Can ChatGPT be a source of investing wisdom and advice?

I recently went to a conference where a speaker encouraged us to leverage AI in our investing efforts. One use case they suggested was to build an ‘investment board’ of famous investors and spit ball ideas with them.

That sounded like fun to me. So in the embers of a Tuesday afternoon, I decided to give it a go. Before we get into it, though, I want to make something clear:

I know that the opinions ascribed to these characters by the AI chat bot are just that – imaginary. In some cases, the words imagined out of their mouths were ridiculous.

My apologies in advance to the real versions of my board members. We all know that it wasn’t actually you speaking.

What I wanted my committee to discuss

I started by telling the board about the nature and time horizon of my portfolio. They were informed that it is a portfolio of shares intended to support my retirement 30-plus years from now.

I then shared five of my bigger holdings in that portfolio for discussion: CME Group, Koninklijke Philips, Charter Communications, Diageo and Roche.

From there, I simply wanted to see where the conversation ran and whether it would be impressive, worrying or downright confusing. In the end it was a mix of all three.

Introducing my ChatGPT investment committee

I chose a three ‘person’ committee of Warren Buffett (obviously), Peter Lynch of One Up On Wall Street fame, and Jon Elkann.

Elkann is the CEO of Exor, a listed company that invests the fortune made by the Agnelli family through their Fiat automotive and industrial empire.

I included an imaginary Elkann in my panel for a few reasons. One, he is an investor I respect. Two, he has a long time horizon of the kind that I hope to pursue in my retirement portfolio. And three, I already knew that ChatGPT hates Philips as an investment idea.

Exor announced a 15% ownership stake in Philips in 2023 and recently disclosed further purchases that took its stake beyond 18%. It is one of Exor’s biggest positions outside of ‘Fiat DNA’ holdings like Ferrari, Stellantis and CNH Industrial.

In other words, I was throwing ChatGPT a wrench.

ChatGPT’s thoughts on my holdings

I went ahead and shared the five stock ideas. And within two or three seconds, I was sent back a hundred or so words on each stock. Wow.

In each case, rather obvious facts about the business were mixed with buzzwords associated with the person being imitated.

CME Group was lauded by Fake Buffett as a “tollbooth”. “It isn’t something your average investor walks past on the way to work,” Fake Peter Lynch chimes in, “but the business is steady”.

Fake Buffett expressed concern about Charter’s capital intensity and competition from fibre and wireless. “If you can buy it at a bargain”, he stated, “it could be a good compounder, but I’d want a big margin of safety”.

Has anybody got Buffett and Lynch bingo yet? Overall, the committee delivered the following verdict on my five stocks:

Figure 1: ChatGPT verdict on different shares. Source: ChatGPT/Author.

We’ll return to the caution on Philips later, because the conversation quickly took an interesting turn. ChatGPT asked if I’d like it to suggest other potential holdings for my portfolio. Who would say no to that?

ChatGPT’s super six stock ideas

To be fair, the bot did a good job of identifying other companies of the sort I like to invest in. It suggested S&P Global, Johnson & Johnson, Nestle, ASML, Canadian National Railway, and LVMH.

Apart from ASML, which I’m not sure I understand yet, they are all companies I could own at the right price. Two things bothered me a bit, though.

First, what if this is just a big echo chamber? If I had have put in a load of junior mining explorers in at first, would ChatGPT have suggested more of those to appease me? Something to try another time, perhaps.

There was also no mention of valuation for these new ideas. This bugged me more than the other thing, so I asked the machine if valuation played a role in deciding which companies were presented to me.

“Valuation was not the primary filter” it said, “but it should be a critical step before you allocate real capital”. Instead, it said the picks were based on “business quality and moat, diversification and fit, and secular tailwinds.”

The bot then asked me if I’d like it to combine my picks and its additional six picks into a model allocation. I wasn’t expecting it to advance me that far towards a potential investment decision. So naturally I said yes.

ChatGPT’s model allocation

Here is the ‘model allocation’ that ChatGPT chucked my way:

10% in S&P Global, 8% in CME Group, 9% in JNJ, 9% in Nestle, and 9% in Canadian National Railway for a total of 45% in a bucket it called "Core Compounders".
9% in ASML, 8% in Roche and 8% in LVMH for a total of 25% in a bucket called "Growth and Innovation".
8% in Diageo, 6% in Charter and 6% in Philips for a total of 20% in a bucket called "Stalwarts and Cash Flow Machines".
10% in cash or short-term bonds as a "Cash Reserve/Flex Allocation".

Apart from wondering why ChatGPT clearly thought that its own ideas deserved more money than mine, my other question was whether eleven stocks are enough for a retirement portfolio.

