7:48 p.m.
Is it because I am too tired already, or should p.37 of the user guide
say "n>=m" instead of "n<=m" in the context of explaining the
fdjac
function? (With respect to the delta method, meaning that you cannot get
out a covariance matrix which is higher-dimensional than the one you
put in.)
On the same page, n is also used as the sample size, but I'm not sure if
that really causes confusion for anybody.
thanks,
sven
8:53 p.m.
On Thu, 17 Jan 2008, Sven Schreiber wrote:
Is it because I am too tired already, or should p.37 of the user
guide say
"n>=m" instead of "n<=m" in the context of explaining the fdjac
function?
(With respect to the delta method, meaning that you cannot get out a
covariance matrix which is higher-dimensional than the one you put in.)
Sure you can: it'll be singular, but who said a covariance matrix can't be
singular? To qualify as a covariance matrix, a square matrix must be
symmetric and positive SEMIdefinite.
In fact, $n$ and $m$ could be any integer. See example:
<script>
function a(matrix *x)
matrix r = x[1]/x[2]
return matrix r
end function
function b(matrix *x)
matrix r = zeros(3,1)
r[1] = x[1]/x[2]
r[2] = x[1]*x[2]
r[3] = x[1]+x[2]
return matrix r
end function
nulldata 3
x = { 1 ; 2 }
Ja = fdjac(x, "a(&x)")
Jb = fdjac(x, "b(&x)")
Ja
Jb
</script>
Fixed the manual on this.
On the same page, n is also used as the sample size, but I'm not
sure if that
really causes confusion for anybody.
Yup, fixed that too. Thanks, Sven.
Riccardo (Jack) Lucchetti
Dipartimento di Economia
Università Politecnica delle Marche
r.lucchetti(a)univpm.it
http://www.econ.univpm.it/lucchetti
9:22 p.m.
Ok, the fdjac function is very cool, I just discovered it this week, and
after some heavy scripting it has given me some very useful results.
Thanks for this great functionality! (And for the excellent
documentation, I wouldn't have even known the function existed without
it. From the start of my involvement with gretl I have always been
impressed by the quality and extent of the docs, it's really a big
argument in favor of gretl!)
Now, it seems to me that fdjac does not accept more than two matrix
arguments. When I had a version where a (redundant) third matrix was
passed --apart from theta and another necessary input matrix-- I just
got a zero Jacobian matrix as a result (mind you, no error, which was
not nice either). Then I removed that redundant argument and everything
was ok.
If so, it should be documented (or fixed?). I know I know, I should give
an example, but right now I can't, maybe later.
thanks,
sven
2:35 a.m.
On Fri, 18 Jan 2008, Sven Schreiber wrote:
Ok, the fdjac function is very cool, I just discovered it this
week, and after some heavy scripting it has given me some very
useful results. Thanks for this great functionality!
Glad you like it!
Now, it seems to me that fdjac does not accept more than two
matrix arguments.
fdjac always accepts just two arguments: a matrix (vector of
parameter values), and a function call. It seems you're saying
that the function call accepts (or properly processes) just two
arguments, but so far as I can tell that's up to the writer of the
function that is called.
I tried introducing a redundant third matrix into the calls to
Jack's "MPC" function (Guide, Example 5.3, "Delta Method"), as in
matrix m = I(2)
mpc = MPC(&theta, &Y, m)
matrix Jac = fdjac(theta, "MPC(&theta, &Y, m)")
(a third matrix parameter is added to Jack's definition of the MPC
function, but the corresponding argument is ignored within the
function).
The MPC script returned the same results as it did without the
redundant matrix.
It's possible that some weird interaction occurs under some
conditions, but we'd need an example to assess that.
Allin.
6186
days inactive
6188
days old
3 comments
3 participants
participants (3)
-
Allin Cottrell -
Riccardo (Jack) Lucchetti -
Sven Schreiber