AI-guided study at IU finds signs of depression common among CEOs

A recent study led by an Indiana University researcher using artificial intelligence to scrutinize corporate earnings calls found CEOs often exhibit vocal markers consistent with depression.

The study used AI to analyze vocal acoustic features—so subtle they are imperceptible to human ears—from CEO presentations of more than 14,608 earnings calls by S&P 500 companies from 2010 to 2021. In 65% of calls, 9,510 instances, the CEOs exhibited vocal markers consistent with depression, according to research led by Nargess Golshan, assistant professor of accounting at the Kelley School of Business at IU Bloomington.

The study noted that depression rate was “much higher than in the general population.” The analysis comprised 826 CEOs representing 421 companies.

Golshan said the study provided an initial examination of depression in CEOs, which can remain hidden because of the high-pressure nature of the position. She added that the emotional states of CEO can significantly impact their careers, firms and the broader economy.

“Our main goal is to really stress that this is very prevalent,” Golshan said. “We hope this study is just opening a lot of other avenues for understanding mental health in the corporate leadership world.”

Golshan co-wrote the study, “Silent Suffering: Using Machine Learning to Measure CEO Depression,” with University of Kentucky doctoral candidate Mark Cheng.

The study, published earlier this year in the Journal of Accounting Research, leveraged machine learning models trained on clinically validated depression indicators, according to IU. Executives’ basic information, such as their gender, age and compensation, as well as company finances are publicly available through regulatory filings.

CEO depression was not related to firm performance in the short or long term. The study did find a lower likelihood of depression among female and older CEOs.

“We didn’t really find any relationship with performance, if there is a depression versus not,” Golshan said. “We found that these depressed CEOs have generally higher compensation levels and also their compensation is more sensitive to their performance.”

Also, the researchers found no evidence CEO depression was associated with CEO turnover. However, they did find turnover-performance sensitivity was higher among depressed CEOs.

The analysis focused on the CEO presentation during earnings calls, which typically happens early in the call, with an average of 4.23 minutes of CEO remarks.

Golshan said she got the idea for the study years ago when she learned of efforts to use technology to analyze vocal features to detect COVID-19. Audio engineers, neuroscientists and startups increasingly are looking at ways to use AI tools to detect potential mental health issues, she said.

Golshan said academic studies look at results only in the aggregate, not identifying individual companies. But she said there are ethical concerns, especially as AI technology is used increasingly by companies.

For example, could a job candidate face an AI-led analysis of their answers to interview questions?

“If it is used by companies, it definitely has to be disclosed and an understanding of all parties involved that this could be a part of the screening,” Golsham said. “Definitely, there are ethical concerns and people and ethics world that are working on these issues.”