A1851
Title: Forecasting inflation
Authors: Michele Lenza - European Central Bank (Germany)
Joan Paredes - European Central Bank (Germany) [presenting]
Marta Banbura - European Central Bank (Germany)
Abstract: What is the best univariate benchmark model for inflation? Does multivariate information help, and under which circumstances? Do stochastic volatility, time-varying means and outlier correction mechanisms matter, and under which circumstances? These three classical questions are reviewed by comparing, on the grounds of their out-of-sample accuracy for headline inflation, specifications of the scalar and vector autoregressive models which switch on and off the different model features which are the object of the comparison. Results are produced that are robust to a specific sample choice and that can be useful also for practitioners who produce forecasts in real-time. Hence, both US and euro area data are considered and, most importantly, the data vintages are used, which were available in real-time to the inflation forecasters in both areas.
