Listen Closely: Using Vocal Cues to Predict Future Earnings
Log in to collectAcademic paper
Listen Closely: Measuring Vocal Tone in Corporate Disclosures
Jonas Ewertz; Charlotte Knickrehm; Martin Nienhaus; Doron Reichmann
- DERuhr University Bochum
- ?Ruhr University of Bochum - Department of Finance and Banking
- ?Ruhr University of Bochum - Department of Industrial Sales and Service Engineering
- DEGoethe University Frankfurt
Strategy in a nutshell
This strategy uses CEOs’ vocal cues from earnings conference calls to predict the direction of one-year-ahead earnings changes. By processing audio with machine learning models like CNNs, wav2vec 2.0, and SVMs, it captures emotional and phonetic signals that are less controllable than verbal content. These signals are then translated into trading strategies, forming portfolios based on predicted earnings increases or decreases. Empirical results show statistically and economically significant out-of-sample excess returns, highlighting the practical investment value of analyzing managerial vocal cues beyond traditional financial metrics.
Economic rationale
This strategy works because managers’ vocal cues reveal genuine emotional signals about future firm performance that are hard to mask. By analyzing these cues using CNN and wav2vec 2.0 models, the strategy extracts predictive information beyond traditional financial data, allowing investors to forecast one-year-ahead earnings changes. Empirical evidence shows these models provide significant out-of-sample predictive power and generate excess returns, demonstrating that vocal cues are an economically meaningful and novel source of information for investment decisions.