The Impact of Misrepresentative Data Patterns on Sales Forecasting Accuracy

Abstract

The development of a sales forecasting system involves three major steps. The first step is to obtain prior sales data and to identify the model that will best forecast the patterns that exist in the data. The second step is to estimate pa rameter values for the selected model by analyzing the prior sales data. The third step is to test the accuracy of the model by use of the prior sales data. Each of the steps requires use of prior data.

In all three steps, there is a basic assumption that the past data represent some underlying process that can be identified and modeled. In some cases the past data may not represent the underlying process, and the forecasting process is seriously distorted. Some frequent causes of dis torted data are 1) accounting methods that are used to record or collect the data, 2) marketing tactics such as promotions which that create outliers, 3) limits on produc tion capacity that cause stockouts.

This paper looks at events and actions that may distort data used for sales forecasting and at the resulting impact the events and actions may have on forecasting accuracy.