Spurious and hidden volatility

This paper analyses the effects caused by outliers on the identification and estimation of GARCH models. We show that outliers can lead to detect spurious conditional heteroscedasticity and can also hide genuine ARCH effects. First, we derive the asymptotic biases caused by outliers on the sample autocorrelations of squared observations and their effects on some homoscedasticity tests. Then, we obtain the asymptotic biases of the OLS estimates of ARCH(p) models and analyze their finite sample behavior by means of extensive Monte Carlo experiments. The finite sample results are extended to GLS and ML estimates of ARCH(p) and GARCH(1,1) models