Forecasting Inflation: An Art as Well as a Science!

Abstract

In this study, we build two forecasting models to predict inflation Harmonised Index of Consumer Prices (HICP) for the Netherlands and for the euro area. The models provide point forecasts and prediction intervals for both the components of the HICP and the aggregated HICP-index itself. Both models are small-scale linear time series models allowing for long-run equilibrium relationships between HICP components and other variables, notably the hourly wage rate and the import or producer prices. The model for the Netherlands is used to generate the Dutch inflation projections for the eurosystem's Narrow Inflation Projection Exercise (NIPE). The recursive forecast errors for several forecast horizons are evaluated for all models, and are found to outperform a naive forecast and optimal AR models. Moreover, the same result holds for the Dutch NIPE projections, which have been provided quarterly since 1999. The aggregation method to predict total HICP inflation generally outperforms the direct method, except for long horizons in the case of the Netherlands.

Keywords: aggregation, model selection, time series models, inflation

JEL Classification: C32, C43, C52, C53