Using Computers to Predict If a CEO is Lying

There is an interesting article in today’s Wall Street Journal discussing an academic paper by a couple of Stanford University Graduate School of Business professors. The article is entitled, “For Lying CEO’s, ‘Team’ Not ‘I’” and refers to a Paper entitled “Detecting Deceptive Conference Calls” available at http://www.gsb.stanford.edu/cldr/cgrp/documents/SSRN-id1572705.pdf. Obviously, for anyone involved in conference calls, either on the corporate or the investing side, this is required reading.

What the professors did was interesting: they examined the Question and Answer sessions of 29,663 earnings conference call transcripts between 2003 and 2007 for language features that predict “deceptive” reporting of financial statements. (Note that they were smart enough to ignore the carefully scripted section of the call.) In the professors’ words, “Our primary assumption is that CEOs and CFOs know whether financial statements have been manipulated, and their spontaneous and (hopefully) unrehearsed narratives provide cues that can be used to identify lying or deceitful behavior.” They then compared their predictive results to whether or not there was a later material financial restatement. And what they found was that their methodology had correctly predicted between 50% – 65% of the conference calls as “deceptive” by virtue of their having later resulted in an earnings restatement.

For those of you who want (or need) to know what the linguistic hot buttons are, here are some of the things to watch out for:

• CEOs that speak in terms of third person plural pronouns or impersonal pronouns are more likely to be deceptive. It seems in this case, honest CEOs tend to speak of I and me while dishonest CEOs prefer to refer to the company or team.
• Expressing extreme positive emotions with words such as fantastic is more likely while making a deceptive claim.
• Longer answers also are more likely to indicate that they are deceptive.
• Lack of hesitation – if a CEO is quick off the mark, the authors hypothesize that the answer is more likely one that he has rehearsed and wishes to answer quickly and move on.
• References to general knowledge such as “you know” also appear more frequently in deceptive responses.
• Lack of mention of shareholder value or value creation are also tip offs.

It will be interesting to see if investors pick up on any of this while parsing conference call answers. The Q & A session is already the portion of the call that gets the most scrutiny, and this research will only help to bring more focus to the area.