daily vs weekly forecasting and the impact of holidays


I have an option to do daily or weekly forecasting of cash deposits and withdrawals at the bank ATMs using the time series data over the past 5 years. Based on the text (https://otexts.com/fpp2/), it seems to me that harmonic dynamic regression using the Fourier terms and neural network are the most suitable options.

The demand for cash is highly correlated with the holidays (Labor Day, Christmas, Easter) some of which move from year to year. Is it something I need to worry about if I use the weekly forecasting or is it an issue mainly for daily (or day of the week) forecasting? Does the harmonic regression take care of it using the Fourier terms automatically?

Any advice is appreciated.

Referred here by Forecasting: Principles and Practice, by Rob J Hyndman and George Athanasopoulos

1 Like

This kind of question would be better on stackexchange statistics page "Cross Validated"

With moving holidays, yes, I think you should be concerned with both daily and weekly forecasts. Weekly is also particularly tricky since some years have 52 weeks and others 53. The holidays, however, you could deal with by introducing a holiday dummy for the day or week the holiday is. In R you can get the dates of holidays using packages like Holidays, bizdays, tis or similar.

1 Like

This topic was automatically closed 21 days after the last reply. New replies are no longer allowed.