Hacker News Reader

Comment by roenxi | Hacker News Reader

We live in an era of 2.0s. This is probably going to turn out to be Phrenology 2.0.

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.

View on HN

Of course they can not predict the future in the strong sense, but they may be able to extract some fraction of a bit of information about the distribution of possible outcomes, slightly reducing the variance. This is enough for a hedge fund to make money. Similar to knowing something like how full a department stores parking spots have been over the last quarter. It won't give you an accurate prediction, but it can give you a slight edge over the competition.

It isn't even a settled debate whether CEOs know what is going on or are a particularly important driver of outcomes.

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.

A lot of things have been studied, the question is what evidence they have that it actually works. It seems similar to one of those things where people have made a lot of claims with only dubious evidence.

Warren Buffett has a famous saying: "I always invest in companies an idiot could run, because one day one will".

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.

Sure tell that to theranos and Enron.

A criminal ceo will often lead to the dissolution of a company so I don’t really see how this follows.

Theranos never had a real business. So it didn't dissolve so much as never existed. Sure, the legal entity was created and dissolved, but there wasn't ever anything approaching a real business.

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.

What’s your definition of a “real business”? Enron had revenue of over 100bn, 30k employees, and delivered actual products and services for many years. How’s that not “a real business”?

In my mind there are multiple types of "not real" business. Two obvious types:

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.

This sounds like AI hucksters justifying their existence… perhaps we’re moving into a later stage in the hype cycle.

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.

It feels like people have been predicting the end of the "hype cycle" for the last decade.

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.

Speaking of phrenology when I was still in the investment business we were trained in the “facial action coding system” which purportedly teaches one to read the facial ticks of your conversation partner to see if they are being honest. I’m not really sure there is a signal there, but it was fun to learn, and fairly trendy in the industry at the time.

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.

Exactly - besides this technology will effectively be moot in the longrun :

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.

or more actively:

CEO ML team reverses model used by Investors and uses it to write own speech

Your CEO brain calipers don’t need to be worth a shit in order for you to make money with them. If the market moves predictably every time you publicly announce your findings, that’s all you need.

The example in the article was a case where information was being hidden. There was a mismatch between content and tone "Everything's fine!" [I know everything is not fine]. Humans can already pick on a mismatch between content and tone. We call it intuition. The things that trigger intuition are quantifiable. Using Machine Learning to systematize things we already do is pretty sane and reasonable.

They don't have to be a driver of outcomes, just an indicator.

Or perhaps, an instigator.

meh, you sre thinking like a 1.0.

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

Reply on news.ycombinator.com