A note on GARCH model identification

Financial returns are often modeled as autoregressive time series with innovations having conditional heteroscedastic variances, especially with GARCH processes. The conditional distribution in GARCH models is assumed to follow a parametric distribution. Typically, this error distribution is selected without justification. In this paper, we have applied the results of Thavaneswaran and Ghahramani [A. Thavaneswaran, M. Ghahramani, Applications of combining estimating functions, in: Proceedings of the International Sri Lankan Conference: Visions of Futuristic Methodologies, University of Peradeniya and Royal Melbourne Institute of Technology (RMIT), 2004, pp. 515–532] on identification of GARCH models to a number of financial data sets.