Local Environment Analysis and Rules Inferring Procedure in an Agent-Based Model – Applications in Economics

The use of agent-based modeling in economics is a step forward enabling a more realistic description of the complex interactions and behaviors occurring in the economic environment. Although it offers increased realism, especially in describing how local characteristics generate global patterns, it suffers from a simplistic approach to modeling local behaviors and rules. From this perspective the paper suggests possible solutions in two directions. First, the paper uses neural networks as an instrument for the agents to scan their local environment and infer possible behaviors. Second, the paper defines and applies an algorithm enabling the agents to understand a subset of rules that are not defined at the beginning of the application. The goal is to see how it is possible to generate new rules with structure and semantics. This would constitute “real” learning, namely defining new rules but not only quantitative variations of the initial rules.