Intelligent forecasting for financial time series subject to structural changes

This paper is mainly concerned about intelligent forecasting for financial time series subject to structural changes. For example, it is well known that interest rates are subject to structural changes due to external shocks such as government monetary policy change. Such structural changes usually make prediction harder if they are not properly taken care of. Recently, Oh and Kim (2002a, 2002b) suggested a method that could handle such difficulties efficiently. Their basic idea is to assume that different probabilistic law (and hence different predictor) works for different situations. Their method is termed as two-stage piecewise nonlinear prediction since it is comprised of establishing various situations empirically and then installing a different probabilistic nonlinear law as predictor on each of them. Thus, for its proper prediction functioning, it is essential to identify the law dictating the financial time series presently. In this article we propose and study a mixing approach for better identification of the presently working probabilistic law.