Reworking the Standard Model of Competitive Markets: The Role of Fuzzy Logic and Genetic Algorithms in Modelling Complex Non-Linear Economic System

Some aspects of economic systems (eg, nonlinearity, qualitative variables) are intractable when incorporated into models. The widespread practice of excluding them (or greatly limiting their role) produces deviations of unknown size and form between the resulting models and the reality they purport to represent. To explore this issue, and the extent to which a change in methodology can improve tractability, a combination of two techniques, fuzzy logic and genetic algorithms, was applied to the problem of how the sellers in a freely competitive market, if initially trading at different prices, can find their way to supply/demand equilibrium. A multi-agent model was used to simulate the evolution of autonomously- learnt rule-governed behaviour, (i), under perfect competition, and (ii), in a more commercially realistic environment. During the learning process, markets may lack a true equilibrium price, and therefore sellers in such a model cannot be price-takers in the conventional sense; instead, it was stipulated that they would set an asking price, buyers would shop around for cheap supply, and the sellers would revise their pricing policy according to its profitability. Each firm's pricing policy was embedded in a fuzzy ruleset; the rulesets were improved over time by successive passes of the genetic algorithm, using profit level as a measure of Darwinian fitness. The simulated evolution was repeated over a random sample of 10 markets. Under perfect competition, sellers' asking prices converged onto the theoretical equilibrium price. This performance was maintained when either uncertainty in demand or a more commercially realistic set of dynamics was introduced. However, when both these features were introduced simultaneously, different, substantially lower equilibrium prices were reached. In both cases, autonomous learning by the sellers suppressed the instability that might have been expected to result from the introduction of a number of nonlinearities. Other possible applications of the methodology are discussed, along with some of its implications.