Hi Houtan,

discretionary_policy should only be used with a linear model and not a linearIZED model (IMHO we could consider imposing in the parser this restriction). Therefor there is no reason to accept option loglinear with discretionary_policy

Best

Michel

Houtan Bastani writes:

Hello,

In working on https://git.dynare.org/Dynare/dynare/issues/1197 I came across a potential bug.

As you can see from lines 22-24 & 39 of https://git.dynare.org/Dynare/dynare/blob/master/matlab/discretionary_policy..., upon exiting discretionary_policy(), `options_ `contains the original values it had upon entering the function with the field `discretionary_policy` set equal to 1.

However, the call to stoch_simul potentially changes options_, as you can see on line 99 of https://git.dynare.org/Dynare/dynare/blob/master/matlab/stoch_simul.m. Here, the field `logged_steady_state` is set to 1 with the idea that this value should be preserved upon exit of the function. However, given the way discretionary_policy has been written above, it is not preserved.

So the question is: is this a bug? If not, that would imply that discretionary_policy is incompatible with the loglinear option of stoch_simul: should I then end processing if the two hold simultaneously? If not, then we should only overwrite the `order` option in discretionary_policy with its original value.

Best, Houtan _______________________________________________ Dev mailing list Dev@dynare.org https://www.dynare.org/cgi-bin/mailman/listinfo/dev

Hi Johannes

the problem with linearizing a model, is that contrarily to Ramsey, the discretionary solution for a linearized model isn't an approximation of the discretionary solution in the original non-linear model (in part because we don't know how to compute the steady state of a non-linear model under discretionary policy).

Is the manual clear enough on that point?

Best

Michel

Johannes Pfeifer writes:

Hi, I would not go as far as Michel. But we explicitly state in the manual: "You should ensure that your model is linear and your objective is quadratic" I don't mind Dynare linearizing the model for me. I consider this a big advantage. But I would provide an error if `loglinear` is used. There are easy ways around this for users, e.g. using an exp()-substitution or appending logged variables to the model. Best, Johannes

-----Ursprüngliche Nachricht----- Von: Dev dev-bounces@dynare.org Im Auftrag von Michel Juillard Gesendet: Mittwoch, 11. September 2019 17:32 An: List for Dynare developers dev@dynare.org Betreff: Re: [DynareDev] Issue #1197

Hi Houtan,

discretionary_policy should only be used with a linear model and not a linearIZED model (IMHO we could consider imposing in the parser this restriction). Therefor there is no reason to accept option loglinear with discretionary_policy

Best

Michel

Houtan Bastani writes:

...

Hello,

In working on https://git.dynare.org/Dynare/dynare/issues/1197 I came across a potential bug.

As you can see from lines 22-24 & 39 of https://git.dynare.org/Dynare/dynare/blob/master/matlab/discretionary_ policy.m, upon exiting discretionary_policy(), `options_ `contains the original values it had upon entering the function with the field `discretionary_policy` set equal to 1.

However, the call to stoch_simul potentially changes options_, as you can see on line 99 of https://git.dynare.org/Dynare/dynare/blob/master/matlab/stoch_simul.m. Here, the field `logged_steady_state` is set to 1 with the idea that this value should be preserved upon exit of the function. However, given the way discretionary_policy has been written above, it is not

preserved.

...

So the question is: is this a bug? If not, that would imply that discretionary_policy is incompatible with the loglinear option of stoch_simul: should I then end processing if the two hold simultaneously? If not, then we should only overwrite the `order` option in discretionary_policy with its original value.

Best, Houtan _______________________________________________ Dev mailing list Dev@dynare.org https://www.dynare.org/cgi-bin/mailman/listinfo/dev

Thanks Johannes. This is fine with me.

Best

Michel

Johannes Pfeifer writes:

Hi Michel, that is a subtle point that is typically neglected in the literature. For that reason, we should elaborate in the manual, but not enforce additional restrictions apart from not allowing `loglinear`. Best, Johannes

-----Ursprüngliche Nachricht----- Von: Dev [mailto:dev-bounces@dynare.org] Im Auftrag von Michel Juillard Gesendet: Mittwoch, 11. September 2019 18:50 An: List for Dynare developers dev@dynare.org Betreff: Re: [DynareDev] Issue #1197

Hi Johannes

the problem with linearizing a model, is that contrarily to Ramsey, the discretionary solution for a linearized model isn't an approximation of the discretionary solution in the original non-linear model (in part because we don't know how to compute the steady state of a non-linear model under discretionary policy).

Is the manual clear enough on that point?

Best

Michel

Johannes Pfeifer writes:

...

Hi, I would not go as far as Michel. But we explicitly state in the manual: "You should ensure that your model is linear and your objective is quadratic" I don't mind Dynare linearizing the model for me. I consider this a big advantage. But I would provide an error if `loglinear` is used. There are easy ways around this for users, e.g. using an exp()-substitution or appending logged variables to the model. Best, Johannes

-----Ursprüngliche Nachricht----- Von: Dev dev-bounces@dynare.org Im Auftrag von Michel Juillard Gesendet: Mittwoch, 11. September 2019 17:32 An: List for Dynare developers dev@dynare.org Betreff: Re: [DynareDev] Issue #1197

Hi Houtan,

discretionary_policy should only be used with a linear model and not a linearIZED model (IMHO we could consider imposing in the parser this restriction). Therefor there is no reason to accept option loglinear with discretionary_policy

Best

Michel

Houtan Bastani writes:

...

Hello,

In working on https://git.dynare.org/Dynare/dynare/issues/1197 I came across a potential bug.

As you can see from lines 22-24 & 39 of https://git.dynare.org/Dynare/dynare/blob/master/matlab/discretionary _ policy.m, upon exiting discretionary_policy(), `options_ `contains the original values it had upon entering the function with the field `discretionary_policy` set equal to 1.

However, the call to stoch_simul potentially changes options_, as you can see on line 99 of https://git.dynare.org/Dynare/dynare/blob/master/matlab/stoch_simul.m. Here, the field `logged_steady_state` is set to 1 with the idea that this value should be preserved upon exit of the function. However, given the way discretionary_policy has been written above, it is not

preserved.

...

So the question is: is this a bug? If not, that would imply that discretionary_policy is incompatible with the loglinear option of stoch_simul: should I then end processing if the two hold simultaneously? If not, then we should only overwrite the `order` option in discretionary_policy with its original value.

Best, Houtan _______________________________________________ Dev mailing list Dev@dynare.org https://www.dynare.org/cgi-bin/mailman/listinfo/dev