1) Accurate | (2) Inaccurate | (3) Haven't Heard |
|---|---|---|---|
| Interaction | LMN real uncertainty | LMN real uncertainty | LMN real uncertainty |
| Controls | Yes | Yes | Yes |
| Observations | 37,445 | 37,445 | 37,445 |
| $$R^{2}$$ | 0.0146 | 0.0146 | 0.0146 |
Appendix Table C.3: Attention with the VIX
Panel A: High Volatility
| VIX | (1) Accurate | (2) Inaccurate | (3) Haven't Heard |
|---|---|---|---|
| (1) $$High\times MPS_{t}$$ | -0.456*** (-4.06) | 0.040 (0.22) | -0.098 (-0.70) |
| (2) $$High\times \Delta$$IP$$_{t}$$ | 0.078*** (2.73) | 0.074 (1.96) | -0.000 (-0.01) |
| (3) $$High\times \Delta\pi_{t}$$ | -0.011 (-0.17) | 0.141* (1.92) | 0.137* (1.83) |
Panel B: Low Volatility
| VIX | (1) Accurate | (2) Inaccurate | (3) Haven't Heard |
|---|---|---|---|
| (4) $$Low\times MPS_{t}$$ | -0.007 (-0.07) | 0.100 (0.79) | -0.179 (-1.45) |
| (5) $$Low\times \Delta$$IP$$_{t}$$ | 0.020 (0.97) | -0.027 (-1.40) | 0.007 (0.35) |
| (6) $$Low\times \Delta\pi_{t}$$ | 0.538*** (11.82) | 0.338*** (6.51) | 0.413* (7.37) |
| VIX | (1) Accurate | (2) Inaccurate | (3) Haven't Heard |
|---|---|---|---|
| Interaction | VIX | VIX | VIX |
| Controls | Yes | Yes | Yes |
| Observations | 37,445 | 37,445 | 37,445 |
| $$R^{2}$$ | 0.0182 | 0.0182 | 0.0182 |
This section presents the full regression tables corresponding to the demographic heterogeneity analysis in Section 4.4. Each table details the complete set of interaction coefficients for the partitions based on stockholding, homeownership, age group, and income quartile, including results for all three accuracy groups and macro control variables.
Appendix Table C.4: Full reports for Stockholding
| (1) Accurate | (2) Inaccurate | (3) Haven't Heard | |
|---|---|---|---|
| (1) $$Stock\times MPS_{t}$$ | -0.410*** (-4.57) | 0.150 (1.20) | -0.299** (-2.38) |
| (2) $$NonStock\times MPS_{t}$$ | -0.228 (-1.42) | -0.047 (-0.21) | 0.034 (0.24) |
| (3) $$Stock\times\Delta$$IP$$_{t}$$ | 0.053*** (2.84) | 0.001 (0.09) | 0.020 (0.91) |
| (4) $$NonStock\times\Delta$$IP$$_{t}$$ | 0.090*** (2.35) | -0.049 (-1.22) | 0.004 (0.11) |
| (5) $$Stock\times\Delta\pi_{t}$$ | 0.394*** (9.67) | 0.258*** (5.92) | 0.377*** (6.89) |
| (6) $$NonStock\times\Delta\pi_{t}$$ | 0.292*** (3.18) | 0.318*** (2.86) | 0.241*** (3.06) |
| Interaction | Stockholding | Stockholding | Stockholding |
| Controls | Yes | Yes | Yes |
| Observations | 37,445 | 37,445 | 37,445 |
| $$R^{2}$$ | 0.0142 | 0.0142 | 0.0142 |
Appendix Table C.5: Full reports for Homeownership
| (1) Accurate | (2) Inaccurate | (3) Haven't Heard | |
|---|---|---|---|
| (1) $$Homeowner\times MPS_{t}$$ | -0.436*** (-5.08) | 0.063 (0.54) | -0.148 (-1.40) |
| (2) $$Renter\times MPS_{t}$$ | 0.026 (0.13) | 0.214 (0.76) | -0.181 (-0.91) |