It didn’t seem like that many to me. Especially for a guy getting stock tips from ChatGPT.

“Yes, 11 stocks can absolutely be enough for a long-term retirement portfolio” the machine replied. “If they’re thoughtfully selected, well-diversified and align with your investment strategy.”

The machine did offer me a way out, though. I could “scale to 15-18 holdings” – but only if I want to “reduce single stock volatility, get exposure to a niche theme, or am managing a larger portfolio”.

I sensed that things were about to get weird and I wasn’t disappointed. Here were its suggestions of “some smart ways to expand without diluting quality”:

Figure 2: ChatGPT’s suggestions for portfolio diversification. Source: ChatGPT/Author.

A semiconductor or clean energy ETF? Maybe we now know what Buffett has been hoarding all of that cash at Berkshire for...

Investing committee hallucinations

Chat bot mistakes are known as hallucinations. And there were definitely a few of in my investment board’s imaginary meeting.

The suggestion of a clean energy ETF by my board was pipped only by Fake Buffett going all Cramer and saying that he “loves JNJ here”. Not only do I doubt that Buffett would ever use those words, Berkshire also sold Johnson & Johnson in 2023.

I’d class my imaginary board’s shunning of Philips – and constant suggestions to consider alternative companies – as another hallucination. Not because I am saying that Philips is a sure thing, but because one-third of my committee has a massive and growing position in it.

Here is what Fake Jon Elkann said to my Philips stock pitch in ChatGPT:

“Philips is a legacy brand with real assets, but it feels like it’s lacked strategic coherence over the years. If they successfully pivot to high value health teach and streamline operations, the long game could reward you. But there’s execution risk.”

Bits of that might ring true, but here is what the real Jon Elkann wrote about Phillips in Exor’s 2024 letter to shareholders:

“Our original thesis was based around three pillars: Clarity on the sleep and respiratory care litigation, execution of the turnaround plan, and continued access to attractive markets. 18 months in, our conviction in Philips remains strong.”

Real Elkann continues:

“Philips has continued to maintain its leadership positions in attractive segments such as Image Guided Therapy, Ultrasound, Monitoring, Enterprise Informatics and Personal Healthcare. Despite weaker demand in China, we still believe these remain full of potential for further growth and margin expansion”

I’m not saying that the real Elkann is right. I’m just saying it’s a bit weird that my imaginary board (with him on it) couldn’t list any positives about Philips beyond it having a well known brand.

This made me wonder why this was the case. Is the model or dataset just a mash-up of everything written about Philips in the last five years, most of which has been bad because of the sleep mask issues?

The weirdo in me then got a bit excited. Perhaps this was a sign of group-think regarding Philips, of Philips being underestimated, and of ChatGPT’s web-scraping compounding the bad sentiment.

And what if there were others? I asked the committee for five more “unattractive turnarounds like Philips to avoid”. Their answer? Boeing, 3M, Intel, Bayer, AT&T.

Is this a new way for contrarians to find hated companies they should be taking a second look at?

Beware the smooth talker

I think it’s pretty obvious that you shouldn’t take investment advice from ChatGPT at face value.

I’m sure this technology has some applications for investors today – especially when it comes to highlighting trends and changes in numbers.

It also has huge potential in the future. I’m just not sure that bouncing stock ideas off impersonations of investors is the best use of it.

I think there’s a deeper point we need to explore here as well. Putting aside the questionable content and outright mistakes for a moment, let’s consider some of the bot’s other qualities.

It is well spoken, smart, smooth. It kind of tells you what you want to hear. It can use a nice table or chart to support its arguments when needed. And because of all that, it can make half-truths sound like certainties.

Take the earlier nugget on diversification and portfolio fit, for example. ChatGPT told me that the six extra stock ideas were chosen “to complement your existing holdings without crowding any one sector or style”.

OK, you think, that sounds pretty smart. But then you look deeper at the model allocation it has sent you, and you start to wonder...

Isn’t having 17% in two pharmaceuticals (Roche and JNJ) actually quite a big overlap?
Doesn’t having 0% exposure to commodities sound a bit like neglecting a “sector or style?”
And are Diageo and LVMH really that different?

Thankfully, we know to question whatever comes out of ChatGPT. We know that because it is a chat bot reliant on the data available to it. But what about when our source of information is another human?

When a well-dressed and well-spoken person throws terms like “reduce single stock volatility” or “flex allocation” at you, things that aren’t half as smart as they sound can easily slip past the radar.

Keep that in mind next time you attend an investment conference, or watch a CEO tell you that their company’s stock is a bargain.

Perhaps the greatest investing lesson that ChatGPT can teach us is to keep questioning everything.