Consider an exponential smoothing model. Initially all parameters are unset, except for the errors, trend, and seasonal parameters that determine what the model can be specialized to, and the constraints parameter that determines whether traditional and/or admissibility constraints are enforced.
If we set the trend parameter to {"A", "M"} (signifying that the trend is additive or multiplicative, but not damped), we will get a warning that the current value is being overwritten. Also, the parameter phi (which determines damping) is set to its default value, .
Consider an exponential smoothing model with parameters where we fix to be .
Let us verify the values of and of the parameters errors, trend, and seasonal.
These are the unassigned parameters: