IN THIS CHAPTER
Understanding when to remove the trend from a baseline
Keeping your baseline from jumping around
Putting the pieces of a baseline back together
Sometimes you’ll decide that you’d prefer to use a single-variable forecast method — for example, one of the two single-variable methods that this book discusses: moving averages and simple exponential smoothing. If you’re going to do that — and there are some pretty good reasons to go that direction, including improved accuracy and the presence of seasonality in the baseline — you should first check to see if the baseline has a trend. As several other chapters discuss, a trend is the tendency in a baseline to move up or down (not usually both) over time. Both moving averages and simple exponential smoothing behave better in baselines that don’t have a trend.
One good way to remove a trend from a baseline is called differencing. If you use differencing, you subtract one value in the baseline from a subsequent value. Doing that subtraction has some consequences for the values you use to forecast from. The decision to use differencing isn’t a slam-dunk, though — some trade-offs are involved.
If you use differencing, you apply your forecast method to the differences. After you’ve got your forecast, you still have to put it back into the original baseline’s scale. This is called integrating, and this chapter shows you how to do it.