in forecasting, typically, we assign a heavier weight to the most current observations. However, I am finding many cases where a "blip" in last month's sales leads to a very pessimistic view about the future 12 months. Blips also lead to wide swings when comparing a forecast generated before the blip vs the forecast updated next month (after the blip happened).

So, my question is whether it is possible to take a look at last month's data within the context of a longer series of observations, and reduce the weight of the most current observation if this deviates substantially from the longer term trend.

I use to consider ARIMA, HoltWinters, TBATS, STL, ETS etc among my model candidates in R, but I haven't found any reference to this being doable in an automated way.

Any ideas that you can think of?


  • $\begingroup$ In forecasting we should identify the relative importance of previous values not assume that most recent values are more important. I would have to actually see your data to precisely answer your question. It is possible to identify "blips" and "seasonal blips" and incorporate them into your equation. If you wish to post your data I and others might try and help you further. $\endgroup$ – IrishStat Feb 10 '18 at 12:55

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.