Title: Disaggregated model-based inflation forecasts: A univariate approach
Authors: Thomas Goetz - Deutsche Bundesbank (Germany)
Jan-Oliver Menz - Deutsche Bundesbank (Germany) [presenting]
Abstract: In the context of the Narrow Inflation Projection Exercise the national central banks of the Euro area are required to deliver various short-term inflation forecasts to the European Central Bank. Apart from the Harmonized Index of Consumer Prices these are particularly five ECB special aggregates (unprocessed food, processed food, energy, industrial goods ex energy and services), which are derived from the roughly 90 COICOP 4-digit-components, the currently deepest available disaggregation level. Naturally, several disaggregation levels and schemes could be considered in-between predicting the target series directly and aggregating forecasts obtained on the deepest disaggregation level. One example is a set of so-called NIPE-aggregates being used in the Deutsche Bundesbank. We aim at systematically assessing which level of disaggregation is optimal for predicting the special aggregates. Therefore we first analyse which of several univariate models (among which are a UCM-SV- and a time-varying AR model) is optimal for predicting which target series for each available aggregation level. We do so by conducting an extensive out-of-sample pseudo-real-time forecast exercise involving over 120 price series for Germany. Given the insights gained in the first step, we then propose a data-driven selection of disaggregate price series in order to obtain a modified set of NIPE-aggregates, which to base our disaggregate inflation forecasts on.