Scaling problems in a dynamic Ramsey model using MPSGE

[Archiver]

Dear Gamsworld members,

I am struggling to solve an MPSGE model with a high level of growth and interest rate. While my own model is quite long, I have been able to replicate the problem in a dynamic Ramsey model (please see the attachment, the model can also be found at http://www.gamsworld.org/mpsge/debreu/p ... tml#move31).

This model solves perfectly when R = 0.05 and G = 0.02, but for higher numbers (R = 0.13 and G = 0.08), it shows 'other error'. The problem in my opinion is that PREF and QREF diverge far away from unity as T progresses, which makes it difficult for the solver to converge (which can be seen in the Jacobian norm results which diverge a lot). In a GAMS code, I could solve it with .SCALEOPT but that does not works with MPSGE. I have found that scaling SAM will also not work because the problem is actually with PREF and QREF which do not really depend on the SAM.

Can anyone suggest me an idea on how to improve scaling or create a workaround. I will really appreciate if you can direct me to a model or an example where the coders incorporated scaling in an MPSGE model.

Please help!! Thanks a lot.

(Please note that while the R and G values I suggested are on the higher side to be theoretically useful, I chose them to demonstrate the problem which exists at smaller values in my code).

--

[Archiver]

Dear Frank,

That sounds like a good strategy. If you have an example code you can share, then I will be highly obliged. It would be helpful to see it working in a correct code to learn. But if you don't have any, that is okay too.

Thanks a lot.

On Tuesday, February 9, 2016 at 11:02:23 AM UTC-5, Frank Ward wrote:

Dear Ashish,



You might try solving the model for a low interest and growth rate, then restart it from the 1st solution using a progressively higher interest and growth rates. I’ve found that sneaking up on the parameter you really want can be a nice strategy for success, rather than hitting GAMS with it right away.



Frank Ward



From: gams...@googlegroups.com [mailto:gams...@googlegroups.com] On Behalf Of Ashish Tyagi
Sent: Monday, February 08, 2016 3:33 PM
To: gamsworld
Subject: Scaling problems in a dynamic Ramsey model using MPSGE



Dear Gamsworld members,



I am struggling to solve an MPSGE model with a high level of growth and interest rate. While my own model is quite long, I have been able to replicate the problem in a dynamic Ramsey model (please see the attachment, the model can also be found at http://www.gamsworld.org/mpsge/debreu/p ... tml#move31).



This model solves perfectly when R = 0.05 and G = 0.02, but for higher numbers (R = 0.13 and G = 0.08), it shows 'other error'. The problem in my opinion is that PREF and QREF diverge far away from unity as T progresses, which makes it difficult for the solver to converge (which can be seen in the Jacobian norm results which diverge a lot). In a GAMS code, I could solve it with .SCALEOPT but that does not works with MPSGE. I have found that scaling SAM will also not work because the problem is actually with PREF and QREF which do not really depend on the SAM.



Can anyone suggest me an idea on how to improve scaling or create a workaround. I will really appreciate if you can direct me to a model or an example where the coders incorporated scaling in an MPSGE model.



Please help!! Thanks a lot.



(Please note that while the R and G values I suggested are on the higher side to be theoretically useful, I chose them to demonstrate the problem which exists at smaller values in my code).

--
To unsubscribe from this group and stop receiving emails from it, send an email to gamsworld+...@googlegroups.com.
To post to this group, send email to gams...@googlegroups.com.
Visit this group at https://groups.google.com/group/gamsworld.
For more options, visit https://groups.google.com/d/optout.

--
To unsubscribe from this group and stop receiving emails from it, send an email to gamsworld+unsubscribe@googlegroups.com.
To post to this group, send email to gamsworld@googlegroups.com.
Visit this group at https://groups.google.com/group/gamsworld.
For more options, visit https://groups.google.com/d/optout.

[Archiver]

Reply-to: gamsworld@googlegroups.com

Hi Frank,

Thanks a lot. You have been a tremendous help. I am sending you an email as you suggested.

Thanks again.

On Thursday, February 11, 2016 at 12:46:16 PM UTC-5, Frank Ward wrote:

Hi Ashish,



I’m attaching a groundwater pumping economic optimization that re-solves for two combinations of two parameters each using a looping approach. Something similar could work in which you start with parameters that work that you’re not interested in, then raise their value gradually to reach the parameter values you’re really interested in.



You might want to send me your email so I can communicate details with you that might not be as interesting to the whole list serve.



FAW



From: gams...@googlegroups.com [mailto:gams...@googlegroups.com] On Behalf Of Ashish Tyagi
Sent: Wednesday, February 10, 2016 4:35 PM
To: gamsworld
Subject: Re: Scaling problems in a dynamic Ramsey model using MPSGE



Dear Frank,



That sounds like a good strategy. If you have an example code you can share, then I will be highly obliged. It would be helpful to see it working in a correct code to learn. But if you don't have any, that is okay too.



Thanks a lot.

On Tuesday, February 9, 2016 at 11:02:23 AM UTC-5, Frank Ward wrote:

Dear Ashish,



You might try solving the model for a low interest and growth rate, then restart it from the 1st solution using a progressively higher interest and growth rates. I’ve found that sneaking up on the parameter you really want can be a nice strategy for success, rather than hitting GAMS with it right away.



Frank Ward



From: gams...@googlegroups.com [mailto:gams...@googlegroups.com] On Behalf Of Ashish Tyagi
Sent: Monday, February 08, 2016 3:33 PM
To: gamsworld
Subject: Scaling problems in a dynamic Ramsey model using MPSGE



Dear Gamsworld members,



I am struggling to solve an MPSGE model with a high level of growth and interest rate. While my own model is quite long, I have been able to replicate the problem in a dynamic Ramsey model (please see the attachment, the model can also be found at http://www.gamsworld.org/mpsge/debreu/p ... tml#move31).



This model solves perfectly when R = 0.05 and G = 0.02, but for higher numbers (R = 0.13 and G = 0.08), it shows 'other error'. The problem in my opinion is that PREF and QREF diverge far away from unity as T progresses, which makes it difficult for the solver to converge (which can be seen in the Jacobian norm results which diverge a lot). In a GAMS code, I could solve it with .SCALEOPT but that does not works with MPSGE. I have found that scaling SAM will also not work because the problem is actually with PREF and QREF which do not really depend on the SAM.



Can anyone suggest me an idea on how to improve scaling or create a workaround. I will really appreciate if you can direct me to a model or an example where the coders incorporated scaling in an MPSGE model.



Please help!! Thanks a lot.



(Please note that while the R and G values I suggested are on the higher side to be theoretically useful, I chose them to demonstrate the problem which exists at smaller values in my code).

--
To unsubscribe from this group and stop receiving emails from it, send an email to gamsworld+...@googlegroups.com.
To post to this group, send email to gams...@googlegroups.com.
Visit this group at https://groups.google.com/group/gamsworld.
For more options, visit https://groups.google.com/d/optout.

--
To unsubscribe from this group and stop receiving emails from it, send an email to gamsworld+...@googlegroups.com.
To post to this group, send email to gams...@googlegroups.com.
Visit this group at https://groups.google.com/group/gamsworld.
For more options, visit https://groups.google.com/d/optout.

--
To unsubscribe from this group and stop receiving emails from it, send an email to gamsworld+unsubscribe@googlegroups.com.
To post to this group, send email to gamsworld@googlegroups.com.
Visit this group at https://groups.google.com/group/gamsworld.
For more options, visit https://groups.google.com/d/optout.