| (3) $$Homeowner\times\Delta$$IP$$_{t}$$ | 0.057*** (3.11) | -0.0000 (-0.00) | 0.027 (1.22) |
| (4) $$Renter\times\Delta$$IP$$_{t}$$ | 0.075* (1.88) | -0.032 (-0.81) | -0.024 (-0.67) |
| (5) $$Homeowner\times\Delta\pi_{t}$$ | 0.386*** (9.61) | 0.286*** (6.43) | 0.334*** (6.61) |
| (6) $$Renter\times\Delta\pi_{t}$$ | 0.297*** (2.67) | 0.166 (1.30) | 0.276* (2.73) |
| Interaction | Homeownership | Homeownership | Homeownership |
| Controls | Yes | Yes | Yes |
| Observations | 37,445 | 37,445 | 37,445 |
| $$R^{2}$$ | 0.0144 | 0.0144 | 0.0144 |
Appendix Table C.6: Full reports for Age Group
| (1) Accurate | (2) Inaccurate | (3) Haven't Heard | |
|---|---|---|---|
| (1) $$[18-34]\times MPS_{t}$$ | -0.613*** (-3.22) | 0.260 (0.82) | -0.006 (-0.03) |
| (2) $$[35-64]\times MPS_{t}$$ | -0.350*** (-3.68) | 0.140 (1.12) | -0.145 (-1.16) |
| (3) $$[65+]\times MPS_{t}$$ | -0.234 (-1.22) | -0.264 (-0.98) | -0.338* (-1.66) |
| (4) $$[18-34]\times\Delta$$IP$$_{t}$$ | 0.110** (2.43) | 0.031 (0.72) | 0.004 (0.10) |
| (5) $$[35-64]\times\Delta$$IP$$_{t}$$ | 0.059*** (2.95) | -0.019 (-0.84) | 0.015 (0.57) |
| (6) $$[65+]\times\Delta$$IP$$_{t}$$ | 0.038 (1.01) | 0.004 (0.13) | 0.022 (0.60) |
| (7) $$[18-34]\times\Delta\pi_{t}$$ | 0.381*** (3.68) | 0.033 (0.27) | 0.206** (2.07) |
| (8) $$[35-64]\times\Delta\pi_{t}$$ | 0.435*** (9.41) | 0.328*** (6.17) | 0.373*** (5.94) |
| (9) $$[65+]\times\Delta\pi_{t}$$ | 0.179** (2.20) | 0.243*** (2.92) | 0.311*** (3.66) |
| Interaction | Age Group | Age Group | Age Group |
| Controls | Yes | Yes | Yes |
| Observations | 37,445 | 37,445 | 37,445 |
| $$R^{2}$$ | 0.0150 | 0.0150 | 0.0150 |
Appendix Table C.7: Full Reports for Income Quartile
| (1) Accurate | (2) Inaccurate | (3) Haven't Heard | |
|---|---|---|---|
| (1) $$YTL1\times MPS_{t}$$ | 0.048 (0.18) | -0.192 (-0.58) | 0.149 (0.71) |
| (2) $$YTL2\times MPS_{t}$$ | -0.669*** (-3.90) | 0.046 (0.19) | -0.212 (-1.12) |
| (3) $$YTL3\times MPS_{t}$$ | -0.361*** (-2.58) | 0.201 (1.04) | -0.119 (-0.77) |
| (4) $$YTL4\times MPS_{t}$$ | -0.298** (-2.47) | 0.132 (0.77) | -0.416** (-2.07) |
| (5) $$YTL1\times\Delta$$IP$$_{t}$$ | 0.024 (0.48) | -0.043 (-0.93) | -0.059 (-1.37) |
| (6) $$YTL2\times\Delta$$IP$$_{t}$$ | 0.142*** (3.43) | -0.062 (-1.49) | 0.017 (0.49) |
| (7) $$YTL3\times\Delta$$IP$$_{t}$$ | 0.024 (0.82) | -0.018 (-0.59) | 0.020 (0.56) |
| (8) $$YTL4\times\Delta$$IP$$_{t}$$ | 0.054** (2.26) | 0.047* (1.80) | 0.069* (1.93) |
| (9) $$YTL1\times\Delta\pi_{t}$$ | 0.427*** (3.30) | 0.220 (1.47) | 0.282*** (2.68) |
| (10) $$YTL2\times\Delta\pi_{t}$$ | 0.265*** (3.16) | 0.313*** (3.12) | 0.271*** (3.12) |
| (11) $$YTL3\times\Delta\pi_{t}$$ | 0.438*** (6.51) | 0.307*** (4.35) | 0.414*** (5.37) |
