Forecasting Time-varying Densities of Inflation Rates: A Functional Autoregressive Approach

신용철

Abstract

This paper utilizes the nonparametric functional autoregressive approach (FAR) to model the time-varying distribution of UK monthly in ation rates using disaggregated cross-sectional data. Our approach is free of any assumptions on the class or structure of the density functions themselves, or the number of dimensions in which the densities may vary. The \pseudo real time" in-sample forecasting evaluation results show that our proposed models track the realized event probabilities fairly closely. Furthermore, out-of-sample forecasting results suggest that the mean is projected to be stable at around 2.5%-2.6% over the period March 2008 - February 2009 whilst the uncertainty bands stay between 1.5% and 4% over the 12-month forecast horizon. In addition, the probability of achieving the 2% in ation target is relatively low.