Reparameterization of student_t(3, 0, 1)


According to Chapter 28.6 the Reparameterization of truncated [0, ] student_t(3, 0, 1) is:

real<lower=0, upper=1> sigma_unif;
real<lower=0> sigma = sqrt(-3+3/sqrt(x));

Is there any advantage in Stan to do so? I got this funnel’s of a multivariate normal in Stan
and I thought the tan transformation of the cauchy results in too long tails.



You should try it for your data. These reparameterizations can vary in effectiveness based on how much the data plus prior constrains the parameter values.