Do predictive AI models belong in investment strategies?

Artificial intelligence has been a major disruptor across industries and their respective workforces, causing quite a stir at the mere mention of “AI anything.” But is AI up to the task of making calculated predictions on the stock market (^DJI, ^IXIC, ^GSPC) and should investors trust those forecasts?

Kaiju Worldwide Global Chair and CEO Ryan Pannell joins Market Domination to discuss predictive AI models’ place in public and private investment strategies.

“It’s not well-suited to that global macro [predictions],” Pannell explains. “Immediate short-term prediction on price action in equities, derivative: Yes, it’s very good at that. But the longer-term stuff, no. And I don’t see that changing,”

For more expert insight and the latest market action, click here to watch this full episode of Market Domination.

This post was written by Luke Carberry Mogan.

Video Transcript

It’s been another A I centric earning season as investors track the next best move in the A I trade.

But what if instead of just investing in A I traders could also use A I to power other public and private investment strategies we’re here to discuss is Kaju worldwide, CEO and global chair, Ryan Pannell Ryan.

Good to have you here.

Thanks, thanks for having me back and it’s nice to be here in person this time.

So, so Ryan, let me ask you maybe just to start your focus is predicted A I correct, right?

As as opposed to gen A I, which we spend a lot of time talking about in this show, right?

And all kinds of companies spending a lot of time, money and effort there maybe start there.

Ryan explain to viewers kind of the differences between the two and then how Kaiju is is really integrating that tech, right?

So I think as you say the the A I that most people are familiar with this generative A I where you’re taking uh a non defined non standard data set in most cases, the internet uh where your ownership of the data to train the models isn’t necessarily well defined.

You’ve got free use versus copyright infringement and you’re trying to answer like any question that somebody could answer and, or ask and then create something new out of it.

Paint me a picture, write me an essay.

Make a you so make me a movie and you get problems with hallucination that come with that.

I mean, no question.

It’s a very powerful technology with a bright future.

Predictive A I is using either easily purchased or collected at source data for us price time and quantity, for example.

And it’s trying to answer a specific question.

So in our case, it’s trying to identify patterns in price time and quantity that immediately precede a price action that we want to take advantage of and we see this over and over again and there’s a reasonable amount of certainty and we can use it to to that end.

Story continues

And so basically, it’s all, it’s not fundamental analysis at all it is just based on.

So it’s effectively predictive A I based on technical analysis to some extent, if you wanna share the the secret sauce, wii I knew that was coming.

Um If you think about it, every positive investment decision results in a transaction, right, it doesn’t matter what the asset class is, you know, real estate derivatives, equities options, you know, you do your analysis, whether it’s fundamental technical, there’s an investment committee and at the end of the day.

If it’s go ahead, you’re going to affect a transaction and we can see the pattern in the transaction, not always, but for specific transactions, you can with some degree of certainty, reverse engineer the investment decision that resulted in the transaction example would be, you know, if you’re watching rotation versus distribution, they’re very different intentions.

And so they are very different patterns, panic looks different, you know, buying quietly because there’s you, you have some certainty there’s going to be a positive earnings and you don’t want to disturb price versus buying indiscriminately because you think in five years the stock is going to be more positive, different patterns, different outcomes.

If you can identify them, then you can predate on them.

I’m I’m interested.

So, you know, it’s interesting, right?

You, you, so you found some tasks where you think?

Ok, predictive A I in my world, it doesn’t make sense.

Are there ones where you think, you know what this tech, this tech is not up to that task ones you wouldn’t want to see it apply to?

Yeah.

Sure.

Absolutely.

Obviously anything with a long term time frame.

I don’t care if it’s human or machine, there’s no person, there’s no system, there’s no predictive A I that’s gonna tell you with any certainty where a stock is going to be in six months from now and anyone who says different is just trying to sell you something.

So it’s not well suited to that global macro, you know, Putin invades Ukraine, what will be the global economic response to that over the next 12 months?

There’s no system that’s going to do that immediate short term predation on price action in equities derivatives.

Yes, it’s very good at that, but the longer term stuff and I don’t see that changing.

So you made a comment at the beginning when you drew the contrast with generative A I and it sounds like you’re a little bit skeptical of that models, the, the way that it collects information or its ownership or lack thereof of the data.

Correct?

So I’m curious your view on whether you think there should be rules around those kinds of things when it comes to gender A I?

Well, I mean, that’s a, it’s a big question to ask and answer in terms of the capacity to potentially answer any question that you might ask.

No, it, it needs to have this latitude.

These large language models are enormous and obviously incredibly costly to run because you can’t determine what someone is going to need, require or want to review it can be information collection, it can be creation.

So you’re going to have to allow it a lot of latitude with predictive A I, whether it’s flying a plane or driving a boat or trading stock, you needed to do one specific task with a high degree of certainty.

And so the hard landscape surrounding those data is important and thankfully that aligns with that type of technology, but for the larger systems, I don’t see how you constrain the data and achieve the same, same performance Ryan.

Thanks for coming in.

My pleasure.

Thank you.