Comparison with rvinecopulib / vinecopulib

rvinecopulib is the R interface to vinecopulib, a header-only C++ library for vine copula models. It is a mature statistical modeling package with high-performance algorithms, parameter estimation, model selection, simulation, visualization, and support for nonparametric and multi-parameter families.

VineCopulas.jl has a different first goal: provide a native Julia construction and evaluation core that composes pair-copulas from Copulas.jl.

Feature	VineCopulas.jl v0.1	rvinecopulib / vinecopulib
Language	Julia	C++ core, R interface, Python interface
Main scope	Explicit construction, evaluation, simulation, transforms	Full statistical modeling workflow
C-vines / D-vines	Yes	Yes
R-vines	Partial/experimental	Mature
Density and simulation	Yes	Yes
h-functions and inverse h-functions	Yes, via Julia API	Yes
Automatic pair-copula fitting	Not yet	Yes
Automatic family selection	Not yet	Yes
Automatic structure selection	Not yet	Yes
Automatic truncation selection	Not yet	Yes
Nonparametric pair-copulas	Not part of stable v0.1	Yes, including tll
Discrete variables	Not part of stable v0.1	Supported in rvinecopulib
Extensibility with Copulas.jl	Main design goal	Not the goal
Bayesian/probabilistic Julia workflows	Natural future direction	External to the package

The honest positioning is therefore:

• use rvinecopulib when you need a mature production-ready fitting and selection workflow today;

• use VineCopulas.jl when you want explicit native Julia vine objects, direct integration with Copulas.jl, transparent conditional primitives, and a research-friendly base for future Julia workflows.

No broad speed superiority is claimed for VineCopulas.jl v0.1. The package now includes local benchmarking scripts under benchmarks/ for comparable explicit C-vine/D-vine operations. These scripts intentionally avoid fitting, family selection and structure selection, because those workflows are not yet implemented in VineCopulas.jl.

The overlapping parametric families targeted by the local benchmark scripts are independence, Gaussian, Student-t, Clayton, Gumbel, Frank, Joe, BB1, BB6, BB7 and BB8. Nonparametric tll is outside the stable v0.1 scope.

Local benchmark results

A reproducible local benchmark summary is available in Benchmarks and numerical validation. The main results are:

• Gaussian D-vines are currently faster in VineCopulas.jl for vectorized log-density in the tested scenarios.

• Clayton and Gumbel D-vines are competitive for log-density, especially in larger truncated vines.

• Frank D-vines are numerically validated but still slightly slower in log-density.

• Student-t D-vines are numerically validated but substantially slower because the current implementation is dominated by scalar Student-t CDF/quantile evaluations.

The benchmark suite compares explicit operations only: density/log-density, Rosenblatt transforms, inverse Rosenblatt transforms, simulation and numerical CDF evaluation. It does not compare fitting, selection, model search or nonparametric estimation, because those workflows are outside the stable v0.1 scope of VineCopulas.jl.