Forecasting Geographic Data
Michael Leonard and Renee Samy, International Symposium on Forecasting, 1998
Virtually all businesses collect and use data that are associated with geographic locations, whether it is sales, customer demand, or profits at different locations. Geographic Information Systems (GIS) provide a means for visualizing and analyzing such data. Collecting each region’s time series into a vector time series forms a geographic time series. Multivariate modeling techniques can then be used to model correlation between geographic regions to improve forecast accuracy. This paper provides a practical demonstration of using a VAR model restricted by an adjacency matrix to forecast and visualize a geographic time series using SAS software. In particular, Base SAS software is used to manage and query the data and generate the adjacency matrix. SAS/ETS software is used to model the VAR process and forecast the geographic time series. SAS/GIS software is used to map the geographic regions, select the regions to analyze, and visualize the forecasts. SAS/GRAPH software is used to plot the time series data. SAS/AF software is used to develop the interactive interface.