Considerable accounting and finance research has attempted to identify whether reported financial statements have been manipulated by executives. Most of these classification models are developed using accounting and financial market explanatory variables. Despite extensive prior work, the ability of these models to identify accounting manipulations is modest. In the paper, Detecting Deceptive Discussions in Conference Calls, forthcoming in the Journal of Accounting Research, we take a different approach to detecting financial statement manipulations by analyzing linguistic features present in CEO and CFO narratives during quarterly earnings conference calls. Based on prior theoretical and empirical research from psychology and linguistics on deception detection, we select the word categories that theoretically should be able to detect deceptive behavior by executives. We use these linguistic features to develop classification models for a very large sample of quarterly conference call transcripts.

A novel feature of our methodology is that we know whether the financial statements related to each conference call were restated in subsequent time periods. Because the CEO and CFO are likely to know that financial statements have been manipulated, we are able to reasonably identify which executive discussions are actually “deceptive”. Thus, we can estimate a linguistic-based model for detecting deception and test the out-of-sample performance of this classification method.

(more…)