192-30: Stationarity Issues in Time Series Models

The decision on whether analyze a time series in levels or differences is an important aspect of forecasting. Visual methods have been around for a long time. Relatively recently, statistical tests for the null hypothesis that the series is nonstationary, meaning that differencing is required, have been developed. This paper reviews the development of these tests, give motivating examples of why they are needed, demonstrates their use, and shows some other related procedures. The paper will be accessible to beginning level SAS programmers, but the reader should have a reasonably strong grounding in statistics.