- You are viewing archived content on our old website. To see events added from March 2014 onwards, please view our new events pages.
Structural Breaks and the Value of Promotional Information in Econometric Models for Forecasting Retailer Sales
Wednesday 3 March 2010, 13:00
LT11, Management School
Abstract: Sales forecasting at the UPC level is important for the retailers in inventory planning. In practice, many retailers produce the forecasts using simple models with subjective adjustment for promotional events. The econometric models, which could be an effective way of incorporating promotional information, are not widely applied. This research examines the performance of the econometric models in forecasting the product sales at the UPC level, asking whether this lack of application arises from poor performance compared to simple benchmarks. Standard models assume constant structure which may subsequently expose the models to structural breaks and then forecast failure. To tackle the problem, we modify the econometric models with three approaches: impulse saturation, intercept correction and estimation window combining. Categories are considered as a basis for generalising the empirical results. 29 categories were segmented using conceptual mapping technique and experiments were conducted for those which are distinguishable and representative. The results show that the econometric models outperform the benchmark (simple) models for most of the categories and their forecasting performance can be further improved by the modification approaches, especially for the non-promoted forecast period. The explanation may lie in the fact that the econometric models take concerns of the competitive promotional information. However, for one distinct category- the “Front-end-candies”, the benchmark models proved superior. For the promoted period, the unmodified econometric models generally yield the best performance. One implication of our research is to use the unmodified econometric models and the modified econometric models to forecast the promoted period and the non-promoted period separately.