Classificação de litígios fiscais em empresas brasileiras com a utilização do aprendizado de máquina

Abstract

This research dealt with the relationship between tax litigation and corporate sustainability. There is a gap that deals with whether companies with a strong commitment to society, enjoy tax avoidance practices (licit act), which may expose them to questioning by the tax authorities. Therefore, the main objective was to determine whether companies certified as socially responsible – Corporate Sustainability Index (ISE) – share the same tax litigation as other companies, both listed on the Brazilian Stock Exchange (B3). For that, 3.653 financial statements of 382 companies were collected in the period between 2010 and 2021. Through a training base composed of 519 frequencies of tax litigation collected and classified manually, a machine learning algorithm was implemented to automated classification of litigation. The results show that ISE companies have, on average, twice as many tax litigation registered and provisioned in their financial statements than non-participating companies. There is a tendency in this environment to associate this result with the item transparency of information, a fundamental characteristic required for companies to be certified as socially responsible. Therefore, this research, in addition to contributing to the literature, becomes relevant for disseminating knowledge about tax avoidance practices adopted by B3 companies, and which tax litigation these companies were subject to.