Hot Topics

Can AI choose your investments?

Go to John Beveridge author's page
By John Beveridge - 
Artificial Intelligence AI investments stocks ChatGPT

There seems to be many skills that Artificial Intelligence (AI) can bring to the table but how does it fare at picking investments?

Well, it seems that it does very well indeed, effectively replicating the task of picking stocks and sorting through a raft of confusing information.

That is perhaps not too much of a surprise – after all, simple computer models such as ETF index funds and even quant investment algorithms have been around for a long time and in many cases have outperformed the majority of human stock pickers.

However, the latest experiments using ChatGPT-4 have shown that AI can even excel at more complicated tasks such as combing through earnings revisions, setting price targets and even making buy, sell and hold recommendations.

Early research shows AI stock picker can beat the market

A team of University of Sydney academics ran the generative AI program across ten years of data from some of the 200 largest companies on the S&P 500 and found it could match or perhaps do better than an expert analyst.

While the research paper is early and has not been peer-reviewed yet, it showed so much promise that the group of academics – PhD student Jason Ming, lecturer Hamish Malloch and professor Joakim Westerholm – put it out in a preliminary form while the research continues.

Macquarie’s Australian equities analysts checked out the research and said the results: “show that ChatGPT’s insights are consistent with analyst behaviour and outperform traditional analyst metrics, allowing for the creation of portfolios that earn abnormal returns.”

Capturing and monetising nuances a key finding

One of the biggest surprises was the extent to which the ChatGPT program could capture nuances in company results and commentary that is valuable in working out where the company’s asset allocation or strategic direction is heading.

It was this skill at detecting nuances and trends that allowed the AI stock picker to outperform a human analyst by 40 to 80 basis points a month.

Mr Ming said the findings from the research which started just before the official release of ChatGPT-4 in March last year showed that ChatGPT could “think like an expert, like an artist”.

While there is obviously a long way to go, it could be that AI will start performing a lot more of the grunt work by automating functions and analysing data while the human analysts are freed up to look for unpublished information or trends to analyse in the search for a competitive investment edge.

Motivation for questions the key

The research used hundreds of earnings calls to train ChatGPT to watch out for the motivation for analyst questions and find out if they were extracting information about merger and acquisition strategy or market expansion or new products.

By coming up with a score for every company, the program then sorted stocks into a portfolio that outperformed the human analysts by 40 to 80 basis points a month – which is a significantly better result if it can be repeated.

What the paper did find is that there is a mine of information released during profit results and earnings calls and that the AI Model was better at picking through all of this information and coming up with a better portfolio.

Interestingly, the approach worked well across all sectors.

Would AI have chosen Nvidia early?

Who knows, maybe it also would even have detected the rise of AI as a stock mover and loaded up on star stocks such as chipmaker Nvidia which has rapidly grown into being one of the most valuable companies on the US exchange.

Or, more impressively, could it have predicted the recent heavy falls and subsequent recovery in Nvidia’s share price?

This model is obviously in the early stages of being developed and needs plenty more testing and more rigorous repetition and peer review but the promise for time strapped brokers is obvious.

Rather than trying to stay across several stocks that are having earnings calls on the same day, it would be possible to leave the AI program reviewing many of them and then check over AI generated reports from the other meetings before compiling an overview.

Which AI program would win?

AI is certainly ushering in a brave new world but it also brings up the possibility of competing AI programs coming to different conclusions about which stocks to buy and when.

Like any stock selection process, it can only be as good as the information fed in and the reasoning behind the conclusions reached, so competition between AI stock pickers would certainly be a likely response.

Even then, it could be difficult to appraise the results if one stock picking program worked well in the short term while another was more plugged in to longer term trends and results.

As always, progress in AI has been very impressive but at this stage some degree of human common-sense appraisal is still needed to account for the risk appetite and investing horizon of the fund that is investing money.

However, AI has a lot to offer for improving the efficiency, cost and returns of stock selection – a role which can also be applied more generally in improving productivity across the board.