The approximation bias of Gram-Charlier Expansion in dynamic higher moments modelling
The Gram-Charlier Expansion (GCE) of the Gaussian density under GARCH framework has been widely used to model the conditional dynamic higher moments. Compared with other generalized distributions, GARCH-GCE models describe the dynamic equations of conditional skewness and kurtosis in a more direct way. While GCE function is not always positive, it is often squared and normalized in empirical studies. However, little attention has been paid to the fact that the higher moments of squared-GCE function are different from the original GCE function.