ST15 Basic Usage of SAS/ETS(R) Software to Forecast a Time Series     Invited

LTC Douglas McAllaster
US Army Logistics Management College
Abstract: This paper is a tutorial on usage of SAS/ETS and SAS/STAT software to forecast a times series. It explains forecasting with PROC FORECAST and PROC REG in detail. It provides a very brief overview of the capabilities and complexities of using PROC ARIMA. PR OC FORECAST is the SAS/ETS procedure which easily provides robust forecasts. It can forecast times series using three exponential smoothing methods: simple, trend, or seasonal. We examine all three in detail. PROC FORECAST also can perform step wise auto r egressive methods, which we examine briefly. PROC REG is SAS/STAT procedure can forecast a time series using linear regression. Although times series often suffer from auto correlation, its deleterious effect is minor. Thus, for quick and dirty forecasts, we use PROC REG on a seasonal times series using indicator variables. Finally, we use a logarithmic transformation to model exponential growth with PROC REG. PROC ARIMA is the SAS/ETS procedure with can identify underlying processes (auto regressive or mov ing average), estimate parameters, and forecast a time series. This complex procedure requires considerable statistical judgment for proper usage.

Biography:
LTC Doug McAllaster teaches management science at the US Army Logistics Management College at Fort Lee (near Richmond), Virginia. He teaches fundamental data analysis techniques (statistics and optimization) to new Army analysts. Previously, Doug was an an alyst at the Pentagon for ten years. There he used SAS to perform statistical analyses of personnel retention and to optimize distribution of manpower assets. He received a bachelor's in civil engineering from the US Military Academy at West Point and a ma ster's in operations research from the University of Texas at Austin. Doug has contributed a number of papers in the statistics and data modeling sections of NESUGs, SESUGs, and SUGIs.