Time: 6 December 2017, 1 - 2 pm
Place: B705


Heteroscedasticity is a property that is apparent in many economic and financial time series and leads to inefficient estimation of parameters. The common way to handle the heteroscedasticity is either to model it assuming that the variance follows some specific structure or to simply ignore it. In this presentation, it is discussed what happens if it is removed instead.

In time series analysis there is a wide variety of filters available that can be used to examine and/or remove certain properties of a time series and stabilize it. However, there are not many filters for stabilizing the variance to be found in the literature. A filter that attempts to do this is suggested; it is a modified version of the filter of Stockhammar and Öller (2012). The filter is designed to remove different kinds of unspecified heteroscedasticity before modelling the mean equation of time series.

A simulation study and a short real data study on the quarterly logarithmic changes in the US GDP is used to illustrate the efficiency gains and usefulness of pre-filtering.