Bayesian DSGE model

[phuong.macro]

Hallo

I am working on an open-economy DSGE model by using Bayesian method. I have some questions that i can not answer myself
1. how to choose an initial value?
2. How we can make sure that the posterior values are acceptable?

I means that the coefficient of inflation in the Taylor's rule, for example, should be greater than 1, then

Firstly, I assign the initial value as 1.2, the result is 1.4
Secondly, I assign the initial value as 1.4, the result is 1.5

Then how can I choose them?

[jpfeifer]

The goal is to find the global posterior mode. What you describe suggests that different starting values led to different (local) modes. Please try different mode-finders and starting values and select the mode with the highest posterior density.

[phuong.macro]

Dear Prof. Pfeifer

Thank you so much for your answer

This means that I have to try the different initial values as many as possible to choose the best posterior. Is this true?

[jpfeifer]

In principle, yes. Ideally, you use a global mode-finder like

Code: Select all

mode_compute=9

[phuong.macro]

Dear Pfeifer

Yes sir, In this case, when I need a stationary data?

and another question is that

I have read your paper, titled: ''a guide to sepcifying observation equations for the estimation of DSGE model''

I know that I have to match the observable variables with the theoretical model variables before estimation

However, there are some complicated models with many variables. Then do we have a general solution to match the observable variables with the theoretical model variables .

In particular, I am working on the paper, titled: ''Optimal monetary policy in an operational medium-sized DSGE model'' (http://onlinelibrary.wiley.com/doi/10.1 ... x/abstract)

Then I read their Technical appendix as the attached file. I have difficulty to derive the measurement equation systems (Section 3.1. Difference Specification, page. 28)

I mean How to derive this measurement equation system?

would you kindly give me some advices about this

Thank you so much indeed

[jpfeifer]

The general rules are shown in the "Guide to specifying observation equation". You have to sit down and think about the mapping between the data and your model objects. Unfortunately, that can be quite cumbersome. with the paper you are citing being the best example. What makes their paper complicated is the fact that there are various relative prices that need to be considered when mapping actual data to the model.

[phuong.macro]

Dear Prof. Pfeifer

I got it

Thank you so much indeed for that