This is a rather basic question, but I would like to save the regression coefficients from nilearn.glm.ARModel. I am unfamiliar with the notation that is being used in the nilearn glm functions. I am used to the regression coefficient values being referred to as the beta-values in neuroimaging or coefficients in sklearn. The theta parameter in nilearn.glm.RegressionResults has the size that I am expecting, but am I correct in assuming that this is equivalent to a beta-value for other fMRI packages?

Having a look at the inline documentation of the code, in likelihood models class, theta values are the coefficients / parameter estimates from the estimated model. I would say you are right.