Replies to This Discussion

Permalink Reply by Dr A J Babu on September 17, 2010 at 2:55am

You need to use hierarchical forecasting method can help you multiple store level forecasting sales.
Do you have SAS HPF Engine? So that you can able to write the macro coding to generate the forecasting results or use SAS Forecast studio is self driven.

Rgds
A J Babu

Permalink Reply by Tom Wolfer on September 17, 2010 at 10:29am

Are you trying to forecast sales for each store individually, or, for all of the stores combined? This will make a difference as far as how you approach the problem.

You also seem to indicate that you are attempting to forecast within geographic areas (for one store or more). Is this also true, because, it will add another layer to your analysis.

Permalink Reply by Ralph Winters on September 17, 2010 at 1:49pm

Tscreg is a time series oriented procedure so I would suggest using that, especially if you intend to build a complex model. Proc Mixed will also work for simple models.

-Ralph Winters

Permalink Reply by DataLLigence on September 20, 2010 at 5:23am

No Babu, I don't have SAS HPF.


Tom: I am trying to forecast sales for each store (individually), at the month level. I will be using MSA (where that particular store is located) level specific information as one of the predictors.

Ralph: I tried using TSCSREG but the model ran into some problem when I used the PARKS option.

Permalink Reply by Tom Wolfer on September 20, 2010 at 7:21am

Are you trying to determine where to put a new store, or, determine sales forecasts within the trade area of an already existing store for inventory/staff planning purposes? This may make a difference in how you approach things.

Here is what I might do (whether using SAS or whatever, automated or whatever), for ONE store, before I start trying to forecast for many all at once depending on what your answer to the above question is:

1. Confirm that you have enough months (rolling, if required) of sales data for the store in question
2. Check for seasonality in the store sales data, and I assume that there will be some
3. Deseasonalize the store sales data by creating a seasonality index
4. Now, you can try various forecasting methods on the deseasonalized data: linear regression, moving average, polynomial etc...determine the predictive power and calculate the error rates for each

What do you think?

Permalink Reply by DataLLigence on September 21, 2010 at 2:51am

Thanks Tom!

That's exactly what I had done :) Just wanted to check with people who had worked on similar projects.

Permalink Reply by tom reilly on December 6, 2010 at 2:28pm

How is that working out? Are you accounting for level shifts in your data? Changes in Seasonality (ie day of the week changes over time)? Trends? Lead/Lag relationships in the causals?

Permalink Reply by DataLLigence on December 7, 2010 at 3:58am

Yes, I took care of seasonality and trends. Day of the week, and week of the year effects were taken into account when I was deriving daily forecasts from the monthly sales/dorecasts.