The surge in demand for data science specialists seems a little like putting old wine in a new bottle. After all, both buy-side firms and investment banks have long used structured datasets to inform their investment decisions, so why the sudden surge in demand for seemingly unobtainable talent?
In a nutshell, artificial intelligence and machine learning are changing everything.
“On the sell side, applications of AI include everything from client acquisition and research to trading to operations to controls and compliance,” said Nikhil Singhvi, the global head of market and client connectivity at Credit Suisse at the Tabb Group Fintech Festival in New York last week. “Each of these has different use cases and different levels of adoption."
When it comes to using AI for investment decisions, investment banks now have a disadvantage, he says. Bulge brackets used to be able to bet their own money through large prop trading desks, which also served as an in-house guinea pig for trying out new ideas. Regulatory crackdowns on banks' prop trading desks after the financial crisis means this is no longer an option.
“With a lot of prop trading going away from the sell side, you have to do more and more testing, whereas in the olden days, when there were prop traders sitting beside you, you could work with them to test the strategies,” he said. “Now you have to do many simulations before people are comfortable and you can put it into production, and there are cost pressures."
More to the point, Credit Suisse is changing the type of people it needs to hire, he said.
“We’ve changed our thinking from a regular stat-based quant to machine learning, which is a big change, and the challenge of finding [and hiring candidates with] the right skill set is a big issue.”
While unstructured data is increasingly in demand, Credit Suisse typically partners with providers who convert data sets from unstructured to structured, rather than bring people in-house to do such tasks.
“From a company’s perspective, how do we want to use this data, and what are the use cases? You still need to be able to make sense of it and read where the signals are pointing,” Singhvi said. “The capacity to be able to build the model – that’s where you need a lot of that data.
“If you don’t have the [cognitive-computing] hardware, then it is difficult to achieve within the bank’s infrastructure,” he said. “You have to push the boundaries with your infrastructure and architecture teams. It requires a combination of the public cloud and a private cloud and a continuous push within the organization to make it happen.”
Photo credit: gorodenkoff/GettyImages