Executive Summary
This study tests the Variance Compression Thesis (Proposition 1 of The Cage): that the IPO, as a formalization event, pushes companies to replace intuitive, novel strategic language with standardized, metric-based language that is easier to defend legally. It analyzes 75 filings across 25 companies and finds that lexical diversity and Shannon entropy decline together after the IPO.
Key findings across five cohorts:
- Legal amplifier: compression is strongest under high liability exposure; Coinbase and Robinhood show a 30–33% drop in lexical diversity in the first annual report
- Founder insulation: dual-class founder control delays but does not prevent compression (Meta 5.8%, Snap 7.2% in year one)
- "Born caged": S-1s from 2015–2020 (e.g., Snowflake, Okta) start with lower variance than S-1s from 1995–2004, suggesting formalization has moved upstream into IPO preparation
- Strategic pivot: radical strategic change (Netflix's move to streaming) can temporarily break the compression trend before it re-forms
Methodology
For each company the study takes the foundational registration statement (S-1 or F-1) and the first two subsequent annual reports (10-K, 20-F, or 40-F), and extracts the full text of "Item 1: Business" and "Item 7: MD&A." Two metrics are computed: lexical diversity (unique tokens divided by total tokens) and Shannon entropy (base 2). Mature public companies (Adobe, Intuit) serve as controls and show stable metrics over the same period.
Because both metrics move in the same direction, the paper argues post-IPO language is not just smaller in vocabulary but structurally simpler and more predictable.
Case Studies
Coinbase: the largest compression in the dataset. Lexical diversity fell 33.0% in the first 10-K, and Shannon entropy fell from 12.48 to 10.21, while the company was under active SEC investigation and litigation.
Robinhood: a nearly identical pattern after an IPO in the aftermath of the GameStop controversy: lexical diversity fell 30.0% and entropy fell from 12.10 to 10.15.
Google: despite a deliberately unconventional S-1 (founder letter and "Owner's Manual"), lexical diversity fell 17.6% in the first 10-K.
Implications
The study supports the Cage framework's claim that the IPO is a quantifiable formalization event. It does not show what the language compresses into; the paper proposes topic modeling and similarity analysis to test whether filings converge on a common profile.
Data Sources
All filings were retrieved from SEC EDGAR; the PDF lists each filing with its CIK.