Modeling and forecasting for heteroscedastic processes

Abstract

A methodology of constructing mathematical models for heteroscedastic processes and its application in describing dynamics of time series are considered. A simplified test for heteroscedasticity is proposed and algorithm for model improvement thanks to taking into consideration the spikes substantially greater than the mean value of a series. The variance forecasting functions are constructed as a measure of risk on the basis of the equations solutions. Some examples of forecasting real time series are given.