11-03-2011 04:05 PM
I have over 100 distributions (Auto dealerships) and I need to use a time series to forecast each dealership's sale for the next 12 months with monthly data I have for the past 10 years. I want to develop a time series model, but first I want to compare dealerships to see if any have the dependent variable (which in this case is the prices) in common (perhaps Proc corr, I don't know) and see if I can group some so that I don't have to model 100 stores, but instead model only on say 60 or 50, or whatever the results show.. I can say 4 stores show similar trends and cycles, seasonality and increase/decrease in price, that I can group Sore 1/2/3/4 together..
How can this be done? Is proc corr the right tool or perhaps a proc glm ??
11-03-2011 04:15 PM
It will be interesting to see what our statisticians have to recommend. I would have thought a discrimant function or factor analysis, or possibly a decision tree or neural net. However, I am definitely not well versed in time series and don't know if there are similar types of analyses that also take time into account.
11-03-2011 06:11 PM
See also http://communities.sas.com/message/39698#39698 for the perspective of Time Series analysts.
I have seen some people use PROC VARCLUS to group similar variables into clusters of variables with similar behavior. They then model a representative variable from each cluster, or construct the "average" of each cluster and model that.