Readme file
SERIES B
 Statistical Methodology
Multivariate bandwidth selection for local linear regression by L. Yang and R. Tschernig, Journal of the Royal Statistics Society, Series B, Statistical Methodology, Volume 61 (1999), Part 4, pages 793 - 815.
Using this software requires the software GAUSS and the GAUSS library OPTMUM.
INSTALLATION:
If the GAUSS library OPTMUM is not installed, you have to install it first.
WINDOWS 95/NT
0) unzip multband.zip. It includes:
multband.tes example file to check and run subroutines
The program multband.tes is a program
file that allows you to test whether
the programs work. For Model 1 and Model 3
in Section 6 in YT you can compute the
scalar and diagonal rule-of-thumb and
plug-in bandwidths as described in YT.
Moreover you can reproduce (except for
different random values) Figure 3e) in
YT. Of course, you can change the file
so that you can use the procedure for
arbitrary real or artificial data.
multband.dec declaration file for GAUSS library multband.src
multband.src GAUSS library for multivariate
plug-in bandwidth estimation
multband.lcg library file of GAUSS library multband.src
density.dll DLLs for nonparam. density estimation
loccubg.dll partial local cubic estimation
locling.dll local linear estimation
of functions and derivatives used by routines
in the GAUSS library multband.src
1) copy density.dll, loccubg.dll, locling.dll into $(GAUSSDIR)/dlib
where $(GAUSSDIR) denotes the home directory of GAUSS
2) copy multband.dec, multband.src into $(GAUSSDIR)/src. All the
subroutines are collected into a GAUSS library called
MULTBAND.SRC which has to be called with the library command
in the main GAUSS routine.
copy multband.lcg into $(GAUSSDIR)/lib
3) copy multband.tes into your working directory. It is recommended
that each directory name is not longer than 8 characters
since some GAUSS versions cannot handle longer names correctly.
4) start GAUSS again and edit the file multband.tes. Although GAUSS
should find the DLLs in the directory $(GAUSSDIR)/dlib, it does
not. Therefore, the path of the DLL files must be included
at the beginning of the file multband.tes or any other main
file that uses multband.src. If gauss_nt is the GAUSS directory,
the command would be
dlibrary c:\gauss_nt\dlib\locling.dll,
c:\gauss_nt\dlib\density.dll,
c:\gauss_nt\dlib\loccubg.dll;
The procedure also requires the optimizing routine of GAUSS (and if
you want to plot the function the publication quality graphics
library). Therefore, you need the command
library optmum,multband,pgraph;
If you run the program, it may take some seconds until you see
the result if n=500.
The procedure was checked with GAUSS for Windows NT/95 Version 3.2.29 ( May 22 1997 )
Sun-UNIX
Be careful that all files are in lower case letters on UNIX.
Otherwise, GAUSS does not find the files.
0) Extract the multband.tar file with the command:
tar xvf multband.tar
Note that some ftp software changes letters to uppercase
when transfering files from DOS to UNIX. So if your
tar command cannot find multband.tar make, check
whether there is a MULTBAND.TAR. Then use
cp MULTBAND.TAR multband.tar
and use again the tar command.
This unpacks the following files for UNIX:
multband.tes example file to check and run subroutines
The program multband.tes is a program
file that allows you to test whether
the programs work. For Model 1 and Model 3
in Section 6 in YT you can compute the
scalar and diagonal rule-of-thumb and
plug-in bandwidths as described in YT.
Moreover you can reproduce (except for
different random values) Figure 3e) in
YT. Of course, you can change the file
so that you can use the procedure for
arbitrary real or artificial data.
multband.dec declaration file for GAUSS library multband.src
multband.src GAUSS library for multivariate
plug-in bandwidth estimation
multband.lcg library file of GAUSS library multband.src
den_c.so DLLs for nonparam. density estimation
loccubg.so partial local cubic estimation
locling.so local linear estimation
of functions and derivatives used by routines
in the GAUSS library multband.src
1) if you do not have a local GAUSS directory in your home directory
with dlib, lib and src subdirectories and if you cannot access
the GAUSSHOME directories for writing, then
create the following directories in your home directory <USERHOME>
with the following commands
mkdir Gauss
cd Gauss
mkdir dlib
mkdir lib
mkdir src
2) if you did 1), then
- copy the gauss.cfg file from the GAUSSHOME
directory into <USERHOME>/Gauss
- copy all *.lcg files from <GAUSSHOME>/lib
into your <USERHOME>/Gauss/lib directory
edit your local gauss.cfg file:
- add
<USERHOME>/Gauss/src
at the end of the path variable "src_path"
- replace the standard path for the path variable "lib_path" by
<USERHOME>/Gauss/lib
3) if you did 1) and 2)
go back into your home directory and edit your .cshrc file. You
need to add two lines that tell GAUSS during starting to use
your local gauss.cfg GAUSS configuration file which contains
your local paths
setenv GAUSS_CFG <USERHOME>/Gauss
setenv GAUSSHOME <GAUSSHOME>
4) copy den_c.so, loccubg.so, locling.so into (USERHOME)/Gauss/dlib
or in the existing dlib directory
5) copy multband.dec, multband.src into $(USERHOME>/Gauss/src or in
the existing src directory.
All the subroutines are collected into a GAUSS library called
MULTBAND.SRC which has to be called with the library command.
copy multband.lcg into $<USERHOME>/Gauss/lib or in
the existing lib directory
6) copy multband.tes into your working directory.
7) start GAUSS again and edit the file multband.tes.
You have to adjust the path for the DLLs at the beginning of
the file multband.tes or any other main
file that uses multband.src
dlibrary <USERHOME>/Gauss/dlib/locling.so,
<USERHOME>/Gauss/dlib/den_c.so,
<USERHOME>/Gauss/dlib/loccubg.so;
The procedure also requires optimizing routine of GAUSS (and if
you want to plot the function the publication quality graphics
library). Therefore, you need the command
library optmum,multband,pgraph;
If you run the program, it may take some seconds until you see
the result if n=500.
USER INSTRUCTIONS are given in the GAUSS library multband.src that contains all
subroutines. The two main routines to call are bandrot.g for rule-of-thumb
bandwidths and bandpi.g for the plug-in estimates. Each subroutine has a header
that explains the input and output variables and its function. The file
multband.tes provides an example how to use the bandwidth estimation routines.
The library does not include a subroutine that produces automatically all
possible blocking schemes. They have to be provided by the user. Examples for
two dimensional models and several sample sizes are given in the multband.tes
file.
If you encounter problems, please write an email to
Rolf Tschernig
Professor of Econometrics
Department of Economics and Econometrics
University of Regensburg
Universitätsstraße 31
D-93040 Regensburg
Tel. (+49) 941/943 2737
Fax (+49) 941/943 4917
email: rolf.tschernig@wiwi.uni-regensburg.de
http://www.wiwi.uni-regensburg.de/tschernig
|
