Computational intelligence for a mechanistic understanding of mass consumer markets
Klaus Jaffe, Universidad Simón Bolívar, Caracas, Venezuela, kjaffe@usb.ve
Abstract:
Mass consumer markets might be viewed using tools developed for the analysis of complex systems such as agent based simulations. Here we used SEMSTAT, which simulates a given number of choosy semi-rational consumer agents and of diverse shopping sites for a given series of products. The simulations closely match the consumer behavior of real populations, as measured by comparing actual sales data of the products under considerations with those predicted by the model. A real life test with markets for beers, soft drinks and cigarettes in two Latin American markets, showed fit with real data with maximal errors of 0.5 % for each product line and product category. This exercise provides new hope for an eventual quantitative detailed prediction of market dynamics.
Introduction:
New
forms at looking into complex economic and financial systems based on
computational tools, such as Sociodynamics, inspired by thermodynamics,
behavioral economics, evolutionary sciences, behavioral ecology, evolutionary
economy, complex systems, and many other disciplines, aims at providing sound
bases for a deeper understanding of the interacting web of biological, social
and economic behavior (1). Sociodynamics aims to be a quantitative experimental
interdisciplinary science for social phenomena, bridging the gap between the mind
of a natural scientist and a social one.
One computational tool for analyzing social dynamics is Sociodynamica,
an agent based model, serving as a metaphor for a society of agents living in a
free competitive market, where each individual agent suffers its own "umwelt"
and interacts with other agents forming a social web and producing a cultural
gestalt. The model allowed exploring the underlying dynamics of the creation of
wealth (2,4), the effect of division of labor on economic growth(2), the working
of cooperation (3), the emergence of social organization through synergisms (1)
and the working of societies using different financial instruments.
Mass consumer markets might be viewed using tools developed for the analysis of complex systems. One such tool, agent based simulations, seems especially appealing due to its adaptiveness and flexibility in addressing general and particular concerns. At the same time, theoretical models of society require means to tests their reliability and need strict and unambiguous protocols for their validation, in order to acquire wider acceptance. The agent based computer program SEMSTAT was developed in order to satisfy simultaneously the two aims presented above.
Methods:
SEMSTAT simulates a given number of consumer agents and of shopping sites for a given series of products. The consumer buys products according to preferences that include availability of the product (distribution), accessibility of the product (price), awareness of the product (promotion), and quality of the product (general quality, including all factors not included in distribution, promotion and price). The model simulates a continuous two-dimensional toroidal world through which different types of agents wandered with Brownian motion. Agents learn and adapt their shopping preferences according to experience.
Results:
The simulation model produces populations of agents, which closely match the consumer behavior of real populations, as measured by comparing actual sales data of the products under considerations with those predicted by the model. For example, the model has been tested in real life scenarios of markets for beers, soft drinks and cigarettes in two Latin American markets, showing adjustments with real data with maximal errors of 0.5 % for each product line and product category.
The simulations, in order to achieve accurate predictions of sales, had to assume certain psychological features of the consumers. These features were fidelity to a brand, effort spend in searching the preferred brand, and susceptibility to prices. These features could be checked independently in two populations giving data that agreed surprisingly well with those predicted by the computer model.
Conclusions:
The results show that the simulation program may serve as a tool for:
1- Visualize the dynamic working of a given market,
2- Detect irregularities in local markets or specific product lines,
3- Detect mismatches and errors in real data,
4- Test specific probable outcomes for plans of action in sales, promotion and distribution policies.
Generalizing, we might affirm with a large confidence that computational intelligence, specifically agent based simulation models of real economies, allows for detailed quantitative predictions in specific consumer markets.
References:
http://atta.labb.usb.ve/Klaus/MonteCarlo%20Explo%20of%20Wealth.htm
3- Jaffe, K. Altruistic punishment or decentralized social investment?
Acta Biotheoretica, 52(3): 155-172, 2004
http://atta.labb.usb.ve/Klaus/altrupun.pdf
4- Jaffe, K. From Sociodynamics to Synergistics: Understanding the Wealth of Nations using Simulations. Proceedings of ESSA 2004, Valladolid, Spain 2004.
http://atta.labb.usb.ve/Klaus/Wealth%20of%20Nations%20Simulations.pdf