糖心原创

Title: Political Competition and Tax Expenditures: A Machine Learning Approach to 121 Years of U.S. Laws

Abstract: Tax expenditures (TEs)—exemptions, deductions, and credits—are a major source of tax complexity. What drives the enactment of TEs? Does political competition play a role? I examine these questions in the context of the U.S using a novel dataset of over 3.1 billion tokens of legislative text built from the full historical archive of State Session Laws spanning 121 years (1900-2020). I use machine learning to identify the frequency of TE provisions and readability metrics to capture their complexity. I find that political competition affects the frequency of TEs. However, the effect depends on which party controls the state legislatures. Republican-led legislatures systematically enact more TE provisions when elections are close, making the tax code more complex over time. The mechanism is obfuscation—politicians appear to tax high-income earners heavily while offsetting their burden through targeted TEs—especially when redistribution is politically costly and campaign contributions are high. These results challenge the conventional view that political competition uniformly improves policy outcomes, showing instead that its effects depend critically on partisan incentives.