A comparative analysis of demand forecasting models: A case study of a Vietnam e-commerce company

No Thumbnail Available
Journal Title
Journal ISSN
Volume Title
School of Business | Master's thesis
Degree programme
Business analytics
Amid the enormous potential of e-commerce sector in Vietnam along with the increasing uncertainty of the economy, an effective demand forecasting undoubtedly plays a significant role in the success of businesses operating in such market dynamics. This research implements a comprehensive comparative analysis of demand forecasting models, tailored to address the unique characteristics and underlying patterns of the demand in Vietnam e-commerce industry. Adhere to a scientific research structure, the study evaluates the performance of three demand forecasting models, namely SARIMAX, Facebook Prophet, and LSTM in predicting online retail customer demand. Exogenous variables such as promotional activities and seasonal fluctuations are also incorporated into these models for better performance. Leveraging historical sales data acquired from a prominent e-commerce enterprise based in Vietnam, the study demonstrates that SARIMAX generates the lowest weekly Mean Absolute Percentage Error (MAPE) of 15% compared to Prophet and LSTM. Despite a complex structure designed to capture non-linear and long-term dependencies in time series, LSTM produces the highest forecast errors, indicating model complexity does not necessarily improve forecast performance. These results lead to the conclusion that SARIMAX is a promising model to be implemented in practice for the company in this case study. Furthermore, the research validates the impact of promotional activities and pricing on demand and their critical role in predicting future demand accurately. Given its solid findings, the study aims at generating valuable insights for e-commerce enterprises in Vietnam, contributing to the existing knowledge related to demand forecasting in e-commerce. The research consolidates the applicability of SARIMAX as proved by previous research while analysing the strengths and limitations of the studied forecasting models. With such findings, businesses can make decisions on model selection, development and performance evaluation. With a well-establish demand forecasting process, companies can further optimize their supply planning, inventory management, and overall operational efficiency. Further research can leverage this study to develop more refined forecasting models for this specific market.
Thesis advisor
Kim, Seongtae
demand forecasting, time series, neural network, e-commerce
Other note