The twinkle package provides an S4 framework for estimating and working with logistic smooth transition ARMAX models with a maximum of 4 states. Models can be specified via the starspec method which returns a STARspec object which in turn can be passed to the estimation method starfit, filter method starfilter or path simulation method starpath. The estimated object of class STARfit can be passed to the forecast method starforecast or simulation method starsim. A number of extraction methods are available, uniform across most returned classes, following the same convention as in related packages such as rugarch.

While the conditional mean equation is defined by the STARMAX dynamics, the conditional variance may follow either a static model, a mixture (upto 4 states) or one of 3 GARCH models (vanilla, exponential or GJR).

The same number of conditional distributions are implemented as in the rugarch package.

The package vignette provides for a detailed and formal treatment of the STARMAX model as well as the finer details of the estimation and forecast methodology implemented.