Vines are nested tree structures. The nodes are univariate marginal distributions (starvine.uvar.uvd.Uvd) and the edges are bivariate copula (starvine.bvcopula.pc_base.PairCopula).

To construct a vine sequentially starting from the top tree:

For each tree level in the vine:
1. Determine the tree structure. In the case of a regular vine, this is the maximum spanning tree with edge weights equal to kendall's tau.
2. Fit copula (maximum likelyhood) at all edges in the current tree
3. Compute the conditional distribution at each edge in the current tree.
4. Use conditional distribution as nodes in the next tree.
5. Next

C-Vine

In each tree level in a C-Vine (Canonical Vine), a central node is selected to which all other nodes are attached via copula. (starvine.vine.C_vine.Cvine)

This dependence structure is favorable if a single primary variable drives changes in all other variables in the system to some degree. This primary variable could then be selected as the "anchor" node.

D-Vine

In a D-Vine, every node is connected with exactly two other nodes via copula.

This dependence structure is well suited when no single primary variable of interest can be identified. It is also suitable for temporally ordered data.

Regular Vines

Regular vines (R-vine) are the superset of all possible vine-copula structures. C- and D-vines are examples of valid regular vines.