I have been recently learning about the Box-Jenkins process for ARMA modeling, and I ran into a bit of a wall when it comes to error analysis. In a lot of my data sets, I have to apply a log transformation to my original data set, followed by a lag difference in order to reach a stationary time series. Following my choice of ARMA model, I am given a time series of standard errors to pair with my forecast. (I am currently using python's statsmodels library for my data modeling.)
My question is this: can I propagate my error using standard techniques such as those listed here in order to reach the standard error of my original time series? I have found almost no resources that answer this question (and those few have so far been very unhelpful).