Thanks for the reply. The problem is that I need the funcionality of forecast menu of the program like UCL, LCL in a "good" plot.

In some packages, like SPSS the re-transformation is given from the software and is supposed that the log model iw evaluated along with the non-log models.

Hi Dimitris and other interested,

a little update on this: I've updated my contributed function package "uniFCextensions" to version 1.2 which now offers the possibility to produce (three different) consistent forecasts of the levels if the model was estimated with the log on the LHS.

However, this does not cover ARMA models, only dynamic linear OLS (including AR terms, obviously). Parameter or model uncertainty are not taken into account, either. (I mention this because of the previously cited remark by BÅRDSEN, G. & LÜTKEPOHL, H. (2011) in a slightly different context that the naive exp-ing of the log forecast might be OK then.)

Feel free to ask if anything remains unclear.

cheers

sven