The best formula to use in excel to forecast daily call volume?

[mdo8105]

I have used single and double exponential smoothing, but did not know if there is a more accurate approach to predict how many calls that can be expected.

 

[Dason]

Single and double exponential smoothing don't account for seasonality. If your call volume seasonal in nature? If so then if you wanted to stay in the exponential smoothing family you'll need to use at least *triple* exponential smoothing.

 

[Miner]

It depends on what components are in your time series (i.e., trends, seasonality, or cyclicity). Each approach has particular strengths and weaknesses in dealing with these components. For example, single and double exponential smoothing handle cyclicity, but not seasonality. Double handles trends, but single does not.


Winters and ARIMA will handle all three, but add additional complexity (particularly ARIMA).

 

[noetsi]

Holt Winters will deal with seasonality and [I believe trend]. ARIMA won't deal with non-linear predictions and you have to have considerable expertise in practice to use it so sticking with some form of exponential smoothing like Winters or Holt Winters is probably desirable. Test of the various predictive methods show it has a good track record and its robust to assumptions.

That said you have to provide more details including context to answer your question. Poison regression or some form of operations research (like queuing models) might predict call rates over time better than time series, particularly if you want to break up call volume in narrower periods of time like hours or minutes. Commonly that is what you use to predict this.

 

[mdo8105]

Thank ya'll for the insight. Right now I am using a multivariate regression to predict the call volume. I am taking my actual calls and and running the log function on them. I am using the log of the actual calls as my Y, then using Holt Winters to produce my Lt and then using the Lt with Mon-Fri (with dummy variables) and Jan through Sept (with Dummy variables) since excel will only allow 16 variables. With my predicted Y I am raising it by 9.8. I tried raising with the base of 10, but found better accuracy with 9.8. I am running an 8 percent MAPE with this equation, but so far for this week I have been sitting at 1 percent MAPE. Does anyone have any ideas on how to better increase the accuracy?

Thank you