Data does not always trump intuition - Evaluating performance of statistical forecasting methods in ecommerce
Zagaria, Delphine
Promotor(s) : Ittoo, Ashwin
Date of defense : 20-Jun-2016 • Permalink : http://hdl.handle.net/2268.2/1498
Details
Title : | Data does not always trump intuition - Evaluating performance of statistical forecasting methods in ecommerce |
Author : | Zagaria, Delphine |
Date of defense : | 20-Jun-2016 |
Advisor(s) : | Ittoo, Ashwin |
Committee's member(s) : | Ghilissen, Michael
Deneye, Pierre Mottroule, Michaël |
Language : | English |
Number of pages : | 73 |
Keywords : | [en] demand forecasting [en] predictive methods [en] Double-Seasonality [en] Holt-Winters [en] ARIMA [en] Regression [en] eCommerce [en] logistics |
Discipline(s) : | Business & economic sciences > Quantitative methods in economics & management |
Target public : | Researchers Professionals of domain Student |
Institution(s) : | Université de Liège, Liège, Belgique |
Degree: | Master en ingénieur de gestion, à finalité spécialisée en Supply Chain Management and Business Analytics |
Faculty: | Master thesis of the HEC-Ecole de gestion de l'Université de Liège |
Abstract
[en] As a global end-to-end eCommerce solutions provider, PFSweb deals everyday with thousands of shipping orders to all over Europe. Warehouse management is indeed an essential function of the organization. eCommerce is moreover a new market that is entering everyone’s day-to-day life. Logistics organizations therefore have to serve more and more customers, which leads to distribution issues like out-of-stock items, manpower shortage, and financial impacts, due to a lack of a demand forecast.
In this thesis, we evaluate the performance of several state-of-the-art quantitative methods in demand forecasting, in opposition to methods based on judgment, in order to optimize and formalize forecasting in eCommerce. A systematic framework is therefore considered, which could:
• Document the forecasting process for sharing purposes among PFSweb’s agents.
• Allow the use of several statistical methods.
• Facilitate traceability.
The development of this structure is based on pertinent scientific readings and a rigorous methodology. The scientific literature brought the necessary knowledge and understanding of the extent of the issue: the importance of forecasting in Business, the challenges in eCommerce forecasting, and the state-of-the-art quantitative time-series methods. On the other hand, the methodology gives an overview of PFSweb’s forecasting approach, before the application of a series of steps that compose the proposed framework.
Based on our findings, we show that while these methods work reasonably well, their use in real-life situations characterized by a fluctuating and double-seasonal demand, do not allow for optimal predictions. The performance of these methods particularly deteriorates in the absence of human domain-knowledge, intuition and experience. We thus recommend that these methods should not be applied on their own; rather, they should be used to complement the skills and expertise of human business analyst.
File(s)
Document(s)
Description:
Size: 1.94 MB
Format: Adobe PDF
Cite this master thesis
The University of Liège does not guarantee the scientific quality of these students' works or the accuracy of all the information they contain.