A0652
Title: A Modified fractionally co-integrated VAR for predicting returns
Authors: Marwan Izzeldin - Lancaster University Management School (United Kingdom)
Xingzhi Yao - Lancaster University (United Kingdom) [presenting]
Abstract: We provide a modified fractionally co-integrated VAR (M-FCVAR) that caters for a system containing both I(0) and I(d) variables. The M-FCVAR restricts shocks emanating from the I(0) variable to only have transitory effects on the I(d) variables within the system, and it also allows for long memory in the equilibrium error. As compared with FCVAR, our M-FCVAR yields less biased co-fractional estimates, especially in the case where the long-memory parameter (d) is higher than the parameter measuring the strength of co-integration relation (b). This feature of the M-FCVAR translates into better performance with respect to return predictability. We verify our claims using both Monte Carlo simulations and empirical analysis based on high frequency equity and index (VIX, SPY, SP500) data observed over the period (2003-2010). Moreover, we find that returns are more predictable during episodes of financial distress when co-fractional relations between implied-realised variance tend to strengthen. Our work allows for a more general and flexible modelling of the co-fractional relations using a mixture of I(0) and I(d) variables which are commonly encountered in empirical applications.
