About the paper
The report examines whether CEO “thought leadership” — voluntary, owned executive communications such as op-eds, speeches, interviews, public statements and shareholder letters — is associated with stock-market value.
It is an associative, observational event-study using 1,034 CEO communications from 357 S&P 500 companies over 26 years, with a narrower 287-event non-shareholder sample for its strongest “canon similarity” finding.
The methodology combines Fama-French three-factor cumulative abnormal return analysis over a one-week window with computational text analysis, including 63 formulaic text features, whole-document embeddings, and semantic comparison against a curated canon of landmark CEO communications.
The geographic scope is U.S. publicly traded companies; the report explicitly says findings may not generalise to private companies or non-U.S. markets.
Length: 24 pages
More information / download:
https://cardinal40.com/2026-alpha-report/
Core Insights
1. What is the report’s central claim about CEO thought leadership and shareholder value?
The report’s central claim is that the quality of CEO thought leadership is associated with measurable differences in short-term stock-market performance. It argues that CEO communications are not merely neutral containers for information already created elsewhere in the business; the way the CEO frames, writes and communicates ideas may itself be associated with value.
The headline number is deliberately attention-grabbing: the report says that the difference between top- and bottom-decile communication quality corresponds to roughly 0.9 percentage points in one-week cumulative abnormal returns. For the median S&P 500 company, it translates this into approximately $367 million in shareholder value; for the average Magnificent Seven company, it says the equivalent would be about $25 billion.
However, the report is careful — at least in several places — to stress that this is not a causal finding. It does not prove that writing a better CEO op-ed, speech or letter will mechanically increase the share price. Rather, it identifies an association between certain kinds of CEO communication and abnormal returns over a short market window. This distinction matters because the report’s own framing sometimes leans rhetorically towards “words create value”, while the research design can only support “words are associated with value”.
2. How does the report define and measure “thought leadership”?
The report defines thought leadership as voluntary public communication authored by, or directly attributed to, a sitting CEO. It includes four categories:
- bylined op-eds
- public speeches and testimony
- interviews and public statements
- and annual shareholder letters.
The communications had to have a known publication or delivery date, and the authors excluded communications that could not be reliably dated or that coincided with obvious market-moving information such as quarterly earnings.
The main financial outcome is the one-week Fama-French three-factor cumulative abnormal return, or FF3 CAR, measured from the trading day of publication through five trading days afterwards. This controls for broad market return, company size and value factors. In plain English, the report is asking: after adjusting for what the stock might normally have been expected to do, did it outperform or underperform in the week after the CEO communication?
The study then uses two samples. The full universe contains 1,034 communications and is used to show the overall spread of market reactions, to test simple text features, and to test broad embedding-based prediction. The narrower sample contains 287 non-shareholder communications and excludes shareholder letters because these are often published alongside annual reports or other financially material disclosures, making it harder to isolate the communication effect.
Methodologically, this is best described as an original, mixed-methods-style empirical report: it combines original data collection, financial event-study analysis, natural language processing, and a professionally curated qualitative benchmark of “great” CEO communications. It is not an expert survey, not a simple secondary analysis, and not a modelling/data pack alone.
3. What does the report find about ordinary writing advice and formulaic text features?
One of the report’s more interesting findings is negative: simple writing rules do not explain the variation in market outcomes. The authors tested 63 text features, including things such as word count, reading level, sentiment, use of data, and pronoun choices. Only three were statistically significant at the 95% level, and none remained significant after multiple-testing corrections.
This matters because it pushes against much conventional communications advice. The report is effectively saying that the market signal, if there is one, is not captured by easy prescriptions such as “make it shorter”, “use more data”, “sound more positive”, or “avoid jargon”. Those features may still matter for readability, reputation, clarity or persuasion, but in this study they did not reliably predict one-week abnormal returns.
The implication is that “good” executive communication cannot be reduced to a checklist. The authors’ interpretation is that quality lives in the whole document: the underlying argument, tone, originality, judgement, structure, substance and credibility of the communication as a complete act of leadership. That is a useful point, although it also makes the report less prescriptive than its commercial framing might suggest.
4. What role do embeddings and the “canon” play in the report’s argument?
After simple text features failed, the report turns to whole-document embeddings. These convert each communication into a position in a 384-dimensional semantic space, intended to capture the overall meaning and character of the document rather than isolated attributes. Using five-fold cross-validation, the embedding model produced a statistically significant but modest relationship between predicted and realised one-week abnormal returns. The reported correlation is around r = +0.079 in the main discussion, meaning the model explains less than 1% of the variance.
That result supports the report’s claim that some textual signal exists, but it does not tell communicators what “good” looks like. To make the finding more interpretable, the authors construct a canon of 33 landmark CEO communications, including examples such as Warren Buffett’s annual letters, Steve Jobs’s Stanford commencement speech, Andrew Carnegie’s “The Gospel of Wealth”, and Marc Andreessen’s “Why Software Is Eating the World”. They then measure how semantically similar each CEO communication is to this canon.
The strongest interpretable finding is that communications closer to the canon are associated with stronger abnormal returns in the 287-event non-shareholder sample. The pooled correlation is modest, r = +0.101, with p = 0.087, which does not meet the conventional 5% significance threshold. But top-quartile canon-similar communications averaged +0.469 percentage points in abnormal return, compared with +0.084 percentage points for the full non-shareholder sample.
This is the report’s most distinctive idea: “quality” is not defined as mimicry of one famous CEO, but as semantic proximity to a broad region occupied by communications that experienced practitioners judge to be exemplary. The report’s page 20 chart visualises a permutation test suggesting that the observed canon relationship is unlikely to be just an artefact of one hand-picked canon, though the canon itself still rests on professional judgement.
5. What are the most important limitations and implications of the report?
The most important limitation is that the study is observational. The authors cannot randomly assign different CEO communications to different companies, so the findings are correlational rather than causal. The report explicitly acknowledges that unobserved firm characteristics, strategic timing, concurrent events or CEO quality more broadly could explain some or all of the observed relationship.
A second limitation is statistical modesty. The embedding result is statistically significant but explains less than 1% of variance. The canon similarity result is directionally consistent and economically interesting, but it does not clear the conventional 0.05 significance threshold. The report itself warns that neither result should be treated as a strong predictive tool.
A third limitation is generalisability. The study is based on publicly traded U.S. companies with sufficient trading data, and on communications that could be systematically collected. It may not apply to private firms, non-U.S. markets, internal communication, political communication, social media, podcasts or other executive-visibility formats not covered by the taxonomy.
The practical implication is not “write like Warren Buffett and your stock will rise”. A more defensible takeaway is that CEO communication quality may be financially material, but that quality is holistic, contextual and difficult to reduce to simple writing hacks. For communications leaders, the report strengthens the case for investing in executive thought leadership as a serious strategic discipline — while also warning against overclaiming what the current evidence proves.