Every few months, the leaders of publicly traded companies get on a phone call with the financial analysts who track their stocks. Management delivers a prepared statement about how the business is doing, then fields questions. Analysts listen closely, then adjust their predictions for the company’s future earnings. But what exactly are they responding to? The hard numbers, or the mood in the room?
A study published in the Journal of Economic Psychology offers evidence that the emotional tone of a manager’s remarks, not just the facts, shapes what analysts predict. And when analysts respond to that tone, their forecasts tend to become less accurate.
The question behind the research
Economists have long wanted to know how people form and share their beliefs about the economy. Traditionally, they have relied on surveys, asking consumers or business owners what they expect. But surveys have limits. They can only reach so many people, response rates can be low, and the act of asking about a topic can push it to the front of someone’s mind in a way that distorts the answer.
Yuting Chen of University College Cork and her colleagues, including corresponding author Maurizio Montone of Utrecht University, wanted a different window into economic beliefs. They turned to earnings calls, which offer a natural, recurring record of how business leaders talk about their companies and how professional analysts react. The team set out to measure the sentiment, meaning the positive or negative emotional tone, embedded in what managers say, and to trace how that sentiment travels to the analysts listening in.
Their reasoning draws on ideas from psychology. Managers control access to inside information about their firms, which can make analysts inclined to trust their framing. The researchers also point to the “halo effect,” the tendency to view people in leadership roles more favorably, and to “framing” effects, where the emotional packaging of information colors how it is interpreted. Because a manager speaks first and sets the narrative before any questions are asked, the authors expected the opening presentation to act as an anchor for the judgments that follow.
Turning millions of sentences into a sentiment score
The team gathered transcripts of every earnings call for U.S.-listed firms from January 2003 through December 2024, excluding financial companies. That came to 196,358 calls covering 6,321 firms across 84 quarters. They split each transcript into three parts: the manager’s prepared presentation, the analysts’ questions, and the managers’ answers during the back-and-forth question-and-answer session.
To score the tone of each section, they used a dictionary-based method. In plain terms, they built a list of words with assigned positive or negative values, tuned specifically for economic and financial language, and adjusted for negations so that a phrase like “not good” counts as negative rather than positive. They then checked their results against FinBERT, a more sophisticated machine-learning tool trained on financial text, and against a large language model similar to the ones behind popular chatbots. The three approaches agreed closely, which gave the authors confidence that the simpler, faster dictionary method was capturing real signal.
The researchers also wanted to know not just how positive managers sounded, but what they were positive about. Using a technique called topic modeling, they trained a program on more than seven million sentences from financial news to recognize ten themes, including company performance, outlook, costs and margins, innovation and technology, risk, and regulation. Training the topic detector on news articles rather than on the earnings calls themselves was a deliberate choice, meant to avoid a statistical trap where a model finds patterns simply because it was built from the same data being tested.
Finally, they matched this textual data to firm financials and to analyst forecasts of earnings per share, allowing them to measure how predictions changed in the days after each call.
What the analysis revealed
The central result is that managerial sentiment is linked to how analysts revise their forecasts. A one-standard-deviation increase in the positivity of a manager’s prepared presentation was associated with a 0.201-standard-deviation increase in upward forecast revisions. That effect was stronger than the sentiment in the analysts’ own questions or in the managers’ spontaneous answers. In other words, the scripted opening remarks carried more weight than the unscripted exchange that followed, consistent with the idea that the first narrative frame sets the tone.
These links held up after the researchers accounted for a firm’s actual fundamentals, including future profitability. To guard against the possibility that upbeat managers were simply better informed about genuinely good times ahead, the team ran a test using only calls made shortly after a new chief executive took over because of a retirement, illness, or death. Because those new leaders had no track record at the firm and no obvious reason to strategically manage their tone, the persistence of the sentiment effect in this group strengthened the authors’ interpretation that something behavioral, not purely informational, was at work.
The second finding is where the story turns. The researchers asked whether these sentiment-driven revisions made forecasts better or worse. They found that revisions following optimistic calls were associated with systematic forecast errors over a one-year horizon. Optimistic presentation tone, in other words, tended to nudge analysts toward predictions that turned out to be too rosy. The authors read this as evidence that narrative tone introduces a kind of non-fundamental noise into professional judgment rather than useful information.
Not all topics are equal
The topic analysis added texture. Sentiment about core business matters, especially costs and margins and overall performance, had the strongest link to forecast revisions. Sentiment tied to more inherently negative themes like risk and competition had weaker effects. The authors interpret this as analysts weighting a manager’s tone according to how relevant it seems to the company’s value, paying most attention when the subject is central to profits.
The broader news environment mattered too. When a topic such as environmental issues or technology was receiving heavy media coverage, managerial sentiment about that topic had a noticeably stronger influence on analyst forecasts. During quieter news periods, the same kind of sentiment had little effect. The researchers connect this to attention: topics that dominate the headlines become more mentally accessible, and analysts appear more receptive to related commentary when those themes are prominent.
Why it matters, and what to keep in mind
For investors, the takeaway the authors point toward is a note of caution. If professional analysts, who are supposed to be sophisticated, absorb the optimism in management’s tone in ways that later prove inaccurate, then stock prices that rest on those forecasts may partly reflect mood rather than substance. The researchers suggest this could contribute, at the firm level, to a misallocation of capital.
A few caveats are worth holding onto. The study documents associations rather than proving cause and effect, and the authors are careful to describe their results as robust predictive relations while calling for future work using more experimental designs. They also note that they cannot say whether analysts are aware of their susceptibility to managerial tone, pointing to research suggesting that even experts often struggle to correct for their own biases under time pressure.
Beyond the finance angle, the authors frame their method as a scalable alternative to surveys for tracking economic beliefs. Because earnings calls happen regularly and cover nearly every listed firm, the approach can be applied backward through past crises and forward as new data arrives, offering a running read on how business leaders talk, and how that talk ripples outward.
