How do I include historical related data and forward looking related data in amazon forecast

In a typical retail case I would see 2 main ways stock level can affect sales/demand:

1. When stock runs out, sales are artificially 0 even though underlying demand may be high.
   • Availability is very important to incorporate in demand forecasting models, to avoid under-forecasting potential sales.
2. If stock is in excess, maybe stores run extra promos to try and shift it?
   • This could actually be a harmful feedback loop retailers might like to avoid? Especially if it results in discounts / reduced margin.

As a result, I would usually suggest users to try a 0-1 "is stock available" feature for demand forecasting - rather than using raw inventory counts. It doesn't necessarily have to be binary - you could set intermediate values e.g. 0.2 when your inventory is so low there might be issues getting it on the shelf or it might be a actually be a mis-count and stock has run out.

I think this would solve your issue because then it becomes very natural to forecast this RTS forward: Just set always 1 in future to encourage your model to predict possible sales (i.e demand) if you have stock available to meet it.

Of course every business is different, so maybe you have some good reasons to include actual stock level in your demand forecast - or other fields that you can't forecast forward. I believe you should be able to just leave these specific fields empty in your CSV? e.g:

timestamp,item_id,stock,price
2050-01-01,P12345,,10.99