Using false alarms to improve forecast accuracy

This paper demonstrates that tracking signals can be used to improve forecasts in an unexpected way; by taking advantage of false alarms. When the demand process has not changed, if the tracking signal exceeds the control limit the resulting ‘trip’ is a false alarm. We show that even a false alarm is cause for concern because it correctly indicates that the forecast is biased. Furthermore, if not corrected, the bias can be expected to persist for some time. False alarms are often caused by statistically unusual patterns, such as a series of unusually high demands, rather than a single ‘outlier’. Thus a correction method requires information from the past; simply ignoring the most recent data point is insufficient, for example. One method for improving the forecast is developed. Tests on a set of simulated demands show that forecast bias, variance, and the false alarm rate are all reduced. The method is easily implemented and requires no new information because it is based on a term that is used already in calculating the tracking signal.