Unbridled Losses: Harnessing Machine Learning for Tax Analysis

Abstract

The Third Circuit’s recent decision in Skolnick once again demonstrates the predictive power of artificial intelligence in forecasting the outcome of tax appeals. Previous commentary analyzed the opinion of the Tax Court in that case, which held that the taxpayers’ horse breeding and racing activities were not engaged in for profit. In the earlier analysis, relying on machine learning, we predicted with 81 percent confidence that the Tax Court’s ruling would be sustained on appeal. Our prediction proved accurate when the Third Circuit sustained the judgment of the Tax Court on March 8, 2023.

In this article we revisit Skolnick, section 183, and the regulations that set out the rules for deductibility of business losses and the test for an “activity not engaged in for profit.” We also review the facts that gave rise to the dispute in Skolnick and recap the Tax Court’s reasons for siding with the IRS. We outline which factors the algorithm indicated would be decisive for the Third Circuit, review Blue J’s original prediction, and assess the extent to which the factors identified using machine-learning techniques were the same as those decisive on appeal.

Keywords: machine learning, income tax, hobby losses

JEL Classification: H0