In this paper we propose a detecting scheme for nonlineartime-series model classification by using a knowledge-basedconstruction and fuzzy statistical decision making. ARCH and bilinearmodels are frequently applied in economic or financial time-seriesmodeling, and both models exhibit certain kind of pattern similarity,such as unusual jumps and a diversity trend. So we take these twomodes as our illustration example for demonstration. Simulationresults presented here demonstrate that our detecting procedure canclassify ARCH and bilinear models effectively. The designed detectionprocess also exhibits a significant rate of correct recognition forsunspot series and exchange rates.
展开▼
