Unobserved components model using the linear innovations state space representation (single source of error).
Key features:
- Estimation using autodiff (TMB)
- Model selection and ensembling
- Regressors in the observation equation (non-time varying)
- Choice of distributions (Gaussian, Student and Johnson’s SU)
- Choice of dynamics in variance (constant or GARCH)
Methods for specification (modelspec), estimation (estimate),
summary with choice of vcov (summary and vcov), diagnostics
(tsdiagnose), online filtering (tsfilter), prediction (predict),
simulation (simulate), backtesting (tsbacktest) and profiling
(tsprofile). Additionally, model selection (auto_select) and
ensembling (tsensemble) is also included.