| (12) $$YTL4\times\Delta\pi_{t}$$ | 0.352*** (6.16) | 0.238*** (3.88) | 0.329* (3.51) |
| Interaction | Income Quartile | Income Quartile | Income Quartile |
| Controls | Yes | Yes | Yes |
| Observations | 37,445 | 37,445 | 37,445 |
| $$R^{2}$$ | 0.0153 | 0.0153 | 0.0153 |
This section tests the robustness of our findings to an alternative definition of the accuracy proxy. Here, we reconstruct the "Accurate" and "Inaccurate" classifications using the three-month change in Industrial Production (IP) instead of the unemployment rate.
Appendix Table C.8: Accuracy Measure with Industrial Production
| (1) Accurate | (2) Inaccurate | (3) Haven't Heard | |
|---|---|---|---|
| (1) $$MPS_{t}$$ | -0.334*** (-3.84) | -0.044 (-0.47) | -0.156* (-1.67) |
| (2) $$\Delta$$IP$$_{t}$$ | 0.053*** (3.32) | -0.003 (-0.16) | 0.013 (0.71) |
| (3) $$\Delta\pi_{t}$$ | 0.336*** (8.26) | 0.351*** (8.90) | 0.325* (7.20) |
| Controls | Yes | Yes | Yes |
| Observations | 37,445 | 37,445 | 37,445 |
| $$R^{2}$$ | 0.0135 | 0.0135 | 0.0135 |
Appendix Table C.9: IP Specification for Stockholding
| (1) Accurate | (2) Inaccurate | (3) Haven't Heard | |
|---|---|---|---|
| (1) $$Stock\times MPS_{t}$$ | -0.398*** (-4.02) | -0.002 (-0.02) | -0.299** (-2.38) |
| (2) $$NonStock\times MPS_{t}$$ | -0.158 (-0.89) | -0.154 (-0.80) | 0.033 (0.23) |
| (3) Stock $$\times\Delta$$IP$$_{t}$$ | 0.053*** (3.07) | 0.0007 (0.03) | 0.020 (0.93) |
| (4) Non-stock $$\times\Delta$$IP$$_{t}$$ | 0.055 (1.47) | -0.018 (-0.45) | 0.004 (0.11) |
| (5) Stock $$\times\Delta\pi_{t}$$ | 0.352*** (8.10) | 0.351*** (8.50) | 0.377*** (6.89) |
| (6) Non-stock $$\times\Delta\pi_{t}$$ | 0.284*** (2.79) | 0.351*** (3.54) | 0.241* (3.06) |
| Interaction | Stockownership | Stockownership | Stockownership |
| Controls | Yes | Yes | Yes |
| Observations | 37,445 | 37,445 | 37,445 |
| $$R^{2}$$ | 0.0138 | 0.0138 | 0.0138 |
Appendix Table C.10: IP Specification for Homeownership
| (1) Accurate | (2) Inaccurate | (3) Haven't Heard | |
|---|---|---|---|
| (1) $$Homeowner\times MPS_{t}$$ | -0.361*** (-3.79) | -0.186* (-1.83) | -0.149 (-1.40) |
| (2) $$Renter\times MPS_{t}$$ | -0.185 (-0.89) | 0.660** (2.56) | -0.182 (-0.91) |
| (3) $$Homeowner\times\Delta$$IP$$_{t}$$ | 0.051*** (2.95) | 0.009 (0.43) | 0.027 (1.23) |
| (4) $$Renter\times\Delta$$IP$$_{t}$$ | 0.061 (1.64) | -0.060 (-1.47) | -0.024 (-0.67) |
| (5) $$Homeowner\times\Delta\pi_{t}$$ | 0.353*** (8.31) | 0.347*** (8.12) | 0.334*** (6.61) |
| (6) $$Renter\times\Delta\pi_{t}$$ | 0.246* (1.89) | 0.369*** (3.56) | 0.275*** (2.73) |
| Interaction | Homeownership | Homeownership | Homeownership |
| Controls | Yes | Yes | Yes |
| Observations | 37,445 | 37,445 | 37,445 |
| $$R^{2}$$ | 0.0142 | 0.0142 | 0.0142 |
