Question about estimation initial values of SW_2007 paper

Estimation
1

Dear Dr.,
May I ask where those initial values of eitimated_params are from in Smets_Wouters_2007 paper? They are very different from prior means.
Thanks!

Best,

2

By default Dynare uses the prior mean as an initial condition for the optimization routine used for maximizing the posterior kernel. The user can change this default behavior by providing initial conditions. The closer to the (unknown) posterior mode are the initial conditions, the easier will be the optimization. Often the optimization fails with the default initial conditions, it is then to the user to find better initial conditions (closer to the posterior mode).

best,
Stéphane.

1 Like
3

Highly appreciate your help!
You said The user can change this default behavior by providing initial conditions. So may I ask what methods users generally apply to find initial values? In SW2007 paper, I think their initial values are computed based on some method. I’m interested in how they did that?

Best,

4

I am not aware of general methods for choosing initial conditions. I don’t know which paper you are exactly referring to, but I suppose that the authors do not give details about this… Generally it’s experience plus try and error. A more systematic approach is to evaluate the likelihood (posterior kernel) on a multivariate grid to identify the regions where the objective function have the highest values.

Best,
Stéphane.

2 Likes
5

Thanks so much, Dr.!

6

I agree with @stepan-a. Mode-finding is hard. Smets/Wouters 2007 in their replication files provide some initial values that are different from the prior mean. But they also provide mode-files. So my guess is that they tested a lot of values and just provided some of the values as their initial values. I don’t think there is anything systematic to them as they are not used most of the time (due to the mode-files)

2 Likes
7

Thank you so much. Now I understand!