Time series analysis in R

Question

I carried out a log transformation on some data about patient admission in the hospital, aiming to generate an additive model. I went ahead to forecast the same data using:

MYFORECAST = forecast.Arima(auto.arima(admission, d=3, D=NA, stationary=FALSE, 
                            seasonal=FALSE, ic="aic", trace=TRUE, allowdrift=FALSE, 
                            allowmean=TRUE)

in R. I got the forecasts but I was left wondering: Should report the forecasts or take exponent of the values?

Answer 1

Yes, do apply exp(*).
Some food for thought: the change of scale is not linear. Prediction errors that seem reasonable in the log-scale often became large in the natural scale (after you apply logarithms), and the predictions tend to be biased. Check the plots of time & (original TS), time & exp( fitted values in the log scale) if possible.