I have one year of daily data for forecasting. while using auto.arima to find the best fit model, it gives me ARIMA(3,1,3). However, when I used tbats to find the best fit model, it gave me the following:


I wonder why they provided different models ( If I get it right, the BATS model is an ARMA(2,2)).

(ARIMA predictions makes more sense as BATS predictions give negative numbers which does not make sense) I also believe that my data have weekly seasonality but non of these models include any seasonality index. Attached is my timeseries plot. enter image description here

  • 2
    $\begingroup$ I suspect you are stripping out the ts characteristics before calling auto.arima or bats. Otherwise they should both identify the seasonality easily. BATS also assumes ARMA errors, so it won't find an ARIMA model. Please provide a reproducible example if you want a proper answer. $\endgroup$ – Rob Hyndman Sep 11 '14 at 23:12
  • $\begingroup$ What do you mean by ts characteristics?Looking at my time series plot, I saw double seasonalities so I used BATS. However the provided BATS model did not include any seasonality parameter. I then used the following command: > x <- msts(NumOrders, seasonal.periods=c(7, 52*7)) plot(x) fit <- tbats(x) .Using msts, it gave me a good predictions with small error values. My biggest question in here is that how I should know that I have to use msts instead of ts? $\endgroup$ – user12 Sep 12 '14 at 12:15
  • 2
    $\begingroup$ I don't see double seasonalities in your time plot. And since there is only one year of data plotted, it is not possible to estimate the annual seasonality without some very strong assumptions. Use msts if there are multiple seasonal periods and ts if there is only one. And give a reproducible example as requested. $\endgroup$ – Rob Hyndman Sep 13 '14 at 2:40
  • $\begingroup$ Thank you Dr Hyndman. I will look for a database having more than at least 2 years of dataset to have a more reliable forecasting model. $\endgroup$ – user12 Sep 14 '14 at 3:54

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.