I have weekly data of yearly seasonality around 3.5 years. By using default seasonal test ocsb or ch test in python pmdarima it is not able to give right D differencing which is resulting in higher errors. I am using exogenous 0/1 dummy categories of major peaks of new year sales ,etc etc .

Can I use the R forecast library seasonal strength formula to determine the D Order .I am doing regression with best suitable dummy 0/1 categories and passing the residuals to STL decomposition and applying R forecast default seasonal test strength formula score (.64 cutoff) to find the right D (using 0 or 1 only) and using that D in auto.arima along with those best suitable dummy categories exogenous. Does this make sense ?

  • $\begingroup$ How do you know it is not giving the "right D differencing"? If you know what the right amount of differencing is, why not just use that? $\endgroup$ – jbowman Feb 27 at 20:34
  • $\begingroup$ It's not giving right because the MAPE errors are coming off high consistently,only when we enforce D as 1 in pmd auto ARIMA it works out.I want some sort of test or score to determine the differencing and not by manual enforcement $\endgroup$ – vij Feb 28 at 1:30
  • $\begingroup$ Any comments or suggestion ? Can R seas test of .64 help ?Also the dummy marker exogenous variables need not be differenced right as in the auto arima code ? $\endgroup$ – vij Feb 29 at 13:58

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.