Error with Bayesian estimation (but not with DSGE VAR)?

Dear jpfeifer,

My coauthor and I have been searching over the forum to try to sort out the following error message. We understand you have advised on several similar posts and suggested the problem could mostly be ‘inconsistency between the data and the observation equations’, or the algorithm ‘had problem finding a mode’… But in our case we find our codes work well if we activate lines 383 and 384, making it a DSGE-VAR estimation (which seems ruling out the data/obs equation inconsistency problem?). Then, if we replace ‘dsge_var=100000’ with ‘dsge_var=inf’ to make it equivalent to a pure Bayesian, the error returns… Are we right to think that this means there is a problem finding the posterior mode when the model is estimated with pure Bayesian (as attached), but not with DSGE-VAR (if lines 383 and 384 are activated)? Is this normal? We really are struggling to understand why this is happening; and we would be very grateful for your advice. The familiar error message is as the following (Thank you!!!):

(It always stops when ‘tuning the scale of the parameters’, just before ‘climbing hills’)

Error using chol
Matrix must be positive definite.

Error in gmhmaxlik (line 197)
dd = transpose(chol(CovJump));

Error in dynare_estimation_1 (line 437)
[xparam1,PostVar,Scale,PostMean] = …

Error in dynare_estimation (line 89)
dynare_estimation_1(var_list,dname);

Error in Formal_DVAR_Lag1_initial_inf (line 675)
dynare_estimation(var_list_);

Error in dynare (line 180)
evalin(‘base’,fname) ;
test1.mod (18.3 KB)
data1.rar (4.73 KB)

There are two issues with the data here:

  1. Many of your series show an extreme seasonal pattern
  2. Some of your series have a mean that is extremely unlikely given the model. The output growth rate has a mean of about 0.02, but according to your prior

TREND_AC , 0.025 ,0.00001, 0.01 , normal_pdf , 0.005 , 0.01 ;
it can be at most 0.01.

[quote=“jpfeifer”]There are two issues with the data here:

  1. Many of your series show an extreme seasonal pattern
  2. Some of your series have a mean that is extremely unlikely given the model. The output growth rate has a mean of about 0.02, but according to your prior

TREND_AC , 0.025 ,0.00001, 0.01 , normal_pdf , 0.005 , 0.01 ;
it can be at most 0.01.[/quote]

Thanks for your quick reply jpfeifer. I will try to revise following your suggestion and get back.

Thanks again,
Zhirong