Personas generated by large language models (LLMs) are candidate heterogeneous agents for economic simulation. We match each persona one-to-one with a respondent in the Bank of Japan's household survey and measure how far its inflation expectations track that respondent's answers. Correspondence is separated into dispersion, attribute-level, and individual-level parts, the last read against the benchmark that attribute-only personas imply. Short-horizon conditional means are partially recovered; dispersion and the five-year average are not. Describing a respondent's awareness of monetary policy can more than double the degree of tracking for one-year expectations, so information frictions can enter a simulation as persona text rather than calibrated parameters.
Keywords: Large language models; Inflation expectations; Agent-based simulation; Information frictions; Heterogeneous agents; Persona prompting
Views expressed in the paper are those of the authors and do not necessarily reflect those of the Bank of Japan or Institute for Monetary and Economic Studies.