Appendix Table C.11: IP Specification for Age Group
| (1) Accurate | (2) Inaccurate | (3) Haven't Heard | |
|---|---|---|---|
| (1) $$[18-34]\times MPS_{t}$$ | -0.439** (-1.99) | -0.236 (-0.95) | -0.007 (-0.04) |
| (2) $$[35-64]\times MPS_{t}$$ | -0.327*** (-3.14) | 0.003 (0.03) | -0.145 (-1.16) |
| (3) $$[65+]\times MPS_{t}$$ | -0.322 (-1.53) | -0.032 (-0.14) | -0.338* (-1.66) |
| (4) $$[18-34]\times\Delta$$IP$$_{t}$$ | 0.106** (2.52) | 0.018 (0.42) | 0.004 (0.11) |
| (5) $$[35-64]\times\Delta$$IP$$_{t}$$ | 0.057*** (2.84) | -0.020 (-0.87) | 0.015 (0.58) |
| (6) $$[65+]\times\Delta$$IP$$_{t}$$ | 0.015 (0.48) | 0.022 (0.53) | 0.023 (0.61) |
| (7) $$[18-34]\times\Delta\pi_{t}$$ | 0.304*** (2.70) | 0.201* (1.74) | 0.206** (2.07) |
| (8) $$[35-64]\times\Delta\pi_{t}$$ | 0.386*** (7.68) | 0.426*** (8.73) | 0.373*** (5.94) |
| (9) $$[65+]\times\Delta\pi_{t}$$ | 0.212** (2.49) | 0.227*** (2.83) | 0.311*** (3.65) |
| Interaction | Age Group | Age Group | Age Group |
| Controls | Yes | Yes | Yes |
| Observations | 37,445 | 37,445 | 37,445 |
| $$R^{2}$$ | 0.0144 | 0.0144 | 0.0144 |
Appendix Table C.12: IP Specification for Income Quartile
| (1) Accurate | (2) Inaccurate | (3) Haven't Heard | |
|---|---|---|---|
| (1) $$YTL1\times MPS_{t}$$ | 0.009 (0.03) | 0.018 (0.06) | 0.148 (0.71) |
| (2) $$YTL2\times MPS_{t}$$ | -0.595*** (-3.18) | -0.227 (-1.03) | -0.213 (-1.13) |
| (3) $$YTL3\times MPS_{t}$$ | -0.248 (-1.59) | -0.145 (-0.88) | -0.119 (-0.77) |
| (4) $$YTL4\times MPS_{t}$$ | -0.358*** (-2.69) | 0.131 (0.89) | -0.416** (-2.08) |
| (5) $$YTL1\times\Delta$$IP$$_{t}$$ | 0.008 (0.19) | -0.066 (-1.27) | -0.059 (-1.37) |
| (6) $$YTL2\times\Delta$$IP$$_{t}$$ | 0.084** (2.25) | 0.014 (0.30) | 0.017 (0.50) |
| (7) $$YTL3\times\Delta$$IP$$_{t}$$ | 0.031 (1.08) | -0.023 (-0.70) | 0.020 (0.56) |
| (8) $$YTL4\times\Delta$$IP$$_{t}$$ | 0.068*** (2.93) | 0.031 (1.10) | 0.069* (1.93) |
| (9) $$YTL1\times\Delta\pi_{t}$$ | 0.313** (2.11) | 0.424*** (3.33) | 0.282*** (2.68) |
| (10) $$YTL2\times\Delta\pi_{t}$$ | 0.297*** (3.25) | 0.345*** (3.75) | 0.271*** (3.12) |
| (11) $$YTL3\times\Delta\pi_{t}$$ | 0.375*** (5.31) | 0.389*** (5.60) | 0.414*** (5.36) |
| (12) $$YTL4\times\Delta\pi_{t}$$ | 0.333*** (5.45) | 0.295*** (5.15) | 0.329* (3.51) |
| Interaction | Income Quartile | Income Quartile | Income Quartile |
| Controls | Yes | Yes | Yes |
| Observations | 37,445 | 37,445 | 37,445 |
| $$R^{2}$$ | 0.0148 | 0.0148 | 0.0148 |
This section provides a further robustness check on the construction of our accuracy proxy. We redefine accuracy using the three-month change in the National Financial Conditions Index (NFCI) as the benchmark, where a rising index signals unfavorable conditions.
