On smoothed probability density estimation for stationary processes (Q1073523)

The authors study estimation of marginal probability densities for stationary processes of both discrete and continuous time parameters. Results on consistency and asymptotic distributions are obtained for estimators defined via delta sequences or delta families under the assumption of long range dependence which is defined, for instance, via the dependence index sequence \[ \beta_ n=\sup_{x,y}\sum^{n}_{i=1}| f_ i(x,y)-f(x)f(y)| \] for discrete-time parameter processes \(\{X_ j,j\geq 1\}\). Here \(f_ j(x,y)\) denotes the joint density of \((X_ 1,X_{1+j})\) and f(x) denotes the density of \(X_ 1\). In the continuous-parameter case, the dependence index function is defined by \[ \beta_ T(\gamma)=\sup_{(x,y)}\int^{T}_{\gamma}| f_ s(x,y)- f(x)f(y)| ds, \] where \(f_ s(x,y)\) is the joint density of \((X_ 0,X_ s)\) and f(x) is the density of \(X_ 0\). It was shown that a ''full rate 1/T'' of convergence of the variance of the estimator to zero can be achieved for a certain class of processes in the case of continuous sampling.