Per title, we're trying to identify a good time series modeling technique for:

  • 70-100 variables of monthly sales or volume data (2015 or 2018 to present)
  • Ability to forecast not only using trend and seasonality of the data but to use econometric factors to fit a model and then the same factors to help forecast (i.e. monthly GDP, personal consumption, industry specific numbers, etc)
  • Be able to change those manually change those econometric future forecasts for scenario modeling and planning

So far we've tried TFT (Temporal Fusion Transformation) but we haven't had luck and it might be due to the fact this data is a little light for it. Also tried Random Forest but we can't seem to capture the trend very well with it as the sales/volume grow over time.

Our preference is using a tool like darts, but open to investigate anything that makes the job easier.



Your Answer

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