Impulse responses, confidence bands and over-identification

[TomDoan]

how i can to solve this problem of graphs and how i can to get good graph according to the economic theory.please reply my questions thanks so much.

Making the the model consistent with economic theory isn't the job of the software.

[shimarats]

yes ,i am trying to get montesvar .but i cannot understand your reply about convergence. this is my question:1) what is the role of convergence in the estimates? 2)if there are any convergence in the estimates are good or bad?3) you told me the sample size is 8.23 are very good then how i can to make this sample size very good? 4) what is difference between bayesian svar techniqe and montesvar?5 )is there special code for bayesian svar?
thanks so much for help me dear tom.

[shimarats]

if my model is over identified in svar method ,is it possible monte svar give me chi-square and significant level?because when i estimate model , give me results of there is no convergence. now i am confuse about the convergence i donot know should there is convergence or no convergence ?is there another results except the exist or lack convergence ? thanks so much

[TomDoan]

yes ,i am trying to get montesvar .but i cannot understand your reply about convergence. this is my question:1) what is the role of convergence in the estimates? 2)if there are any convergence in the estimates are good or bad?3) you told me the sample size is 8.23 are very good then how i can to make this sample size very good? 4) what is difference between bayesian svar techniqe and montesvar?5 )is there special code for bayesian svar?
thanks so much for help me dear tom.

You control the number of iterations. The default is 100. If the model hasn't converged in 100 iterations but appears to be making progress, just increase the ITERS option. You shouldn't report results that are based upon non-converged estimates.

To increase the effective sample size, you can try increasing the scale up on the FAXX matrix or decreasing the degrees of freedom (NU) or both. However, it might be difficult to get good results which such a large parameter space. However, it may also be affected by running without a converged model. Of course, you really shouldn't be trying to do the MC integration until you get the base model working. I have no idea what someone might mean by "Bayesian SVAR"---that could mean either Bayesian analysis of an SVAR, or structural analysis of a Bayesian VAR.