Appendix Table C.13: Accuracy Measure with NFCI
| (1) Accurate | (2) Inaccurate | (3) Haven't Heard | |
|---|---|---|---|
| (1) $$MPS_{t}$$ | -0.336*** (-3.88) | -0.002 (-0.03) | -0.155* (-1.66) |
| (2) $$\Delta$$IP$$_{t}$$ | 0.039** (2.46) | 0.021 (1.14) | 0.013 (0.70) |
| (3) $$\Delta\pi_{t}$$ | 0.315*** (7.98) | 0.375*** (8.95) | 0.325* (7.21) |
| Controls | Yes | Yes | Yes |
| Observations | 37,445 | 37,445 | 37,445 |
| $$R^{2}$$ | 0.0135 | 0.0135 | 0.0135 |
Appendix Table C.14: NFCI Specification for Stockholding
| (1) Accurate | (2) Inaccurate | (3) Haven't Heard | |
|---|---|---|---|
| (1) $$Stock\times MPS_{t}$$ | -0.340*** (-3.51) | -0.028 (-0.26) | -0.299** (-2.38) |
| (2) $$NonStock\times MPS_{t}$$ | -0.336* (-1.74) | 0.069 (0.40) | 0.033 (0.24) |
| (3) $$Stock\times\Delta$$IP$$_{t}$$ | 0.038** (2.20) | 0.026 (1.23) | 0.020 (0.92) |
| (4) $$Non-stock \times \Delta$$IP$$_{t}$$ | 0.046 (1.17) | 0.003 (0.09) | 0.004 (0.11) |
| (5) $$Stock\times\Delta\pi_{t}$$ | 0.308*** (7.43) | 0.390*** (8.62) | 0.377*** (6.89) |
| (6) $$Non-stock \times \Delta\pi_{t}$$ | 0.341*** (3.28) | 0.329*** (3.33) | 0.241* (3.06) |
| Interaction | Stockownership | Stockownership | Stockownership |
| Controls | Yes | Yes | Yes |
| Observations | 37,445 | 37,445 | 37,445 |
| $$R^{2}$$ | 0.0137 | 0.0137 | 0.0137 |
Appendix Table C.15: NFCI Specification for Homeownership
| (1) Accurate | (2) Inaccurate | (3) Haven't Heard | |
|---|---|---|---|
| (1) $$Homeowner\times MPS_{t}$$ | -0.430*** (-4.65) | -0.020 (-0.20) | -0.148 (-1.40) |
| (2) $$Renter\times MPS_{t}$$ | 0.170 (0.71) | 0.086 (0.43) | -0.181 (-0.91) |
| (3) $$Homeowner\times\Delta$$IP$$_{t}$$ | 0.046*** (2.71) | 0.021 (1.02) | 0.027 (1.22) |
| (4) $$Renter\times\Delta$$IP$$_{t}$$ | 0.005 (0.15) | 0.024 (0.64) | -0.024 (-0.67) |
| (5) $$Homeowner\times\Delta\pi_{t}$$ | 0.317*** (7.84) | 0.395*** (8.73) | 0.334*** (6.61) |
| (6) $$Renter\times\Delta\pi_{t}$$ | 0.296** (2.19) | 0.258** (2.32) | 0.275*** (2.73) |
| Interaction | Homeownership | Homeownership | Homeownership |
| Controls | Yes | Yes | Yes |
| Observations | 37,445 | 37,445 | 37,445 |
| $$R^{2}$$ | 0.0141 | 0.0141 | 0.0141 |
Appendix Table C.16: NFCI Specification for Age Group
| (1) Accurate | (2) Inaccurate | (3) Haven't Heard | |
|---|---|---|---|
| (1) $$[18-34]\times MPS_{t}$$ | -0.167 (-0.68) | -0.379* (-1.83) | -0.006 (-0.04) |
| (2) $$[35-64]\times MPS_{t}$$ | -0.363*** (-3.50) | 0.068 (0.34) | -0.145 (-1.16) |
