Case Study 14—Data Analysis Tools for Forecasting Sales
An average monthly sales figure derived from 12 months of annual sales is not useful for a seasonal business, as it overstates sales for a peak summer season business that has a loss period in midwinter, and it will understate the loss for that same business in midsummer. Even for a nonseasonal business, monthly average sales computed from earlier years' data can be misleading if there has been an upward or downward trend over time. If the measure of damages is the lost value of the business, then forecasting sales using a five-year simple or weighted average of past sales will misstate next year's expected revenue if historical sales have exhibited a trend in either direction. So, are there any situations that allow for the use of averaging techniques in determining damages?
Need for Analytical Tests
Yes, there are such techniques that are appropriate for stationary time series, that is, sales over time that exhibit no significant upward or downward trend, and that have equality of means and variances throughout the length of the series. First, though, in this chapter we will present some more sophisticated and complex analytical tools than those we introduced in Chapter 2 to determine stationarity, independence of observations, autocorrelation among the observations, and test for trend and seasonality. In subsequent chapters, we will describe the techniques for forecasting with random models, introduce techniques ...