Norberto
Rodríguez & Patricia Siado
Abstract
This paper contains the results of a non parametric
multi-step ahead forecast for the monthly Colombian inflation, using Mean
conditional kernel estimation over inflation changes, with no inclusion of
exogenous variables. The results are compared with those from an ARIMA and a
non-linear STAR. The nonparametric forecast over perform the others two, as
well as being the only, from the three, that statistically improved the naďve
forecast given by a random-walk model.
Key words: Nonparametric
forecast, Kernel estimation, Forecast evaluation, Bandwidth selection, Rolling
forecast.
PDF (Spanish)