Macroeconomic and Financial Forecasting in the Aftermath of the Covid Shock: Inflation Forecasting With Large Language Models

We compare the inflation forecasting performance of large language models (LLMs) to traditional models and the Kalman filter approach developed by Hall et al. (2026). Using monthly data for inflation in both the United States and the euro area, we forecast over the period 2020:M1 to 2025:M9; our findings show that all the LLMs significantly improved forecasting performance against the traditional models.