On Earnings Calls, Do Executives Mumble on Purpose?

Earnings calls are meant to be a simple, direct way to get information to investors, but sometimes the audio leaves listeners struggling to understand. Could that be deliberate?

Research by Seoul National University’s Bok Baik, Chicago Booth PhD student Alex G. Kim, MIT PhD student David Sunghyo Kim, and Korea University’s Sangwon Yoon suggests it might. They explored how the quality of the vocal delivery on earnings calls can influence investors’ trading in the moment, and find that market reactions tend to be more subdued when delivery is poor and listeners struggle to understand what is being said. The researchers contend that managers may even deliberately modulate the quality of their vocal delivery on the basis of their own economic incentives.

A variety of prior studies have established that the tone of earnings calls, and even the obtuse language and erroneous expressions that executives sometimes use, affect investor reactions. Baik, Kim, Kim, and Yoon instead focused on qualities that can hamper the clarity of vocal delivery, such as mumbling, mispronunciation, and lazy diction.

The researchers analyzed nearly 29,000 quarterly earnings calls occurring between 2008 and 2020 from almost 2,000 companies, limiting the calls to those that took place during market trading hours and focusing on the portion of the calls in which managers presented results.

They used an open-source model based on a deep-learning algorithm that converted audio files to letters, which were then combined into words and ultimately text. When doing this, the model generated a probability value, which represented the certainty of each audio-to-letter conversion. When there was a high probability for one particular letter, the researchers inferred clarity in the audio. But when the probabilities were more diffuse, they deemed the text to be less listenable and harder to understand. (For example, if the letter “a” was pronounced clearly, the model gave it a probability of 1—or else it might give it a 50 percent chance of being an a and a 50 percent chance of being an e.) The researchers took advantage of the probabilities that the model calculated to develop a score for audio clarity.

Using a regression analysis, they find that when the quality of vocal delivery fell, trading activity was more subdued than normal. On the other hand, clear delivery was associated with more active market reactions. When vocal-delivery quality was high, analysts were more likely to issue early forecast revisions even before an official earnings call transcript was released. High-quality vocal delivery also appeared to strengthen the media hype (measured by text sentiment analysis of coverage), as well as the positive relationship between earnings surprises and abnormal market reactions.