Computer Aided Tax Evasion Policy Analysis: Directed Search using Autonomous Agents

摘要

Abusive tax shelters implemented through partnerships and S corporations have become increasingly popular amongst tax planners, helping high-income taxpayers to underreport an estimated $91 billion of income annually in the US alone. The most challenging problems for tax collection agencies in this respect are a) the recent upswing in large, tiered partnership structures and b) the evolving nature of tax evasion schemes in response to auditing policy. By representing tax evasion schemes as sequences of financial transactions, we are able to conduct a directed combinatoric search that can find effective abusive tax shelters, given an initial ecosystem of taxable entities and their respective portfolios. Assigning auditing likelihoods to certain types of transactions allows us to consider policies that would result in increased compliance. We accomplish this by considering each tax plan and auditing policy as individual agents.