A company's lengthy, back-and-forth explanation of its finances is not always a sign that managers are hiding bad news. In a study of more than 13,000 corporate filings, a Cornell accounting researcher found that disclosures that return to related subjects in different contexts were more common among struggling firms that later recovered.
The finding comes from research by Nicholas Guest of Cornell and Jiawen Yan of the National University of Singapore, published Aug. 17 in The Accounting Review. The authors call the pattern “circuitousness.” It describes how related information is spread through a narrative, not simply how long the document is or how hard its words are to read.
Measuring a winding narrative
Guest and Yan analyzed management discussion and analysis sections of annual reports filed with the U.S. Securities and Exchange Commission from 1997 through 2019. They divided each section into chunks of about 250 words and used natural-language processing and machine learning to measure how the discussion moved among related topics.
Their measure compares the path a document takes through its ideas with a shorter path that could have covered the same points. A higher ratio means the writing circles back to topics rather than grouping each one in a single place. For example, a company might discuss revenue in the context of financing, customer relationships and regulatory demands at different points in the report.
The researchers focused particularly on firms with negative earnings. Their hypothesis was that a company nearing a recovery might need to explain several interacting changes, making its discussion more circuitous. They compared the measure with document length, repetition and the Fog index, a common measure of linguistic complexity.
“CEOs shouldn't necessarily try to be as brief as possible,” Guest, an assistant professor of accounting at Cornell's Samuel Curtis Johnson Graduate School of Management, told the Cornell Chronicle. A fuller discussion may be useful to an investor willing and able to work through it, he said.
What the pattern did — and did not — show
Higher circuitousness was associated with underperforming firms that went on to report earnings upturns, the researchers found. They did not find the same association for firms headed toward downturns. The paper's abstract says circuitousness was more consistently predictive of a turnaround than the related text measures the authors tested, including repetition and the Fog index.
The paper also reports more downloads of those filings from the SEC's EDGAR system and more analyst forecast revisions, signs that market participants spent effort processing the information. Initial stock-price reactions were faster but incomplete, according to the authors, consistent with a complicated disclosure taking time to understand. These are associations in historical filings, not proof that a winding report causes a recovery or that any one stock will rise.
Guest contrasted circuitousness with obfuscation, or writing meant to make information harder to understand. Returning to a topic can add context, he said, although executives could also have reasons to withhold or carefully stage good or bad news. The study measures text patterns and subsequent financial results; it cannot establish a manager's intent from prose alone.
The historical scope matters for readers considering the result today. The filings ended in 2019, before the pandemic and more recent shifts in disclosure and automated text analysis. The research offers a way for analysts, journalists and investors to ask sharper questions of a report, not a stand-alone investment rule.
Guest said an investor might use the measure as an initial screen for firms that could be approaching a turnaround. An analyst or reporter could also use it to identify topics a company's leadership appears to be working especially hard to explain, then check those claims against the underlying numbers and later results.
Yan, a Cornell doctoral graduate, is now an assistant professor of accounting at the National University of Singapore. The research received partial support from a Dyckman Research Grant awarded to Yan for his doctoral proposal.




