Prior distribution with lower bound of 0.01 but posterior distribution in negative range

Estimation ML/Bayesian estimation
1

Hello everyone,

I have the following question: Assume I have a prior distribution with positive support. In my case, these are the relevant lines from the estimated_params block:

@{co}_gammaimc1,(p_@{co}_gammaimc1),0.01,15,GAMMA_PDF,4,1;
@{co}_gammaimi1,(p_@{co}_gammaimi1),0.01,15,GAMMA_PDF,4,1;

I thought that this means that the posterior includes only positive values as well, but this is not the case, as you can see from the attached graph. The posterior covers a negative segment. I am using the slice sampler, I don’t know if that matters.

Many thanks in advance

Estimation_slice_LCPsep_PriorsAndPosteriors5.fig (151.0 KB)

for your help!

Best,

Ansgar

2

That is not supposed to happen. Can you provide me with the codes to replicate the issue? It looks like a bug in Slice.

3

Hi Johannes,

many thanks! The strange thing is that the traceplots for that parameter actually never become negative. Here are the files.

data.xlsx (1.5 MB)

Estimation_slice_LCPsep_mean.mat (86.5 KB)

readestimates_20211007.mod (695 Bytes)

POW_symdecls.mod (7.7 KB)

init.mod (400 Bytes)

modeqs_CES_eopb_firmsharestraded_d.mod (53.7 KB)

Estimation_slice_LCPsep.mod (5.7 KB)

4

Thanks. The eagle_steady.txt is missing.

5

Oh sorry, here it is.

eagle_steady.txt (27.1 KB)

6

From what I can see, estimation works fine but the Kernel density smoother with a Gaussian kernel has a hard time with your data as there is some data close to the bound.

7

Thanks a lot. Just to clarify, you write

I assume you meant write parameters, not data?

8

I meant the data for which we compute the kernel density. In this case, it refers to the parameters.