Measuring the tone of central bank communication in Mexico and its effects on inflation expectations

Abstract

This thesis investigates the influence of central bank communication on inflation expectations in Mexico. Departing from traditional sentiment analysis methods, this study employs two Large Language Models (GPT-4o and LLaMA 3.2) to construct tone indices that capture the hawkish or dovish stance of monetary policy statements issued by Banco de México. Then, we use a linear local projection model to estimate the dynamic response of inflation expectations to a shift in tone, complemented by a non-linear extension that accounts for monetary policy regimes based on the relationship between the policy rate and the estimated natural rate. Results show that a hawkish tone generally raises inflation expectations in an accommodative policy environment but may reduce them in a restrictive regime. These findings highlight the nuanced effects of central bank communication and the value of LLMs in economic research.