I'm trying to forecast sales time-series data across a product line (in R). I can identify groups of products that have very similar seasonality profiles that I would like to apply to others in a couple of situations:

  • Forecasting new items that have no history.
  • Forecasting Individual products where I want to use an aggregate of multiple products to smooth out the seasonality.

In the past I have typically used something like a classic multiplicative decomposition and applied that to a subjective number of recent periods for the item I'm trying to forecast to determine a deseasonalized trend which I can then apply a model like exponential smoothing.

I would like to use STL to identify the seasonal components (allowing for seasonality shifts over time). But I can't figure out how to scale either the seasonal component from STL or my target product to combine them.

Is there a way to do this, or should I just use simple multiplicative decomposition? FYI, the data I'm currently working has very high seasonality, but I'm not sure that matters.


Your Answer

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

Browse other questions tagged or ask your own question.