Analyzing Speech to Detect Financial Misreporting

We examine whether vocal markers of cognitive dissonance are useful for detecting financial misreporting. We use speech samples of CEOs during earnings conference calls and generate vocal dissonance markers using automated vocal emotion analysis software. We begin by assessing construct validity for the software-generated dissonance markers by correlating them with four dissonance-from-misreporting proxies obtained in a laboratory setting. We find a positive association between these proxies and vocal dissonance markers generated by the software, suggesting the software's dissonance markers have construct validity. Applying the software to CEO speech, we find that vocal dissonance markers are positively associated with the likelihood of irregularity restatements. The diagnostic accuracy levels are 11% better than chance and of similar magnitude to models based solely on financial accounting information. Moreover, the association between vocal dissonance markers and irregularity restatements holds even after controlling for financial accounting and linguistic based predictors. Our results provide new evidence on the role of vocal cues in detecting financial misreporting