As one who used to earn a living by writing copy to guide executives through their quarterly earnings calls to investors and the media, this new study summarized in The Economist is equal parts revealing and horrifying:
David Larcker and Anastasia Zakolyukina of Stanford’s Graduate School of Business analysed the transcripts of nearly 30,000 conference calls by American chief executives and chief financial officers between 2003 and 2007. They noted each boss’s choice of words, and how he delivered them. They drew on psychological studies that show how people speak differently when they are fibbing, testing whether these “tells” were more common during calls to discuss profits that were later “materially restated”, as the euphemism goes.
Since not all executives write their own remarks for those calls, it would be fair to question the study’s viability. Except that the researchers know that, and so they only examined the Q&A portion of the calls, which takes so much time to script that it’s not worth bothering to try.
The article does correctly observe that PR practitioners are bound to get their game-theory on and avoid precisely the things the research says indicate deception. In that vein, here’s a cheeky handy list of things to do when drafting earnings scripts for the 3Q10 period:
To be serious for a moment, the payoff for everyone else is that, as claimed by the authors, “linguistic features statistically improve the out-of-sample performance for a traditional accounting-based model that uses discretionary accruals. These performance results suggest that it is worthwhile for researchers to consider linguistic cues when attempting to measure the quality of reported financial statements.”
So they’re saying the models are reliable data-points for traders and other investors on the Street?
A working version of the full paper appears after the jump; it’s great background for IR types.







