A0810
Title: Noncausal AR processes driven by causal GARCH volatility
Authors: Jean-Michel Zakoian - CREST (France) [presenting]
Daniel Velasquez-Gaviria - Maastricht University (Netherlands)
Abstract: The purpose is to study the introduction of causal conditional heteroskedasticity in noncausal autoregressive (AR) models. It is demonstrated that large shocks to the independent innovation that drives the GARCH error term of a noncausal first-order AR model result in heightened volatility following a bubble crash. The non-coincidence of the information sets generated by past observations and past values of the GARCH process makes estimation non-standard. In particular, the full quasi-maximum likelihood estimator (QMLE) is generally inconsistent. The asymptotic properties of three-step weighted least squares estimators of the AR coefficient and the QMLE of the volatility coefficients are investigated. Findings are illustrated via Monte Carlo experiments and real financial data.
