Credit where credit is due: what investors can give AI credit for

91% of investment managers have used and plan to use artificial intelligence in their work, cites Mercer in its 2024 survey. AI is already analyzing data, identifying market trends and even forming portfolios. Ekaterina Smirnova found out what advantages and disadvantages an AI investor has.

In 2024, JPMorgan began testing AI as a "co-pilot" in its investment division. As wrote Business Insider at the time:  "JPMorgan plans to process nearly 50% more files under the Know Your Client program, using 20% fewer employees." A year later, the bank introduced its own genAI platform to more than 200,000 employees.  

GenAI can already now be given the task of compiling a portfolio according to its criteria, Samit Yakovlev, partner and investment director of ArtQuant Global  tells Oninvest. He recalls a super-successful case when the AI itself chose Activision Blizzard shares:  "AI has never bought shares of companies engaged in game design, development and promotion before. To our surprise, at the end of December 2021, several publicly traded companies from this sector appeared in the AI's portfolio. For the first time in 5 years."   "We rejoiced like children when on January 18, Microsoft announced the acquisition of Activision Blizzard   [developer of Call of Duty, Warcraft and King (Candy Crush Saga - Oninvest note], the entire sector's stock soared. Activision itself rose in price by 40%," Yakovlev says.

A year after the successful experience with the creator of Call of Duty, the AI failed, recalls Yakovlev:  "2022, our AI was constantly buying Intel stock thinking it was a solid and steadily growing company, and after the first quarter results, Intel stock dropped 15% in one day. To our surprise, the AI bought the stock before the second quarter results. The results were again disappointing, with Intel down 18%." Overconfidence in the algorithm and in the algorithm can turn into risk, he reminds us.  

But if you decide to put together a portfolio using artificial intelligence,  help it with natural intelligence, he advises:

1. Select the country and investment sector that you think should grow.

2. Ask AI to find the best hedge funds specializing specifically in stock piсking in this sector;

3. Ask to select those that publish a statement of their assets every quarter.

4. Ask AI what stocks they bought and sold this quarter, look at the composition of these funds' portfolios.

5. Of the stocks you like, you can ask AI to keep a portfolio based on certain criteria, such as the lowest volatility of the entire portfolio. 

Yakovlev says traditional analytics no longer offer an advantage.  Funds like Orbital Insight and RS Metrics rent satellites to count the number of cars in Walmart, Tesla or IKEA parking lots.  For example, if a parking lot is 20 percent fuller than last quarter, that signals sales growth. That's why before Tesla reports, analysts compare car pileups at factories in Shanghai and Texas, predicting production volumes. Customers pay for benefits, not outdated methods, Yakovlev explains, and without AI, you don't see 90% of the signals.

When the developers of the investment analytics platform PRAAMS calibrated their model and compared it to benchmarks, they asked ChatGPT to compare the credit qualities of JP Morgan and Citigroup. "About JPMorgan, we got an acceptable answer, a paraphrase of the S&P credit report," says PRAAMS co-founder and CEO Rinat Kirdanj about working with AI. - And it predicted Citigroup's imminent and imminent default: poor asset quality, illiquid balance sheet, problems with profitability. When I asked about the rationale, I was told that Citigroup's credit rating was only a C. It turned out that ChatGPT had confused Citigroup's ticker symbol (NYSE: C) with a C credit rating."  

AI is really good at processing large data sets, it can save weeks of human labor, Kirdan says, but you have to remember that it's only a third as good as it is for now.

As the expert explains, the AI architecture in an investment context conventionally consists of three components. 

First, the mathematical model is a complex neural network structure that requires large computing power.  

Second, the dataset (training sample and relevant information) - the data on which the AI is trained and which it uses for answers. This component determines 80-90% of the model's quality. In investments, it can be transaction histories, portfolio statistics, economic events, market behavior patterns, prices, reports, news. In universal AI like ChatGPT, the dataset is not specialized: it contains a lot of general texts and relatively little profile market data. This reduces accuracy and creates "information noise".

Finally, the methodology of application - how the neural network interprets the data and draws conclusions: what parameters it considers significant, how it selects analogs, what it proposes as an investment solution. In Kirdan's opinion, this is the most vulnerable component of current models.

For example, ChatGPT has excellent math," the expert evaluates. - Dataset is a three: too many irrelevant things. The methodology is a one. In order to trust AI to make decisions, all three components must be A+.

In addition, if you speak politely to ChatGPT, it's more likely to give you a nod, wanting to evoke positive feelings from you, writes MarketWatch.

General AI is not yet robust enough to be trusted with critical tasks such as building market or credit risk models," confirms Rinat Kirdan, "but specialized AI models like ours have made significant progress;

The role of neural networks in investing is to save time, says Alexei Kozhukharev, founder and owner of Quanta Nova, an algorithmic fund that trades derivatives on the US market. A neural network can collect, structure, compress and analyze a lot of data. But it cannot yet come up with a really working investment strategy that will outperform the market (generate alpha);

To date, according to CREATE-Research, only 3% of asset managers have already implemented generative AI in their operations, but 26% are working on implementation. 

Hiring for entry-level positions at investment banks has already slowed. What's next for university graduates will be even more difficult. The head of investment bank Moelis & Co. Navid Mahmoodzadegan put it bluntly: AI can greatly "narrow the pyramid at the lower levels" - reducing the number of junior employees. No more torturing newbies with 20-hour workdays: AI will process data and create a Power Point presentation faster than a trainee.  Financial Times reminds, one entry-level analyst costs an investment bank about $200,000 a year. Nothing compared to top salaries. But Jamie Dimon can't be replaced by AI. Not yet.

This article was AI-translated and verified by a human editor