Data Analysis with Microsoft Excel: Updated for Office 2007

Chapter 11 Times Series 453

To decrease the value of the smoothing constant:

1 Click the down spin button repeatedly to reduce the value of w to

0.03. The forecasted values and the weight assigned to prior obser-

vations change dramatically. See Figure 11-13.

Figure 11-13
Reducing the
w value
to 0.03

forecasted values

are more heavily

smoothed

observations

further back in

time are weighted

more heavily

With such a small value for w, the smoothed value has a long memory. In

fact, the fi nal forecasted value, 0.023, is based in some part on observations

spanning the entire time series. A consequence of having such a small value

for w is that individual events, such as the large drop-off in the middle of

the time series, have a minor impact on the smoothed values. The line of

forecasted values is practically straight. Note as well that the standard error

has declined from 0.293 to 0.281. In that case, the time series data is best

estimated by the overall average or smoothed value that has a long memory.

Now increase the value of the smoothing factor to make the forecasts

more susceptible to unit-by-unit changes.