A recent technical development touted by JPMorgan could provide institutional investors an advantage in today’s algorithmically-driven markets. The world’s largest bank based on revenue is rolling out an automated stock execution platform that uses machine learning techniques to route orders along the most cost effective path, the Financial Times first reported. The technical mechanism for trading large blocks of stock orders without disrupting market pricing is “significantly better” than an unspecified pricing benchmark and high-frequency trading systems, beating the traditional human method to execute large institutional stock transactions, according to a JPMorgan executive.

Get The Full Ray Dalio Series in PDF

Get the entire 10-part series on Ray Dalio in PDF. Save it to your desktop, read it on your tablet, or email to your colleagues.

We respect your email privacy

high-frequency trading systems
HypnoArt / Pixabay

Is JPMorgan technology turning the tables on high-frequency trading systems?

If there is one distinction that most can agree on regarding modern high-frequency trading, it is that technology is used in a creative fashion to provide advantages to liquidity providers in new and innovative ways.

Critics of high-frequency trading have charged that modern technology has gamed the traditional market making system to create a technical arms race that prices out traditional market makers – and does so using ethically questionable technical methods that disadvantage investors. Supporters say the practice provides consistent liquidity at the lowest transaction costs in the history of stock trading.

JPMorgan’s latest advancement appears to be turning the tables on the high-frequency trading world by using technology to provide institutional investors the advantage in determining when and how to execute their trades.

“Such customisation was previously implemented by humans, but now the AI machine is able to do it on a much larger and more efficient scale,” David Fellah, of JPMorgan’s European Equity Quant Research team, was quoted as saying.

Observers have waited for the day algorithmic systems would give institutional investors the advantage

In part, many of the high-frequency trading systems utilize speed to identify large institutional block orders and then beat the order to the exchange where that order is going to be placed to engage in what critics call "front running." This time-honored practice, executed in both physical pits and electronic exchange, occurs when a large order is known by a limited group of participants to be entering the trading arena, providing them an opportunity to trade in front of the order to obtain a slight price advantage.

A key point of "price improvement" in routing large institutional orders has focused in this area.

While the exact methodologies and practices of JPMorgan’s AI technology are not specifically known, a trading algorithm could, in theory, effectively assist institutional investors through algorithmic order placement techniques based on the very actions of the high-frequency trading systems. Rather than the high-speed trading mechanism watching the institutional trader to gain an advantage, the institutional trader could now be given the advantage by watching the market maker activity and directly routing orders to exchanges at multiple locations simultaneously.

“Best execution is becoming more and more important to clients,” Daniel Ciment, JPMorgan’s head of global equities electronic trading, told the FT, noting that a “Deep Reinforcement Learning” method was used to invest in what was characterized as “pioneering technology” that can help the bank win new clients.

Not only will the new system provide a method to obtain price improvement over the high-frequency trading systems, but it eventually could be used to consider individual client needs and customize trading methods accordingly, a trading technique that would include human oversight.

“Any customisation would only be if the client agrees to that,” Ciment said, pointing to the limit of responsibility the bank might have in using such an algorithm.

Trading algorithms gone bad can turn costly. In August of 2012 Knight Capital lost $440 million when a trading algorithm went rogue during testing and entered live trades. A Goldman Sachs algorithm was blamed for trading losses in 2013 as well.

“The machine is restricted in its trading behaviour, as it learns under, and operates within, our general electronic trading risk framework, which is overseen by internal control groups and validated by regulators,” Fellah reassured.