R-vines

RVineCopula represents a general regular vine. It supports standard R-vine structures, matrix exchange, density evaluation, simulation, Rosenblatt transforms, fixed-structure fitting, and automatic Dissmann-style structure selection.

An R-vine is the most flexible public structure in VineCopulas.jl. It can represent C- and D-vine-like models, but its main purpose is data-driven branching structure.

Fixed structure

Construct an RVineStructure and fit its pair copulas:

st = RVineStructure(order, struct_array)
model = fit(RVineCopula, U; structure=st)

The structure can also be paired directly with an explicit edge array:

model = RVineCopula(st, edges)

Automatic structure selection

model = fit(
    RVineCopula,
    U;
    family_set=:default,
    tree_criterion=:tau,
    tree_algorithm=:kruskal,
    selection_criterion=:bic,
)

The first tree is a maximum spanning tree using absolute dependence scores. Higher trees are selected under the R-vine proximity condition and fitted sequentially. tree_criterion chooses the dependence statistic behind those scores, or takes a weight function, and groups restricts the first tree to the spanning trees in which every group of variables is connected; see Fitting and selection.

Warning

R-vine flexibility is useful, but it also makes introspection more important. Inspect order(model), struct_array(model), edges(model), and rvine_matrix(model) before treating a fitted structure as substantively meaningful.

Matrix exchange

M = rvine_matrix(model)

Matrix representations can differ across packages even when they encode the same vine. For interoperability, compare the represented edges and conditioning sets rather than raw matrix strings.