Sorry, I don't understand your design. I try to show, what/how I best
understand and what questions are open:
Ronald Bloom schrieb:
...
> So let's say I have 51 variables. I collect them 30 days at a time.
I assume the following scheme:
data of 30 days data of 30 days data of 30 days
var01 ! ! ! ! ! !
var02 ! ! ! ! ! !
var03 ! ! ! ! ! !
... ! ! ! ! !
var50 ! ! ! ! ! !
var51 ! ! ! ! ! !
+-------------------+ +-------------------+ +-------------------+
D1 D2 D3
>
> Every time I collect a new set of observations I want to assess
> its "surprise value", in respect of the recent past.
... a surprise-value of the whole new set D2 respectively to D1...
>
> So I have a monthly covariance matrix of rank 30. There should be
> 30 non-zero eigenvalues.
cov01 = D1*D1'/30
cov02 = D2*D2'/30
...
> If I get one more observation (of 51 items)
"one more observation" ????
???? a.1) One single observation or
a.2) a new set of 30 days-observations?
> can I map it into the reduced dimension subspace of size 30,
???? b.1) you try to find a composition of D1-days to match your new single
observation
b.2) you try to match the factor-structure of the new set of observations
D2 with that of D1 ?
> and evaluate it for "extremity" using a multivariate prediction
> "region" using the 30x30 diagonal matrix of eigenvalues in
> the standard "hotelling form" on the reduced space?
b.3) you try to match the new eigenvalue-structures of D2 to the
old of D1 ??
>
> I'm just trying to translate theory into something I can
> actually practise.
>
Gottfried Helms
--
Gottfried Helms
Univ. Kassel
.
.
=================================================================
Instructions for joining and leaving this list, remarks about the
problem of INAPPROPRIATE MESSAGES, and archives are available at:
. http://jse.stat.ncsu.edu/ .
=================================================================
