AI detects CEO depression through earnings calls

Corporate leaders could give away signs of emotional stress during earnings calls with investors, according to a new study co-written by a researcher at the University of South Florida St. Petersburg.

The study, titled “Silent Suffering: Using Machine Learning to Measure CEO Depression” and published in the Journal of Accounting Research, highlighted how artificial intelligence can analyze subtle vocal patterns. Professor Sung-Yuan (Mark) Cheng and his co-author came to their conclusions after recording and analyzing over 14,500 earnings calls.

Their artificial intelligence (AI) model does not diagnose clinical depression. However, the research helped identify factors, including compensation and company uncertainty, that likely contribute to mental health challenges among executives.

“With that knowledge, people can intervene in one way or another,” Cheng told the Catalyst. “For example, they can definitely suggest doing therapy or things like medication to help deal with the situation.”

Cheng, a faculty member in the Lynn Pippenger School of Accountancy, and his co-author, Nargess Golshan at Indiana University, based their machine learning system on a large audio model originally developed by Google.

The researchers utilized recordings from patients who completed professional mental health assessments to train their AI. Their model then learned frequent speech patterns in people with depressive symptoms.

Cheng noted that medical studies have previously demonstrated that “it’s possible to use a patient’s voice recording to try and identify signs of depression or other mental health issues.” The idea for applying that research to CEOs originated with Golshan.

“So, we know from medical research that this is possible,” Cheng said. “The other kind of angle is that there are surveys showing that CEOs or other executives suffer from mental health issues at, maybe, surprisingly high rates.”

The two researchers hoped to compile additional data that supported those theories. They succeeded.

Cheng designed an algorithm that isolated the CEO’s voice from over 14,500 S&P 500 earnings call recordings. He called eliminating other speakers a “key challenge.”

The machine learning model identified signs of depression in over 9,500 of the chief executives studied. Cheng said that was a higher rate than initially expected.

“That’s one surprising fact,” he added. “The other is that, at least on average, it doesn’t really affect the performance of the firms. It seems that there is no significant difference between firm performance when comparing depressive CEOs versus non-depressive CEOs.”

Professor Sung-Yuan (Mark) Cheng is a faculty member in the Lynn Pippenger School of Accountancy at the University of South Florida St. Petersburg campus. Photo: USF.

However, the study found depressive signals increase when a CEO’s company faces lawsuits, volatile stock returns or disappointing financial results. Another surprising pattern was that business leaders who are more likely to suffer from depression often receive higher compensation.

Cheng said that company boards, whether consciously or not, may increase financial incentives for CEOs who appear to be succumbing to the pressures of running a company. “I have to be very clear that our method just provides a way to identify signs of depression,” he emphasized.

“It’s not really that definitive.”

The researchers also realized that women and older CEOs appear less likely to exhibit depressive vocal patterns. Potential explanations include a greater stress resilience among those who face a more difficult path to the top and model bias.

Cheng explained that real-world speech differs from clinical voice recordings collected in a healthcare setting. The AI model is “informative, but far from perfect.”

Governing boards could potentially use the model to ensure CEOs receive mental health care. The system could also help investors select which companies and people they want to entrust with their money.

Cheng said he plans to research how depressive signals impact corporate decision-making. The Economist and Fortune shared the study’s results, which he credited to a lack of data on chief executive mental health and the current focus on AI.

Golshan, in an interview with Fortune, said he hoped to highlight the prevalence of mental health issues among business leaders. “Of course, it is important for the personal health of these executives, but it also has far-reaching implications for the organization, the employees, the investors and the broader economy.”