On 3/2/2003 I posted minor updates to the code, documentation, and demos. The last substantial update to the code was on 12/1/00 (improved handling of user-defined weighting matrices, some additional diagnostics, and modification of the way the gradient (derivative of objective function) is calculated).
The software below is provided as is. Although I believe it is bug-free, there are no guarantees. I will try to maintain a list of known problems and suggested remedies. Please report any suspected errors to Mike Cliff mcliff@vt.edu.
If you send an e-mail with the subject "Matlab User" I will add you to a mailing list to notify you of enhancements and additions. Version numbers appear in brackets after function name. All files in the GMM and MINZ packages were numbered version 1.1 on 11/6/99, regardless of the data the function was originally written or the number of previous revisions. I will try to keep track of version changes and repost files.
If you have an older version of the software you should consider upgrading all files as several enhancements have been made to the GMM library. The handling of the weighting matrix is different now. The old procedure gmmW.m is replaced by an enhanced routine gmmS.m. The former file calculated the weighting matrix assuming the user wanted to use the ``optimal'' matrix (inverse Spectral Density matrix). Now the handling of the weighting matrix and the Spectral Density matrix are more decoupled, allowing for pre-specified weighting matrices. The Spectral Density matrix now handles the Andrews/Andrews-Monahan procedure, in addition to the White, Hansen (truncated), Newey-West (Bartlett), and Gallant (Parzen) procedures. Other enhancements include better printing facilities (can now easily direct output to screen, file, or suppress), plotting diagnostics, and more efficient coding in some instances.
The software is designed for use with Matlab 5.x. It WILL NOT work with Matlab 4.x since I make heavy use of the structure variables. It may require some of the Matlab Toolboxes, I was not particularly careful in keeping track. I also make use of several functions I have created and I may have inadvertantly excluded them below. If you have error messages about missing .m-files, check that your put the files below in your search path. If it still seems that there are missing files, let me know and I will see what you need.
The files for GMM and MINZ are packaged into two files for convenience. Each of these files is available in .tar.gz form for unix and .zip form for M$ PCers. Since the .zip file was made in unix I can't guarantee it will sucessfully unzip on a PC but it is supposed to. To unzip the files in unix issue the commands
gunzip gmm.tar.gz
tar xvf gmm.tar
gunzip minz.tar.gz
tar xvf minz.tar.
The library program files will be put in the current directory (mine is called gmm) and the demos will be put in a subdirectory gmmex. The PC files should be unzipped with command like pkunzip gmm.zip and pkunzip minz.zip from the directory where you want the files to go. Make sure you update the search path in Matlab. For my filestructure the Matlab command is path(path,'/home/mcliff/matlab/metrics/gmm').
GMM Documentation
GMM files
MINZ files
Misc Utilities (needed for GMM)
Hansen-Jagannathan and Efficient Frontier Programs
Here is a listing of the individual files contained in the zip files above. The links below do not work. DO NOT email me to tell me that!