I am just wondering that, in terms of the multi-seasonal time series forecast, what is the difference between

  1. using auto.arima find the ARMA order, then fit arima and include xreg=fourier in.

  2. using tbats

As ARMA+Fourier can also takes other covariates into account while tbats cannot, why people tend to use tbats one? Better performance in terms of forecasting?


I had the same question, and found this link here https://robjhyndman.com/hyndsight/dailydata/. In short, if you want to account for other regression coefficients (i.e. explicitly model a given holiday period), then you use Fourier and then ARIMA+regressor.

Note that if you want to fit a Fourier, then you will need to do some 10-fold cross-validation to get the best order of the Fourier. For example,

z <- fourier(ts(x, frequency=365.25), K=5) ## K is the parameter to be tuned.

| cite | improve this answer | |

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

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