Research

This paper surveys the major facts from research on disagreement between households on what they expect inflation to be. We document them using figures and correlations that capture: the statistical regularities on the observable drivers of disagreement, the measurement of residual disagreement, the usefulness of disagreement to forecast inflation, the response of disagreement to shocks, the disagreement between households and professionals, and the relation between disagreement, risk, and uncertainty.

Households overweight their personal price experience when forming inflation expectations, and information provision does not fix this. Among the NY Fed SCE panel respondents, who are interviewed monthly for up to twelve months, forecast accuracy improves substantially over their tenure in the panel. We show that this is not because households acquire better information about inflation during or between interviews: forecast variance falls 66% while the covariance of forecasts with realized inflation does not change. Instead, experienced respondents reweight away from their personal price experience toward the public CPI, reacting more strongly than newcomers to the same monthly CPI release. Within-person dynamics during the 2021–22 inflation surge corroborate the mechanism. A Bayesian model in which agents learn the precision of their private price signal rationalizes the evidence. 

Surveys increasingly elicit households' inflation expectations as probabilistic density forecasts, which are often summarized by a mean and a variance under the assumption of a single-peaked belief. Multimodal responses are overlooked as inattention or reporting noise, rather than read as genuine within-individual disagreement across competing scenarios or regimes. Yet they are not rare. In the New York Fed Survey of Consumer Expectations, around 11 percent of households report multimodal forecasts in a given month, and about a third do so at least once over the panel, more often among lower-income and less-educated respondents. I show this multimodality is a genuine belief rather than an artifact: it is domain-specific, responds to CPI dynamics, and rose specifically during the post-2022 disinflation, when high-inflation and return-to-target scenarios were both plausible. It also shapes behavior in a way that marks it as a distinct dimension of uncertainty: holding the mean, variance, and higher moments fixed, multimodal households report a marginal propensity to spend out of a windfall about 2 percentage points lower. A simple regime-learning model shows that their forecasts update like a rational learner’s, shifting probability toward the mode that realized inflation validates and eventually resolving onto it.

Standard practice in household expectations surveys discards "don't know" responses, which account for roughly 10% of answers to the Michigan Survey of Consumers' one-year inflation expectations question. I show this convention biases inference. Non-respondents are concentrated among women, lower-income, less-educated, and older households --- the same characteristics that predict higher reported expectations --- so dropping them induces sample selection. Three complementary strategies pin down the direction of the bias: regressions on observable correlates, multiple imputation from each respondent's other survey answers, and a natural experiment in the Bank of England Inflation Attitudes Survey in which the "don't know" option was briefly removed in May 2020. All three point the same way; the imputation exercise puts the average gap at about 1 percentage point. Re-running tests of full-information rational expectations on the multiply-imputed sample further reveals substantial underreaction to current inflation among "don't know" households --- underreaction that is statistically undetectable in the conventional sample. Because non-response declines when inflation rises, the share of households whose expectations are well-measured is itself cyclical, with implications for the time-varying traction of expectations-based monetary policy.

To what extent do the risks of starting and running a business lead to resource misallocation? Individuals face a variety of risks when starting a business---a key one is the difference in earnings relative to staying in salaried employment. Starting a business may offer a higher probability of faster earnings growth, but can also increase the probability of very low earnings realizations. We provide novel evidence on the earnings dynamics of business owners compared with employees using rich administrative data from the UK. We complement these findings by implementing a new survey to elicit the risk preferences and constraints faced by potential and current business owners. Informed by these findings we develop and estimate a model of occupational choice to quantify the extent to which earning risk discourages entrepreneurship and study the different ways policy can alleviate these risks. 

Lower-wage workers report substantially higher subjective probabilities of job loss, but whether this gradient reflects objective risk or systematic belief biases remains an open question. This paper uses the French RÉPONSE survey, which interviews employees, management, and employee representatives from the same establishments, matched to administrative employer-employee earnings records. The design separates three inputs into a worker's layoff expectation: the employer's own account of recent activity and economic outlook; the employee representatives' account of the information disclosed by management about the latter; and the worker's administratively measured history of employment and earnings, as well as workplace-related feelings and beliefs. I document that the wage gradient in subjective layoff risk is large and is not accounted for by the observable job- and firm-level risk factors. I then ask how much of the residual gradient is attributable to the workplace information environment, and how much to workers' own past exposure to earnings instability. Income-based heterogeneity in belief biases emerges as a distinct mechanism through which inequality can perpetuate itself, with implications for consumption, job search, and wage bargaining.