ESTIMATION OF A MULTIVARIATE DENSITY
ESTIMATION OF A MULTIVARIATE DENSITY
复制标题
DOI:
10.1007/bf02869528
复制
发表时间:
1966-01-01
影响因子:
1
通讯作者:
CACOULLOS, T
中科院分区:
文献类型:
--
作者:
CACOULLOS, T
Parzen [2] gave the asymptotic properties of a class of estimates f~(z) of a univariate density function f (z) on the basis of a random sample XI,..-, X~ from f (z). f~(z) is of the form n where h= h (n)---> O as n--> oo and K (z) a bounded function such that l~: K (z) dx= l and [ylJK (y) l---> O as [yJ~ oo. Our purpose is to indicate how the f,(z) can be adapted to provide estimates of a multivariate density. Actually here the extension is carried out in two directions corresponding to the two general forms of kernels K, as given in Theorems 2.1 and 4.1.The results concern'ng the consistency, asymptotic unbiasedness, and bounds for bias and mean square error of fn follow very easily by using Theorem 2.1 below and straightforward modifications of those in [2]. With respect to asymptotic normality we give here a stronger result, namely, the joint asymptotic normality of the estimates f~ at continuity points of f (Theorem 3.5). Finally, the interesting case of estimates based on product kernels is studied in Section 4.