Conflict with model spans and log-transformation in RegArima models

[palatej]

Source: Alina Sadovska

When the model span is used, a log-transformation can be chosen, even if some figures outside the model span are negative or 0. In that case, incoherent results will appear (the pre-adjustment factors correspond to the log-transformed series, but the linearized series is not log-transformed, due to the negative values).

Possible solutions:

• put missing values at the positions corresponding to negative values (but quid if seasonal adjustment)
• consider the log-transformation test on the complete series (not on the model span), at least for the presence of unacceptable figures.
• ...

[palatej]

The second solution is probably the best one.