Evaluating the accuracy of business forecasts : a case study. Analysis of a variety of forecasting methods in actual use by a company with an evaluation of their relative accuracy

Abstract

In the last decades the business environment of many companies has become more volatile, more unpredictable and it changes faster. Indicators for these changes are for example an increasing product variety, shortening product lifecycles, the demand for shorter delivery times, the increasing number of companies with global ambitions and alliances and more complex supply network relationships. In order to defy these new rules of competition, companies face numerous problems in developing accurate product demand forecasts. Especially, in business-to-consumer environments these customer demands have to be estimated in many cases. This thesis takes a look at different ways to estimate the future customer demand using different methods. The methods are on the one hand driven by historical demand data which is used by different models to develop the expected future demand. On the other hand qualitative methods are used which are mainly based on the experience and the knowledge of experts or groups of experts. The historical data driven methods used in this thesis can be divided into two different groups. The first group contains four smoothing methods and the second group three methods which consider the distribution of the historical demands. Next to the comparison of the different methods the thesis also evaluates the performance of a forecasting process. Usually, forecasting processes consist of different steps. These steps are analyzed and it is checked if they are adding value to the examined forecasting process or not. The value adding steps are revealed using a forecasting value added analysis. Another aspect examined is which general characteristics of historical demand patterns have a connection to forecast accuracy archived in forecasting processes. The results obtained from this analysis show that out of the characteristics examined the two with the highest impact on forecast accuracy are the volatility of the historical demand and the appearance of periods without any demand. The analysis carried out on the accurateness of the different methods suggests that the best results can be obtained when the smoothing methods are combined with qualitative methods. The forecasting value added analysis suggests that the methods which consider the historical distribution of the demand do not add a lot of value to the examined forecasting process. As a conclusion, it can be said that forecasts made are always wrong, but in most cases how wrong they are can be limited by combining results generated by statistical models and expert opinions.