Initial values in osr

Optimal Policy
1

Dear all,

I am wondering what a role do the initial values play when using the osr command? It seems that different starting values in osr could result in quite different optimal simple rule parameters. Then how to set correct initial values?

I also have the following two questions about using “osr” in a simulation (note that the parameters used in the simulation are posterior means of an estimation):

(1)should I have to set the initial values of osr to be the estimated means of the empirical monetary policy rule?

(2)In the estimation, I get an estimate of the monetary policy shock (estimates of the standard deviations). But should I still keep that monetary policy shock in the “stocks” block before using osr? I ask this question because I feel this is more like a contradiction. Can a monetary policy rule be optimized to fight against its own shock?

(3) This question is related to question (2). If I change the specifications of the optimal simple rules, do I have to add a monetary policy shock to it? If yes, how can a policy shock estimated from the empirical rule be directly applied to another policy rule?

I wish I have made my questions clear. …May this question is naive, but sometimes I really feel a little bit confused…

Please give me some hints, thanks very much!

Thanks for helping!

1 Like
2

Hi,

The optimization routine to compute OSR requires to choose starting values in order to make it converge more or less quickly (depending on their accuracy).
Overall, if different guess values yield different optimal weights, it certainly means you have multiplicity of solutions. Usually, they should lead to the same result.

Besides, I don’t see any contradiction with calibrating your parameters to their estimated value (as the estimated value of the MP shock) and
optimizing the parameters of your policy rule.

Hope this helps

3

In reality, I encounter this problem these days. The results are very close to the initial values. I tried to estimate the reaction coefficients but the results were very bad.
Finally, I choose the values in literature to be the initial values. Do you think it is reasonable?

4

No, that is not reasonable. You need to find a global optimum. Being constrained by starting values means you are only searching locally.