Predicting Volatile Consumer Markets using Multi-Agent Methods: Theory and Validation

Predicting Volatile Consumer Markets using Multi-Agent Methods: Theory and Validation

Abhijit Sengupta (Unilever Research and Development, UK) and Stephen E. Glavin (Unilever Research and Development, UK)
DOI: 10.4018/978-1-4666-2011-7.ch016


A behavioral model incorporating utility-based rational choice enhanced with psychological drivers is presented to study a consumer goods market, characterized by repeat purchase incidences by households. The psychological drivers incorporate purchase strategies of loyalty and change-of-pace, which affect the choice set of consumer agents in an agent-based simulation environment. Agent specific memories of past purchases drive these strategies, while attribute specific preferences and prices drive the utility-based choice function. Transactions data from a category in a supermarket is used to initialize, calibrate, and test the accuracy of predictions of the model. Results indicate that prediction accuracy at both macro and micro levels can be significantly improved with the incorporation of purchase strategies. Moreover, increasing the memory length beyond a certain limit does not improve predictions in the model, indicating that consumer memory of past shopping instances is finite and low and recent purchase history is more relevant to current decision making than the distant past.
Chapter Preview


Sengupta and Glavin (2010) focused on developing a theoretical model of behavior, which took into account the heterogeneity in tastes within the consumer population and illustrated the link between this heterogeneity and the resultant volatility in overall market shares of brands and specific product characteristics. In order to carry out this analysis, the authors developed a rigorous methodology, which focused on out of sample predictions of both macro level market share movements and micro level household level choices. Results showed a reasonably high degree of accuracy with which market share movements as well as choices made by individual households could be predicted out of sample. These results were compared with a benchmark model, where agent level heterogeneity was ignored, and it showed that the former far outscored the latter at both the macro and micro levels.

Complete Chapter List

Search this Book: