2024-02-08-IMF-U.S._Inflation_Expectations_During_the_Pandemic_49页_3mb

The COVID-19 pandemic triggered significant changes in U.S. inflation expectations, particularly from 2021 onwards. Using data from the University of Michigan Consumer Survey, this working paper shows that the distribution of inflation expectations widened, with the mean increasing and higher heterogeneity observed among consumers with lower education and income levels. These groups exhibited higher and more volatile expectations.

Empirical analysis rejects the rational expectations hypothesis with simple rationality tests, which failed to explain the observed inflation forecast errors. Conversely, a reduced-form adaptive learning model involving recursive least squares and constant gain learning fits the data better, suggesting that consumers update expectations based on past forecast errors. Low-income, less educated, and female consumers tend to use simpler forecasting models (e.g., AR(1)), less frequently update their expectations, and are more backward-looking, often placing less weight on recent data.

A larger-scale DSGE model incorporating adaptive learning, estimated using a long-term dataset including the COVID period, improves both in-sample fit and out-of-sample forecasting performance compared to models assuming rational expectations. Under adaptive learning, inflation provenved to respond more persistently to innovation, including the large price mark-up shock interpreted as driving the 2021 inflation surge. This requires stronger and more sustained monetary policy responses to return inflation to the target.

In conclusion, the widening distribution of inflation expectations during the pandemic is largely driven by differential information sets and expectation-formation mechanisms, such as reliance on more complex models among higher-income and more educated consumers versus simpler models among lower-income and female consumers. The findings highlight the heterogeneity in expectation formation and the role of learning mechanisms in modeling inflation dynamics.