A new study by the German Institute for Economic Research (DIW Berlin) shows that artificial intelligence can improve the accuracy of forecasts for European Central Bank (ECB) interest rate decisions by analysing the tone of its official communications. The AI-supported model, developed using ECB statements from January 2019 to March 2025, categorises messages as restrictive, expansive, or neutral, and incorporates this analysis into a broader forecasting framework. The research introduced a “communication rate indicator” that tracks the overall direction of ECB communication over time. When combined with other economic indicators such as inflation, economic policy uncertainty, and previous interest rate decisions, the model improved prediction accuracy for rate changes from around 70% to 80%. According to the study, 11 of the last 14 ECB decisions were correctly anticipated using the full model, compared to 10 without communication analysis. “Central banks use carefully chosen language to signal their monetary policy intentions,” said Kerstin Bernoth, lead author of the study. “Our AI-based approach captures these signals and can meaningfully enhance forecasts.” The study uses the RoBERTa language model to evaluate ECB speeches, press statements, and interviews. The model classifies the tone of individual sentences and tracks changes in messaging over time. For example, ECB communication was generally expansionary in 2019 and intensified during the pandemic in 2020. As inflation rose toward the end of 2020, the ECB maintained an expansionary stance, drawing some criticism. A shift toward a more restrictive tone began in late 2021, peaking in 2022, alongside a series of interest rate hikes. Since mid-2024, ECB messaging has shifted toward a more neutral tone. For the upcoming ECB Governing Council meeting on 17 April 2025, the AI model suggests a strong likelihood of an interest rate cut. This forecast is supported by slowing economic activity and inflation rates that are now within the ECB’s target range. However, the model does not account for recent developments, such as new U.S. trade policies and tariffs announced by President Trump, which could influence monetary policy decisions in the short term. “Given the increased economic uncertainty, it may be necessary for the ECB to pause and assess these external impacts before proceeding with further rate changes,” Bernoth noted. The study highlights the value of integrating advanced language models into economic forecasting, particularly in times of policy uncertainty.