Predicting fraud from quarterly conference calls: A small-sample study of scripted language

This study introduces a method of detecting deceptive and fraudulent executives by using the information contained in quarterly conference calls and financial statements. I argue that executives in fraudulent companies will adhere more closely to a script as way to minimize the risk of disclosing negative or inconsistent information, and that the MD&A section of the corresponding 10-K or 10-Q statement is a valid proxy for this script. I use a small sample of companies from the financial services industry during the financial crisis of 2007-2008. Using tf-idf term weighting and cosine similarity, I compare the executives' language in the conference calls to the corresponding MD&A statement. The results indicate that executives in fraudulent companies use quarterly conference call language that is more similar to the corresponding MD&A statement than those in non-fraudulent companies. This is a research-in-progress, so I discuss the next steps for this project.