Resumen de The Forecasting Performance of Single Equation Models of Inflation.

David Norman, Anthony Richards

• In the US academic literature, it is now widely accepted that it is difficult to generate a more accurate forecast of inflation than a simple autoregressive model, following Atkeson and Ohanian's seminal work. In central banks, however, the conventional view that inflation is ultimately determined by excess demand in the economy is still widely accepted, and research into activity-based models has continued. We add to this research by examining a range of activity-based models, with the aim being to determine whether any of these can outperform the autoregressive benchmark, and if so, which performs best. Our results show that an expectations-augmented, unemployment-based Phillips Curve can produce significantly superior forecasts of Australian trimmed mean inflation to both the autoregressive benchmark and others at a range of forecast horizons, while mark-up type models also produce superior results at longer horizons. This outperformance is shown to come from the period in which inflation rose to relatively high rates. These results do not, however, translate to headline consumer price index inflation, suggesting that inflation in the items that are 'trimmed' largely reflects 'noise', as opposed to any signal of trend inflation.