Density forecasts of inflation: A quantile regression forest approach

Abstract

Inflation density forecasts are a fundamental input for a medium-term-oriented central bank, such as the European Central Bank (ECB). We demonstrate that a quantile regression forest, which captures general non-linear relationships between euro area inflation (both headline and core) and a broad set of determinants, performs competitively against state-of-the-art linear and non-linear benchmarks and judgmental forecasts. The median forecasts generated by the quantile regression forest exhibit a high degree of collinearity with the Eurosystem inflation point forecasts, displaying similar deviations from “linearity”. Given that the Eurosystem’s modeling toolbox predominantly relies on linear frameworks, this finding suggests that the expert judgment embedded in the projections may incorporate mild non-linear elements. Finally, we provide a real-time application illustrating how the model is employed to assess risks surrounding the Eurosystem inflation projections in the context of the recent euro area disinflation path.