BVARResult._summary_table

BVARResult._summary_table()

Build the structured summary every renderer draws from.

The single description of this result that summary(), print(res), a bare res in a notebook, and res.summary().to_pandas() all render, per the _SummaryMixin contract. The metadata block carries the specification and the posterior’s accounting – model, prior label, draws, variables, observations, dummy rows, trend, and the log marginal likelihood to three decimals. The table has one row per equation with that equation’s own first-lag coefficient: its posterior mean and 68% interval from credible_interval() at label "{name}.L1", four decimals, or dashes when order is zero. Own first lags are the one coefficient every equation has and the one the Minnesota prior shrinks toward \(\delta\), so the row reads as how far the data moved each equation off the prior mean; every other coefficient is available through credible_interval().

The notes state what a reader coming from a likelihood-based result would otherwise look for and misread: that the draws are independent so no convergence diagnostics apply, what the log marginal likelihood is and is not, the posterior probability of stability from stable_share, the absence of llf and information criteria, and that forecast() is the full predictive. Building the table evaluates stable_share, so it costs one eigendecomposition per draw.

Return type:

SummaryTable

Returns:

The SummaryTable with title "BVAR({order}) Results", columns ("equation", "own L1 mean", "68% interval"), one row per name, and the notes above.