Researchers from Lancaster Centre for Forecasting present at INFORMS conference
Published 6 August 2010
Researchers from the Lancaster Centre for Forecasting recently presented some of their work at the INFORMS annual Marketing Science Conference in Cologne (Germany 17th-19th of June 2010). Two papers were presented at the conference.
Retail Promotions
Tao Huang, Professor Fildes and Dr Didier Soopramanien, presented a paper on “Structural breaks and the value of Promotional Information in Forecasting retailer sales”. In this paper they initially demonstrate that information on marketing and promotional activities significantly improves retailer forecasts of SKUs compared to the benchmark models (such as exponential smoothing) which are typically used by retailers. A more specific contribution of the research focuses on structural breaks in the time series sales data of SKU and how this can potentially affect the forecasts. A structural break in the data series of sales of an SKU can be defined as fundamental change in the time series characteristic of the data. If, for example, we consider any particular brand of bottled juice, the introduction of a strong competitor brand in that product category may fundamentally affect the sales of other existing brands. An interesting and novel insight from the study is that this research demonstrates it is important to test for the presence of structural breaks in the data series and that, if they are present, these breaks can significantly change the way in which we attribute the effect of promotions on sales. The presentation can be accessed through http://forecasters.org/isf/past-isfs.html
Car Demand in a dynamic market
Lixian Qian and Dr Didier Soopramanien presented a paper on “Forecasting Car Market Demand in China under Dynamic Consumer Preferences”. Forecasting car demand in China is particularly challenging in terms of data availability given the short history of car ownership in that market. Given the state of that market, the reasonable assumption is that most potential buyers are new “adopters of cars”. In this research therefore, the researchers test whether diffusion models that have been applied to model new product adoption can be used to forecast car demand in China. The main diffusion models (Bass, Gompertz and Logistic) are employed and are compared to benchmark models and a linear econometric model of car sales. To evaluate how the different models fare, a rolling forecast methodology is used. The extended specification of the logistic diffusion model, which includes both trend and GDP, outperforms the other diffusion models and the benchmark models. The results also show that the Logistic model is better than the linear econometric model.
If you would like to know about both studies please contact either Professor Fildes; R.Fildes@lancaster.ac.uk or Dr Didier Soopramanien email: d.soopramanien@lancs.ac.uk
