Random Sampling of Almost Periodic Stochastic Processes

Abstract

We study the effect of random sampling on stochastic processes whose finite-dimensional distributions are Bohr almost periodic under time shifts. The sampling times are generated independently by a two-sided random walk with positive increments or by a two-sided subordinator. Under a non-degeneracy condition we show that the shifted finite-dimensional distributions of the sampled process converge to those of a strictly stationary process. Consequently, if the sampled process is itself almost periodic in distribution, then it must already be strictly stationary. We further prove an $L^2$-law of large numbers for sampled averages under covariance decay and a uniform second-moment bound, and identify its limit with the mean of the stationary limit process. Finally, for $m$-dependent, $L^2$-uniformly integrable almost periodic processes, we establish a central limit theorem for the centered sampled sums, whose asymptotic variance is given by the autocovariance series of the stationary limit.