Ordered Multidimensional Arrays

Hi,
I am inquiring as to what are the best practices with respect to storing and
manipulating ordered multi-dimensional arrays.  For example, suppose I have
a sequence of time-varying covariance matrices of asset returns.  The data
is ordered, but the ordering is not necessarily regular (e.g. daily data
omitting weekends and holidays, etc.).  The data array is say, N x N x T.
For example, the first two elements may look as follows:

  

*result$covariance[,,1:2]
    

, , 1*
*             [,1]         [,2]         [,3]         [,4]
[1,] 1.511137e-06 1.918668e-06 1.201553e-06 3.205271e-06
[2,] 1.918668e-06 7.488916e-06 6.593317e-06 1.203421e-05
[3,] 1.201553e-06 6.593317e-06 1.305861e-05 2.132272e-05
[4,] 3.205271e-06 1.203421e-05 2.132272e-05 4.571225e-05*
*, , 2*
*             [,1]         [,2]         [,3]         [,4]
[1,] 1.500858e-06 1.905574e-06 1.193412e-06 3.183290e-06
[2,] 1.905574e-06 7.444871e-06 6.555459e-06 1.195876e-05
[3,] 1.193412e-06 6.555459e-06 1.297075e-05 2.117777e-05
[4,] 3.183290e-06 1.195876e-05 2.117777e-05 4.551706e-05*

I would like to be able to partition this sequence of matrices by date and
by individual element.  Partitioning by individual elements is trivial;
however, partitioning by time stamp is not (especially if the partitioned
data set must be carried through a number of downstream calculations).  I
could carry the data in a list complete with a date vector and the data
array, and partition the list as I go, but this seems somewhat clunky.  Any
ideas?  A "zoo"-like package capable of handling multidimensional arrays
would be optimal, but I don't believe this exists.

Thanks,
Derek

	[[alternative HTML version deleted]]

The old fashioned solution is to have the N x N x T
array and use character strings of the dates as the
dimnames on the third dimension.

Is there something you think you need to do that is
hard with such a setup?


Patrick Burns
patrick at burns-stat.com
+44 (0)20 8525 0696
http://www.burns-stat.com
(home of S Poetry and "A Guide for the Unwilling S User")

Derek Schaeffer wrote:

Hi,
I am inquiring as to what are the best practices with respect to storing
and
manipulating ordered multi-dimensional arrays.  For example, suppose I
have
a sequence of time-varying covariance matrices of asset returns.  The data
is ordered, but the ordering is not necessarily regular (e.g. daily data
omitting weekends and holidays, etc.).  The data array is say, N x N x T.
For example, the first two elements may look as follows:

*result$covariance[,,1:2]

, , 1*
*             [,1]         [,2]         [,3]         [,4]
[1,] 1.511137e-06 1.918668e-06 1.201553e-06 3.205271e-06
[2,] 1.918668e-06 7.488916e-06 6.593317e-06 1.203421e-05
[3,] 1.201553e-06 6.593317e-06 1.305861e-05 2.132272e-05
[4,] 3.205271e-06 1.203421e-05 2.132272e-05 4.571225e-05*
*, , 2*
*             [,1]         [,2]         [,3]         [,4]
[1,] 1.500858e-06 1.905574e-06 1.193412e-06 3.183290e-06
[2,] 1.905574e-06 7.444871e-06 6.555459e-06 1.195876e-05
[3,] 1.193412e-06 6.555459e-06 1.297075e-05 2.117777e-05
[4,] 3.183290e-06 1.195876e-05 2.117777e-05 4.551706e-05*

I would like to be able to partition this sequence of matrices by date and
by individual element.  Partitioning by individual elements is trivial;
however, partitioning by time stamp is not (especially if the partitioned
data set must be carried through a number of downstream calculations).  I
could carry the data in a list complete with a date vector and the data
array, and partition the list as I go, but this seems somewhat clunky.
 Any
ideas?  A "zoo"-like package capable of handling multidimensional arrays
would be optimal, but I don't believe this exists.

Thanks,
Derek

       [[alternative HTML version deleted]]