0

I have data about sale as example follows:

Date Product Category Sale Promotion of product 1 Promotion of product 2 Promotion of product 3
01/01/2020 Product1 A 3 1 0 0
01/01/2020 Product2 A 4 0 1 0
01/01/2020 Product3 B 2 0 0 1
02/01/2020 Product1 A 7 1 0 0
02/01/2020 Product2 A 4 0 1 0
03/01/2020 Product1 A 2 0 0 0
03/01/2020 Product2 A 1 0 0 0
03/01/2020 Product3 B 9 0 0 1

Suppose I have 50 products and because of the hypothesis that the sale and promotion of one product can impact to others. So, I would like to forecast each product in parallelly. (I mean that model forecast whose the outputs are the sale of each product) What are the methods to solve this problem?

Pitouille
  • 1,506
  • 3
  • 5
  • 16
Sherry
  • 1
  • How mNY DIFFERENT PRODUCTS? Maybe look at [hierarchcal forecasting](https://www.google.com/search?channel=fs&client=ubuntu&q=site%3Ahttps%3A%2F%2Fstats.stackexchange.com+hierarchcal+forecasting) – kjetil b halvorsen Oct 01 '21 at 07:46
  • Yes! There are more than 50 products for each category. I also consider about Hierarchical forecasting, but i dont know what is the package which can be used for this methods in python with exogenous variable. – Sherry Oct 01 '21 at 08:03

1 Answers1

0

Similar questions have been asked and given fairly comprehensive answers here and here.

A few comments:

  • The main keywords you will want to look for are i) 'Hierarchical Modeling' and ii) 'Panel Modeling'.

    • Panel modeling is concerned with modeling multiple time series at once.
    • Hierarchical modeling makes use that some products belong to the same category.

  • If you have a large number of products, it may make sense to engineer the promotional features. For example, instead of having 1 column per product, maybe try columns like:

    1. Is this product on Sale?
    2. How many other products are on sale?
    3. Are products from the same category on sale?
Adam Kells
  • 908
  • 1
  • 12