Weekly SSRN Tax Article Review And Roundup: Layser Reviews Predictive Analytics And The Tax Code By Soled & Thomas

This week, Michelle Layser (San Diego) reviews Jay A. Soled (Rutgers; Google Scholar) and Kathleen DeLaney Thomas (UNC; Google Scholar), Predictive Analytics and the Tax Code, 51 Fla. St. U. L. Rev. __ (2023).

I am obsessed with ChatGPT. If you haven’t seen it yet, I suggest that you finish grading and then go check it out (in that order). It is at once fascinating and terrifying, and it leaves little doubt that the artificial intelligence (AI) tools of the future will dramatically impact most aspects of the legal profession. And the future may not be so far off. In a forthcoming article, Professors Jay A. Soled and Kathleen DeLaney Thomas argue that today’s predictive analytics tools are already capable of fundamentally changing the application of the tax code’s civil tax penalty regime.

To demonstrate how, the authors begin with a review of current theory about taxpayer compliance and the civil tax penalty regime. 

On the one hand, taxpayer compliance reflects a cost-benefit analysis of the severity of tax penalties, the chances of detection, and the economic benefits associate with noncompliance. On the other hand, even when taxpayers’ noncompliance is detected, the IRS often has discretion over whether to impose a penalty. Consider taxpayers whose deductions are subsequently disallowed in court. Such taxpayers will always owe the additional taxes due, but they will not necessarily be subject to a penalty. Whether a taxpayer is subject to a penalty frequently depends on how likely they were to prevail in court (even if the taxpayer actually loses!).

For example, a taxpayer may be subject to a penalty for negligent reporting. The Code defines negligence as “any failure to make a reasonable attempt to comply” with the tax law. The term “reasonable” is vague, but it has been interpreted as a “reporting position that has at least a 20 percent or greater chance of prevailing in a judicial action.” In other cases, a taxpayer may be subject to a penalty for a substantial understatement of taxes due. However, the taxpayer can often avoid the penalty by showing that they had “substantial authority” for their reporting position. The substantial authority standard has been interpreted as: “a 40 percent or greater probability that a court would uphold the taxpayer’s position if the IRS were to challenge it.” Alternatively, the taxpayer can show they had a “reasonable basis” for the position—a 20 percent or greater chance of prevailing in court—and adequately disclosed it.

If a taxpayer can show that they had a high enough probability of winning, then they usually will not be subject to a penalty. Of course, it is not always easy for taxpayers or the IRS to determine their chance of prevailing in court. Enter AI. The authors argue that predictive analytics can help both taxpayers and the IRS by producing probabilistic predictions of future events. They explain that today’s machine learning programs “can analyze big data and look for patterns quickly, making connections that humans may miss.” In the context of tax law, “a machine learning program might find correlations between particular facts of cases and outcomes of those cases” and, by analyzing those patterns, “the computer program could then predict the outcome of future cases.”

For example, a taxpayer and the IRS could use predictive analytics to determine whether the taxpayer’s chance of succeeding on the merits met the 20 percent threshold needed to demonstrate they had a “reasonable basis” for a position. Not only could the taxpayer avoid the penalty, but the use of predictive analytics would “obviate the need for the taxpayer and the IRS to expend time and resources litigating this issue.” This potential is more than hypothetical. The authors provide several examples in which predictive analytics tools have been used to estimate the chances of prevailing in cases related to the deductibility of corporate management fees, the determination of workers’ employment status, and the existence of a trade or business. They argue that the IRS can leverage tools like these to save time and litigation costs associated with discretionary penalties, and taxpayers can use them to determine what positions to take on their tax returns.

The authors conclude that predictive analytics would result in “enormous efficiency savings” and “enhanced tax compliance, leading to more revenue collected, because taxpayers should be more likely to avoid taking aggressive tax positions for which they cannot demonstrate sufficient legal authority.” They may be right, but I would be curious to know more about how taxpayers’ risk tolerance affects the positions they take on their returns. It is clear that people over-estimate some risks and under-estimate others, and predictive analytics could help take some of the emotion out of the risk analysis. It’s less clear which direction that pushes, from a tax compliance and revenue collections standpoint.

Some taxpayers may be less likely to take aggressive tax positions when they see their risk of receiving a penalty is higher than they thought. But more conservative taxpayers may be surprised to learn that they are unlikely to suffer a penalty, even if does turn out their position was wrong.  In the latter case, a taxpayer that would have erred on the side of reporting additional income may take a more aggressive position instead. In other words, predictive analytics would probably increase voluntary compliance among some taxpayers, but it may reduce compliance among others—and I would love to hear more about the authors’ thoughts about those trade-offs.

This is a fun and timely article that makes important contributions to the tax compliance and enforcement literature. I recommend it to anyone interested in tax administration, compliance, or technology and taxation.

Here’s the rest of this week’s roundup:

• Fabian Barth (Bournemouth University), Evolving Law in AI’s Hands? – Preliminary Experiments, Thoughts and Observations on the Basis of Chat GPT (April 27, 2023).
• Alexandra Braun (Edinburgh), Private Purpose Trusts: Good for Scotland? (May 10, 2023).
• Wei Cui (Univ. British Columbia), The Mirage of Mobile Capital (May 7, 2023).
• Samay Jain (Independent) & Vaanya Mathur (Independent), The Intersection of Angel Tax, FEMA, and Income Tax: Navigating the Regulatory Landscape for Indian Startups (April 25, 2023).
• Yaoting Lei (Nanchang Univ.), Hong Liu (Washington Univ. St. Louis) & Jing Xu (Renmin Univ. China), Inflation and Tax Timing (May 8, 2023).
• Joseph Liberman (AQR Capital Management, LLC), et al., Beyond Direct Indexing: Dynamic Direct Long-Short Investing (May 3, 2023).
• Doron Narotzki (Akron) & Vered Kuperberg (Independent), The Potential Federal Income Tax Liability of Foreign Digital Nomads, 179 Tax Notes Federal (April 3, 2023).
• Ganesh Rajgopalan (Independent), Tax Treaties and the MLI - Interpretation and Interplay, 27th International & Tax Conference of the Bombay Chartered Accountants' Society (BCAS), Mumbai, (April 6, 2023).
• Natalya Shnitser (Boston College), Retirement Plan Reforms in the Absence of a Retirement Policy, in The Cambridge Handbook of Investor Protection 117–133 (Arthur B. Laby ed., 2022).
• Palma Joy Strand (Creighton) & Nicholas A. Mirkay (Hawaii), Creating a Pro-Tax Story for Racial Equity, Tax Notes State (April 24, 2023), University of Hawai’i Richardson School of Law Research Paper.
• Samuel Tuyisenge (Independent), Advance Rulings on Classification, Origin and Valuation (May 10, 2023).