What’s happening with inflation expectations?

By: Yuriy Gorodnichenko and Olivier Coibion

Economic policymakers spend a lot of time dissecting different measures of inflation expectations in the economy, since these affect prices, wages, consumption and many other outcomes. Over the last year or so, a major gap has appeared between two of the leading surveys of U.S. consumers’ inflation expectations: the New York Federal Reserve’s Survey of Consumption Expectations (SCE) and the Michigan Survey of Consumers (MSC). The latter displays high levels of expected inflation since the inflation surge whereas the former reports low “anchored” levels, as shown in Figure 1. Which is correct? The answer is the Michigan survey.

Figure 1: Surveys of Household 1-Year Ahead Inflation Expectations

How do we know this? We use two strategies. The first is to bring in other surveys of consumer expectations to determine whether it is the MSC or the SCE that is the outlier. The second is to dig into the surveys to understand why they seem to point to different answers. Both approaches lead us to the same conclusion.

Other surveys of inflation expectations of households confirm the Michigan survey. Figure 1 plots results from surveys of Nielsen Homescan panels we run every quarter, covering around 20,000 representative households per quarter (much larger than Michigan or NY Fed). It also plots the Indirect Consumer Inflation Expectations survey from the Cleveland Fed. Both give approximately the same dynamics as Michigan. What is clear from this figure is that it is the reported data from the NY Fed survey that is the outlier, not the Michigan survey.

The NY Fed applies different methodological choices: The NY Fed survey results reported in Figure 1 are different from those of the other surveys in three ways, each of which we think is a problem.

a. Medians vs means: The NY Fed series is for the median instead of the mean: this throws away all the information in the right tail, which is where the action is when it comes to inflation expectations. The mean is more volatile but the resulting volatility is not noise. See Figure 2 for example: most of the high-frequency variation in mean forecasts is associated with gasoline (oil) price variation that consumers observe. We should not be throwing away this information by relying on the median. Since the NY Fed provides individual-level responses prior to the last 12 months, we can easily construct the mean, with the exception of the most recent data.

Figure 2: Household Inflation Expectations and Gasoline Prices

b. Distribution questions vs point forecasts: Whereas the other surveys use point forecasts, the NY Fed presents responses from a “distribution” question where respondents assign probabilities to a pre-defined set of possible outcomes. This induces framing effects (it’s centered around zero) and introduces a bias as the inflation rate rises (because the bins are fixed). Fortunately, the NY Fed also asks respondents for point forecasts, so this is easy to address.

c. Panel conditioning effects: The NY Fed has respondents participate every month for a year. This is a problem because people learn from participating in the survey: their responses in the wave are usually ~2p.p. higher than in their third wave (Kim and Binder 2023). In other words, after the first wave, they are no longer representative and should not be used. Since the NY Fed provides individual data with information about how many times they have participated (albeit with a one-year delay), we can easily address this issue by focusing on the subset of respondents who are responding for the first time.

If we correct for these three differences using the NY Fed micro-data, we get the results in Figure 3 below, which mirror what we get from the other surveys. In other words, the surveys actually all tell the same message: year-ahead inflation expectations are currently around 5-6%. The same logic applies to LR expectations: you should rely on mean 5-year ahead expectations of the Michigan survey, and these are currently very high and unambiguously unanchored.

Figure 3: Uncorrected and Corrected NY Fed Survey Expectations

Conclusion:

Expectations of households and firms, following the inflation surge, have settled at new higher levels after the surge than they were prior to the surge. This suggests that the inflation surge experience has further unanchored their expectations. Since these feed into consumer spending as well as firms’ pricing, employment and investment decisions, we should expect inflationary pressures to be particularly high, complicating the job of achieving the “last mile” of inflation reduction and responding to new inflationary pressures from tariff policies.