Optimal simple rule - approximation order

[lja]

Hi,

I am wondering about the Taylor approximation order used in the osr command.

In the manual of osr, it is said that "This command accepts the same options as stoch_simul". I take this so that one can specify the Taylor approximation order in the call, say "osr(order=2,irf=40)", which is then used *both* in osr calculation (optimizing the Taylor rule parameters vis-à-vis a second-order model) and subsequently in model simulation using its second-order approximation and optimized simple rules parameters.

However, it seems that on line 22 of osr.m the order is explicitly set to 1 (options_.order = 1). Is this deliberate? What if I comment that line out, hence allowing the user to call osr with approximations of order 2 (or even 3)?

I am using Dynare 4.4.1.

Best,
Lari

[lja]

Hi,

Could you please comment on this? In brief, I am just wondering if it is possible to provide the option "order" to the osr command. I have tried this, but see no difference, neither in the optimized Taylor rule coefficients nor in simulated IRFs.
While investigating Dynare code, the underlying reason for this seems to be that the order is fixed to equal 1, regardless of the order given in the call.

[MichelJuillard]

You are right: the manual is misleading. OSR works only with first order approximation. Thanks for pointing the issue.

Best

[jojokre]

I have a follow-up question on Michels answer: "OSR works only with first order approximation" does not mean that Dynare conducts only a first-order approximation? In the Dynare manual, it is stated that "This command computes optimal simple policy rules for linear-quadratic problems". Therefore, the function I state as objective function should be automatically approximated up to second order. Is my understanding of how OSR works correct?

[jpfeifer]

With osr, you provide weights for the unconditional (co-)variances of your target variables via the

Code: Select all

optim_weights ;

block. Dynare does not take an objective function it would approximate.