stepan
/
dynare
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity

Changed priors in tests/optimizers/fs2000_* integration tests. 

To avoid asymptote at zero on the autoregressive parameter rho.
time-shift
Stéphane Adjemian (Charybdis) 2017-02-08 12:36:55 +01:00
parent 5b73d7d59c
commit b14cf33c8f
1 changed files with 14 additions and 5 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

19
tests/optimizers/fs2000.common.inc
View File

@ -1,7 +1,7 @@
/*
* This file replicates the estimation of the cash in advance model described
* Frank Schorfheide (2000): "Loss function-based evaluation of DSGE models",
* Journal of Applied Econometrics, 15(6), 645-670.
* This file replicates (except for a small difference in the priors)the estimation of
* the cash in advance model described Frank Schorfheide (2000): "Loss function-based
* evaluation of DSGE models", Journal of Applied Econometrics, 15(6), 645-670.
*
* The data are in file "fs2000/fsdat_simul.m", and have been artificially generated.
* They are therefore different from the original dataset used by Schorfheide.
@ -17,7 +17,7 @@
*/
/*
* Copyright (C) 2004-2010 Dynare Team
* Copyright (C) 2004-2017 Dynare Team
*
* This file is part of Dynare.
*
@ -106,7 +106,16 @@ alp, beta_pdf, 0.356, 0.02;
bet, beta_pdf, 0.993, 0.002;
gam, normal_pdf, 0.0085, 0.003;
mst, normal_pdf, 1.0002, 0.007;
rho, beta_pdf, 0.129, 0.223;
/*
* The prior on the autoregressive parameter rho is different from the one considered in the original paper.
* We lowered the prior variance to ensure that the prior does not have an asymptote at zero, which causes
* troubles for some optimizers. For instance, mode_compute=12 (which relies on Mathworks' particleswarm
* routine) finds a posterior at zero for rho (with a much higher posterior density than the one reported
* by csminwell or other optimization routines). Because zero is on the boundary of the prior, this is an
* an issue when we have to compute the hessian matrix (which is not full rank) at the mode or if we try to
* use this solution as an initial guess for csminwell.
*/
rho, beta_pdf, 0.129, 0.100;
psi, beta_pdf, 0.65, 0.05;
del, beta_pdf, 0.01, 0.005;
stderr e_a, inv_gamma_pdf, 0.035449, inf;