Goldman Sachs Exec: Markets Still Too Complex for AI Trading

The rise of the machines may be the topic du jour, but a senior executive at one of the world’s largest trading firms say the human factor is still a limiting one.

Buy-side firms must process and manage vast volumes of data daily in a bid to forecast industry trends and develop sophisticated, predictive trading models. However, the application of artificial intelligence (AI) capabilities into investment processes have proven progressively complex due to the diverse nature of current market structure, the influx of data sources and the existence of multiple asset classes.

“In my experience, these applications of AI tools are very limited in their capacity, and you cannot run $200 billion of equity investments on pure AI; there are other factors that come into place,” said Michael Steliaros, global head of quantitative execution services at Goldman Sachs.

Steliaros was speaking on a panel during TradeTech Europe, held in Paris on April 25.

Decades ago the traditional method of placing orders or making decisions was dependent on the memory and expertise of individual traders. Translating that knowledge and information into electronic systems to source liquidity and achieve best execution has become a significant undertaking for the buy side, according to the panelists.

The consensus during the talk was that modern-day AI tools are unable to cope with the unpredictable nature of market microstructure and events related data. Machine-learning technologies are programmed to follow a set of rules based on the coded models which can then learn from vast sums of data. However, corporate events, news or social media feeds don’t follow pre-programmed outputs. Because of that, investment firms must consider what information or strategies to build their models on, as traders often make split-second decisions based on undefined market knowledge and instinct.

In many cases, the technology “is limited by the human that is coding it” explained Steliaros.

He highlighted that AI is more effective at describing or identifying complicated relationships between data rather than predicting sources of alpha. Since the emergence of AI, it has generated significant hype surrounding its potential but has yet to materialize successful use cases for best execution or outperforming the market.

Daniel Leon, deputy global head of trading and securities financing at Axa Investment Management explained that although it is worth trialing and testing AI for seeking liquidity, for some, that it is more effective as a long-term strategy to also use it to drive efficiency and automate practices, in an attempt to cut bottom-line costs.