A1940
Title: Fractional integration in mixed causal-noncausal models
Authors: Sean Telg - Vrije Universiteit Amsterdam (Netherlands) [presenting]
Sebastien Fries - Vrije Universiteit Amsterdam (Netherlands)
Jean-Michel Zakoian - CREST (France)
Jorik van der Oord - Vrije Universiteit Amsterdam (Netherlands)
Abstract: The notion of fractional integration is introduced into the mixed causal-noncausal autoregressive (MAR) model. It is shown that the new model, called FIMAR, is able to generate baseline paths that exhibit combinations of exponential and hyperbolic growth as often observed in speculative bubbles. For a general class of error distributions, stationarity conditions are derived and the existence of a purely causal second-order equivalent (SOE) form of the FIMAR model. Using this SOE representation in combination with a Whittle-type estimator, the performance of the model selection for FIMAR models is demonstrated.
