In next month’s Expert Systems with Applications, Nikolaos Goumagias (Northumbria University), Dimitrios Hristu-Varsakelis (University of Macedonia), and Yannis Assael (University of Oxford) have “Using Deep Q-Learning to Understand the Tax Evasion Behavior of Risk-Averse Firms.”

Deep Q(uality)-Learning is a form of artificial intelligence, where the AI agent lacks a model of the environment they are in, but uses machine learning to discover a optimal policy from trying to collect rewards from different actions. Or so I gather from this site showing a Deep-Q agent learn to play Doom. What they do in the paper is take a sample of Greek firms and attempt to figure out their degree of relative risk aversion in the Greek system. This is not a program designed to detect tax evasion, but rather evaluate tax policy in terms of tax revenue and evasion when confronted with new tax policies. You can think of this also as differing from similar normative public finance policy studies where utility and profit functions of agents are first defined and parameterized by the researcher. Here is the abstract from the Goumagias et al. paper:

Designing tax policies that are effective in curbing tax evasion and maximize state revenues requires a rigorous understanding of taxpayer behavior. This work explores the problem of determining the strategy a self-interested, risk-averse tax entity is expected to follow, as it “navigates” – in the context of a Markov Decision Process – a government-controlled tax environment that includes random audits, penalties and occasional tax amnesties. Although simplified versions of this problem have been previously explored, the mere assumption of risk-aversion (as opposed to risk-neutrality) raises the complexity of finding the optimal policy well beyond the reach of analytical techniques. Here, we obtain approximate solutions via a combination of Q-learning and recent advances in Deep Reinforcement Learning. By doing so, we (i) determine the tax evasion behavior expected of the taxpayer entity, (ii) calculate the degree of risk aversion of the “average” entity given empirical estimates of tax evasion, and (iii) evaluate sample tax policies, in terms of expected revenues. Our model can be useful as a testbed for “in-vitro” testing of tax policies, while our results lead to various policy recommendations.

I was also interested to discover from this that tax evasion is a popular applied subject for AI work, which makes sense considering it is a natural candidate for policy relevance. Here are some pieces I turned up, by title:

–Data Mining Application Issues in Fraudulent Tax Declaration Detection
–Tactical Exploration of Tax Compliance Decisions in Multi-Agent Based Simulation
–Tax non-compliance detection using co-evolution of tax evasion risk and audit likelihood
–Tax Evasion Behavior using Finite Automata: Experiments in Chili and Italy
–Detecting Corporate Tax Evasion Using a Hybrid Intelligent System: A Case Study of Iran
–The Computational Rules Extractor in the Detection of Tax Evasion
–Detecting Tax Evasion: A co-evolutionary approach