Effects of outliers on the identification and estimation of GARCH models

• DOI: 10.1111/j.1467-9892.2006.00519.x
• Publication Date: 2006-11-09T10:26:24Z
• Authors: M. Angeles Carnero; Daniel Peña; Esther Ruiz
• Keywords: Homoscedasticity; Ordinary least squares

Abstract. This paper analyses how outliers affect the identification of conditional heteroscedasticity and estimation generalized autoregressive conditionally heteroscedastic (GARCH) models. First, we derive asymptotic biases sample autocorrelations squared observations generated by stationary processes show that properties some homoscedasticity tests can be distorted. Second, obtain finite ordinary least squares (OLS) estimator ARCH( p ) The results are extended to (GLS), maximum likelihood (ML) quasi‐maximum (QML) estimators GARCH(1,1) Finally, estimated standard deviations biased estimates deviations.