0
$\begingroup$

My data contains no trend or seasonality, but contains noise. While I use ARIMA to forecast it, the model is (1,0,0). In addition to this if I use dayOfWeek and a promotion predictor I get a forecast with a sequence repeating itself. I don't trust this forecast. Is there a better model for forecasting my data? Below is a STL decomposition plot of my data. First plot attached is a promotion data. enter image description here

STL decompose

$\endgroup$
  • $\begingroup$ You really need to be more specyfic for this to be answerable. Please edit to add more details. If your data is noise-only, then probably the best you can get is to use the mean as your forecast. $\endgroup$ – Tim Oct 25 '17 at 8:48
  • $\begingroup$ Thanks Tim for the revert. Added to the question a plot of STL decomposition of my dataset. Does this help specifying my question? $\endgroup$ – Sid Verma Oct 25 '17 at 9:14
  • $\begingroup$ It does help. So your data is daily, and the "time" probably refers to weeks? Can you please add a plot indicating when promotions happened (was there only a single kind of promotion), plus the forecast of your model including day of week, and the output of your model? $\endgroup$ – Stephan Kolassa Oct 25 '17 at 9:40
  • $\begingroup$ Yes, my data is daily and "time" refers to weeks. I have added a plot of the promotions values. My model is (1,0,0) with following coefficients: dayOfWeek as 0.3920 and promotion as 0.3070 $\endgroup$ – Sid Verma Oct 25 '17 at 9:59
  • 1
    $\begingroup$ Please post the actual data for both Y and the promotion. $\endgroup$ – IrishStat Oct 25 '17 at 10:31

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.