Racing with Wall Street's Robots

The biggest names and deepest pockets on Wall Street do not have the high-frequency trading (HFT) technology arms race to themselves. Serial trading tech entrepreneur George Kledaras is touting a solution that he and his clients say is less costly and clunky than, and enables risk management at least on a par with, those of the industry's biggest guns.

Kledaras' Atreyu is an access point on Amazon Web Services (AWS). Powered by cloud computing along with programming innovations, the platform supports trading at the millisecond speeds that are HFT table stakes. It is also low latency, meaning there is minimal delay between an instruction and its execution.

The bottom line for clients, Atreyu Group Holdings CEO Kledaras says, is significant cost reductions and access to computing power unimaginable in decades past.

“This world is a different workflow,” says the 56-year-old son of Greek immigrants who started out on Wall Street as a computer programmer in 1987 and founded his first start-up, Javelin Technologies, in 1996. “We're building stuff for robots. We don't have to react. We're not holding the steering wheel. We're going to let our clients trade automatically - let their machines trade automatically.”

Atreyu Group, which was formed in 2015 and has 16 employees, pulled in $7.5 million in revenue from 160 clients last year. They are charged transactional fees - there are no minimums or connection fees - which might run from as low as $3,000 a month up to $20,000, depending on their trading strategies.

“We're a small high-tech company,” Kledaras explains in an AWS case study, “but we've processed billions of shares so far.” A typical day of more than 100,000 orders and 20 to 25 million shares “puts us up there with some of the biggest firms in the space based on volume. We could never do this without having scalable systems, because the moment one thing goes wrong, hundreds of thousands of things will go wrong. That's why it's critical to be able to start from zero and to be able to scale our systems.”

Supercomputing Scale

Kledaras tells GARP Risk Intelligence that users “are writing algorithms, using machine learning or data science or 'big data' techniques” that were beyond their reach five years ago - “server farms that give them 100,000 GPUs to crank out very, very difficult statistical computations. We're talking about super-heavy-duty, supercomputing power that they're renting for very, very low cost.”

The Atreyu platform's lean-and-mean feel appeals to small to medium-size quantitative investment managers and family offices that need to automate in order to compete. In place of teams of portfolio managers, they use data-driven strategies to automate their investment processes.

In interviews, clients say they find Atreyu's user-friendly interface appealing, allowing for faster trade executions at competitive commissions when compared with major broker-dealers. In effect, that lowers a barrier to entry. The holding time in equity positions was reduced from 250 milliseconds to “a lot less,” one client says. And the system's scalability makes it easy to start with small orders and move to larger ones.

Proprietary API

Atreyu doesn't require FIX (Financial Information eXchange) connections, the standard communications protocol that propelled electronic trading growth starting in the 1990s. Atreyu's proprietary application programming interface (API), FLIRT, is easier to integrate with commonly used computer languages, such as Python and C++.

The FLIRT API “is more elegant in an engineering sense,” says Jared Broad, CEO of Quantconnect, which provides algorithmic tools to break up large trades into smaller ones.

“I do a lot of my own coding in Python, so I was looking for a way to connect without having to write my own FIX engine,” says an individual trader with a limited liability company who asked not to be named. “Pretty much every prime brokerage I looked at wanted you to connect with FIX. None of them offered the kind of API support I wanted.”

Atreyu's platform could still be faster. “AWS is fast, but it doesn't give you the speed you need,” says Torsten Wegner, associate partner at McKinsey & Co.'s Risk Dynamics unit.

An eight-millisecond delay that Kledaras noticed on a network between Northern Virginia and Secaucus, New Jersey, is “a huge thing,” says the individual trader. “I can fire up a server in NY 5 [the Secaucus data center], where all the exchanges are located and Atreyu has a presence, and I can connect in two milliseconds there.”

Keeping Pace with Risk

Atreyu's clients say the platform is fast enough for them to navigate the risks of equity investing.

