Notes for Archive item TPMH0101

tpmh0101.zip contains a paper by  Mark Horridge and Ken Pearson, "Systematic
Sensitivity Analysis with Respect to Correlated Variations in Parameters and
Shocks" which explains how to do Systematic Sensitivity Analysis (SSA)  with
GEMPACK, when you want to
* simultaneously vary parameters AND shocks; or
* assume that variationss in parameters and/or shocks are correlated
  according to a specified covariance matrix.

tpmh0101.zip also contains two further zip files:

* gq.zip contains an example TABLO program, gq.tab, created to
  experiment with SSA possibilities.

* corssens.zip is a complete RunGTAP version archive
         -- its use is described in the document.

GEMPACK must be installed to run the gq programs.

gq.zip must be unzipped in a fresh folder, and the examples run from the
command line. The script DOGQ.BAT runs the program 4 times, using the following
four input files, each containing all samples of a dummy parameter length 7
from a particular SSA run:
   uniliu.HAR         uniform distribution, Liu quadrature
unistroud.HAR         uniform distribution, Stroud quadrature
   triliu.HAR      triangular distribution, Liu quadrature
tristroud.HAR      triangular distribution, Stroud quadrature
The above files were producing by 
(a) asking the SSA procedue to save SLC files,
(b) combining the SLC files into one, using cmbhar.sif, and 
(c) extracting just the relavant header.
It is interesting to compare the different samples created by Stroud and Liu.

The gq program also reads from parm.har which contains target means and
covariances.

The gq program also shows how to use the Cholesky decomposition of the
correlation matrix, instead of the Cholesky decomposition of the covariance
matrix.

Note: Facilities for doing SSA have been much improved in the version of RunGEM
prepared for GEMPACK Release 11 (avail mid-2011), and in the matching version
of RunGTAP. However, examples supplied here work OK with GEMPACK Release 10.   

Mark Horridge December 2010

