These are the replication files for Zito & Kowal (2025 arXiv). We propose a dynamic Gaussian factor copula (DGFC) model for jointly modeling a collection of time series that include mix of data types (discrete versus continuous) and/or non-Gaussian distributional features (skew, multi-modality, etc).
The two main commands in the source code are
DGFC.mcmc(Y, k.star, prior, init, ndraw, burn, thin)
DGFC.forecast(H, draws, use_spline)
DGFC.mcmc runs the Gibbs sampler (Algorithm 3) for approximating the pseudo-posterior
in the DGFC, and given the posterior draws, DGFC.forecast simulates the model forward
to forecast H steps into the future (Algorithm 2).