Tax enforcement in an agent-based model with endogenous audits

We generalize the classic Allingham and Sandmo’s model of tax evasion considering heterogeneous agents with different degrees of tax morale and matchable, as opposed to non-matchable, income. The Tax Agency evolves its control scheme, maximizing the revenues from fines, and takes into account some minimal information on the taxpayers. We compare different audit policies and find that the most effective scheme remarkably depends on the way agents update the subjective probability of being audited, on the distribution of matchable income in the population as well as on the level of tax morale. Hence, different features of societies and taxpayers’ behaviors not only affect the compliance rate, as expected, but require the Tax Agency to alter its audit policy in a context-dependent way. In particular, high revenues are obtained performing random audits when agents think they are directed towards peculiar individuals and, conversely, should be biased towards low declarations when taxpayers believe audits are nonspecific or random.