Navigating Uncertain Times — the ECB's Machine-Learning Model for Real-Time Inflation-Risk Tracking

Central banks traditionally rely on linear econometric models that produce single-point inflation forecasts, which can miss tail risks and shifts in the distribution of possible outcomes.

The ECB sought a complementary analytical tool that could flag upside and downside inflation risks in real time, alongside — not instead of — its traditional forecasting models.

Since the end of 2022, the ECB has used a quantile regression forest (QRF), a machine-learning model, as part of the analytical toolkit informing Governing Council monetary-policy decisions. The model draws on roughly 60 indicators covering inflation expectations, cost pressures, real economic activity and financial conditions, and is updated several times a quarter to produce point forecasts and a full distribution of inflation risks. The methodology builds on ECB working-paper research on density forecasting and complements 'Project Spectrum', a joint initiative with the Bank for International Settlements and Deutsche Bundesbank applying text-embedding and machine-learning techniques to inflation nowcasting. The tool has since been extended from inflation forecasting to nowcasting euro-area GDP growth.

In the second and fourth quarters of 2025, the QRF flagged upside risks to core inflation that subsequently materialised; final readings came in roughly 20 basis points above the official Eurosystem staff projections, which did not incorporate the AI signal.

The ECB frames the QRF as a 'second opinion' complementing, not replacing, traditional econometric models. Because the disclosure comes from the ECB's own blog and internal deliberations remain confidential, external verification of the model's real-time predictive edge is limited to the institution's own account and follow-on financial press coverage — it is nonetheless one of the more concretely evidenced examples of an operational AI tool embedded in a major central bank's live decision-making.