nb_bvar: Notebook generator: BVAR NiW
In ooelrich/oscbvar: Stuff for Bayesian vector autoregressions

Generates a notebook for a BVAR model with a Minnesota-flavoured NiW prior..

nb_bvar(
  data = oscbvar::macrodata[, 1:7],
  agc = list(5, 60, TRUE, 1),
  lags = 1,
  overall_tightness = 0.2,
  lag_decay = 1,
  include_intercept = TRUE
)

`data`	Dataset from which to generate the notebook. Should include only the variables used by the model. Defaults to the macroeconomic dataset from this package using all varaibles.
`agc`	List of atomic prediction generation controllers. The first element of the list gives the starting time (ie what observation is considered as t = 1), the second element is the minimum window length used for estimation, and the third one is a boolean indicating if the estimation window is rolling or not, the forth indicates which variable is the response variable and it defaults to 1.
`lags`	The order of the VAR model. Defaults to 1.
`overall_tightness`	Overall tightness (pi_1 in Sunes notation). Defaults to 0.2
`lag_decay`	The lag decay rate. Defaults to 1.
`include_intercept`	Whether or not to include an intercept. Defaults to TRUE.

Generates a notebook of predictions for the decision maker to use. Uses a version of the Minnesota prior: the prior for the variance is data based with nu_0 set to the number of time series plus two, and S_0 obtained by running a simple AR(4) model and extracting the diagonal elements. The prior for the regression coefficients is tweaked using two hyperparameters: the overall tightness (defaults to 0.2) and the lag decay rate (defaults to 1). The cross-variable tightness is set to 1 to retain the Kronecker structure required for conjugacy.

ooelrich/oscbvar documentation built on Sept. 8, 2021, 3:31 p.m.