Title: Can CEO vocal cues help to predict future firm performance? Authors:  Zihao Liu - Tilburg University (Netherlands) [presenting]
Abstract: Artificial intelligence (AI) algorithms on CEO voiceprints of earning conference calls are applied to predict the firm's future performance as measured by analyst recommendation changes, unexpected earnings, and cumulative abnormal returns. Tailored deep learning models based on convolutional neural networks (CNNs) are used. An out-of-sample evaluation (which is standard in AI) predicts a firm's future performance above the benchmark level is predicted after controlling for firm characteristics and the textual sentiment. In other words, how firm information is communicated in addition to the content can affect the market perception of a firm. The purpose is to add new evidence to audio recordings of conference calls containing valuable information about a firm's fundamentals, incremental to quantitative earnings information, and qualitative "soft" information conveyed by textual content. The economic mechanisms are further investigated using innovative techniques such as visualization of neural networks and classical methods such as control variables and experiments.