2
$\begingroup$

I am working on call center load forecasting. I have daily call volume for past 24 months along with few variables which does affect call load. Though none of these variables have come significant in ARIMA model.

My management has asked me to create forecast for two months in advance. I am concerned that just by using call center daily call volumes in Arima model my prediction will not be accurate so far in future.

Autocorrelation of daily call volume deteriorates drops to less than 0.5 in 5 days. I don't think that it will be very accurate for predicting values 2 months from now. What are your thoughts on this.

$\endgroup$
  • $\begingroup$ It sounds like you are well on top of the analysis side of your problem. Now you need to need to ask at explainingstatstodummies.stackexchange.com about how to communicate that effectively to management! $\endgroup$ – Bogdanovist Oct 12 '12 at 9:36
  • $\begingroup$ I am no expert on time-series, but assuming that you and @Bogdanovist are correct, which seems likely, what you could do is make a plot, based on predicting from the past, of how accurate the predictions are. $\endgroup$ – Peter Flom - Reinstate Monica Oct 12 '12 at 10:29
2
$\begingroup$

Whatever method you use - be it ARIMA forecasting, or a plain, old exponential smoothing - be sure to place confidence limits around your forecasts. This will show management how accuracy diminishes as you move into the future.

$\endgroup$
1
$\begingroup$

In addition to confidence intervals, you could also examine the peaks in volume and hypothesize causes that are not quantified in your existing data. Examples will depend on your industry, but could be new product launches, recalls, updates to partner products, or Mondays. This is where I usually find a subject matter expert in the business to help brainstorm if I don't have sufficient visibility. These could then be codified into a generic "event" variable, or simply layer on top of a volume graph to help illustrate how other factors really drive the variation in call volume.

$\endgroup$

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.