I am doing a forecasting project for school and trying to predict air passenger numbers based on 18 months of historical data from the airport.

I have considered applying Holt-Winters seasonal method due to the observed seasonality in the series. But wouldn't the limited amount of historical data cause an issue? Perhaps someone could recommend an appropriate forecasting method?


1 Answer 1


The best approach is to use an ITERATIVE scheme to form a model while incorporating any identified trend , seasonal dummies , level shifts , pulses that can be identified. A HW additive model is a good starting point with so little data. I suggest that you post your 18 historical values.

  • $\begingroup$ Another approach would be to borrow seasonality from another series. If you have monthly data, you could see whether the seasonal factors you could derive from the famous airline data calcnet.mth.cmich.edu/org/spss/Prj_airlinePassengers.htm look like they would fit your data (a bit of a judgment call). After you deseasonalize, you could use a variety of methods, Holts being one of them. Then reseasonalize to get your forecast. $\endgroup$
    – zbicyclist
    Nov 30, 2017 at 5:11
  • $\begingroup$ autobox.com/pdfs/vegas_ibf_09a.pdf showing why logs are unnecessary when you identify the recent anomalies in the airline series. stats.stackexchange.com/questions/18844/… nicely summarizes when to use a power transform. $\endgroup$
    – IrishStat
    Nov 30, 2017 at 14:44
  • $\begingroup$ Thank you very much for your comments and for sharing the resources. I was given a dataset with almost hourly data - will aggregate! Perhaps you could advice how would borrowing seasonality work? Could you assume the same seasonality in your data because it is observed in another dataset with similar series? I have already observed the weekly, quarterly seasonalities. Does "deseasonalize" mean "seasonally adjusted" in this context? Does iterative scheme refer to using different forecasting methods? Sorry for my "forecasting" vocabulary being quite basic - I am new in this subject. $\endgroup$
    – mage10af
    Dec 2, 2017 at 14:22
  • $\begingroup$ When the reflection was made about copying seasonality it was based on the ARIMA structure for the airline series. I don't believe this is applicable to you due to you only having 18 months of data. Deseasonalize essentially means a contrivance to seasonally adjust data . Iterative modelling suggests adding and deleting structure as needed much like step-down and step-forward regression to customize model form/parameters. $\endgroup$
    – IrishStat
    Dec 2, 2017 at 15:13
  • $\begingroup$ Since you have hourly data I would strongly suggest that you aggregate to the daily level and present a csv file in that format. Make sure you specify the starting date and the country as holidays usually have an effect, Daily data can often be used to make weekly and monthly and quarterly predictions. Make sure that every day is reported , filling in 0.0 for non-reported days. $\endgroup$
    – IrishStat
    Dec 2, 2017 at 15:18

Your Answer

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

Not the answer you're looking for? Browse other questions tagged or ask your own question.