Proyección de la inflación en Chile: una visión sectorial

Abstract

We explore the advantages of using Phillips curves in the inflation forecast. To do this, we use conventional, flexible specifications that adjust the data from four definitions of inflation: headline, core (excluding foods and energy, EFE) and a partition of the latter into EFE goods and services. We design an out-of-sample projection exercise with moving windows to compare (statistically) the forecast errors of the Phillips curves and other control models. For headline inflation, we find that the time series models have a significantly lower forecast error. Core inflation forecasts using Phillips curves only 12 months out are significantly better. Finally, we show that the combined forecast method for EFE inflation based on the projections of EFE goods and services deliver even better forecasts 12 months out. At shorter horizons, the time series models do better.