Density forecasts of inflation : a quantile regression forest approach

Density forecasts of inflation are a fundamental input for medium-term oriented forecasters, such as National Statistic Institutes or Central Banks. We show that a quantile regression forest, capturing a general non-linear relationship between euro area (headline and core) inflation and a large set of determinants, is competitive with state-of-the-art linear benchmarks and judgemental survey forecasts. The median forecasts of the quantile regression forest are very close to the ECB point inflation forecasts, displaying similar deviations from “linearity”. Given that the ECB modelling toolbox is essentially linear, this finding suggests that the expert judgement embedded in the ECB forecast may be characterized by some mild non-linearity