I have a question regarding forecast. I'm building an inventory model around warehouses, where all warehouses have multiple customers/countries assigned. I have data on sales for all countries separately, so I can perform my forecast on this data to get a forecast for the country demand. I can do the same at the warehouse level, i.e. first sum all historic sales per warehouse, and then use this for the forecast.

I need both forecasts. Inventory is decided on the warehouse forecast, but other costs are determined based on the countries, as well as a specific service measure I use.

Can I forecast for each country seperately, and then add the forecasted demand to get the forecast for the warehouse? Or, is it also okay to forecast everything individual? Is it then still in lign with each other?


country 1: 0 2 0 1 1 5 0 3  
country 2: 1 1 4 3 0 3 2 0 
wh:        1 3 4 4 1 8 2 3 

So, the wh level is how many inventory we need. The country level, e.g. is for the distribution costs which depend on countries. Do I forecast for country 1 and 2, and add the forecast and use this as input. Or do I forecast for all separately. And what if on the country level it may be intermittent demand but results in smooth demand at wh level?

EDIT: I tried something, using HoltWinters and predict in R. The SSE for forecasting on warehouselevel is 15410 and the SSE for forecasting at country level and then adding is 18576, so you would say the first one is better (in this particular example). But is it then still possible to also forecast for both countries seperately and use this as forecast to determine the country based costs?


You can sum them, but as described by Hyndman et al in the Optimal combination forecasts for hierarchical time series paper (see also their hts package, the fpp2 chapter and those slides), you can do better by using hierarchical forecast that makes forecasts on all the levels (upper and higher) and then combines them.

| cite | improve this answer | |
  • $\begingroup$ Thank you! I will look in to this. Had no idea such a thing existed. $\endgroup$ – pk_22 Dec 13 '17 at 10:35

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

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