Hi! I'm an economist at the Federal Reserve Bank of San Francisco.
I'm interested in how people form expectations, how we forecast the economy, and how monetary policy is designed and implemented.
I'm co-organizing the Expecations Workshop on November 2. See the program here!
Here is my CV.
Email: yeji.sung@sf.frb.org
Measures of beliefs, sentiment, and narratives often send recession signals that differ from those in hard data, defined as conventional economic and financial indicators. Using a real-time forecasting framework, we compare how soft and hard data predict recessions from one to twelve months ahead. Forecasts based on soft data are more responsive to rising recession risk: they identify more downturns, but also produce more false alarms. Even with far fewer inputs, soft-data forecasts remain competitive with hard-data forecasts out of sample, especially at shorter horizons. Combining hard and soft data often improves forecast performance, suggesting that the two types of information are useful complements.
(with Nicolas Petrosky-Nadeau and Daniel J. Wilson)
FRBSF Working Paper, July 2026
Firms frequently revise not only their expectations, but also how uncertain they feel about those expectations. Using the U.S. Survey of Business Uncertainty, we study perceived uncertainty about firms’ own sales and employment growth. Reported uncertainty rises after larger revisions to firms’ point forecasts, with the strongest response to the most recent revision. This recency pattern remains visible outside elevated sectoral-volatility episodes. We develop a model in which agents learn about a constant-volatility process but recall older observations noisily. Noisy recall gives recent surprises disproportionate influence, even when objective volatility is constant.
(with Miguel Acosta)
FRBSF Working Paper, July 2026
Standard models of information frictions explain sluggishness in forecasts but struggle to account for excessive sensitivity. I argue that this gap reflects a narrow definition of information frictions: these models constrain the processing of new information while assuming agents can access and use previously acquired information without error. I propose a broader framework with two constraints: attention frictions introduce errors in how agents use new information, and memory frictions introduce errors in how they use previously acquired information. This unified model better matches observed forecast biases and jointly identifies attention and memory frictions using survey forecasts.
FRBSF Working Paper, updated March 2026
Conditionally accepted, American Economic Review
We propose a model of optimal decision-making subject to a memory constraint in the spirit of models of rational inattention. Our theory differs from that of Sims (2003) in not assuming costless memory of past cognitive states. The model implies that both forecasts and actions will exhibit idiosyncratic random variation; that average beliefs will exhibit a bias that fluctuates forever; and that more recent news will be given disproportionate weight in forecasts. The model provides a simple explanation for the overreaction to news observed in the laboratory by Afrouzi et al. (2023).
(with Rava Azeredo da Silveira and Michael Woodford)
American Economic Review, October 2024
According to household surveys, inflation expectations tend to move together with expectations for gasoline price changes. The relationship is not symmetrical, though: Inflation expectations rise when households revise up their outlook for gas price growth but change little when they revise it down. This pattern is strongest among lower-income and less-educated households. Analysis also suggests that increases in expected gas price growth are associated with greater perceived inflation uncertainty. However, quantitatively, the evidence suggests a moderate relationship between changes in expected gasoline prices and households’ inflation expectations.
(with Taerin Kim)
FRBSF Economic Letter, August 2026
This work evaluates the impact of the second series of Targeted Longer-Term Refinancing Operations (TLTRO2) on credit market conditions for Italian firms. The estimates use a difference-in-differences approach on quarterly data, including term loans, interest rates, and bank and firm characteristics, between the start of 2015 and the end of 2017. The TLTRO2 had a positive impact on credit, encouraging lending to firms and reducing interest rates. The impact on the amount of credit was stronger for less risky firms, while that on interest rates was stronger for riskier ones. Smaller firms benefited more in terms of both loan amounts and interest rates.
(with Lucia Esposito and Davide Fantino)
Bank of Italy Temi di discussione, February 2020
SF Fed Workshop on Expectations and Behavioral Macroeconomics, October 27, 2025 (program)