Recomendded prior for the correlation of two exogenous variables

Estimation ML/Bayesian estimation
1

Hi everyone,

I am estimating a model with correlated shocks using bayesian estimation with the command:
corr e_x e_y, somepdf, expected value, sigma;
Assuming I have no informative prior, what are the recommended distributions for the correlation parameter? I tried with a normal distribution and got the warning:
“your prior allows for correlations between structural shocks larger than ±1 and will not integrate to 1 due to truncation”.
What are the bounded distributions you will recommend?

Yaakov

2

Most often, you would use a generalized beta distribution or just live with a truncated normal distribution.

1 Like
3

Generalized beta distribution? Do I need to specify more than one prior parameter after “beta_pdf”? Is this also applicable for negative correlation?

4

Yes, the third and fourth optional parameter are for the lower and upper bound. When not specified, they are [0,1], but you could stretch the distribution to the interval [-1,1].

1 Like
5

Great. Thanks!

By the way, is there a documentation for the available pdfs and the parameters’ description? I could only find a list of them in the manual in section 4.15, but without the parameters’ description.

6

See

1 Like