| (3) $$[65+]\times MPS_{t}$$ | -0.372* (-1.81) | 0.079 (0.34) | -0.338* (-1.66) |
| (4) $$[18-34]\times\Delta$$IP$$_{t}$$ | 0.059 (1.40) | 0.084* (1.86) | 0.004 (0.10) |
| (5) $$[35-64]\times\Delta$$IP$$_{t}$$ | 0.031 (1.57) | 0.026 (1.14) | 0.015 (0.57) |
| (6) $$[65+]\times\Delta$$IP$$_{t}$$ | 0.056* (1.72) | -0.024 (-0.61) | 0.022 (0.60) |
| (7) $$[18-34]\times\Delta\pi_{t}$$ | 0.046 (0.41) | 0.440*** (3.84) | 0.206** (2.07) |
| (8) $$[35-64]\times\Delta\pi_{t}$$ | 0.418*** (8.34) | 0.391*** (7.72) | 0.373*** (5.94) |
| (9) $$[65+]\times\Delta\pi_{t}$$ | 0.145** (1.96) | 0.287*** (3.12) | 0.311* (3.65) |
| Interaction | Age Group | Age Group | Age Group |
| Controls | Yes | Yes | Yes |
| Observations | 37,445 | 37,445 | 37,445 |
| $$R^{2}$$ | 0.0147 | 0.0147 | 0.0147 |
Appendix Table C.17: NFCI Specification for Income Quartile
| (1) Accurate | (2) Inaccurate | (3) Haven't Heard | |
|---|---|---|---|
| (1) $$YTL1\times MPS_{t}$$ | 0.118 (0.40) | -0.155 (-0.54) | 0.148 (0.71) |
| (2) $$YTL2\times MPS_{t}$$ | -0.563*** (-2.81) | -0.271 (-1.40) | -0.212 (-1.13) |
| (3) $$YTL3\times MPS_{t}$$ | -0.418*** (-2.75) | 0.240 (1.44) | -0.119 (-0.77) |
| (4) $$YTL4\times MPS_{t}$$ | -0.271** (-2.03) | 0.024 (0.17) | -0.416** (-2.07) |
| (5) $$YTL1\times\Delta$$IP$$_{t}$$ | -0.003 (-0.08) | -0.025 (-0.49) | -0.059 (-1.37) |
| (6) $$YTL2\times\Delta$$IP$$_{t}$$ | 0.040 (1.14) | 0.067 (1.36) | 0.017 (0.49) |
| (7) $$YTL3\times\Delta$$IP$$_{t}$$ | 0.013 (0.43) | 0.003 (0.10) | 0.020 (0.56) |
| (8) $$YTL4\times\Delta$$IP$$_{t}$$ | 0.076*** (3.11) | 0.026 (1.00) | 0.069* (1.92) |
| (9) $$YTL1\times\Delta\pi_{t}$$ | 0.490*** (3.14) | 0.273** (2.17) | 0.282*** (2.68) |
| (10) $$YTL2\times\Delta\pi_{t}$$ | 0.268*** (3.16) | 0.343*** (3.51) | 0.271*** (3.11) |
| (11) $$YTL3\times\Delta\pi_{t}$$ | 0.319*** (4.94) | 0.464*** (6.09) | 0.414*** (5.37) |
| (12) $$YTL4\times\Delta\pi_{t}$$ | 0.277*** (4.64) | 0.352*** (5.66) | 0.329* (3.51) |
| Interaction | Income Quartile | Income Quartile | Income Quartile |
| Controls | Yes | Yes | Yes |
| Observations | 37,445 | 37,445 | 37,445 |
| $$R^{2}$$ | 0.0151 | 0.0151 | 0.0151 |
This section contains a set of robustness checks. We re-estimate our baseline micro-level specification using household-head weights to ensure population representativeness, add controls for gasoline price changes to account for their salience, and use the Economic Policy Uncertainty (EPU) index as an alternative measure for the state-dependence analysis.
