The Sample Moments Integrating

Normal Kernel (SMINK) density

estimator
Let X1; :::; Xn be a random sample from a k-variate absolutely continuous distribution with density f(x); expectation ; and non-singular variance
matrix : Let x(i) be the i-th component of x, and Xi;j the i-th component
of Xj: The SMINK density estimator of f(x) takes the form:

n
k
Y
1X
b
f(xj)
=
I(x(i) 6= Xi;j) fbn;j (xj);
n j=1 i=1

where
fbn;j(xj) =

1
p k
b
2 det

exp

2X

xfb(xj)dx

= X;

1 2 X = 2 ;

P
b = (1=n) Pn
with X = (1=n) nj=1 Xj and
j=1 Xj X
For 2 (0; 1] we have
Z

q
q
1
x 1 2 Xj 1 1 2 X
2

b
xx0 f(xj)dx

b 1

Xj X :

n
1X
=
Xj Xj0 :
n j=1

Let n 2p(0; 1] be a sequence of non-random numbers such that limn!1 n =

0; limn!1 nnk = 1: In particular, let
n =

p =k

; 0 < < 1:

EasyReg International will ask you to specify (the default value is 0:5).
Moreover, let n be a sequence of (random) numbers such that n 2 [n ; 1]
b
(a.s), (p) limn!1 n = 0: Then f(xj
n ) is uniformly consistent:

p n!1
lim fb(xjn ) f (x) = 0:
1

(1)

Furthermore, if we choose
b
n = arg min Q();
2[n ;1]

where

then

R b
f(xj

b
Q()
=
2

n ) f(x)

fb(xj)2dx 2

n
1X
b
f(X
j j);
n j=1

dx is (approximately) minimal, and (1) carries over.

Reference: Bierens, H.J. (1983), Sample Moments Integrating Normal

Kernel Estimators of Multivariate Density and Regression Functions, Sankhya,
45, Series B, 160-192.