BofA reports its use of algorithms that study language in earnings calls to inform buy and sell decisions yielded a 6+% improvement in performance.
According to the Financial Times, “…unclear language is being identified as a consistent signal to sell a company’s stock.”
In other words, BofA has invented a robot BS Meter.
Natural language processing is one way of capturing information outside of the numbers that appear on financial reports, otherwise known as “unstructured data.” The BofA analyst quoted in the FT called it “thematic sentiment analysis.”
I wonder if there isn’t a low-fi corollary to the tool? Here’s my equation:
First, count the number of buzzwords that appear in the description of the business overall (AI, machine learning, innovation, disruption, startups, things like that).
Second, count the times a verb is attached to no concrete, measurable object, so statements like “we created value” or “increased customer engagement” would qualify. Multiply the number by the first total.
Third, multiply the total by 2 if the company reveals 1) It has or will advertise on the Super Bowl, 2) Has opened an innovation center, and/or 3) Hired someone to lead “transformation” or “digtial somethingoranother” (who usually dresses in black and comes from a consulting firm).
The higher the number, the more BS that’s behind the reporting, which could mean a nice opportunity to sell the stock. Conversely, a lower BS number might reveal a company that actually knows what it’s doing, or at least respects us enough to talk to use about it honestly, which may be a buying opportunity.
Getting a robot to sense BS is cooler, of course. But we all know BS when we see it.