Study Group report
This is the final report on the modelling of consumer behaviour, brought to ESGI49 by Unilever Corporate Research.
Pablo Casas (Universidad Politécnica de Cataluña)
Jon Chapman (University of Oxford)
Robert Hunt (University of Cambridge)
Gregory Kozyreff (University of Oxford)
Andrew Lacey (Heriot-Watt University)
Emeline Larrieu (University of Cambridge)
Robert Leese (Smith Institute)
Tiina Roose (University of Oxford)
David Schley (University of Southampton)
Lydie Staron (University of Cambridge)
The Study Group looked at a variety of modelling frameworks for investigating market behaviour in fast-moving consumer goods. They included:
• Deterministic models of consumer loyalty and psychology. Key psychological features such as ‘minimising anticipated regret’ are included in the models. It is seen how the introduction of a new product can explain behaviour such as the ‘decoy effect’.
• Models that add sociological aspects, for example the positive influence of friends already buying particular products, and show how they can lead to the phenomenon of ‘lock-in’ if the social interactions are sufficiently strong.
• A probabilistic approach, based on Markov chains, to the study of both psychology and sociology. The qualitative conclusions of the earlier deterministic models are confirmed, namely that new products can give rise to a decoy effect, and social interactions can lead to lock-in.
• A framework for studying social interactions, in which consumers are explicitly given individual psychological characteristics and so behave in intrinsically different ways. In this model, the probabilities of individuals buying particular products change over time, under the influence of other members of the population. Simulations, supported by analysis, again exhibit lock-in for sufficiently strong interactions.
• Models of the decision process of individual consumers, as a sequence of pairwise comparisons of different products. It is found that a twolevel decision tree can lead to a decoy effect. When loyalty is strong, the decoy effect persists, even though the decoy product gains almost zero market share. In contrast, the standard Logit model for consumer choice does not exhibit the decoy effect. Preliminary network simulations using a two-level decision process exhibit lock-in, which can be prevented by advertising.
• Network effects at the propagation of consumer behaviour and the way in which new products gain market share. Simulations show how the final market shares depend heavily on the position of the new product relative to the existing ones in quality space.
The report also gathers together some further ideas for modelling frameworks. In a particle-dynamics model, consumers could be modelled as particles moving in product space (corresponding to changes in their preferences), under the influence of ‘potentials’ that are due to the available products. Again, a decoy effect seems possible. Another possibility is to model a continuum of products, so that market shares are replaced by a density in product space, which evolves over time.