Appendix Table C.18: Household Head Weight
| (1) Accurate | (2) Inaccurate | (3) Haven't Heard | |
|---|---|---|---|
| (1) $$MPS_{t}$$ | -0.298*** (-3.56) | 0.097 (0.81) | -0.154 (-1.52) |
| (2) $$\Delta$$IP$$_{t}$$ | 0.061*** (3.50) | -0.014 (-0.76) | 0.008 (0.40) |
| (3) $$\Delta\pi_{t}$$ | 0.357*** (8.85) | 0.266*** (5.73) | 0.300* (6.20) |
| Controls | Yes | Yes | Yes |
| Observations | 36,565 | 36,565 | 36,565 |
| $$R^{2}$$ | 0.0130 | 0.0130 | 0.0130 |
Appendix Table C.19: Including Gasoline Price Controls
| (1) Accurate | (2) Inaccurate | (3) Haven't Heard | |
|---|---|---|---|
| (1) $$MPS_{t}$$ | -0.193** (-2.49) | 0.170 (1.55) | -0.088 (-0.94) |
| (2) $$\Delta$$IP$$_{t}$$ | -0.001 (-0.09) | -0.063*** (-3.45) | -0.030 (-1.51) |
| (3) $$\Delta\pi_{t}$$ | 0.035 (0.83) | 0.061 (1.27) | 0.112** (2.21) |
| (4) $$\Delta Gas_{t}$$ | 0.042*** (14.62) | 0.033*** (10.84) | 0.026*** (8.18) |
| Controls | Yes | Yes | Yes |
| Observations | 37,445 | 37,445 | 37,445 |
| $$R^{2}$$ | 0.0283 | 0.0283 | 0.0283 |
Appendix Table C.20: Attention with Economic Policy Uncertainty
| EPU | (1) Accurate | (2) Inaccurate | (3) Haven't Heard |
|---|---|---|---|
| High Uncertainty - (1) $$MPS_{t}$$ | -0.651*** (-5.69) | -0.242 (-1.29) | -0.267 (-1.81) |
| High Uncertainty - (2) $$\Delta$$IP$$_{t}$$ | 0.056** (2.53) | 0.011 (0.50) | 0.010 (0.38) |
| High Uncertainty - (3) $$\Delta\pi_{t}$$ | 0.324*** (5.08) | 0.383*** (5.91) | 0.292*** (4.26) |
| Low Uncertainty - (4) $$MPS_{t}$$ | 0.168 (1.58) | 0.276** (2.17) | -0.007 (-0.07) |
| Low Uncertainty - (5) $$\Delta$$IP$$_{t}$$ | 0.051** (1.98) | -0.025 (-0.98) | 0.009 (0.37) |
| Low Uncertainty - (6) $$\Delta\pi_{t}$$ | 0.373*** (7.79) | 0.223*** (3.98) | 0.342*** (5.75) |
| Interaction | EPU | EPU | EPU |
| Controls | Yes | Yes | Yes |
| Observations | 37,445 | 37,445 | 37,445 |
| $$R^{2}$$ | 0.0167 | 0.0167 | 0.0167 |