The European Central Bank published an ECB Blog post describing how it uses an artificial intelligence-based model to monitor risks around euro area inflation forecasts in near real time. The tool is designed to complement the baseline path in ECB/Eurosystem projections by estimating how likely inflation is to turn out materially higher or lower than expected. The approach is built on a quantile regression forest model that produces both point forecasts and density forecasts, drawing on a broad set of routinely monitored indicators, including wage developments and selling price expectations. The post argues that the model’s ability to process many variables and capture non-linear patterns helps interpret conflicting signals in volatile conditions, and notes that it has been part of the Eurosystem’s broader analytical toolkit for monetary policy preparation since the end of 2022. As an illustration, the authors highlight that in 2025 the model pointed to wages and selling price expectations as key drivers of revisions to projections for core inflation (Harmonised Index of Consumer Prices excluding energy and food, HICPX), and that for the second and fourth quarters of 2025 the model flagged upside risks that coincided with HICPX inflation outcomes 20 basis points above the ECB/Eurosystem projections. The post concludes that machine learning tools can provide timely information on the magnitude, direction and determinants of inflation risks, can be updated several times within a quarter versus quarterly projection rounds, and may play a growing role in forecasting and monitoring inflation and output, while noting that the views expressed are those of the authors.
European Central Bank 2026-04-21
European Central Bank outlines how a machine learning model is used to assess inflation risks around its baseline projections
The European Central Bank published an ECB Blog post on its use of an artificial intelligence-based quantile regression forest model to monitor euro area inflation risks in near real time, complementing baseline ECB/Eurosystem projections. In use since end-2022, the model processes indicators such as wage developments and selling price expectations to generate point and density forecasts, and has highlighted upside risks to core inflation (HICPX) in 2025 that aligned with outcomes 20 basis points above projections. The post concludes that such machine learning tools can provide more timely information on inflation risks and may play a growing role in forecasting and monitoring inflation and output.