newsfilter
← Latest News
Federal Reserve FEDS Notes·

Assessing Monetary Policy Globally: Evidence from LLMs and Semi-structural Models

  • Central bank communication has become a critical tool for shaping global expectations on inflation, growth, financial stability, and future policy directions.
  • Traditional semi-structural models evaluate policy stances using macroeconomic variables but face limitations regarding data lags and the inability to process unstructured text.
  • Large language models (LLMs) offer a method to extract timely, qualitative, and contextual information from central bank speeches that keyword-based approaches often miss.
  • This study compares two sentiment measures derived from BIS speeches (1996–2025): a lexicon-based approach and an LLM-based approach using Claude Haiku 4.5.
  • Both measures track each other closely before the Global Financial Crisis but diverge afterward, with the LLM returning to neutral while the lexicon remains persistently dovish.
  • The global aggregate score represents a communication-weighted tone where major economies with frequent speeches exert greater influence on the signal.
  • Semi-structural models define the policy stance as the gap between the actual policy rate and the country-specific neutral nominal interest rate.
  • Around major turning points like the GFC and the post-pandemic tightening cycle, LLM measures shift direction more rapidly than semi-structural models, which adjust gradually.
  • The LLM factor captures forward-looking guidance and risk assessments, often signaling policy turns before they are reflected in macroeconomic fundamentals.
  • Dynamic factor models show that both LLM and semi-structural factors exhibit common variation across countries, with the LLM factor displaying greater short-term volatility.
  • The two approaches are complementary: LLMs provide timely signals of intended policy direction, while semi-structural models offer a benchmark grounded in economic data.
  • Divergence between the measures can indicate periods where central banks are preparing markets for future policy changes not yet visible in realized macro data.
  • The study concludes that LLM-based indicators should be used alongside traditional semi-structural measures to provide a richer assessment of the global monetary policy environment.