The motivation for writing this paper comes from Gaussian
quasi maximum likelihood estimation (QMLE) for GARCH (generalized
autoregressive conditionally heteroscedastic) processes with regularly varying
noise; we refer to Section 4 for a detailed description of the problem.