TracePlot vs Geweke (1992)

Hi everyone

I have a question related to test for convergence by using Traceplot and Geweke (1992)

  1. Traceplot: If the MH converges to a stable distribution, the trace plot has to be stable (no trend in the sampled value). Accordingly, I can conclue the following figure rhog.pdf (93.3 KB)
    alpha.pdf (91.4 KB)
    gamma.pdf (92.5 KB)
    rhoe.pdf (92.4 KB)
    rhopii.pdf (93.3 KB)
    rhoa.pdf (195.9 KB)
    They converge. Is that correct?

  2. the Geweke (1992) use the two sample t test to evaluate the MH convergence. The null hypothesis is that the two mean of two sample equal. In Dynare, it reports as the file Geweke 1992.pdf (93.5 KB)
    Accordingly, I would say I reject the null hypothesis of equality of means.
    If such, there is confiction with the Trace plot. So that my conclusion is correct?
    And could you explain the meaning of p-val 4% Target in Dynare report about Geweke(1992) convergence test
    Thank you so much

The trace plots are an easy visual check while the Geweke statistics is a formal statistical test. Usually, one uses the trace plots to see why a particular parameter did not converge. In your case, the Geweke test rejects you cannot reject the null of equal means except for SE_epsr. But that is a parameter for which you did not provide the trace_plot. For all other parameters, both the trace plots and the Geweke statistic suggest convergence.
The 4% refers to the taper window used for computing the standard error in the presence of serial correlation (Newey-West standard errors). One most often looks at the 15% version.

Dear Prof @jpfeifer
Thank you so much for your answer
I would ask more about the meaning of p-val 4% Target in Dynare.

In OLS, for example, we have to recompute the Newey West Standard Error (NWSE) in the presence of serrial correlation in error term. In particular, in the following link, it shows how to compute NWSE when there is serrial correlation in error term

On the other hand, your explation

So I still confuse

would you give me a document or link or formula to better understanding how to compute NWSE?

Thank you so much for that

It’s all documented in Geweke’s (1999) “Using Simulation Methods for Bayesian Econometric Models: Inference, Development, and Communication”. See formula 3.8.1

Dear Prof @jpfeifer

Thank you so much for your suggestion

With best regards