Analysis options for logging, differencing, and seasonal adjustment in Statgraphics:

When you are using the Time Series procedures in Statgraphics, transformations such as logging, seasonal adjustment, and nonseasonal and seasonal differencing can (and usually should be) performed INSIDE the procedures as right-mouse-button "analysis options." Here is the analysis options panel for the Descriptive Methods procedure. Setting the "Nonseasonal order" of differencing to 1 causes a first-difference transformation to be applied to the data. Similarly, setting the "Seasonal order" of differencing to 1 causes a seasonal difference transformation to be applied. (The number of periods in a season that was specified in the "Seasonality" field on the Data Input panel comes into play here.) Also, note that a log transformation, if any is needed, can be specified here as well.

Here is the analysis options panel for the Forecasting procedure. Notice that all the transformation that were available as options in the Descriptive Methods procedure--logging, deflation at a fixed rate, seasonal adjustment, differencing, etc.--are also available here. The one exception is that the fields for specifying the order of differencing are not active unless the model type is set to ARIMA. If your Descriptive Methods analysis suggested that an order of nonseasonal differencing and/or an order of seasonal differencing is needed to stationarize the series and/or remove the gross features of the seasonal pattern, then the same orders of differencing probably would be appropriate IF you decide to fit an ARIMA model to the data. The assumption is that if you don't fit an ARIMA model, then you will not want to apply an explicit differencing transformation. Other model types such as "random walk" or "exponential smoothing" use differencing tranformations implicitly, so you would not want to difference the data before applying one of these models. Conversely, if the model type is set to ARIMA, the "seasonal adjustment" options are inactive, the assumption being that if you are fitting an ARIMA model, you will deal with seasonality by using seasonal differences and seasonal coefficients (SAR or SMA terms) rather than by explicit seasonal adjustment.