About

Overview

How uncertain are households about future inflation? This is an important question. There are many theories about the role of uncertainty in the economy. Uncertainty about inflation, in particular, has interested economists for decades. Inflation uncertainty is of particular interest for monetary policymakers as they attempt to shape and monitor the expectations of the public and establish credibility of their inflation target.

But a major challenge in studying of uncertainty is finding a way to measure it. In "Consumer Inflation Uncertainty and the Macroeconomy: Evidence from a New Microlevel Measure" (Binder 2017, Journal of Monetary Economics), I develop a new measure of households’ uncertainty about inflation based on their rounding behavior when they report their inflation expectations on the Michigan Survey of Consumers. I provide support for the validity of the measure and document its time series and demographic properties, including differences in uncertainty across demographic groups. I use the measure for several applications—to study durables consumption, inflation dynamics, and monetary policy credibility.

What is uncertainty?

Uncertainty is a property of an individual’s subjective probability distribution over a future outcome. A wider, higher variance distribution corresponds to higher uncertainty. Since different people have different beliefs, uncertainty is inherently a micro-level concept. Common time series proxies for uncertainty include measures of disagreement or volatility, which are related to uncertainty but theoretically distinct. The New York Fed's Survey of Consumer Expectations began collecting probability density inflation forecasts from consumers in 2013. The interquartile range of each respondent's probability distribution is a measure of her uncertainty. The uncertainty measure in Binder (2017) is a proxy for this uncertainty constructed from survey data that begins in 1978.

Methodology

To construct the new measure, I use microdata from the Michigan Survey of Consumers, a monthly survey of about 500 households since 1978. Respondents provide their point forecast for inflation over the next year, and they must provide an integer response. A point forecast by itself doesn’t seem to tell us anything about the variance of the consumer’s probability distribution. But if you look closely at the forecasts that are provided, a striking feature is that about half of all responses are a multiple of five! Every single month, you see far more consumers forecasting 5% inflation compared to 4 or 6%, and far more forecasting 10% than 9 or 11%, and so on. In the literature in psychology, cognition, and communication, researchers have found that round numbers play a special role in that we tend to use them to express uncertain estimates. This idea is even given a name, the RNRI principle, which stands for Round Numbers express Round Interpretations. It should seem fairly intuitive that people making a 5% forecast are on average more uncertain than people who say 4% or 6%. I use this insight as the basis for constructing a micro-level inflation uncertainty proxy.

Not all round number responses are always equally likely to indicate high uncertainty. For example, in times during the 1990s when inflation was near 5%, a 5% forecast may not have indicated that the consumer was very uncertain. I want to try to quantify the uncertainty associated with different round number responses in different months. To do so, I design a maximum likelihood estimation framework. I assume that each consumer has some probability distribution over future inflation, and that if the variance is high enough, they will report their forecast to the nearest multiple of five, while if the variance is lower they will report their forecast to the nearest integer. When I see a forecast that is not a multiple of five, I know that it comes from a lower variance, lower uncertainty consumer. If I see a forecast that is a multiple of five, I don’t know for sure whether it comes from a low uncertainty type or a high uncertainty type. But I can estimate the probability that it comes from the high uncertainty type. The way I do this is to note that each month, the responses come from a mixture distribution, mixing the distribution of integer forecasts from the low uncertainty types and the multiple of five forecasts from the high uncertainty types. I use maximum likelihood to estimate the parameters of each distribution including the mixture weight. The estimated parameters allow me to compute the probability that a particular response comes from the high uncertainty type. This probability is what I use as my micro-level uncertainty measure.

I also construct an Inflation Uncertainty Index, a monthly time series taking values between 0 and 1 that indicates the fraction of highly uncertain consumers each month. Consumers are also asked about their inflation expectations over the five- to ten-year horizon, and I use these responses to construct analogous uncertainty measures at the long horizon.

The details of my methodology can be found in Binder (2017) "Consumer Inflation Uncertainty and the Macroeconomy: Evidence from a New Microlevel Measure," Journal of Monetary Economics.