I have sales data for last four year. I want to divide my data in two parts so that I can have base volume and promotional volume. I am thinking to use " Decomposition of Time Series " approach such that the base volume would be trend component and seasonal and error component would be promotional volume.

Is this method would be good? Is there any other method/way to break down the data in two parts.

P.S : promotional volume is the volume that comes from promotions during festival seasons.


1 Answer 1


Promotional volume that arises as a result of festival seasons is better to be handled by dummy variables. The error term is the reminder after trend, seasonality and such dummy.

  • $\begingroup$ Thanks for you reply...How to define dummy variables..can you please elaborate it more...it will be very helpful. Thank you $\endgroup$
    – Arushi
    Commented Dec 24, 2013 at 21:15
  • $\begingroup$ At times you know there were festival events that might have affected the series you put binary variable - 1 if there was a feast, 0 otherwise. Then you will perform regression with that variable included. It will estimate the impact of the festival event on the sales. $\endgroup$ Commented Dec 24, 2013 at 21:21
  • $\begingroup$ so you mean , the dependent variable would be my actual sales data and independent would be the the binary variables like you have defined and no other independent variables...Please reply as I like your approach... Thanks $\endgroup$
    – Arushi
    Commented Dec 24, 2013 at 21:34
  • $\begingroup$ of the form e.g.: y(t) ~ a + bsin(2*pit/365) + ccos(2*pit/365) + d*(IF festival at time t THEN 1 ELSE 0) $\endgroup$ Commented Dec 24, 2013 at 21:40
  • $\begingroup$ Why these sine and cos terms and why are you dividing by 365, I have monthly data for last 4 years... $\endgroup$
    – Arushi
    Commented Dec 24, 2013 at 21:46

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

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