Intermittent Demand Forecasting

10Judgement, Bias, and Mean Square Error

10.1 Introduction

In Chapters 6 and 7, we examined statistical approaches to forecasting. This emphasis is appropriate, particularly in the context of intermittent demand, because the sheer volume of stock keeping units requires an automatic solution. In practice, though, software packages almost always allow for user intervention. This means that users of forecasting software packages may override statistical forecasts with forecasts of their own. Whether such overrides are beneficial or harmful to forecast accuracy has been a matter of debate. The evidence will be reviewed in this chapter.

One of the potential disadvantages of overriding statistical forecasts is that the new forecasts may be biased. In fact, the statistical forecasts may be biased themselves if they are not well chosen. Performance measures for bias, including the mean error, were reviewed in Chapter 9 . In this chapter, we examine bias in more detail, focusing on its monitoring, and its expectation.

The mean square error (MSE) is a scale dependent measure and is not robust to outliers, as noted in Chapter 9. Nevertheless, it is useful in gaining insights into the components of forecast error, including the expected bias, and is discussed in some depth in this chapter.

The structure of this chapter follows the themes outlined above. We begin with a discussion on judgemental forecasting, reviewing the evidence of its effectiveness for fast‐moving products and for products ...