Gianni Amisano and Maria Letizia Giorgetti, "Entry in Pharmaceutical
Submarkets: A Bayesian Panel Probit Analysis", Journal of Applied
Econometrics, Vol. 28, No. 4, 2013, pp. 667-701.

There are two zip files. The file AG-text.zip contains Matlab program
(.m) files and data files (.csv) in CSV format. The file AG-xls.zip
contains data files (.xls) in XLS format. The .m and .csv files are
ASCII in DOS format, so Unix/Linux users must use "unzip -a". The .xls
files are binary, so the "-a" options must not be used.

The structure of the directories underneath the top-level directory
(AG-text or AG-xls) is the same, except that AG-text contains a
directory called BPP, and AG-xls does not.


How to replicate the results:

The first thing to do is to download the zip files and extract
everything. Then copy the .xls files from their directories under
AG-xls to the corresponding directories under AG-text.

The directory BPP (Bayesian panel probit) contains all the routines
needed to run the model. You need to tell Matlab where to find these
routines. It will automatically find them if they are in the current
directory, or you can use the "path" or "addpath" commands within
Matlab, or you can set the MATLABPATH environment variable.

Besides the BPP directory, there are three other directories:

AE (any entry)
GF (Greenfield entry)
NGF (non Greenfield entry)

These directories contain the datasets and codes necessary to produce the
results described in the paper. The first two directories contain both
UR (unrestricted) and RE (restricted) directories. These
subdirectories refer to the unconstrained and constrained estimations
contained in the paper. The third (NGF) contains only a UR directory.

Typically, for each set of analyses, there will be 16 different
submarkets, and for each of them there will be a main *.m Matlab
script which carries out the computations calling a particular *.xls
dataset. Each of the .xls datasets has a corresponding .csv file,
included to meet the JAE requirements but not needed for carrying out
the computations (the .xls file is enough). Please note that the .csv
files have the same structure as the .xls files.

It is possible to control the specification of the probit model to be
estimated by directly editing lines 2 and 3 of each .xls data file.
For instance, if you want to exclude a particular covariate, set to
zero the element of line 2 corresponding to that covariate. You can
also accommodate different ways to account for initial conditions in the
Wooldridge (2005) approach that we use (see paper). This is done by
using line 3. For each covariate, the corresponding cell on line 3 can
be set to:

1 if sample mean of variable is used to condition random effects on
  (use this only on truly exogenous variables)

2 when only the initial condition is used to condition random effects on.

3 when no conditioning on that variable is done (use this for
  variables having only time variation).
 
Please, let us know whether there are issues in running the code. 

Drop us a line at:

  gianni.amisano [AT] ecb.europa.eu
  or
  amisano [AT] eco.unibs.it

Have fun!
Gianni (on behalf of Letizia too).