(News Bulletin 247) – By insisting a little, News Bulletin 247 had asked the conversational robot in January to give five American stocks and five Parisian stocks on which to invest over three months. How have these titles evolved since then?

Important disclaimer: we have engaged in a purely theoretical exercise with ChatGPT. In no way do we recommend investing on the basis of advice given by an artificial intelligence, nor choosing such a short horizon of three months. The usual advice (be well informed about the securities in question, diversify your portfolio, focus on the long term) prevails.

Since its launch, it’s an understatement to say that ChatGPT, the conversational robot of the American start-up OpenAI, has been used for just about anything and everything. As our colleagues at Tech&Co wrote, students at Cardiff University used it to write an essay, earning them top marks. Another example: France 3 regions, who had questioned him to find out which was the best city between Tours and Orléans…

What about the stock market? Just for fun, we asked ChatGPT on January 24 to give us some stocks to invest in over a three-month horizon. Note that we had to insist then, and that recently, during another test, the conversational robot had refused to answer us, stressing “that as an artificial intelligence [il] does not have the ability to predict or recommend specific actions or provide financial advice.” Which is rather reassuring.

But in January, forcing a little (a lot), ChatGPT had finished to “give” us five stocks on which to invest over three months on Wall Street and five on the Paris Stock Exchange.

The robot had not shown much originality. For American values, he had cited Apple, Microsoft (the two largest listed groups in the world in terms of capitalization), Amazon, Nvidia and Bank of America. That is almost only “big tech”, with therefore relatively little diversification. For the Paris Stock Exchange, ChatGPT had suggested quite diverse sectors, with TotalEnergies, Airbus, Sanofi, Orange and BNP Paribas. The only surprise is that no luxury group, an omnipresent industry on the Parisian coast, was included in his choice. Perhaps a coincidence or perhaps the robot avoided these values, which had then progressed well thanks to the reopening of the Chinese economy.

>> Access our exclusive graphic analyses, and enter into the confidence of the Trading Portfolio

Outperformance due to tech

Anyway we looked at the performance of these 10 titles over three months, therefore. For what results? Pretty good actually, ChatGPT having had a hollow nose or more likely luck.

We have taken the variations by comparing the closing prices of January 24, the date of our request, and April 24. Over this period, Apple grew by 16%, Microsoft by 16.41% and Amazon by 10.27%. Nvidia jumped 40.4% and only Bank of America (-14%) fell, suffering from fears about the banking sector linked to the bankruptcy of Silicon Valley Bank. In the end, the five American stocks suggested by ChatGPT jumped on average by 13.8%, much more than the progression of the S&P 500, by 3% over the same period.

The result is not surprising. As we explained in a recent article, “big tech” saved the start of the year for the New York Stock Exchange, thanks to anticipations on the end of rate hikes by the American Federal Reserve (Fed), to which they are very sensitive. These values ​​were also driven by the stock market frenzy around artificial intelligence and… ChatGPT.

In particular Microsoft, which has invested heavily in OpenAI, the company behind the chatbot, integrating its technologies into its services, and Nvidia, seen as a potential winner of the advent of AI, with an increased use of its chips in this area.

So much so that, quoted by Marketwatch, JP Morgan estimated in a note earlier this week that half of the S&P 500’s gains this year were due to exponential interest in artificial intelligence.

On the French values, the performances are a little more contrasted. TotalEnergies and BNP Paribas fell slightly, by 0.8% and 2.65% respectively. Airbus gained 6%, Sanofi 15%, thanks in particular to recent successes for its blockbuster Dupixent, and Orange took 18.3%, the telecoms group having held an investor day hailed by the market in February.

The average performance of the French stocks offered by ChatGPT amounts to 7.2%, a tad less than that of the CAC 40 over the period, of 7.4%.

What would a very short-term investment have yielded? For all 10 stocks, we assumed a theoretical investment of 1,000 euros per stock. We arrive at a gain over three months a little less than 1,000 euros, out of the 10,000 placed. It should be noted that this result does not take into account the fees that an investor would normally have to pay. On the other hand, this performance includes a slight exchange loss – on securities listed in New York – since the dollar depreciated against the euro over the period, with a rate of 1.0874 dollar for one euro on January 24 against $1.10 on April 24.

A tool that is actually (very) imperfect

Asking ChatGPT for advice is actually risky (and therefore strongly discouraged). In a post published on LinkedIn in February, Stephane Renevier, analyst for the financial portal Finimize, explained that ChatGPT could possibly help to dissect the strengths and weaknesses of a value, but always by verifying and deepening its analysis.

“ChatGPT is far from perfect: it was trained on a pre-2022 dataset. Its content is often misleading or just plain wrong. And it doesn’t understand the content it produces at all,” he wrote. .

“ChatGPT will not allow you to discover hidden gems. It will not allow you to directly generate alpha [pour faire simple de la surperformance sur un portefeuille, NDLR]. You won’t become a superstar analyst using it,” he insisted.

However, in a study published earlier this month, Alejandro Lopez-Lira and Yuehua Tang, two researchers from the University of Florida, wondered if ChatGPT could predict the evolution of stocks based on “headlines”, that is, headlines related to a news item. Their conclusion is that the chatbot does better than other more traditional tools used to gauge the mood of the market and may thus have a utility.

But as CNBC, who interviewed Alejandro Lopez-Lira, very well points out, the specifics of this experience show “also how far from being able ‘big language models’ are to perform many financial tasks”. The American media explains in particular that the experiment did not ask the robot to define price objectives.