Bayesian Estimation : Posteriors

I did my first Bayesian estimation of my model. Do not have any experience in interpreting the posteriors though.
my posteriors on the shocks look unusually tight around the mode.
I have attached the posteriors (graphs) in a zip file.

My questions is:
Are the posteriors for the shocks

attached to this post normal ?
What is the interpretation for such graphs?

thnx (75.6 KB)

Except for the rhoIS with its two peaks they look pretty ok. You might want to look at the trace_plot for these parameters. Is there anything that makes you worry? Are the mode_check plots ok?

Thanks Prof. Pfeifer,

I had assumed they would look not much different from the prior. They looked very tight so I thought sth could be wrong with them.

The most worrying thing was that the posterior for persistence parameter for one of the shocks

looked close to zero. Does it mean the shock dies away the very next period ?? !!!
Would that sound ok ?

thanks again for your taking your time to comment!

ps. I could fix the problem regarding rhoIS in a different version though.

If the posterior looks very different from the prior, the data is very informative. This is perfectly fine unless you have well-founded suspicions that this cannot be correct.
It’s hard to have a good intuition on what a realistic shock autocorrelation is. Is there a reason to think that shock must be really persistent? Also note that ideally the model should provide propagation without relying on persistent shocks because we want to explain data with the model, not with exogenous stuff outside of the model.

Following your comment

the shock

refers to a shock on banking incentive parameter of Gertler Karadi (2011). and probably fits with the idea that it is a one time shock that is propagated through the model, hence does not need a high persistence.

May I please ask your opinion on a different type of posterior and diagnostics ?
Attachéd to this comment there is a ZIP folder with posteriors and diagnostics .

  1. In graph

. I find it difficult to interpret

2. How close should the blue and red line be in diagnostics of Brooks&Gelman(1998) to be considered ok.
are the ones I have attached close enough ??

thanks again!
very much appreciated! (365 KB)

  1. Maybe you still have a scaling issue in the model. Having a standard deviation differ by two orders of magnitude is indeed unusual.
  2. There is not rule here, but I would say they are reasonably close.