jmobrien/macrofinancer

Time-Varying Structural Vector Auto-Regression for Macrofinancial Analysis

macrofinancer's Issues

documentation

Still need to document in fit_tvv():

• tvA (seems like a boolean for time-varying A matrix)
• nlt
• noLmd (also rename)
• seed params
• hyper params
• some options (steps, etc)

Document other functions (rfvar3, bvwrap, etc.) with appropriate exporting/not

major task - consolidate to model object

General placeholder issue for consolidating behavior across the many nested functions.

Idea: should we change this to an environmental structure (vs. list?)? Worth time to compare to how done in other modeling paradigms (BHSBVAR, glmmTMB?)? (probably not under contract, but writing here for future consideration)

model object passing

Trying to make the structure follow with a unified model object as much as possible

• bvarwrap (lhfcn) / csminwelNew functions need to be updated to accept the object
  • deeper functions also
• rfvar3 also gets called independently for use in seed setup, review this one.

Consolidate bvarwrap functions

bvarwrap5() and bvarwrap_tvA() are highly similar; their separation requires an aliased function call that gets nested deeply and makes things harder. Can we unify these and simplify everything?

Checks needed

(extend as we go):

• Correct mode/formatting for arguments, general
• Variables in data as expeted
• overlapping arguments (e.g., cos_prior & mn_prior)
• nlt specified with hyperparams
• match priors, etc to model variable size (e.g., v_prior has right length

prior parameters

a_diag, ur_lambda, & ur_mu were amendable by the previous pparams() function (now embedded, may be separated again later), but were not addressable by the previous wrapper function TvvDir(). Still fixed in fit_tvv(), should this be changed?