“My main way of managing risk is to be fast,” says Michel Watteyne, an Atreyu client who manages funds for Gabriel Petricca, a Canadian investor. “You get an enormous amount of slippage if you're not fast, whereas now I can get in and out of a position before the market changes in a millisecond.”

Clients are confident that Atreyu's accounting systems accurately record thousands of trades made in a day, including records of the shares each trader owns at any particular millisecond. That's a luxury that at least one client says he didn't enjoy at brokerages E*TRADE and TradeStation, with which he no longer does business. “They lose track. They think you have shares when you don't have shares. It's an accounting nightmare,” the client says. E*TRADE and TradeStation did not respond to requests for comment.

Trading Risks and Regulations

Although not formally trained in risk management - Kledaras has a BS degree in electrical engineering from Lehigh University and MS in mathematics from New York University - he keeps a laser focus on it.

“There are several points of risk that we're constantly looking at, because our particular client set is doing heavily automated trading driven from brand new technologies,” Kledaras states. He prioritizes regulatory risk, pre-trade risk, margin risk, portfolio risk and volatility risk - all of which become interconnected.

A stark example of potential exposure is SEC Rule 15c3-5, which requires a broker-dealer with market access to have risk management controls and supervisory procedures in place against “financial, regulatory, and other risks, such as legal and operational risks.” The rule effectively bans “unfiltered” or “naked” sponsored access by requiring these controls on a pre-trade basis.

To ensure 15c3-5 compliance, Kledaras has developed a pre-trade risk check for code errors that could wrongly send hundreds of thousands of orders in a fraction of a second. Resulting trading disruptions could bring stiff fines from securities exchanges. Atreyu has built-in throttling and checks for such activity, and it offers trading simulators for clients to test their code.

“We have to go beyond Rule 15c3-5 because now we're talking about some tricky workflows coming in from machines that are writing code themselves,” Kledaras says. “We have a very sophisticated way of slowing down the trading. We throttle the trades based on message rates.”

Compounding the regulatory risk concerns are FINRA 5270, which bans front-running block transactions, and 5320, which prohibits trading ahead of customer orders. “This might happen when you anticipate orders based on behavioral patterns,” McKinsey's Wegner explains.

Margins and Pricing

Atreyu's clients appear confident in the regulatory risk features. As an extension of Portfolio Margin, which aligns margin requirements with the overall risk of a portfolio, regulators have approved Atreyu for certain margins that usually take brokers longer to get approval “because the regulatory agencies feel more comfortable with what they are doing,” says an Atreyu client, who asked not to be named.

To manage portfolio risk, Atreyu built a price-risk tool on AWS, which was completed in March. This shortened the time it takes to compute price risk across “thousands of portfolios” from 10 minutes to “a second or two,” says Kledaras.

For volatility risk, Kledaras gauges worst-case scenarios from multiple sources including the Office of the Comptroller of the Currency (OCC), historical volatility of stocks, and the options market. Atreyu analyzes kurtosis risk, which is based on measures of the combined weight of a distribution's tails relative to the center of the distribution. Atreyu does not use a normal or Gaussian distribution, which appears as a bell curve in Black-Scholes models for pricing options contracts and in value at risk (VaR) models.

Kledaras believes his system is better for managing tail risk. “I'm capturing 100% of the tail risk,” he says, “yet I'm not taking these arbitrary haircuts as a big firm would typically do.”

Further Acceleration

The continued evolution of high-speed trading and the rise of robots will only bring greater risk management challenges. “It is extremely difficult to navigate all of the regulatory risks in all markets,” Wegner comments. “The humans have left the sea.”

Kledaras has been helping to accelerate trading with start-ups Javelin, whose Coppelia was the first high-speed FIX engine built entirely in the Java language, and FIX Flyer, which followed in 2006. With Atreyu backed by Decathlon Capital Partners and National Innovation Fund, Kledaras could be destined to raise more venture capital. He suggests that his next project will be in decentralized finance (DeFi).

“I like to work on the institutional level so I can make the most impact for the entire industry,” says Kledaras. “I like to help investors do better.”