D.A.T.A.'s technology is grounded in our 10+ years researching the 100+ years of findings of deception science. We then use NLP to rapidly look for known behavioral differences between deceivers and truth-tellers.
Deception And Truth Analysis - D.A.T.A.
HOW D.A.T.A. WORKS
Behavioral Differences Examined. Our algorithm works by examining how language is used, and not what language is used. Deception And Truth Analysis (D.A.T.A.) scrutinizes over 30 unique behavioral indicators, making it exceedingly difficult to spoof our algorithm. Other text-based deception detectors look for the presence of "red flag" words. Sadly, recent research (Cao, Sean; Wei Jiang; Baozhong Yang; Alan L. Zhang. "How to talk when a machine is listening: Corporate Disclosure in the Age of AI." SSRN. October 2020) shows that the advantage from searching texts for these words has created an arms race, with companies now hiring speech and language coaches in order to excise offending "red flag" language from their texts and speech.
Algorithm Developed Out-of-Sample. D.A.T.A.'s algorithm was developed out-of-sample. That is we evaluate how people deceive using language, in general, rather than how they use language to deceive within a specific context. This means that our technology is general enough to be applied in multiple domains.
Scientifically Demonstrated Accuracy. Our work is supported by scientific research into the ability of computers and text-based analyses. D.A.T.A.'s ability to surface deception is statistically significant and in excess of the capabilities of people. Our double-blind scientifically-tested accuracy is 88.4% for communications in excess of two pages in length. Our Type I error is 11.3% and our Type II error is 14.3%. While our p-value < 0.0001 and Cohen's d is 4.23. This accuracy compares to a January 2021 study that placed expert human capabilities of detecting deception in texts at an even 50.0% chance level. Additionally in experiments D.A.T.A. is able to identify 9 of the top 10 largest scandals in global corporate history and with an average lead time of 6.2 years. Last in a test of Deception And Truth Analysis used for investment selection resulted in an average annual outperformance vs. the equal-weighted DJIA of 37 bps, outperformance in 9 of 12 years, and compounded outperformance of 4.45%. See our Insights section for more information on each of these validation methods.
No overfitting. As former investment managers ourselves we are dubious of back-testing and data-mining. We have literally seen and heard it all. In short, we are not fans. Consequently, in reporting Deception And Truth Analysis' tests we have tested our technology in multiple settings very different from one another to ensure its general use. First, D.A.T.A. has demonstrated its success in uncovering known deception, lies, and fraud. Second, we have tested our ability to discern between deceptiveness and truthfulness in high stakes real world political decision-making. Third, it may be that our ability to tell the difference between deception and truth is irrelevant to financial markets. But instead we have found that markets do care about, and do price deception. Different settings, same algorithm, same success.