October 11, 2004

This package contains two files:

1. break.prg 

This is the main program where you set up your data and options.
The current setup is to test for and estimate shifts in means 
for the US real interest rate series (see Garcia nd Perron (1996), 
"An Analysis of the Real interest Rate under Regime Shifts,"
Review of Economics and Statistics, 78, 111-125).

There are no real limitations on the length of the data set, the number of
breaks, the number of regressors, etc. It depends on the amount of memory
available.  In many cases, the sample size can be a constraint since the
program uses a T(T+1)/2 vector as a storage device to compute global
optimizers (but the constraint is binding, in most cases, only if T is
greater than 500).

The structure of the program is such that what to do should be self
explanatory. The main part is to set up your data in three parts: 

1) the dependent variable (y), 

2) the regressors whose coefficients are allowed to change (z, dimension q)

and 

3) the regressors whose coefficients are not allowed to change (x, dimension
p). Note that even if there are no x's, you should initialize x to something
like x=0. With these you need to specify p, q, and bigt (the effective
sample size).

You also need to specify m, the maximal number of breaks allowed (usually 5
if you want to use the table of critical values in Bai and Perron (1998)).
Also h, which is the minimal length of a segment. When constructing the
various tests, you should use h=int(eps1*T) where eps1 can take values .05,
.10, .15 , .20 and .25 (critical values are available and printed as output
for these cases). If estimation is the sole concern h an be set to any 
value greater than q. If you are using options like robust=1, hetvar=1
hetomega=1 or hetq=1, h or eps1 should be larger.

To have the program perform some task or not, simply put a 1 or a zero to the
corresponding dummy.

2. brcode.src

This is the main code. You should not have to modify it.

Use the file real.dat, which contains the US real interest rate series
mentioned above, to run the program once to see if everything is OK. This
file is included in bp-data.zip.

*****

Note on the October 2004 version: This new version contains a correction for
the construction of the confidence intervals of the estimates of the break
dates as discussed in Zeileis, A. and C. Kleiber (2004), Validating Multiple
Structural Change Models: A Case Study, University of Vienna, Austria, which
is forthcoming in the Journal of Applied Econometrics.

*****

This program is distributed freely for non-profit academic purposes only.
For other uses, please contact Pierre Perron at perron@bu.edu. A lot of
effort has been put to construct this program and we would appreciate that
you acknowledge using this code in your research and cite the relevant
papers on which it is based:

Bai, J. and P. Perron (1998), "Estimating and Testing Linear Models with
Multiple Structural Changes," Econometrica, 66, 47-78.

and

Bai, J, and P. Perron (2003), "Computation and Analysis of Multiple
Structural Change Models," Journal of Applied Econometrics, 18, 1-22.

Although a lot of efforts has been put in constructing the program, we
cannot be held responsible for any consequences that could result from
remaining errors.

Comments about errors, possible improvements and so on are most welcomed and
should be directed to Pierre Perron at perron@bu.edu.

Thank you for your interest and good luck with the program.

Pierre Perron
Department of Economics
Boston University
270 Bay State Road
Boston, MA, 02215
