I have been an independent stock trader for the majority of my adult life. I have no formal finance education. I was an English major.
Approaching stock evaluation from a linguistic standpoint, as it turns out, was a novel way to go about things. I quickly learned that companies had to file annual reports (10-Ks) and quarterly reports (10-Qs), and while I knew nothing about finance, I did know how to read. But man…these things were horrible to read. Very long (and always getting longer) and stupendously boring, when they were comprehensible at all.
Oh, how I wished reading them was as engaging as reading Charlie and the Chocolate Factory. But if Charlie actually did own a publicly traded chocolate factory and he screwed up in major fashion, he’d probably hide a sweet nugget like this on page 95 of his company’s 10-K:
|Management has identified a material weakness in its internal control over financial reporting that existed as of December 31, 2015, solely related to its accounting for cocoa derivative financial instruments pursuant to the provisions of FASB ASC Topic 815, Derivatives and Hedging (ASC 815). This material weakness arose in the quarter ended October 4, 2015, and was the result of changes the Company made in the third quarter of fiscal 2015 to its hedging program and related controls. Specifically, the control designed to monitor compliance with the Company’s hedge documentation did not operate as designed during the third and fourth quarters of fiscal 2015. As a result, instances of non-compliance with our hedging program related to cocoa derivatives occurred and were not detected timely and, therefore, we did not meet the technical requirements to qualify for cash flow hedge accounting treatment under ASC 815.|
Make no mistake, any company’s “material weakness in its internal control over financial reporting” is a giant, flaming red flag. So, how many investors in Hershey stock read 95 pages or more of that 10-K to find that important piece of information? Very few. The best CFOs in the business agree there’s no way to get through the filings they publish.
As a starry-eyed entrant into this brave new world, I had the advantage of not knowing just how daunting a task it would be to make a living of parsing SEC filings. Armed with this profound naïveté, a desire to make a decent living, and a penchant for absorbing lengthy, dense material, I trudged ahead. It was intensely unpleasant work. But eventually, over the course of years, I learned how to find and decipher the information many companies were hiding. Often, they were doing this in language so convoluted, so intentionally disguised as meaningless boilerplate, that any investor would have a nearly impossible time finding it.
For 23 years I built my own neural network of “tells”… indicators that companies were intentionally obfuscating the timeline or extent of negative events, or even their own systemic fraud. Specific combinations, in various densities, set off different kinds of alarms. SEC filings, by dint of their combination of meaningless government-mandated boilerplate language and similar-sounding, though materially meaningful information added by the issuing company (and designed to protect them from future litigation), are impenetrable thickets of legalese written in overlapping conditional clauses. A Rubik’s Cube, in a hall of mirrors, housed within a giant shell game.
While I was sharpening and parlaying my unique and unenviable skill set into a career path, algorithmic trading was simultaneously dismantling the edge the vast majority of my math-expert competitors once had. But my edge continued to endure unscathed, because no machine could do what I had trained myself to do. That is, of course, until now.
Now, a quarter century later, I am fortunate to find myself teamed up with super smart people, on the verge of disrupting how 10-Ks are read and, more importantly, understood. Topos Labs is an AI company that has built a cognitive text mining engine called Gracie. Together, we are training Gracie to highlight important, more overt disclosures (like the Hershey’s example above) while also unraveling the code of intentional linguistic opacity that most often accompanies fraud. And financial world be warned, she’s learning fast.
As Gracie slices her way through the pablum, boilerplate, and pleasantries of these lengthy documents to serve up a visual and numerical cross-section of corporate disclosure, she represents a fascinating new chapter in this arena. All of my strategies and knowledge about how to do this work are now replicable, as she learns to identify this pervasive plague of malfeasance, hiding in plain sight in over 21 million documents.
Admittedly, I was skeptical at first. It was hard for me to imagine a machine successfully navigating these intentionally complex documents. Fortunately, our starting point was Gracie’s unparalleled ability to understand unstructured textual language. Working closely with the Topos team, we taught Gracie the language of opacity, as well as the ability to identify and ignore trivial language. The results are nothing short of amazing. I’m a believer.
Understanding how and where companies bury bad news in SEC filings had been my life’s work. Yet Gracie somehow gained this knowledge in a fraction of the time, and she’s getting smarter by the minute. She doesn’t suffer from eye strain. She never gets distracted or bored.
I don’t feel guilty about assigning her the task.
Gracie is like having a better version of 1,000 of me at your beckon call, 24/7/365, helping to ingest and interpret these documents with speed and accuracy impossible for humans. Make no mistake, Gracie is taking this job away from me. Thank heaven!