Long run behaviour of the autocovariance function of ARCH() models

The asymptotic properties of the memory structure of ARCH( ∞ ) equations are investigated. This asymptotic analysis is achieved by expressing the autocovariance function of ARCH( ∞ ) equations as the solution of a linear Volterra summation equation and analysing the properties of an associated resolvent equation via the admissibility theory of linear Volterra operators. It is shown that the autocovariance function decays subexponentially (or geometrically) if and only if the kernel of the resolvent equation has the same decay property. It is also shown that upper subexponential bounds on the autocovariance function result if and only if similar bounds apply to the kernel.