System approach to forecasting using time series models

Abstract

Constructing of forecasting functions is considered for the following classes of processes: stationary autoregression and autoregression with moving average part, processes with deterministic and stochastic trends, heteroscedastic and cointegrated processes. The forecasting functions are given derived with and without the difference equation solution. To describe the stochastic trend, the random step model with noise and drift and the model of linear local trend are used. The basic types of equations are given for describing heteroscedastic and cointegrated processes.