rollcast: Probabilistic Forecasting with Adaptive Mixtures of Rolling
Statistics
Implements a probabilistic time-series forecasting framework
based on adaptive mixtures of rolling statistical anchors. Rolling means,
medians, minimum and maximum values, regression endpoints, and
user-specified quantiles define candidate forecast locations. A
proper-score gating model assigns state-dependent mixture weights,
optional state-conditional residual
sampling adds local dispersion, and recursive simulation produces
marginal and joint predictive distributions. Numeric hyperparameters can
be supplied as scalars or candidate vectors for causal validation-based
selection.
Documentation:
Downloads:
Linking:
Please use the canonical form
https://CRAN.R-project.org/package=rollcast
to link to this page.