ABSTRACT
This chapter focuses on outliers and improving forecasting when outliers exist in the data. There are several ways of categorizing outliers. Outliers can be classified by their source: the randomness of the underlying process, acts of nature such as floods or fires and human activities such as accounting techniques, promotions, or price cuts. Sales time series are frequently subjected to distortions caused by advertising, competition, promotions, accounting systems, and stockouts. These distortions are called outliers in statistical literature. Outliers can be classified as predictable, such as those due to promotions, and unpredictable, such as those due to randomness in the data. Returned products can be handled by accounting systems in two ways that result in a distortion of the data pattern that forecasters need to deal with. Two approaches to dealing with outliers are possible: automated data modification and manual intervention. In many instances, empirical information about the impact of an atypical situation may be available.
