Time-Series Forecasting: Exponential Smoothing Part 2

Like smooth jazz floating gracefully out of a sweet saxaphone, the forecasting series continues today with part II of exponential smoothing. Previously we looked at the equation which will help us find the projected sales or demand for the following period.

That equation is:The formula for smoothing is:


Let's see how that would work in action, shall we. Consider this simple data set with the demand for Oranges:

Oranges sold (actual):
Jan	7
Feb	5
Mar	10
Apr	14
May	11
Jun	12

Next we will want to find out what the 2 month moving average of each of these are. This will provide us with a forecasted value.

Oranges sold (actual):		2 mo avg
Jan	7
Feb	5
Mar	10	6
Apr	14	7.5
May	11	12
Jun	12	12.5

Now that we know our 2 month moving averages, we can use those as our original forecasts. Now we can apply a smoothing constant. If we use a smoothing constant of .8, inserting it into the equation above, we can see how our forecast is affected. Note that January, February, and March are not able to do an forecast using "exponential smoothing" since we do not have a previous period forecast to work from.

Oranges sold (actual):		2 mo avg	Smoothing Constant	Equation	New Forecast
Jan	7		0.8
Feb	5
Mar	10	6
Apr	14	7.5		6+0.8(10-6)	9.2
May	11	12		7.5+0.8(14-7.5)	12.7
Jun	12	12.5		12+0.8(11-12)	11.2

Ultimately, the end result of this will be a smoother and hopefully more accurate forecast. We only looked at a few months, but in reality these forecasts would continue on as long as necessary.

This will move us into the the next part in my series: Forecasting Errors. Using the mean absolute deviation, we should be able to compare the accuracy of forecasts. Thanks for visiting and I will see you next time.
-----Sincerely, Trevor Stasik.

---

To return to initial post about forecasting, click HERE.
To visit the next post in this forecasting series, click HERE.

---

Financial Forecasts, Exponential Smoothing, HTML Table, Sales Demand