What sounds like a future scenario is already reality at JP Morgan AM. There the KI Themebot is looking for attractive companies. That can lower the costs for investors, but it is not worthwhile for every topic.
“Unnatural Selection”, in German: “Unnatural Selection”, is the name of a documentary on gene therapies running on Netflix. Scientists, hobby biologists and people suffering from a genetic disease are accompanied in it. In a very dramatic way, a picture of the future is drawn in the series in which gene therapy could create supermen and people modify genomes as they please. The discovery of the enzyme Crispr-Cas9 in 2012, better known as gene scissors, marked the beginning of a new era for medicine for many researchers. More and more companies have also heard the wake-up call.
There are now a number of listed companies that conduct research on the subject of gene therapy and are therefore also of interest to investors. They include Editas Medicine and Crispr-Therapeutics, among others. “There is huge customer interest in the topic,” says Sherene Ban, Product Specialist at JP Morgan Asset Management. She oversees one of the first actively managed funds to be put together by artificial intelligence (AI), the “Genetic Therapies Fund”. An “unnatural selection” also takes place in the pre-selection of companies for the theme fund. These are found by an AI.
Search stocks like on Google
But how does such a program work exactly? While most conversations about AI in asset management remain very vague and the exact use often remains nebulous, Sherene Ban turns on the projector that is connected to her laptop during the conversation with the FAZ. You can explain the program, says the Princeton graduate. Showing it is more catchy – and more impressive.
Ban types the term “Genetic Therapies” into the search line of the program. In doing so, she not only selects the overarching topic of gene therapies, but also which terms she wants to include or explicitly exclude. Themebot, that’s the name of the thematic investment system from JP Morgan AM, then selects the desired companies from an investment universe of more than 10,000 stocks and five terabytes of data. The system filters news articles, company profiles, research results and approval documents, among other things. In the end, there are only around 200 companies left to be shortlisted.
“In principle, Themebot works like a search on Google,” says Ban. The desired investment theme is entered into the search field and the machine spits out all possible answers. Except that the AI thinks like a human, adds Ban. Because it weights and evaluates the company directly – according to two criteria: relevance in terms of content and turnover. Depending on how these turn out, the papers of the respective companies are then either overweighted or underweighted.
The fact that both criteria are being searched for is a result of the possible pitfalls, adds Ban, and says that one wants to avoid so-called false positives. For example, if the name of a company appears particularly often in the search due to a certain report. “For example, Facebook has repeatedly been the victim of cyber attacks,” she explains. Accordingly, the two terms often appeared in tandem: Facebook and “Cyber”. If you are looking for a cyber security company, Facebook suddenly becomes particularly important, even though the group is of course not a cyber security company at all. Therefore, Themebot is also looking for another criterion, the turnover that the company generates in the area in question.
Funds possible on all topics
But even the sales alone are not meaningful enough, says Ban. Weighting companies only according to their total sales yields a distorted picture. Then large pharmaceutical companies like Roche and Novartis would quickly have a correspondingly high weight in their portfolios. How much they earn exactly with the genetic engineering sub-area, however, is not clarified. In many conglomerates, it is not always clear how much is earned in the individual sectors. Themebot estimates the revenue.
In addition, there are no limits to customer requirements with the theme offer. You don’t follow a particular philosophy, says Ban. Almost every investment topic can be implemented with the program, not just gene therapies. Examples are esports and education. Theme funds are usually actively compiled and managed by fund managers. The search is too detailed and fragmented. The human component makes active management comparatively expensive. For customers, the ongoing fees for the gene therapy fund are somewhere between active and passive management; they are 1.02 percent.
AI cannot replace people
AI can do a lot of the work for fund managers. But never everything, adds Ban. “It is a supporting tool and does not replace the portfolio manager.” Ban remains pragmatic: It is a scalable tool and is much more efficient than humans. That way you can create customized portfolios faster.
However, not all of the wishes and demands of customers make sense for this type of fund management, emphasizes Ban. Using Themebot to set up a large-cap fund with large companies is nonsense. “We’re looking at small companies, the work is already done with the big ones.” So it is not worth using the theme offer for mainstream funds, but it is for investment topics such as gene therapy, says Ban. At this point in time, it is not yet possible to say who exactly will win the race and prevail over all other companies, says Ban. “We invest in the topic and not in individual values.”