CHAPTER 3Understanding Your Software's Bias and Accuracy Measures

3.1 INTRODUCTION

Before we look at how forecasting software packages turn data into forecasts, we need to look at how to interpret bias and accuracy measures that are usually reported by software. Bias in forecasts usually refers to a persistent tendency for the forecasts to be too high or too low (though there are other types of bias). Detecting bias can be particularly important in learning how forecasts might be improved. One research study, by Nada Sanders and Gregory Graman in 2009, suggested that high levels of bias in forecasts can add considerably to a company's costs.

As we'll see, accuracy measures come in a variety of forms. They can represent the typical closeness of the forecast to the actual sales, measured in actual sales units or as percentages, or they can provide comparisons of the accuracy of a chosen forecasting method with a benchmark method. It's important to remember that there is almost certain to be a difference between a forecast and actual sales because of noise. A difference does not necessarily imply that the forecasting method has failed in some way. As we saw earlier, forecasting methods are not intended to anticipate noise. Some sales histories consist largely of noise, so large differences are to be expected since the product's sales are largely unpredictable. We should only consider changing a forecasting method after monitoring its performance over several periods and comparing ...

Get Profit From Your Forecasting Software now with O’Reilly online learning.

O’Reilly members experience live online training, plus books, videos, and digital content from 200+ publishers.