=======================================================================
Version: 22 February, 2008
-----------------------------------------------------------------------
Ley, E. and M.F.J. Steel (2008) "On the Effect of Prior Assumptions
in Bayesian Model Averaging with Applications to Growth Regression"
Journal of Applied Econometrics
=======================================================================
0. Files
ls6bmalps.f 140Kb f77 code
ls6bma.par 4Kb parameter file
grk41t72.dat 20Kb data file from FLS
grk54t93.dat 68Kb data file from MP
grk67t88.dat 80Kb data file from SDM
1. Code
In order to reproduce the results in the paper, youÕll need to compile the
f77 file ls6bmalps.f and generate an executable, say, bma.exe. Then youÕll
need to place a data file (*.dat) and a parameter file (ls6bma.par) in the
same directory. The file flsbma.par controls some options as explained
below.
The successful execution will always produce an output file *.out, and (if
the option to draw inference subsamples is set to TRUE) then it will also
produce additional output files.
2. Parameter File
The flsbma.par file controls some of the execution time parameters.
[1] The first line sets the name of the different output files (the LPS
*.dat files will have Ô 1.datÕ and Ô 9.datÕ appended to this name. (These
dat files were later processed with R to produce the graphs in Figures 6
and 7 in the paper.)
[2] The second line must contain the exact name of the data file.
[3] The rest of the parameters are fairly self-explanatory. If in doubt,
check the routine setup in flsbmalps.f to see what the parameters do. For a
description of prediction issues and the G&M convergence estimate refer to
FLS, and random theta priors are discussed in LS6. Jointness issues are
described in LS5.
An example of a flsbma.par file follows:
k41RthetaMH OUT namefile, change at wish
grk41t72.dat DAT namefile
-212166542 random number seed, enter any negative integer
9 integer (1-9) specifiying prior, see routine computefj
100 warmup draws in thousands (use 1 for initial testing)
500 chain draws (use 5 for initial testing)
F standardise Xs??
F Loop to compute LPS?
100 If qbove yes, then # LPSloop?
0.85d0 real taking care of sample split
F wrpost?
F dogm---do G&M stuff? convergence
F dojoint---do jointness stuff?
F thetafixed: T for fixed theta, F for random theta
20.5 emodsize---expected model size
3. Data
There are 3 datasets (plain text files):
grk41t72.dat [FLS]
grk54t93.dat [MP]
grk67t88.dat [SDM]
The 3 *.dat files are plain ascii files with the following structure:
Line 1..................int T: the number of observations
Line 2..................int K: the total number of possible regressors in Z
Lines 3 to (K+2)........char*8: the varnames for the K regressors
Lines (K+3) to (K+2+T)..each line contains an obs of (y,Z) in free format
(i.e., K+1 numbers, y and the K vars in Z)
4. References
[FLS] Fernandez, C., E. Ley and M.F.J. Steel (2001) "Model Uncertainty in
Cross-Country Growth Regressions" Journal of Applied Econometrics, 16:53-76
[LS5] Ley, E. and M.F.J. Steel (2007) "Jointness in Bayesian Variable
Selection with Applications to Growth Regression" Journal of Macroeconomics
29(3): 476-493
[LS6] Ley, E. and M.F.J. Steel (2008) "On the Effect of Prior Assumptions
in Bayesian Model Averaging with Applications to Growth Regression" Journal
of Applied Econometrics
[MP] Masanjala, W. and C. Papageorgiou (2005) "Initial Conditions,
European Colonialism and AfricaÕs Growth," unpublished (Baton Rouge:
Department of Economics, Louisiana State University)
[SDM] Sala-i-Martin, X.X., G. Doppelhofer and R.I. Miller (2004)
"Determinants of long-term growth: A Bayesian averaging of classical
estimates (BACE) approach" American Economic Review 94: 813-83
==========================================================================