I need help to predict future values for a time series (26 observations of consumer prices). This time series is not stationary (i.e I have increasing values of price). What is the best method knowing that I want to take the time series to it original scale at the end?

For now, I run this code on stata. But, all my future values are decreasing which is a non sense. So, i'm probably

tsappend, add(20)
twoway (tsline consumerpriceavocadogermany)

gen log_cp_avocado_germany = -log( consumerpriceavocadogermany )
ac consumerpriceavocadogermany
pac consumerpriceavocadogermany
arima consumerpriceavocadogermany, arima(1,0,1)
predict cp
predict cp_dynhat, dyn(2018)
list if inrange(time,2015,2025), clean
  • $\begingroup$ If you think the process is non-stationary, change to arima(1,1,1). But in that case you should also choose the orders p and q of the AR and MA-parts by plotting the ac and pac of the differenced series. $\endgroup$ Jun 1 '17 at 8:49
  • $\begingroup$ hello Jarle Tufto, thank you for your response. I did this but the problem is that it changes the way my time serie looks like at first. I want to keep my original time serie with predicted value ( like the world bank does in its researches). $\endgroup$ Jun 1 '17 at 8:49
  • $\begingroup$ @poofidoudou Then you just need to "integrate" it back to its level by writing a piece of code that sums the first observation to the first fitted difference, iteratively, up to the last forecast. $\endgroup$ Jun 1 '17 at 8:57
  • $\begingroup$ @lucasfariaslf this is my first attempt to estime a time serie. I am not familiar with this kind of code. Can you help me ? $\endgroup$ Jun 1 '17 at 9:01
  • $\begingroup$ But this is not an estimation problem. Once you have the estimates, you just need to figure out how to do what I said in Stata. This might help you: stats.stackexchange.com/questions/126525/… $\endgroup$ Jun 1 '17 at 9:04