Beyond Boilerplate: Lessons from the Sec’s Regulation of Standardized Disclosure

Abstract: In this paper I analyze how the use of boilerplate language in securities disclosure has evolved in light of the SEC’s attempts at regulating it, and argue that the of use algorithmic language processing and machine learning methods that drive the country’s most innovative companies could also vastly improve the SEC’s disclosure regime. In particular, topic modeling can be used to understand the revealed preferences of market participants with respect to certain kinds of boilerplate disclosure, and enable to SEC to offer a menu of disclosure options for issuing companies. The analysis in this paper suggests that the SEC has had limited success regulating boilerplate because it treats such language in a one-size-fits-all manner. However, as the analysis shows, different boilerplate language performs different functions: some truly is meaningless, and most likely the product of over-conservative drafting, other language is used for efficiency, and still other language is used by issuers to be strategically vague. Natural language processing can be leveraged to identify which boilerplate is useful and which is not, to allow for more nuanced, targeted regulation. In particular, the empirical analysis in the paper does three things: First, it documents and analyzes the mixed results of the SEC’s past efforts to control boilerplate in IPO disclosure. Second, it provides evidence that the use of boilerplate impacts investors, and draws conclusions about the usefulness of various kinds of boilerplate language. Third, it demonstrates how a more targeted approach to boilerplate could make for more successful regulation, as well as more comprehensible disclosure for investors.