0

Like the title suggests I'm having difficulty deciding the next steps I should take in terms of my current project. At the moment I have about 4 years worth of daily sales data dealing with "Christmas" items.

Obviously there is a very strong seasonal component to my data, and have used several different methods of trying to model it correctly.

The issue in my data seems to come from the fact that there are long stretches of time throughout the year with almost no demand at all, followed by large spikes in demand at the end of the year. To that end, I have been attempting to use methodologies like Croston's that deal with intermittent demand, but I feel like Im out of my depth in terms of what to do next. Any suggestions or help would be greatly appreciated!

tmoriss
  • 1
  • The proposed duplicate is about Croston's method, which is inappropriate in your case... just as in the duplicate, because neither your nor their time series is intermittent in any meaningful sense of the term. Take a look at my answer, which proposes using a simple seasonal decomposition method to forecast. – Stephan Kolassa Jun 29 '23 at 02:02
  • Hi Stephan! Yes I took a look at your post and it was actually exactly the information I was looking for thank you! I guess I was just confused by the flood of information about intermittent demand. I assumed that it was a similar case because my data has a lot of zero values in it, but now I see that it’s essentially just a seasonal time series problem. I’ve started to run STL, TBATS, and dynamic harmonic regressions on my data and have gotten some decent predicted values. If you have the bandwidth for it, do you recommend any methods for fine tuning my models for better predictions? Thanks! – tmoriss Jun 30 '23 at 04:23
  • Glad to hear that thread helped! As to fine-tuning your forecasts, you may find this thread helpful. Essentially, are there any predictors you could use, like promotions, price changes or marketing spend? Or any other external data that you think could be informative? Then you could regress the sales series on these and run the time series forecasting methods on residuals. However, daily sales around Christmas are always a headache, because with Christmas shifting around, the normal intra-weekly pattern gets all messed up... – Stephan Kolassa Jun 30 '23 at 07:07

0 Answers0