New methods for forecasting inflation and its sub-components: application to the USA

Forecasts are presented for the 12-month ahead US rate of inflation measured by the chain weighted personal consumer expenditure deflator, PC, and its three main components: non-durable goods, durable goods and services. Monthly models are estimated for 1974 to 1999, and pseudo out-of-sample forecasting performance is examined for 2000-2007. Alternative forecasting approaches for a number of different information sets are compared with benchmark univariate autoregressive models. In general, substantial out-performance is demonstrated for the aggregate and components models relative to benchmark models. The combination of equilibrium correction terms, which bring gradual adjustment of relative prices into the inflation process, and non-linearities, to proxy state dependence in the inflation process, is shown to contribute importantly to this out-performance. There is also evidence that forecast pooling or averaging improves forecast performance. The indirect forecasts constructed by weighting the three component forecasts are compared with the direct forecasts from the aggregate PC. In most cases, the indirect method outperforms the direct method. A key innovation is to compare standard AR or VAR methods of using an information criterion to select the large length, with a parameterization in which longer lags appear in parsimonious forms. Another is to compare general unrestricted models with corresponding parsimonious models selected by Autometrics, Doornik (2008).