Random sampling of continuous-parameter stationary processes: Statistical properties of joint density estimators (Q1120234)

This article considers consistency in quadratic mean and the CLT for kernel estimators of the joint density of a process \(X_ t\) observed on a sequence of stopping times \((t_ j)\). Hypotheses on \(X_ t\), besides those related to the continuity of joint densities, are i) \(X_ t\) is asymptotically non correlated and its coefficient \(\rho\) satisfies integrability with respect to the renewal density of the process \((t_ j)\), or ii) \(X_ t\) is \(\alpha\)-mixing and the behaviour of its mixing coefficient is similar to that of \(\rho\) in i). The proof of consistency in quadratic norm under the above condition uses estimator covariance calculations. For the CLT, big and small block techniques are used, considering a modification of Volkonskij-Rozanov's classical result on the approximation of the sums of \(\alpha\)-mixing random variables by independent ones.