with density f, to estimate the density f from the data, Naito (2004) provided a class of
semiparametric density estimators that have multiplicative adjustment, including estimators
proposed by several authors (Hjort and Glad, 1995, Hjort and Jones, 1996) as special cases.
In the proposed approach, a parametric density estimator g(x,
ˆ
θ) is utilized, but it is seen
as a crude guess of the true density f. This initial parametric approximation is adjusted via
multiplication by an adjustment factor ξ = ξ(x). ξ is determined by minimization of the
empirical version of the function
Q(x, ξ|α) =
Z
K
h
(t − x)
[f(t) − g(t,
ˆ
θ)ξ]
2
g(t,
ˆ
θ)
α
dt, (1.3)
for a fixed target point x, where α is a real number called the index. This method is called
the local L
2
-fitting criterion. After omitting the irrelevant term, the empirical version of
(1.3) can be expressed as
Q
n
(x, ξ|α) = ξ
2
Z
K
h
(t − x)g(t,
ˆ
θ)
2−α
dt −
2ξ
n
n
X
i=1
K
h
(X
i
− x)g(X
i
,
ˆ
θ)
1−α
.
The minimizer can be easily determined as
ˆ
ξ =
ˆ
ξ(x) = arg min
ξ
Q
n
(x, ξ|α) =
n
−1
P
n
i=1
K
h
(X
i
− x)g(X
i
,
ˆ
θ)
1−α
R
K
h
(t − x)g(t,
ˆ
θ)
2−α
dt
.
Using this
ˆ
ξ, a class of semiparametric density estimators is obtained by
ˆ
f
α
(x) = g(x,
ˆ
θ)
ˆ
ξ = g(x,
ˆ
θ)
n
−1
P
n
i=1
K
h
(X
i
− x)g(X
i
,
ˆ
θ)
1−α
R
K
h
(t − x)g(t,
ˆ
θ)
2−α
dt
. (1.4)
Theoretical comparison reveals that the estimators in this class are better than, or at least
competitive with, the traditional kernel estimator in a broad class of densities.
3
http://www.paper.edu.cn