Buy-Side Firms Turn to AI for Efficiency Amid Barriers to Adoption

Buy-side firms are allocating more of their budgets to manage the influx of complex data—aggregated across various business lines throughout the front-to-back office and in multiple formats. Managing the data is proving to be a colossal challenge for the industry, bringing with it a significant drain on time and resources.

“It’s the million-dollar question. I think it’s so complex, and we make it more complex. The world is evolving. It used to be [whether you] could deal with voice [-related data] but now you have to manage voice, video, WeChat, Symphony, and all these other platforms,” said Phil Fry, vice president of product strategy at Verint, during an artificial intelligence (AI) and automation panel talk at TradeTech Europe on April 24.

The general consensus during the discussion was that modern-day data challenges cannot be effectively resolved without the implementation of AI or machine learning capabilities. Trading firms must capture, clean and manipulate vast amounts of data in order to extract valuable and tangible insights—a task that is proving increasingly difficult to manage via humans or traditional IT infrastructures.

Jas Sandhu, head of global equity execution algorithms at RBC Capital Markets, explained during the session that AI enables buy-side firms to scale and carry out complex, cumbersome tasks more effectively. “Its capabilities can be used to decode inputs or accelerate dimensionality reduction techniques,” he said.

As well as building proprietary AI technologies in house, some firms are turning to third-party experts for help. JP Morgan Asset Management, for instance, is looking to leverage cloud providers’ advanced AI toolkits, which are built into virtual environments.

“Services like transcribing, taking audio and being able to pull that data—a lot of those things are easily available at our fingertips, so being able to leverage the public cloud to its full potential and then to deploy that relatively easily across our business is what I see as the next step in where we want to take this,” said Ashwin Venkatraman, head of equity trading execution technology at JP Morgan Asset Management.

Lacking the Human Benchmark

The adoption of AI technology has proven revolutionary across various industries outside of finance, including commercial technology, medicine, and even self-driving cars. The common denominator among each of these use cases is an existing human benchmark to build AI models.

“Doctors are able to detect diseases most of the time and humans can drive cars most of the time. So we have benchmarks to train our models against or even go beyond that in terms of performance, but it is unclear what that would be in the financial markets,” said Frank Steffen, managing director and co-founder at CapTec Partners.

With active managers failing to meet their own performance benchmarks, according to the latest S&P 500 report published in March, building successful AI trading systems has proven incredibly complex. Today the industry lacks a standardized model whereby AI subsets such as machine learning, deep learning, natural language processing, and others can be applied to buy-side trading practices.

Another major obstacle impacting the adoption of AI-powered trading includes exposure to unpredictable or non-traditional datasets. In many cases, AI systems could be easily disrupted due to the erratic nature of some alternative data sources such as social media. Programming a machine to make a trading decision based on Twitter or live feeds is not only complex but also opens the door to bias and issues around governance.

Under the Watchful AI of the Regulators

Although AI technology and the exploration of intelligent trading tools have been around for some time, the growth in their adoption has caused regulators to sit up and take notice. As buy-side firms are responsible for justifying their best-execution methods under regulations such as Mifid II, one concern is that AI trading strategies and sophisticated algos could become too complex to explain how they work to investors and regulators. During the discussion, the panelists highlighted the danger of creating an “AI black box” and the issues of effectively controlling or managing a technology that is too difficult to understand.

Institutional firms may have to implement systematic processes for benchmark testing or verifying algos and AI systems in the future. Gerard Walsh, head of business development, institutional equities at Northern Trust, explained that if regulatory changes or standardized methods of implementing the technology come into play, firms will have to adapt their practices, as well as allocate teams to manage the verification processes or use a third-party provider.

“The last decade [involved] hiring quants to build algos and smart technologies, now we are going to have to hire more clever quants to check the quants. So you would need properly built system surveillance, built by people with a naturally cynical mindset,” he said.