Help: estimate negative correlation between two shocks in Bayesian estimation

Estimation ML/Bayesian estimation
1

Hi all,

I am working on a DSGE model with two shocks that are potentially correlated, and I expect the correlation to be negative. I read in the Dynare manual that it is possible to estimate the correlation between shocks in the Bayesian estimation by specifying a prior and a proper distribution.

However, I am not sure what prior distribution can be used to estimate a negative correlation. I tried both normal and uniform distributions as priors for the correlation parameter, but both caused an error during Bayesian estimation:

Error using chol
Matrix must be positive definite.
Error in simult (line 80)
chol_S = chol(DynareModel.Sigma_e(i_exo_var,i_exo_var));
If I remove the correlation prior line (attached file, line 522;
corr eta_pds, eta_c, -0.3, , normal_pdf, -1, 1;), the model runs without any issues.

I would appreciate any advice on how to properly estimate a negative correlation between shocks in the Bayesian setup.

Any suggestions or help would be very welcome.

Thank you so much!

Best,
Yamoi
param_and_ss1.mat (6.0 KB)
model5.mod (17.9 KB)
data.xlsx (17.5 KB)

2

Your syntax for the distributions was wrong. You set the prior mean to -1. Use

corr eta_pds, eta_c, -0.3, ,, uniform_pdf, , , -1, 1;
3

That’s helpful. It works perfectly. Thank you, Prof. Pfeifer.

Best,
Yamoi