Executive Summary

Anticipating
Tax Avoidance
Before It Happens

We aim to anticipate tax avoidance strategies by employing cutting-edge computational tools, so that for every new provision attempting to affect the tax system — such as tax reforms, legislative proposals, candidate platforms — we leverage computational power to identify potential loopholes arising from the interactions with existing laws, as well as to simulate strategic behaviors that could exploit them.

Scroll
Preliminary Results

Research
Outputs

◈ Paper JURISIN 2026 · Gunma, Japan · 7 June

A Computational Framework to Uncover Gray Areas in Tax Legislation

We trained a model to find Gray Areas in tax law — provisions that could potentially give rise to multiple interpretations, including ambiguities, open-ended lists, interpretable terminology and discretional powers. We used NLP techniques to represent tax laws computationally and conducted a classification task on a sample of Colombian Free Trade Zones legislation.

The model achieved an accuracy of approximately 80% in identifying gray areas.

jurisinformaticscenter.github.io/jurisin2026
Paper cover — Gray Areas in Tax Legislation
Confusion matrices — MEL model results
◈ Paper GECCO 2026 · San José, Costa Rica · July 13–17

Evolutionary Computation for Tax-Minimizing Strategies in Special Economic Zones

We built a mathematical model of the economic environment of production in Special Economic Zones (SEZs), incorporating tax differentials and price-setting between related companies. We used genetic algorithms — a class of optimization methods inspired by natural evolution — to simulate sequences of transactions within this environment.

The model was able to identify strategies that minimized tax liabilities, where production decisions were driven solely by the objective of reducing tax payments.

gecco-2026.sigevo.org
Paper cover — Evolutionary Computation for Tax-Minimizing Strategies
Algorithm performance results — GECCO 2026

Project
Goals

01
Tax Administration · DIAN

Smarter Audit Allocation

We aim to provide the tax authority DIAN with an operational input to enhance audit efficiency. By delivering a theoretical exploitation model of legal loopholes — one that illuminates how transactions and operations are structured to exploit ambiguities — auditors can move from reactive enforcement toward intelligence-driven resource allocation, focusing efforts precisely where risk is highest.

Direct impact on audit prioritization and cost-effectiveness
02
Deterrence · Compliance

Deterring Avoidance Behavior

By demonstrating that computational tools can systematically anticipate loopholes before they are exploited, we aim to signal to taxpayers that the cost-benefit calculus of avoidance has shifted. When the state can model the full strategy space of a new provision as soon as it is published, the window of opportunity narrows — and the rational incentive to exploit gray areas diminishes.

Behavioral deterrence through credible computational capacity
03
Policy · Legislation

Anticipatory Filter for Policy Makers

We aim to offer legislators and policy designers a pre-promulgation stress test: a computational filter that identifies weak points, exploitable interactions, and potential abuses within a draft norm before it becomes law. This enables more robust legislation from the outset, closing loopholes proactively rather than patching them reactively after revenue losses have materialized.

Upstream policy robustness — legislate with foresight
Work in Progress

We are currently exploring Reinforcement Learning approaches to learn and refine avoidance strategies autonomously, in close collaboration with subject matter experts — pushing the anticipatory power of the framework further.

Initializing…
Research Team

The People
Behind the Work

More
Information

Download Technical Overview PDF · 4 pages · English