A New Proposal for Forecasting Inflation in the Eurozone: A Global Model

ABSTRACT This paper evaluates the forecasting performance of the relatively new machine learning Global Unrefined (GlobalUN, hereafter) model with respect to inflation in the Eurozone. In this global pooled neural network framework, we use a quarterly panel dataset covering 20 euro‐area countries (2001Q1–2025Q1) together with the EA‐20 aggregate, which includes key variables such as HICP, energy prices, food, and others. Thus, the network remains simple yet flexible enough to absorb heterogeneity across countries. Our work's contribution is crucial since monetary policy in the Eurozone hinges on accurate inflation forecasts (i.e., as ECB decisions target expected rather than current inflation). Our findings are crystal clear. The GlobalUN model outperforms all other benchmark models and the advanced machine learning XGBoost model in almost all Eurozone countries and horizons (i.e., the naïve model seems to outperforms in a few cases). These results are useful for policymakers, central banks, and fiscal institutions, as they should take the GlobalUN model into account as part of their arsenal.