Investors use AI to analyse CEOs’ language patterns and tone
reuters.com
reuters.com
It isn't even a settled debate whether CEOs know what is going on or are a particularly important driver of outcomes. It is unlikely that NLP models can foresee the future, even if bankrolled by hedge funds.
It's pretty clear that a negligent/criminal CEO will lead to the bankruptcy of a company and a loss for investors. So yes, you can say at the extreme end that CEO's are drivers of important outcomes.
Now whether it's effective I think is another subject all-together. My own opinion is that micro-expressions, body language, etc. have been studied and used by the FBI/CIA for their field work. The idea being that once you establish a baseline behavior, you can notice "clusters" that deviate from the baseline based on verbal+non-verbal cues. So I don't see why an AI couldn't do the job.
There are many companies where it's likely that a negligent CEO (even a criminal one) wouldn't lead to the bankruptcy or even close. Many companies might be better off with a negligent CEO rather than one actually trying to do much. Right now Google could be run by a person on the beach sipping margaritas. Many other big public companies with very strong competitive positions probably could too.
A criminal ceo will often lead to the dissolution of a company so I don’t really see how this follows.
Most of Enron (some was actually ok) wasn't a real business either.
It follows like this:
A great business can often be run by a fool and be fine, because the business is just so good. Think Google, Coke, and almost all newspaper companies in the 20th century.
The opposite is not true - a terrible business is hosed no matter how good the CEO is. Think farms, most retail businesses, newspapers in the 21st century.
And most companies sit somewhere in the middle - the CEO can make a difference.
i) A business where the true state of it is being hidden to raise money to keep it afloat and it's not sustainable without raising said capital. Enron did this, raising lots of debt financing. Theranos did it with equity.
ii) A business which isn't sustainable over any time period without external money to keep it afloat. It need not be fraud, it could just be stupidity on part of investors, executives. Many internet bubble businesses were this type of "not real".
Enron had a lot of businesses under the corporate umbrella. Some of them were real (they owned hard assets, pipelines, energy generation assets). However, a lot of the revenue from other businesses was fake - derivatives revaluation accounting tricks (like, "hey this derivative contract is now worth $50 mill more because of some analysis we did, up goes revenue"), debt hidden via special purpose entities, and other similar things. The "not real" part of Enron was so big that it's debts brought down the rest of it. The chapter 11 process sold the real assets and the creditors got some money back and some employees stayed with those business.
I suppose you could have a portfolio of AI tools that would alternate between telling you “stonk go up” and “stonk go down”. Blame the customer for choosing the wrong one if they are unhappy.
I agree that this particular application seems... questionable, to say the least, but finance is probably the last place where data scientists need to work very hard to justify their efforts. Statistical modeling has been a core part investing for a very long time, and ML is just a subset of statistical modeling.
Of course we also did nlp to identify companies that were using language associated with negative returns in their quarterly conference calls. This was very successful, mostly by identifying pieces of shit we had not heard of yet. Once identified there were usually much stronger red flags. But the performance of that strategy was good. It definitely was not phrenology. Around 2001-2005.
If it works well enough, firms will put their CEOs speeches through models to ensure that their speeches rank well, and that will cause it to fail.
CEO ML team reverses model used by Investors and uses it to write own speech
what matters isnt what their intent is, what matters is how the greater fools will react.
as such, figuring out what the fools will do can lead to successful pumping and dumping, adding more liquidity skimming.
they did that successfully with trump
If I recall: Steve Jobs rarely attended IR events, Jeff Bezos just sent a letter in advance (note: beautifully written), but barely made live comments. Future guidandes are often released as `something between +30% increase or -30% decrease`, meaning useless.
My impression is that nowadays most PRs and releases are already meticulously reviewed. Quite an interesting paradox: The more regulations, rules, controls and surveys are introduced less detailed and exciting releases and disclosures are..
Interesting example of what I see as counterintuitive behavior, which brings up a serious question.
If companies are prevented from making anything other than dry financial disclosures every quarter, how is a normal individual investor who is not an insider then supposed to judge investment opportunities?
Personally, I feel like the markets would operate better if companies, big shareholders, etc were always 100% free to make public comments without worrying about the SEC breathing down their neck micro-analyzing every statement. As an investor, I want to hear more from companies, not less.
That’s my point. I’d like to hear more forward looking statements of the kind that the SEC frowns on.
Of course these insights are statistical in nature, not definitive.
The fantastic aspects of such "solutions" for the stock-market is, that they're probably as reliable as most other solutions, since they all try to conclude something from incomplete and ambigous data-points.
Their purpose is probably not to be fully accurate, but just to give more comfort when taking the final buy/sell decision.
We're probably already in the first stages of that journey, complete with the unreliability and all.
