Elon Musk’s “Taking Private” tweet analysed by ‘AI’
It looks like the Tweet that caused so much controversy is actually not as “binding” as you might think. We used word-embeddings to help keep Elon out of jail.
Last week $TSLA stock soared on the news that Elon Musk was considering to privatise the company. It would be the largest buyout of its kind in history!
Once the news was confirmed via other official channels it didn’t take long for people to start questioning the legitimacy of the Tweet. Did Elon Musk just manipulate the stock market and is there a chance he could be facing an investigation by the SEC or even criminal prosecution?
Update: SEC has reportedly served Tesla with a subpoena
The question wether Elon’s Tweet was indeed “criminal” comes down to three parts:
- Did Elon buy or sell Tesla shares in conjunction with the Tweet?
- Is Elon’s Tweet a legitimate substitute for something like a K8-filing with the SEC. Was information distributed in line with the “spirit” of SEC regulations?
- Is the phrase “Funding Secured” misleading information for investors?
The first question (although speculative) is probably answered rather quickly. No, I personally highly doubt that Elon would risk it all to make a few extra million with a blatant pump-and-dump scheme. Although…Elon has been known to run a lot of those “Ethereum giveway” schemes lately 🤔
All joking aside, let’s leave this question for people with the legal right to subpoena bank-account statements to investigate.
For the second question I would love to refer to this short summary post in Business Insider in which Nimish Patel is quoted on his expertise in the field. His conclusion is that SEC regulations state that
“Within 24 hours of making a disclosure to a select group of people that could affect shareholders, companies must make sure that information is “broadly disseminated,”
and as such Elon’s Tweet can be seen as having achieved that goal.
Which leaves the third and final question…
Is the phrase “Funding Secured” misleading?
If a Tweet could be seen as being of “equivalent” value as a K8 filing, then did Elon Musk portray false information to investors by saying that funding was secured. Especially since it was discovered after the tweet that no agreement had been signed yet but that Elon was only in negotiations with a potential buyout partner.
“I left the July 31st meeting with no question that a deal with the Saudi sovereign fund could be closed, and that it was just a matter of getting the process moving. This is why I referred to “funding secured” in the August 7th announcement.” — Elon Musk
The controversy arose around the word secured. A lot of people thought it meant a contract had been signed. And indeed if we look at the definition of the word in a legal dictionary we find that:
Oh dear! So for Elon to avoid jail time we need to see if there is enough of a legal precedence to interpret his words to mean “we are extremely confident about funding” instead.
Luckily we had already built some technology to analyse legal text (for our app Donna). To help with this analysis we trained something called Word Embeddings. This means you teach a Neural Network the meaning of a word by having it predict other words that are likely to appear in the same sentence.
As humans, by looking at “I petted a cat” and “I petted a dog” we can easily see that “dog” and “cat” must be somewhat similar because we can both “pet” them. If we were asked to assign a unique number to each word so that similar words are close together we would keep the number for “dog” and “cat” close together (but not quite identical, since most people don’t “walk” their “cat”)
This is essentially how we trained our AI as well. But instead of just one number we pick about 450 numbers. These numbers have been shown to be a rather accurate representation of the semantic meaning of a word.
For instance if we take the numbers for the word “letter” and subtract the numbers for “paper” and add the numbers for “computer” we get almost identical numbers as the word “email”. Doing math with words! 🤯
letter -paper + computer = email
By combining historic SEC filings and our proprietary dataset of legal texts and agreements (given by a few data partners) we had over 100GB of legal text to analyse. This gave us some extremely accurate legal Word Embeddings that learned to capture the semantic meaning of each legal term. Here’s a rather beautiful picture to visualise the result:
By looking at the differences between the numbers of each word we found the following conclusions:
- The Webster dictionary definition seems to be correct as the word “secured” has about an 80% similarity with the word “guaranteed”. But…
- There are quite a few words that have a less stringent definition that seem to also be used in the same “spirit” as the word secured. Therefore one could argue that, based on our dataset, in this context the word “secured” could reasonably be interpreted as “pledged (78%)” or “indebted (74%)”.
So it seems that Elon’s statement, although not in line with a dictionary definition, actually has quite a bit of precedence in legal text and other SEC filings.
Final Notes
I want to stress that these similarity-scores and explanations are simplified for the purpose of this article. We have actually been running more extensive (and scientific) tests and might post an in-depth article if there is interest…or if Elon actually needs some help convincing the SEC.
I also want to disclose that at the time of writing I personally own shares in Tesla but have no further affiliations. Nor does our company have any affiliations with Tesla.