Forecasting inflation

This paper provides an overview of conventional and recent models used to forecast inflation. The evaluated models include autoregressive, leading indicator, heterogeneous autoregressive, Phillips Curve, factor and several threshold factor models. Forecasts of 1, 6, 12 and 24 month horizons by these models using up to 126 different macroeconomic predictors are evaluated over the period of 1990 until 2009. The forecasts by the different models are combined by constraint least squares regression to evaluate the possibility of a superior combined forecast. Conclusions include superior forecasting by factor models, especially by stage-wise regression using a least angle regression and by including an autoregressive term. Combining forecasts does not improve forecast quality because of individual model superiority.