On the oscillation of infinitely divisible and some other processes (Q921711)

The authors study path properties of the class of stochastic processes X(t) which have the same finite-dimensional distributions as the series expansions \(\sum^{\infty}_{k=1}g_ k(t,{\mathbb{Z}})\), where \({\mathbb{Z}}=(Z_ 1,Z_ 2,...)\) are i.i.d. random variables. The assumptions in the paper allow for partial dependence of the components of the expansion. Sufficient conditions for the oscillation process \(W(t)=\limsup_{u,v\to t} | X(u)-X(v)|\) to be nonrandom are given. The result is then used to study path properties of stochastic processes. In particular, sufficient conditions are given for the alternative that with probability one paths are either continuous, or locally unbounded. Path properties of infinitely divisible processes that can be represented as stochastic integrals are analyzed in terms of the kernel function. In particular, for such processes the sufficient condition for nonrandom oscillation is shown to be necessary, and in Corollary 1 the authors show that under some additional assumptions almost sure continuity of paths is equivalent to continuity of the kernel function (up to a modification). The techniques of the paper can also be applied to study path differentiability.