Predicting Economic Substance Cases with Machine Learning

Journal of Tax Practice & Procedure, 2020

7 Pages Posted: 25 Jul 2020 Last revised: 13 Oct 2020

See all articles by Benjamin Alarie

Benjamin Alarie

University of Toronto - Faculty of Law; Vector Institute for Artificial Intelligence

Abdi Aidid

University of Toronto - Faculty of Law; Centre for Ethics

Date Written: May 20, 2020


The economic substance doctrine is intended to prevent taxpayers from engaging in transactions that have no substantive purpose other than to obtain a tax benefit. As one of the judicially-created substance-over-form doctrines, its meaning is almost entirely derived from the series of court decisions articulating its application. Unfortunately, these cases have not always been clear; scholars have long noted that courts have applied the doctrine inconsistently, judges have criticized it for being tantamount to a “smell test” and even derided it as the government’s “trump card.” Recent scholarship suggests that the doctrine’s codification in 2010 in Section 7701(o) of the Internal Revenue Code might have contributed to even more confusion.

For practitioners, this creates a major challenge. How do you advise clients as to whether their transaction has economic substance when the case law is “confusing and conflicting,” particularly when you know that clients expect confident, near-certain advice? Fortunately, advances in computing power offer an opportunity to cut through the morass and more fully understand the law. Alongside colleagues at the University of Toronto and Blue J Legal, we have developed artificial intelligence systems that identify and assess patterns in tax cases. Using techniques in supervised machine learning, we analyze historical case law to surface hidden insights and predict how future courts will respond to new tax situations. Sophisticated tax practitioners can make use of these insights to provide better advice, structure tax-optimal transactions, respond effectively to tax authorities and resolve disputes more efficiently.

In this note, we take a brief look at how machine learning can improve our understanding of the economic substance doctrine. In Part I, we discuss the importance of prediction in tax practice, noting that the emergence of computational technologies has made the age-old desire to predict legal outcomes considerably easier. In Part II, we discuss the economic substance doctrine in detail and identify areas of confusion that cannot be resolved by conventional legal research methods. In Part III, we apply machine learning to two recent Tax Court cases – Cuthbertson (2020) and MCM (2019) – to demonstrate how algorithms can correctly predict the case outcomes and give practitioners opportunities to refine their understanding and ultimately provide better tax advice.

Keywords: economic substance, machine learning, tax avoidance, business purpose, tax

JEL Classification: h2, h26

Suggested Citation

Alarie, Benjamin and Aidid, Abdi, Predicting Economic Substance Cases with Machine Learning (May 20, 2020). Journal of Tax Practice & Procedure, 2020, Available at SSRN:

Benjamin Alarie (Contact Author)

University of Toronto - Faculty of Law ( email )

Jackman Law Building
78 Queen's Park
Toronto, Ontario M5S 2C5
416-946-8205 (Phone)
416-978-7899 (Fax)


Vector Institute for Artificial Intelligence ( email )

Abdi Aidid

University of Toronto - Faculty of Law ( email )

78 Queen's Park
Room J364
Toronto, Ontario M5S 2C5
416-978-4150 (Phone)

Centre for Ethics ( email )

6 Hoskin Avenue
Toronto, Ontario M5S 1H8

Do you have negative results from your research you’d like to share?

Paper statistics

Abstract Views
PlumX Metrics