All of the stuff that is being signalled here can also be worked out by just reading the financial reports (the issue with a lot of these models is that information leaks through...ofc, companies that announced downsizing are more likely to go bust...if you need NLP to work this out, finance is not for you...the example of semis is the same, everyone knew that ppl were uncertain about supply, you didn't need NLP for that). The only reason these results are in any way remarkable is because academics convinced everyone that there is no "signal" in conference calls...right, everyone who has ever invested money knows this is false, this isn't surprising.
And, just like value factors, this will go badly wrong because computers can't actually value businesses by themselves, they can't do fundamental analysis. Attempting to shortcut this is not smart.
Consider me a skeptic.
AI NLP models have way too many false positives and negatives for this to be workable. Maybe in 10 years, but definitely not now.
Perhaps the reason why detecting rascism or hate speech is so hard for an AI is that what is considered racism or hate speech is a moving target.
Off the top of my head, racism is often so casual that stochastic AI models may have difficulty discerning a difference in syntactic structure or other linguistic features compared to a similarly casual statements of a non-racist nature.
But you're right to be sceptical. Hedge funds are largely a sales job. How can you convince gullible investors to overlook the terrible performance of your sector and invest in you anyways, and then take a large slice of the profits when you get lucky? Buzzwords and neat sounding strategies help.
"...we went from a negative growth in Q4 in storage revenue on a year-over-year basis. We now have flat..."
"DELL – Q1 2022 Dell Technologies Inc Earnings Call"
https://investors.delltechnologies.com/static-files/d70442f3...
Some NLP related problems can be solved at a similar level of quality of humans - (humans actually make a lot mistakes as well).
This is just wrong. State of the art NLP models are pertectly able to identify racism and hate speech if they are trained to do so.
The issue is just that most of the times political correctness is not in the training objective. In fact, general language models are trained to reproduce what they read as closely as possible, hence the racism/hate speech.
Again, the problem is how they are trained, not what they could/could not achieve.
> AI NLP models have way too many false positives and negatives for this to be workable.
Citation needed.
why wouldnt it be possible to train on it?
Why, what does it say? Can you log that in a reproducible Notebook with Docs and Test assertions please?
Or are we talking about maybe a ScholarlyArticle CreativeWork with a https://schema.org/funder property or just name and url.
Maybe we'll end up preferring text communication. About bloody time, voice chatting is needlessly overused. But I'm sure extroverts will solve that another way however it suits them.
So the people who are good at faking as well as reading those signals get an additional benefit over those who aren't.
That already exists, it's called the financial markets and there are daily battles between competing quant strategies.
But NFTs can.
In my humble opinion, it seems kind of unlikely that AI will outperform people in solving this problem.
You talk into the phone and it removes any trace of sentiment or tone or better still allows you to choose and then plays that version out to the listener.
IKR.
Pennebaker thinks so. He's a big proponent of "computational linguistics".
https://en.wikipedia.org/wiki/Computational_linguistics
I recently read his book Secret Life of Pronouns.
http://secretlifeofpronouns.com
I thought it'd be fun to replicate the book's claims. Like maybe analyze political speeches.
But Pennebaker's current rulesets apparently aren't available. And I didn't have the gumption to figure how to use the tools which are available.
Another book about truthiness is Everybody Lies by Seth Stephens-Davidowitz. https://www.amazon.com/Everybody-Lies-Internet-About-Really/...
There's got to be some overlap, right?
However, the same hedge fund consistently beating the index fund, that's unlikely. But things like ARKK certainly do exist and are funds that can beat the market for a few years, then they tend to regress to the mean.
https://www.reddit.com/r/market_sentiment/comments/p0emqj/do...
- Focus on a particular sector, if a biotech-focussed hedge fund is up 4%, SPY is up 7% and biotech stocks in general are up 1%, they have still outperformed. Clients will want exposure to that sector.
- Tail-risk hedging, losing small amounts of money most of the time to make huge amounts during unlikely events.
- Lower volatility, e.g. a fund which underperforms SPY slightly, but hedges against dramatic downturns in the market, so timing is not as important for redemptions.
Sophisticated investors (the only people allowed to invest in hedge funds anyway) generally have more complicated requirements than throwing their money at an index fund and waiting decades to retire on the returns.
ETFs/Hedge funds can have particular features, like equity protection or exposure to a certain market or markets, instead of growth. You wouldn't expect a bond fund to out perform the total market index, that's not it's purpose. It's purpose is to protect your wealth in case of a downturn.
Similarly, some hedgefunds (private equity and arkk are probably two good examples) that seek to find that rare 10X company that they can make millions or billions from, and are okay with generally losing most of their bets.
"The Renaissance Technologies Medallion Fund has produced some of the greatest returns in the history of the markets. "
I was curious about how much money Renaissance manages: it’s about two orders of magnitude less than Blackrock.
RTC manages $165 billion Blackrock manages nearly $7 trillion