UBS trials Netflix-style algorithms for trading suggestions

Picking an investment may soon start to be more like choosing a TV show on Netflix or finding new music on Spotify.

UBS is looking at applying recommendation algorithms to suggest trades to its asset management and hedge fund clients, similar to those used by a host of consumer technology companies.

The move is in its early stages of development, with the task of reliably recommending interesting trades to investors posing very different challenges from suggesting a new indie band to listen to or what fresh comedy show to watch.

Nonetheless, the broad theory behind the initiative is the same, as finance increasingly borrows from the innovations of Silicon Valley’s technology companies. Increasing cost pressure on banks, coupled with the rise of computer-powered market players, has pushed banks to boost investment in their technology and explore new ways to automate some of their core businesses.

“Imagine what the world looked like when you watched television and had to scan through channels, whereas now it is not only on demand, it is presented to you so you easily find what you are looking for,” said Giuseppe Nuti, who heads up data science in UBS’s FX, rates and credit Strategic Development Lab. “That’s what we are trying to do for our clients, presenting them with a choice of likely, interesting trades.”

Just as people once took recommendations for TV shows or new music from friends or industry critics, a bank’s clients have often listened to recommendations from trained salespeople. UBS is hoping to make this process more automated, taking inputs of a client’s previous trading behaviour to assess whether they might be interested in a specific transaction. And it’s not just for clients.

The technology should help pinpoint investors that will want to buy something UBS is trying to sell, and vice versa. “A good sales person calls his or her client and goes over what they think will be interesting for the day,” said Mr Nuti. “We are trying to automate that.

It means the job of a sales person, which in many ways has remained immune to the technology revolution, will likely change.” The algorithm is currently being trialled in the bank’s corporate bond trading business but there are hopes to roll it out to other asset classes as well.

At the moment, recommendations produced by the technology are sent to salespeople to decide whether or not they should be passed on to clients, but over time the plan is to eliminate the middle man. The challenge is finding enough data to plug into the recommendation engine for it to produce reliable results, said Mr Nuti.

Each new TV show watched, or song listened to, is indicative of the type of music or television that a person likes. But it is harder to group trades into similar themes, says Mr Nuti, because people execute transactions for very specific reasons.

Other electronic trading experts have also urged caution. “While it sounds intelligent and advanced, a client should definitely ask a lot of questions about the construction of such a recommendation algorithm,” said Christian Hauff, co-founder of Quantitative Brokers, a trading algorithm company. “Creating and managing a financial instrument portfolio is not the same as creating and managing a playlist.”