Nonparametric empirical Bayes improvement of shrinkage estimators with applications to time series (Q2325380)

The authors consider the problem of estimating a vector \(\mu=(\mu_1,\ldots,\mu_n)\) based on independent observations \(Y_i\in N(\mu_i,1),\; i=1,\ldots,n,\) and possibly extra structural assumptions. The performance of the estimator \(\delta=\delta(Y)\) is measured by its square error \(E||\delta-\mu||^2.\) The aim is to improve the performance of some classical estimators which asymptotically have the presentation \(\hat{\mu}_i=\alpha\tilde{\mu}_i+(1-\alpha)Y_i+\xi_i=\tilde{\mu}_i+(1-\alpha)(Y_i-\tilde{\mu}_i)+\xi_i,\) where \(\alpha\in [0,1]\) and \(\tilde{\mu}_i\) may depend on the data, but is not a function of \(Y_i,\) and \(\sum \xi_i^2=o_p(n).\) The authors consider the optimal estimator of the form \(\tilde{\mu}_i+g(Y_i-\tilde{\mu}_i)\) where \(g\) is a general function and approximate it using nonparametric empirical Bayes ideas and techniques. The results are demonstrated on the case where \(\hat{\mu}_i\) are Kalman filter estimators.