# What functional forms can I use to model a time series that is not stationary but has a unit root process?

I'm trying to choose the best forecast for this blue line (time series with 550 observations). Its not stationary but it does have a unit root. AR processes dont seem like the best choice. Could you please help me brainstorm some functional forms that incorporate the unit root idea? Thanks!

• Have you tried ARIMA or exponential smoothing? E.g. the automated functions ets and auto.arima from the "forecast" package in R? These can give you more or less solid benchmarks you could try to beat with other models of your choice. Also, visually it does not necessarily look as a unit-root process, so you can still consider other possibilities. – Richard Hardy Oct 14 '17 at 19:07
• Thanks a lot Richard! This is very helpful! I will try those out. – JorgeT Oct 16 '17 at 17:55