Jefferies' Quant Team Builds Chatbot For Faster Equities Trading

The quantitative arm of American investment bank Jefferies has built a chatbot, JEFQuants, to assist in equity trading by providing anyone—traders, clients, or portfolio managers—with real-time pre-trade information, such as bid-ask spreads, spots to find liquidity, or whether a stock is a historically volatile name.

JEFQuants, developed in partnership with Symphony Communication Services, acts as a single point of reference for data that traders typically source using a combination of different avenues, like checking exchange datafeeds, calling up a broker, or sifting through pages within a Bloomberg Terminal, says Jatin Suryawanshi, head of global quantitative strategy at Jefferies.

“One of the things we have done for our clients and traders is create a pre-trade application, which we can call on-demand, [and say] ‘I have interest in XYZ, and I want to buy a million shares, what would you recommend?’” he says. “And at the backend sits the quantitative server, which has tons of data—execution data, big data, real-time data, historic data—and the server is going to number-crunch all of this and come up with the best possible quantitative solution to recommend to the trader.”

As an example, a trader may want to know how two stocks correlate if they want to trade the stocks as a pair. All that trader would need to do, then, is put a request for that information into JEFQuants and calculate the correlations that come back.

There are three guiding principles that must be met to ensure the bot remains value-additive to the trading desk: the information it gives back must be unique and not readily available anywhere else; it must be able to process requests and return answers within minutes, no matter how complex the query is; and it has to be accurate, Suryawanshi says.

JEFQuants, now aged almost two years since its conception, will likely not be confined to equities for long. Jefferies’ fixed-income desk recently reached out about possibly incorporating the bot into its own operations, he adds, and it is suitable for trading within any asset class.

While more asset classes would be a natural extension, what is perhaps surprising is that JEFQuants has found use-cases beyond trading, in educating the company’s interns and new joiners. Especially in the remote-work environment wrought by Covid-19, it has acted as an interactive FAQ bot for recruits to learn about business strategies and company philosophies.

Almost everyone in the market has easy access to static data—five-day averages, yesterday’s news, and yesterday’s analyses of the news. JEFQuants differentiating quality is that it should already understand the implications of today’s news, which is key to making good trading decisions.

“Is it going to cost you more in terms of market impact if you execute this order because [for example] there’s no liquidity on a Friday afternoon? And would I be able to quantify the additional impact caused by 35% less liquidity than normal? Is the stock volume curve different today because everybody’s almost gone home? These things are very easy for our traders to now be able to detect,” Suryawanshi says. “All of this information is available to them within milliseconds if they’re looking for it.”