Detecting Deceptive Discussions in Conference Calls

We estimate classification models of deceptive discussions during quarterly earnings conference calls. Using data on subsequent financial restatements (and a set of criteria to identify especially serious accounting problems), we label the Question and Answer section of each call as \truthful" or \deceptive". Our models are developed with the word categories that have been shown by previous psychological and linguistic research to be related to deception. Using conservative statistical tests, we find that the out- of-sample performance of the models that are based on CEO or CFO narratives is significantly better than random by 4%- 6% (with 50% - 65% accuracy) and provides a significant improvement to a model based on discretionary accruals and traditional controls. We find that answers of deceptive executives have more references to general knowledge, fewer non-extreme positive emotions, and fewer references to shareholders value and value creation. In addition, deceptive CEOs use significantly fewer self- references, more third person plural and impersonal pronouns, more extreme positive emotions, fewer extreme negative emotions, and fewer certainty and hesitation words.