Model with collateral constraints

Hello everyone!

I am trying to construct a model close to Iacoviello and Neri (2010) but including also a collateral constraint for capital goods purchases. It would be also similar to the model of Gerali et al. (2010) but without credit supply rigidities. When I use the commands “check” and “steady” everything seems correct and I am able to perform stochastic simulation (resulting in strange IRFs) but when it comes to estimation the Blanchard Kahn conditions are not satisfied.
I have checked the timing and the parameters for stochastic simulation are calibrated at the prior mean. At this point, I really do not understand where this problem is coming from. I would really appreciate any insight.

estimationDATA.mat (16.1 KB)
macroprudential.mod (13.3 KB)
macroprudential_steadystate.m (6.9 KB)

You are missing the Excel file.

So sorry, I have edited the post and include the mat file instead of the m file.
Thank you very much for your response.

Why do you provide explicit starting values? You should simply use the calibration that works to start mode-finding.

I have deleted the starting values as you suggest. Now, I get the error:
initial_estimation_checks:: The forecast error variance in the multivariate Kalman filter became singular.

I have one more shock than observables and as far I understand there is not any single equation that imply an exact linear combination of an observable. Anyway, I reduced the number of observables and the mode routine 1 is able to start but it stops after few iterations at a huge (negative) log posterior.
Moreover if I run the identification(advanced=1) command I see that the collinearity across most of the parameter is close to 1.

I guess that all these things are signs that there is something wrong in the model but I do not know where to look at because at this point every equation seems correct to me. I am sure that any clue from you will be very helpful.

Thank you so much in advanced.

The updated .mod file:
macroprudential.mod (13.3 KB)

Mode-finding for such a large model takes time and a lot of iterations. But you limited the number of iterations. With mode_compute=5, there was a significant improvement, but still the maximum number of iterations was reached:
macroprudential_mode.mat (1012 Bytes)