|
If you want to build time series model under ARIMA framework, you have to have a stationary series before doing next steps, because all assumptions about ARIMA model are based upon stationarity. To get a stationary series, you can take difference of the original series. If ADF tests still show non-stationary, you may think of where this non-stationarity comes from. Is it from seasonality, or some outliers? If there is seasonality, you need make necessary seasonal adjustments. If it is because of outliers, you can replace those outliers with exponential smoothing estimates, which can be done by SAS JMP. After getting stationary series, then you can play from there.
|