I need to automatically identify if a time series is intermittent or not. Depending on the result I'll use one or another method for forecasting it.

Is there any test to detect intermittent time series?


Just set a threshold like 30% and if the number of "zeroes" exceeds this threshold then declare it to be an intermittent demand series. For guidelines to deal with "unusual demands" rather than believing them and Level Shifts ( n.b. A level Shift is not a time trend ) . Also since intermittent demand can yield rates that are auto-regressive ( i.e autocorrelated ) models like the Poisson Model or the Croston approach are of limited value. Please see the discussion Please see my comments in How to forecast based on aggregated data over irregular intervals? regarding this.

| cite | improve this answer | |
  • 1
    $\begingroup$ Hi Dave. I never heard the term intermittent time series before. Does that mean that it is 0 for a time then behaves like a statioanry time series then go back to 0 for a while and so on. $\endgroup$ – Michael R. Chernick May 8 '12 at 0:26
  • $\begingroup$ @MichaelChernick yes .. exactly , It arises when there is sales. Consider a time series of the number of gallons of gas that you put into your car on a daily basis. $\endgroup$ – IrishStat Jun 11 '12 at 18:41

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

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