Dear All,
I have just pushed a fix to the diffuse filter which I think is a bug.
When estimating models with unit roots, it can happen that for some combinations of deep params line search tries to violate the Blanchard Kahn condition passing through an additional spurious unit root which cannot be killed with the available data. The error message breaks the estimation, while the optimizer can be able to move away from that boundary and converge to a local optimum.
I would like to back port this to 4.3.1: do you agree?
thanks a lot for your assistance best Marco
Marco Ratto marco.ratto@jrc.ec.europa.eu writes:
I have just pushed a fix to the diffuse filter which I think is a bug.
When estimating models with unit roots, it can happen that for some combinations of deep params line search tries to violate the Blanchard Kahn condition passing through an additional spurious unit root which cannot be killed with the available data. The error message breaks the estimation, while the optimizer can be able to move away from that boundary and converge to a local optimum.
I would like to back port this to 4.3.1: do you agree?
If this change is a bugfix, then it should be backported.
OK, I will test it. I think I will also have to use the penalty and not the NaN for fval. Possibly I will commit also this fix later on. thanks Marco
On 9/17/2012 10:31 AM, Sébastien Villemot wrote:
Marco Rattomarco.ratto@jrc.ec.europa.eu writes:
I have just pushed a fix to the diffuse filter which I think is a bug.
When estimating models with unit roots, it can happen that for some combinations of deep params line search tries to violate the Blanchard Kahn condition passing through an additional spurious unit root which cannot be killed with the available data. The error message breaks the estimation, while the optimizer can be able to move away from that boundary and converge to a local optimum.
I would like to back port this to 4.3.1: do you agree?
If this change is a bugfix, then it should be backported.
Dev mailing list Dev@dynare.org https://www.dynare.org/cgi-bin/mailman/listinfo/dev
Dear All,
I have pushed on my personal repo the fixes to diffuse filter in estimation and manually cherry-picked to 4.3 thanks a lot for your assistance best Marco
On 9/17/2012 10:53 AM, Marco Ratto wrote:
OK, I will test it. I think I will also have to use the penalty and not the NaN for fval. Possibly I will commit also this fix later on. thanks Marco
On 9/17/2012 10:31 AM, Sébastien Villemot wrote:
Marco Rattomarco.ratto@jrc.ec.europa.eu writes:
I have just pushed a fix to the diffuse filter which I think is a bug.
When estimating models with unit roots, it can happen that for some combinations of deep params line search tries to violate the Blanchard Kahn condition passing through an additional spurious unit root which cannot be killed with the available data. The error message breaks the estimation, while the optimizer can be able to move away from that boundary and converge to a local optimum.
I would like to back port this to 4.3.1: do you agree?
If this change is a bugfix, then it should be backported.
Dev mailing list Dev@dynare.org https://www.dynare.org/cgi-bin/mailman/listinfo/dev
-- Marco Ratto, Joint Research Centre The European Commission, TP 361, 21027 ISPRA(VA), ITALY Tel: +39 0332 78 9217 Fax: +39 0332 78 5733 marco.ratto@jrc.ec.europa.eu
Marco Ratto marco.ratto@jrc.ec.europa.eu writes:
I have pushed on my personal repo the fixes to diffuse filter in estimation and manually cherry-picked to 4.3
Merged, thanks.
-
Marco Ratto -
Sébastien Villemot -
Sébastien Villemot