What Companies Don’t Say (But Do Write)
Corporate taxes often seem like an obscure and technical topic, far removed from the concerns of everyday life. Yet, they affect nearly everything: how much governments can spend on schools or infrastructure, how shareholders value companies and even whether a business is playing fair in the economic game. Taxes represent one of the most substantial costs facing companies today, with important implications for their business operations. The problem? Understanding a company’s tax behavior is surprisingly difficult, even for professionals who do it every day.
One reason is that companies are not exactly eager to “spell things out”. Talking too much about their tax strategies can draw unwanted attention from competitors and tax authorities alike. As a result, what is publicly disclosed about taxes in corporate reports tend to be cryptic, usually highly standardized and buried in dense language. But what if, hidden in all that text, companies are telling us more that we think? In a study forthcoming in the Review of Accounting Studies, together with Olga Bogachek and Antonio De Vito, we examine whether recent advances in natural language processing and machine learning can help corporate stakeholders better forecast companies’ tax outcomes. We looked at 14 years of US corporate filings and used topic modeling to “read between the lines” of thousands of annual reports and to quantify what companies were saying in those reports. Such measures of textual content were then added to machine learning models developed to predict tax outcomes like companies’ effective tax rates, cash taxes paid and amounts paid as settlements to revenue authorities.
The results were striking. Even when companies don’t say much directly about taxes, the way they talk about other things — internal controls, corporate structure, mergers or regulatory risks — can reveal a lot about their likely tax outcomes. In fact, when we used these indirect clues to forecast real-world tax results, we were able to cut prediction errors by more than half.
Why does this matter?
First, for investors and lenders, it means better information for evaluating financial risks. For policymakers and tax authorities, it offers a new lens to spot aggressive or unsustainable tax practices before they make headlines. And for society at large, it hints at a path toward greater accountability: if stakeholders can extract meaningful insights from what companies do say, even if indirectly, it could pressure firms to be more transparent and responsible in their tax behavior. Second, today’s corporate reports are long, complex and often overwhelming. That is partly why many readers skip over them. But with the right tools, we can turn this ocean of words into actionable information. Our work shows that artificial intelligence and machine learning are not just buzzwords: they can help make sense of complexity and bring clarity to topics that affect us all.
In a time of growing concern over corporate fairness and fiscal responsibility, understanding how companies manage their taxes is no longer just a job for specialists. It’s a matter of public interest. And if companies won’t always tell us directly, we now have better ways to listen carefully to what they’re saying between the lines.
