Bayesian benchmarking of the Fay-Herriot model using random deletion
Section 3. Random deletion methodology
As remarked earlier, the random deletion methodology is
obtained by introducing a new variable which takes the values
with equal, possibly different, probabilities
(weights). In Section 3.1, we show how to construct the joint posterior
density of the parameters of the BFH model under random deletion, and in
Section 3.2, we show how to sample from the joint posterior density.
3.1 Construction of the joint posterior density
The Basic Benchmarking Theorem is motivated by the work
reviewed in Section 1. The next goal is to construct the joint prior
density of
subject to the benchmarking constraint that
where
is a known external target. The joint prior
density of
will be used to complete the constrained
Fay-Herriot model for the last one deletion benchmarking (see Section 1).
Theorem 2
Let
Then, under the
constraint,
where
is constant,
letting
be the vector
of all the
except the last
one, the joint density of
is
and
are
respectively a
matrix and a
vector of ones.
Proof of Theorem 2
See Appendix C.
The proof of Theorem 2 uses the multivariate normal
distribution, and it will be used to prove the more general theorem when the
prior is adjusted to delete any area.
The constrained BFH model is
The constraint on
the prior on
essentially
adjusts the joint prior density. However, to incorporate the constraint, we
will use Theorem 2 to adjust the posterior density under the unconstrained
model.
For
let
and
Also, let
Then, under the unconstrained model the joint
posterior density is
We now extend the result of Theorem 2, to reflect
our interest in the constrained BFH model. Theorem 3, below, is used to
construct the random deletion benchmarking method.
Theorem 3
Using a general notation, let
Let
denote the
vector of all the
except the
one. Also, let
and
Under the
constraint
with
Then,
for
and
Proof of Theorem 3
The proof of Theorem 3
is similar to Theorem 2. See Appendix C.
In what follows, one of the
area parameters will be deleted randomly
(i.e., with probability
Let
represent the county that is deleted. That is,
Then, under the constrained BFH model, using
Theorem 3, the joint posterior density is
where
and for
and
3.2 Sampling the joint posterior density
Unlike the BFH model, the constrained model in (3.2)
cannot be fit using random draws; we use a Gibbs sampler. The joint conditional
posterior density (cpd) of
is
where
denotes the
vector of
with the
component
deleted, and
and
are defined in
Theorem 3. Then, the joint conditional posterior density of
is
It is straight forward to sample the cpd’s of
and
However, it is not so simple to sample the
cpd’s of
and
that we will next discuss.
First, to obtain the cpd’s of
and
we define
Second, for
let
and
Let
and
respectively denote the vector with entries
excluding the
component and the matrix with columns
excluding
Now, let
where
is the covariance matrix without the
row and
column. Third, with a multivariate normal prior on
of the form
where
and
are specified,
we let
thereby offering
some protection against posterior impropriety. It follows that
We can eliminate the
prior of
by letting
(i.e.,
noninformative prior) to get
as in the BFH
model.
Finally, we consider the cpd of
Let
and
Then, the cpd of
is
where
denotes
the vector of the
excluding the
component and
is a
prior on
As in
the BFH model (see Section 2), we assign the prior density
to
Because
the baseline BFH posterior density is proper, the constraint BFH posterior
density will also be proper.