Back in February, I “tricked” ChatGPT into picking stocks to buy. Using creative prompt engineering, I nudged the ordinarily tight-lipped GPT-3.5 version to create a list of 30 stock picks.
The recommendations were surprisingly good.
The chatbot created a list of growth, value and short stocks that essentially distilled the knowledge of the internet. Any investor who followed along would have returned 7.2% in the past two months!
But the list lacked a sense of timely insight. ChatGPT could only “see” news through September 2021, making its recent performance more about luck than skill. The same list of stocks badly underperformed in 2022. The artificial intelligence (AI) chatbot also lacked any notion of risk management. An investor following AI’s recommendations since February would have seen their margin accounts blow up when picks like AMC Entertainment (NYSE:AMC) jumped on a short squeeze.
GPT-3.5 was also limited to picking popular stocks on the internet. Not one contrarian choice showed up in the entire list.
So, when ChatGPT’s “GPT-4” upgrade was released last month, I was excited to try again.
And it didn’t disappoint.
Can AI Pick Stocks? GPT-4 Says Yes.
The introduction of GPT-4 and competing services like Alphabet’s (NASDAQ:GOOG, NASDAQ:GOOGL) Google Bard adds three essential elements to AI investing.
First, they’re far happier to give answers. This time, no fighting, negotiations or prompt engineering was required to coax the AI to give stock suggestions. I assume the search race between Microsoft (NASDAQ:MSFT) and Google has caused every player to lower their guard on that front.
Second, the chatbots can now “see” current news. Recent upgrades now allow GPT-4 to browse the internet, making its picks far more timely than before.
Finally, the newest version of ChatGPT seems happier to defer to human insight when it can’t figure things out, reducing the amount of hallucinations.
Still, GPT-4 has some shortcomings.
Because let’s take a look at what it said…
The 13 Favorite Stock Picks of ChatGPT
On the tech side, ChatGPT was quick to offer several promising high-growth companies:
- Amazon (NASDAQ:AMZN)
- ASML Holding NV (NASDAQ:ASML)
- Intuitive Surgical (NASDAQ:ISRG)
- Meta Platforms (NASDAQ:META)
- Palo Alto Networks (NASDAQ:PANW)
- ServiceNow (NYSE:NOW)
On the one hand, these companies are actually quite compelling. Names like Amazon and Meta trade for fractions of their former values and the latter is seeing a resurgence in digital ad spending. The most recent version of ChatGPT also avoided only picking popular mega-cap companies; ASML and ServiceNow are smaller firms with excellent growth rates. If an intern came to me with this list, I’d try to negotiate a full-time deal.
On the other hand, though, GPT-4 had no idea why it was picking these stocks. When asked to elaborate, it could only give the same canned responses:
- Amazon: a “massive and diversified customer base.”
- Meta: a “massive and engaged user base.”
- Palo Alto Networks: a “large and loyal customer base.”
And so on.
In other words, GPT-4 missed every nuance of why it did anything. It’s the same problem that all large language models (LLMs) seem to have — they’re only designed to predict text, not understand it.
That meant ChatGPT had no idea of fundamental analysis or using a proven strategy. It was only a rehash of what others had already said before. Most alarmingly, the AI still failed to understand valuations. Most online stock analysis skips past their discounted cash flow (DCF) models, which means GPT-4 will always do the same.
More Stock Picks From ChatGPT
Then there were ChatGPT’s other choices.
In energy:
And in miscellaneous industries:
Again, these are relatively strong picks. For energy stocks, Devon Energy and Exxon are low-cost hydrocarbon producers, giving them a significant advantage over their higher-cost peers. Historically, these firms tend to outlast competitors, leading to substantial long-term outperformance.
In non-energy stocks, H&R Block is also a solid enterprise priced at historical valuations; I estimate the firm has a 30% upside over the next three years. And Meritage Homes is an above-average homebuilder trading at below-average prices.
But the same issues of insight persist. The AI chatbot gave “strong financial performance” as the key reason for buying Devon, Exxon and Meritage Homes. Only Duke Energy garnered a different response, with ChatGPT calling its financial performance “stable” instead of “strong.”
Not exactly a compelling reason to invest.
Can You Us AI to Pick Stocks?
The financial industry is no stranger to using machine learning to trade stocks. High-frequency traders profit from minor pricing differences across exchanges. Their algorithms also help predict moves in the milliseconds following specific trades.
Meanwhile, quantitative traders use every imaginable shred of information in their high-powered models. Some use investor sentiment on social media, while others use weather data to help predict crop outputs. Even the weather at the New York Stock Exchange has some predictive ability on stocks.
ChatGPT is an extension of this strategy. Rather than rely on fundamental data or complex models, the AI chatbot mines the internet for insights into what others are writing about.
Usually, it’s a surprisingly powerful way to invest. Companies with high share prices can raise cheaper capital, making it easier to fund projects. It’s a self-fulfilling prophecy that benefited startups like Tesla (NASDAQ:TSLA) and Amazon.
But it’s still only a regurgitation of the internet. As compelling as ChatGPT might sound on paper, it’s only as insightful as the web articles it’s hoovering up. Until someone decides to plug GPT-4 into a hedge fund’s algorithm, investors must continue making their own investment decisions.
On the date of publication, Tom Yeung did not hold (either directly or indirectly) any positions in the securities mentioned in this article. The opinions expressed in this article are those of the writer, subject to the InvestorPlace.com Publishing Guidelines.
