Prognozowanie dynamiki gospodarczej Niemiec z pomocą modeli warunkowych (1992-2014)

Abstract

A great diversity characterizes economic dynamics of Germany over a long period of time. This refers to many time series: in some periods, they show large volatility which then moves into stability and stagnation phase, generating specific difficulties in a long-term forecasting of economic dynamics. The aim of the research is the attempt to determine the prognostic efficiency of conditional modelling and to answer the question whether or not conditional errors are significantly smaller than the unconditional ones in long-term forecasting.The research showed that conditional errors (root mean square errors RMSE) of an ex- post forecast did not differ significantly from the unconditional RMSE. The decreasing RMSE of the ex-post forecast for Germany’s individual economic processes (with the assumption that an intercept occurs in the ARMA procedure) was correlated more strongly with the procedure of filtering economic time series than with the application of the conditional maximum likelihood method (ML) and robust procedures. The relationship between a decreasing RMSE of the ex-post forecast and the application of conditional ML methods occurs in ARMAX forecasts (with exogenous processes) for data filtered with Hodrick - Prescott (HP) filter. It is worth pointing out that a relatively high prognostic efficiency of the robust (resistant) estimation of quantile regression occurs for the economic series linearized with the help of the TRAMO/SEATS method.