Partially observable Markov methods in an agent-based simulation: a tax evasion case study

Defining and testing a policy on a socioeconomic system is one of the main problems addressed by agent-based modelling. While research continues to be conducted to come up with hybrid frameworks that tackle the complexity of different problems, no model explicitly integrates computational replications of multi-agent systems, particularly in dealing with partially observable situations. We show in our work how a Markov based reinforced learning and partially observable computations in the behaviour of a taxpayer agent can contribute to refining the analysis of an audit policy.