Getting started

Installation

using Pkg
Pkg.add("VineCopulas")

Then load the package:

using VineCopulas
using Distributions

VineCopulas.jl reexports Copulas.jl, so the pair-copula families used on vine edges are available from the same session.

Construct an explicit vine

A three-dimensional D-vine has two pair copulas in tree 1 and one conditional pair copula in tree 2:

C12 = GaussianCopula(2, 0.5)
C23 = ClaytonCopula(2, 1.5)
C13_2 = FrankCopula(2, 2.5)

vine = DVineCopula([1, 2, 3], [[C12, C23], [C13_2]])

Evaluate or simulate it with the standard distribution interface:

u = [0.2, 0.5, 0.7]
logpdf(vine, u)
pdf(vine, u)
U = rand(vine, 1_000)

Fit a vine

If the structure or ordering is not supplied, the fitting layer can select it from the data:

fitted = fit(
    RVineCopula,
    U;
    family_set=:default,
    selection_criterion=:bic,
    tree_criterion=:tau,
    allow_rotations=true,
)

For a fixed structure, pass the desired order or RVineStructure explicitly. See Fitting & Selection for the complete controls.

Data layout

All fitting and multivariate evaluation routines use p × n matrices:

rows    -> variables
columns -> observations

If your data are stored as n × p, transpose them before fitting.