How French banks are using AI to help salespeople and regulators
Machine learning and AI are talking points in many different industries. Finance is of course no exception. Like other industries we are trying to grapple with how they could impact us and as such these topics are regular talking points at many industry conferences.
At the recent Paris market microstructure conference, a panel sought to understand how machine learning and AI could impact market structure. The panellists were drawn from differing sectors of the finance industry, Laurent Carlier (BNPP), Laurent Fournier (Euronext) and Franck Railton (AMF). Moderating the discussion was Charles-Albert Lehalle from CFM, a large quant fund based in Paris.
Carlier noted that historically quants have been focused mostly areas related to trading. This has ranged from areas such as pricing derivatives to risk management, and more recently, in electronic trading. However, there has been one area, where they have been conspicuous by their absence, namely in sales. As a result, two years ago, BNP created a data lab to use machine learning and AI to change how sales work. Carlier said that 90% of what his group did, was on “applied” machine learning, that is doing projects which had been sponsored by the business.
The idea was to change the way sales people managed their relationships with clients. Part of this, involved making them more efficient. Another part of this was trying to be predictive, trying to understand what type of information could be useful to a client. There were of course lots of challenges, notably that clients behaved very differently from one another.
However, it was very much a step by step process to build up trust. He noted that in areas, where machine learning had been very successful, such as computer vision, the ground truth is known. A human can easily judge whether a computer algorithm has correctly identified an object in a picture. All the information is there. In finance, by contrast, we have incomplete information. Carlier noted how a complicated approach proved difficult, because it was tricky to explain the results, when an algorithm had too many inputs. There was a trade-off, where the performance of certain “black box” techniques, like neural networks could be at the cost of being explainable. In practice, using a simple approach, which was more easily explainable to clients proved better at first.
Working in a regulator, Railton had a different perspective on how machine learning and AI could be used to look at their dataset. He noted that since the start of the year, the AMF have had access a much richer dataset of transactions, which now also includes the final beneficiaries of a trade. With more granular data it would be possible to detect market abuse quicker. He also noted how clustering could be used to detect outliers and errors of declarations. Clustering was also used to group together member firms, so that their behaviour could be compared to their peers.
Fournier explained how machine learning and AI could be used to help his exchange in many areas. As an exchange they wanted to be able to explain price movement. Furthermore, a key objective of an exchange was finding ways of improving liquidity for its participants and also increasing market share.
French banks are not alone in using artificial intelligence, but they are pushing the envelope in applying the technology away from the trading function. As AI becomes more predominant across finance, other banks are likely to do the same.
Saeed Amen is a systematic FX trader, running a proprietary trading book trading liquid G10 FX, since 2013. He developed systematic trading strategies at major investment banks including Lehman Brothers and Nomura, and runs Cuemacro, a consulting and research firm focused on systematic trading.
Have a confidential story, tip, or comment you’d like to share? Contact: firstname.lastname@example.org in the first instance. Whatsapp/Signal/Telegram also available.
Bear with us if you leave a comment at the bottom of this article: all our comments are moderated by human beings. Sometimes these humans might be asleep, or away from their desks, so it may take a while for your comment to appear. Eventually it will – unless it’s offensive or libelous (in which case it won’t.)