2
$\begingroup$

I'm currently a MS student in statistics with very limited forecasting experience. I would like to find a model to predict grocery store department sales by week. I have a data set with about 2 years of weekly sales for 5 different grocery stores that I would like to fit the model on.

My question is, how do I handle the multiple departments? Do I need to use separate models on them? Or are there techniques for a multivariate model that I can use?

If you know of any good books or websites that have something to start with that would be a huge help.

Thanks!

Improve this question
$\endgroup$
3
  • $\begingroup$ If you're new to forecasting, you'll find this book of help: otexts.org/fpp. Make sure you take a look at it. $\endgroup$
    – mugen
    Sep 28, 2015 at 14:38
  • $\begingroup$ Also, Diebold "Forecasting in Economics, Business, Finance and Beyond" is a nice resource (the textbook has incidentally been updated today). $\endgroup$
    – Richard Hardy
    Sep 28, 2015 at 19:49
  • $\begingroup$ Thanks for the book suggestions. I'll check both of them out. $\endgroup$
    – Kristofersen
    Sep 29, 2015 at 4:47

1 Answer 1

Reset to default
1
$\begingroup$

Practically speaking you are better of dealing with each department separately for each separate store. In terms of frequency you might be much better off using daily data data and then (if you wish) sum the daily forecasts to weekly forecasts. The reason for this is that holidays will skew your weekly series. Long weekends around holidays can be treated with a daily model. More generally what we do on Monday is similar to what we did last Monday whereas what we do on week x is not necessarily similar to what we did on week x last year. That plus the confusion due to 52/53 weeks in the year results in generally poor accuracy using weekly data. In terms of a web example of daily data take a look at http://www.autobox.com/cms/index.php/afs-university/intro-to-forecasting/doc_download/53-capabilities-presentation slide 45 as it lays out some ideas that you might find illuminating.

Improve this answer
$\endgroup$
3
  • $\begingroup$ Thanks for your reply. The holiday comment definitely makes sense for Christmas and other holidays that will land on the same day each year. But what about holidays like Memorial Day when the holiday can land in different weeks. $\endgroup$
    – Kristofersen
    Sep 29, 2015 at 4:49
  • $\begingroup$ That is why you would prefer a daily model because the specificity of the daily indicator would point to the precise day. In addition to daily effects one might also need weekly effects and/or monthly effects. $\endgroup$
    – IrishStat
    Sep 29, 2015 at 11:37
  • $\begingroup$ Okay, thanks. It looks like I will have a lot of reading to do. $\endgroup$
    – Kristofersen
    Sep 29, 2015 at 23:25

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.