Exponential Smoothing for time series with seasonality

[geomat0626]

Does vensim have a Holt-Winters type triple exponential smoothing function for forecasting a times series that has seasonality.

Seasonality of the data series is critical to my model hence I want to be able to forecast this as accurately as possible within the exogenous data time series.

Any comments or advice is welcomed.

George

[tomfid]

I did this once using a macro, so it certainly can be done, but I'm away from my office at the moment so I can't post it.

[geomat0626]

Thanks Tom. I am looking forward to see this when you are able to post it.

George

[tomfid]

Here it is. I couldn't find my original, so I had to reconstruct it.

This is similar to https://www.otexts.org/fpp/7/5, but that's discrete time and dimensionally inconsistent unless you add a bunch of implicit time constants.

I think my approach is mathematically very similar, if not identical, when TIME STEP=1.

Hope this is helpful.

[tomfid]

I should add that this requires arrays for the seasonal bins. If you're only smoothing, or forecasting at integer multiples of the seasonal period, you could get by with a DELAY FIXED instead of the array, which would permit making this into a macro.

[geomat0626]

Hello Tom,

Thanks for the help.

I did not see a file/function to look at, only the web link explaining the method was visible.

[tomfid]

Oops. Duh.

seasonal smoothing.mdl

(9.48 KiB) Downloaded 343 times

Sorry!

[geomat0626]

Thanks Tom. Am going to have a look at it.

[Xuanao Zhou]

I should add that this requires arrays for the seasonal bins. If you're only smoothing, or forecasting at integer multiples of the seasonal period, you could get by with a DELAY FIXED instead of the array, which would permit making this into a macro.

Hi Tom,

Thank you for sharing. I'm reaching out to inquire about the feasibility of implementing seasonal forecasting functionality on the PLE version. You had mentioned using the DELAY FIXED in this context; could you please share more details on how to utilize it?

The aim of my model is to predict the next shipment time so as to set the safety stock accordingly. Shipment times are influenced by a variety of factors and show seasonal trends, being longer in summer and winter, and shorter in spring. Therefore, I am looking to embed a seasonal forecasting module (using the Holt-Winters method) into the model to predict the next shipment time based on historical shipment data.

Moreover, the variables I have set up to record shipment times include instances of zero data, indicating periods without shipping tasks. In such cases, how should I go about selecting non-zero data for prediction purposes?

Please find details about my base model in the attachment.

Looking forward to your insights.

Best, Xuanao