The proof of this proposition (3.1.1)
follows a technique used by
Parzen (1962) in the setting of density estimation.
Recall the definition of the kernel weights,
Consider the denominator and numerator separately. I show that
|
(3.3.14) |
|
(3.3.15) |
From (3.1.9) and (3.3.15) it follows by Slutzky's Theorem
Schönfeld (1969, Chapter 6), that
Only (3.3.14) is shown; the statement (3.3.15) can be proved very
similarly.
Note that
where denotes the joint density of the distribution of
Conditioning on gives
since
Using integration by substitution it can be shown (see Lemma
3.1 in these
Complements) that for
The last two terms of this bound tend to zero, by (A1) and (A2), as
.
Now let tend to zero; then the first term
by continuity of will tend to zero.
This proves that
, as .
Now let
. Use integration by substitution and the
above asymptotic unbiasedness of to see that the variance
of is
This is asymptotically equal to
using the techniques of splitting up the same
integrals as above.
Observe now that the variance tends to zero as . This
completes the argument since the mean squared error
as
.
Thus we have seen that
This implies
(see Schönfeld; 1969, chapter 6) proof can be adapted to kernel estimation with higher
dimensional .
If is -dimensional, change to , where :
and the ratio in the argument of
has to be understood coordinatewise.
LEMMA 3.1
The estimator
is asymptotically unbiased
as an estimator
for
.
Use integration by substitution and the fact that the kernel integrates
to one to bound
The first term can be bounded in the following way:
The third term
The second term can be bounded as follows:
Note that the last integral exists by assumption (A3) of Proposition
3.1.1
The derivative estimator
is asymptotically
unbiased.
using partial integration, (A0) and (A4).
The variance of
tends to zero if
, as the following calculations show: