I knew that tbtas could not consider regressor in r but I knew that Dr Hydman group were working to include this feature. I just wanted to know if it is still under work.



closed as off-topic by Stephan Kolassa, Andy, gung, Nick Cox, StasK Sep 26 '14 at 0:57

  • This question does not appear to be about statistics within the scope defined in the help center.
If this question can be reworded to fit the rules in the help center, please edit the question.

  • 1
    $\begingroup$ This question appears to be off-topic because it is about how to use R, but not about programming in R per se & does not have a reproducible example (& thus would be closed on Stack Overflow). $\endgroup$ – gung Sep 25 '14 at 17:11
  • $\begingroup$ This question appears to be off-topic because it is about a time line of somebody's development of an R package. It does not belong anywhere in the SE universe -- an email to the developer(s) should have sufficed. $\endgroup$ – StasK Sep 26 '14 at 0:57
  • $\begingroup$ The author of TBATS Rob Hyndman has addressed this question in his recent blog $\endgroup$ – forecaster Oct 6 '14 at 20:11

The help page to tbats() notes that you can specify parameters that are passed through to auto.arima(), which models the errors.

EDIT: I assume the OP contacted Rob Hyndman, the maintainer of the forecast package. Rob writes that this information from the help page is actually misleading. tbats() does not and likely will never include covariates - it will not complain if they are provided, but they will be silently ignored.

Rob recommends using auto.arima() if external regressors are required to be used. The help page to auto.arima() notes that you can specify external regressors by feeding a vector or matrix to the xreg parameter.

  • $\begingroup$ Thank you for your note. I already implemented your code for my data and it improved my forecasts a little not significantly. Although I included daily seasonality to capture lower values for weekend, but it seems that the model is not properly smoothing the forecast for weekend. I dont know if I have to include dummy variables for days of a week rather than including daily seasonality. $\endgroup$ – user12 Sep 25 '14 at 17:14
  • $\begingroup$ On the one hand, you can try multiple different models, as you propose. On the other hand, sometimes it simply is hard to improve on a given forecast, if the series is inherently not very forecastable. Good luck! $\endgroup$ – Stephan Kolassa Sep 25 '14 at 17:34
  • $\begingroup$ Stephan I tried tbats with regresor for multiple data sets and it did not work. I mean the fitted model with regressor and no regressor were the same. Are you sure tbats work with regressor? $\endgroup$ – user12 Sep 26 '14 at 14:17
  • $\begingroup$ It's possible that auto.arima() removes regressors, e.g., by AIC. A quick glance at the paper referenced in the auto.arima help page does not tell how external regressors are handled. You may possibly have found a bug. Best to step through the source code and/or contact the package maintainer, Rob Hyndman. $\endgroup$ – Stephan Kolassa Sep 26 '14 at 14:30
  • $\begingroup$ Honestly I saw in one of Dr Hyndman's post that he said right now tbats can not handle regressors and they are working on it. So I thought its still under work and they have not published the code yet but Im not sure about it at all. $\endgroup$ – user12 Sep 26 '14 at 14:32

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