Informativeness and Timeliness of 10-K Text Similarity for Predicting Tail-Risk Comovement

Abstract

We measure a bank's connectedness by constructing a measure of its text similarity with other banks based on 10-K business description and MD&A discussions. We find that tail-risk comovement between a given bank and the banking system is increasing in the bank's average similarity. We also construct groups of connected peer banks, finding that banks co-move significantly more in the tails with their highest similarity peers. Finally, we separate 10-K text into boilerplate and non-boilerplate components. We find that both boilerplate and nonboilerplate similarity have incremental information about future tail comovement. However, nonboilerplate similarity is significantly timelier than boilerplate, consistent with non-boilerplate similarity capturing commonalities across banks in currently evolving fundamentals and boilerplate similarity capturing commonalities in structural features that evolve slowly over time.

Current Market Top Business Scopes Trend—A Concurrent Text and Time Series Active Learning Study of NASDAQ and NYSE Stocks from 2012 to 2017

“…This method was then widely used in text similarity analysis. Bushman [15] calculated a daily market return by summing the abnormal returns; these analysis considered a bank's average text similarity and tail co-movement with all other banks. Since news may contain more information, also full of diverse expressions, Bushman's method cannot be guaranteed in terms of validity.…”

Shared Exposures or Management Fashions? Drivers of Cross-Industry Convergence of Textual Risk Disclosures

“…We suggest that a firm's 10-K report -the document that all listed companies have to submit yearly to the US Securities and Exchange Commission (SEC) -can play such a role. In fact, 10-Ks have been extensively used as data sources for describing managers' and firms' behaviours (e.g., Dutt and Joseph, 2019;Bushman et al, 2017;Bao and Datta, 2014;Guo et al, 2017;Cazier et al, 2021;Mai et al, 2019). The 10-K reports are structured in several sections, describing a company's business, the risks it faces, and the operating and financial results for the fiscal year.…”

“…We use Kendall's rank correlation because it is robust to extreme observations (Lindskog et al, 2003). We preferred it over tail risk measures (e.g., Bushman et al, 2017) for reasons of estimation stability, as we are not specifically focused on extreme levels of stock co-movement.…”

“…The insurance and banking industries have different business models, but at the same time, are subject to common pressures from the economic and broader environment, while their adjacency means that they are often able to enter each other's markets. For example, the near-collapse of the insurance giant AIG in 2008 demonstrated interconnectedness across industry boundaries, with insurers investing in mortgage-backed securities, which generated shared vulnerabilities and, ultimately, extreme downside risks (Billio et al, 2012;Bushman et al, 2017). Furthermore, the cultural commonalities of the financial services industries make issues such as groupthink and herding behaviour an important concern (e.g., Trueman, 1994;Clement and Tse, 2005), with such co-behaviors causing systemic risk (Piškorec et al, 2014;Haldane and May, 2011).…”

“…Specifically, we use companies' 10-K reports, spanning the 2006-2018 period. Such data have been widely used as a data source in management (e.g., Guo et al, 2017;Dutt and Joseph, 2019), accounting (e.g., Cazier et al, 2021), risk management (e.g., Bao and Datta, 2014), bankruptcy prediction (Mai et al, 2019), and -importantly -systemic risk measurement (Bushman et al, 2017). In contrast with previous literature, we focus on convergence in attention across industries, which contrasts with a growing literature using financial reports to extract information about individual firms or risks arising within a single industry (e.g., Bushman et al, 2017;Mai et al, 2019;Gupta et al, 2021).…”