Estimation of a symmetric density function

From MaRDI portal
Publication:3680076





Let \(X_ 1,..., X_ n\) are iid rv's with common density f(x-\(\theta)\). Let \(f_ n(x)\) be the classical kernel estimator of f. In the case when \(\theta\) is known, let \[ (1)\quad f_{n,\theta}(x)=(1/2)(f_ n(x)+f_ n(2\theta -x)), \] if \({\hat \theta}\) is an estimator of an unknown \(\theta\), set \[ (2)\quad f_{n,{\hat \theta}}(x)=(1/2)(f_ n(x)+f_ n(2{\hat \theta}-x)). \] The main result of the paper is the following theorem: Let f be a symmetric density about known \(\theta\), f' and f exist and are bounded, \(0<\int^{\infty}_{-\infty}| f(x)| dx<\infty\) and \(0<\int^{\infty}_{-\infty}t^ 2w(t)dt<\infty.\) Then \[ MISE_{f_{n,\theta}}(h)=E_{\theta}\int^{\infty}_{- \infty}(f_{n,\theta}(x)-f(x))^ 2dx= \] \[ (2nh)^{- 1}\int^{\infty}_{-\infty}w^ 2(t)dt+(h^ 4/4)(\int^{\infty}_{- \infty}t^ 2w(t)dt)^ 2\int^{\infty}_{-\infty}(f(x))^ 2dx+o((1/nh)+h^ 4). \] Moreover, it is shown, that the same result holds when f is a normal density with unknown \(\theta\), if \(\theta\) is replaced by \(\bar X_ n\). A Monte Carlo study illustrates the theoretical conclusions.











This page was built for publication: Estimation of a symmetric density function

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q3680076)