A large-scale agent-based model of taxpayer reporting compliance

Abstract

This paper describes the development of the Individual Reporting Compliance Model (IRCM), an agent-based model for simulating tax reporting compliance in a community of 85,000 U.S. taxpayers. Design features include detailed tax return characteristics, taxpayer learning, social networks, and tax agency enforcement measures. The taxpayer's compliance reporting decision is modeled as a partially observable Markov decision process that takes into account taxpayers' heterogeneous risk profiles and non-stationary beliefs about the expected costs associated with alternative reporting strategies. In order to comply with legal requirements prohibiting the disclosure of taxpayer information, artificial taxpayers are created using data from the Statistics of Income (SOI) Public Use File (PUF). Misreported amounts for major income and offset items are imputed to tax returns are based on examination results from random taxpayer audits. A hypothetical case study illustrates how IRCM can be used to compare alternative taxpayer audit selection strategies.