476 Statistical Methods
The output does not give the month for each forecast, but you can easily
confi rm that observation 145 in column A is January 2008 because observa-
tion 144 on the Liquor Sales worksheet is December 2007. On the basis of
the values in column B, you forecast that in the next year, sales will reach a
peak in December (observation 150) with a sales fi gure of $5,093.5 million.
The 95% prediction interval for this estimate is about $4,932.9 million to
$5,254.2 million. In other words, in December you would expect sales of
not less than 4,932.9 million dollars or more than 5,254.2 million dollars.
You could use these estimates to plan your sales strategy for the upcom-
ing year. Before putting much faith in the prediction intervals, you should
verify the assumptions for the smoothed forecasts. If the smoothing model
is correct, the residuals should be independent (show no discernible ACF
pattern) and follow a normal distribution with mean 0. You would fi nd for
the liquor sales that these assumptions are met.
Scrolling back up the worksheet, you can view how well the smooth-
ing method forecasted liquor sales in the previous years, as shown in
Figure 11-32.Figure 11-31
Forecasted
sales values
with 95%
confidence
region