The quant investor harnessing the power of ants

Swarm Technology designs network of trading algorithms that mimics hive mind of insects

Engineers have long looked to the natural world for inspiration. Japanese high-speed trains were modelled on kingfisher beaks. Architects have studied the structure of termite mounds to improve ventilation in buildings.

Financial engineers, too, are using designs and practices from nature. UK hedge fund Swarm Technology runs a $95 million systematic trading strategy based on the way ants interact.

The trading programme, known as Swarm XVI, consists of individual algorithms that share information with each other, rather like an ant colony.

The approach has proven profitable. In the 40 months from inception through to October 2022, Swarm XVI generated a return of 57.29%. The S&P 500 over the same period gained 30.55%.

Charlie Drew is the co-founder and chief operating officer at Swarm. He explains that ants are adept at communicating simple information and, crucially, making decisions that benefit the group. Humans, by contrast, fall victim to biases and self-interest.

“The idea of decision-making intelligence fascinates me,” says Drew, who previously worked at a biotech company that converted food waste into proteins using fly larvae. “How does a creature with a few hundred thousand neurons compared to the 87 billion neurons that us humans have, make decisions so differently?”

The Swarm XVI strategy is powered by 16 trading algorithms, each of which uses a medium-term trend-following system to take long and short positions in four liquid futures markets: equity indexes, interest rates, commodities and currencies.

Drew compares the approach to a group of 16 traders working for the same dealing room. Each trader uses their own strategy and trades a separate market. They might discuss the goings-on in their market with colleagues over lunch or during meetings but each trader shares, at best, just a slither of this knowledge with their co-workers.

Even if a trader were able to digest all the information available to their colleagues, they wouldn’t necessarily use it to trade on their own book.

14.61%

Swarm XVI year-to-date return

“We have the luxury of a central brain on a nervous system that allows us to solve complex problems by ourselves,” says Drew. “Indeed, most actors in financial markets are human decision-makers, or humans that programmed algorithms to trade… we’re all driven by this mammalian survival instinct driven by inputting information and making individual-based decisions.”

Ant colonies, by contrast, share reams of information constantly. Although they only tend to communicate with their immediate neighbours, information quickly cascades throughout the colony. Decisions are made and tasks allocated, with some ants even taking a sacrificial role for the good of the collective.

Swarm aims to apply this concept to its own investing technique.

Network effect

The Swarm XVI programme is based on a strategy that Swarm co-founder and chief executive officer Pan Yiannakou used when running a $2.5 billion managed futures fund as part of Man Group’s associated managers programme.

In setting up the new fund, Yiannakou says he was searching for an edge: “My job is to make decisions, and I want to consider different ways of making decisions. Is there a way that I can make a decision that is different from my colleagues but will still lead me to profitable outcomes?”

To answer this question, Yiannakou looked to ant colonies.

“A single ant cannot survive and prosper by itself,” says Yiannakou, “yet as a species it is extremely successful.”

Swarm XVI’s 16 algorithms – known as agents – share four pieces of information with each other: volatility, momentum, their current position, and its performance. The data is exchanged in every possible combination, amounting to around 65,000 virtual meetings.

Drew likens these 65,000 information exchange nodes to a neural network: “In effect, this becomes the brain.”

How does a creature with a few hundred thousand neurons compared to the 87 billion neurons that us humans have, make decisions so differently?

Charlie Drew, Swarm Technology

Each agent uses the information received at these meetings to determine its position. For example, the agents know both gold and the US 10-year Treasury’s current momentum. This is used to produce a synthetic market, GoldxUS10-year. Swarm XVI then calculates the momentum of this new synthetic instrument.

As a result, the position of the agent trading gold is determined not just by conditions in the gold market, but by conditions in other markets too, Yiannakou says.

Agents can take a long or short position with up to three times leverage. The leverage is determined by identifying patterns in those data points that correspond with times when the different agent models worked well historically.

Just as a proportion of an ant colony remains inactive at any one time, Swarm’s agents can simply choose to avoid making a trade on any given day if they lack sufficient conviction on their respective market.

But if one agent decides to sit out a trading session, then another can ramp up its own position.

Deciding to simply do nothing “is not a mammalian instinct”, says Drew. “You will rarely see traders stand up and say, ‘I don’t have a particularly strong view, I’m going to give up my conviction to one of the other systems that has a stronger view or appetite for risk.’”

It is this unusual feature that Drew believes is responsible for the majority of the strategy’s alpha.

Agents can also trade in the opposite direction to others, to provide diversification or to ward off a potential reversal.

Slow to react

Unusually, ants don’t act on information that arrives from one part of a colony quicker than another. Instead, decisions are made based on the information received from the whole.

Yiannakou has attempted to mimic this feature by feeding the programme just one piece of price data each day. The advantage is that it limits the degree to which the programme is influenced by market noise.

For example, if a trader sees an announcement that a central bank is suppressing government bond yields and decides to take a position using this information, they risk being whipsawed by another announcement that reverses this trend later that same day.

Agents “make decisions based on the information that they’ve received from the whole, rather than trading very high frequency with lots of noise and potentially getting caught out,” says Drew.

But allowing agents to glimpse just one piece of price data per day means that Swarm XVI can react sluggishly to unexpected events in financial markets.

Photo: Flickr/Hungarian Snow

The aerodynamic quality of Japan’s bullet trains was inspired by kingfisher beaks

September was “a particularly poor month for us,” says Drew. The strategy lost 5.42% in September, putting a dent in annual returns which stood at 14.61% net of fees as of end-October. September’s loss was a result of the model building up a significant risk-on position in equities at the start of the month. All four of the agents covering stocks held long positions for the second time in three years.

This caught the model off-guard. A higher-than-expected US inflation print on September 13 resulted in a repricing of rates, currencies, commodities and stocks. In the following 14 days, the strategy’s monthly return went from +5% to -5%.

The strategy’s return profile is different from that of other systematic approaches, though. Drew points to the model’s performance in November 2021, which gained 3.71% net of fees when the CTA Index return fell 3.72%.

Swarm uses in-built and discretionary risk management tools to keep risk equal between the four markets it trades and to monitor overall risk levels, tweaking each agent’s position size weekly. A variety of other metrics are also checked each quarter – such as daily profit and loss and a market’s volatility – to ensure that they remain within historic parameters. If volatility strays near or above precedented levels, then the entire portfolio is degeared in 25% increments.

Blunter discretionary tools allow Yiannakou and his team to manually degear the portfolio if they feel that it is necessary, a process that is used around five or six times per year. Clients investing via managed accounts also have the option of setting a variety of hard stops, such as a percentage daily loss, sectoral limits, or a ceiling on margin, which are enforced by the fund manager or broker.

Sideways expansion

The hedge fund recently sold an additional $300 million of capacity for the strategy to two investors. The firm expects the extra investment during the next 18 months. Yiannakou and Drew are confident that the programme can grow into a multi-billion-dollar business.

Rolling out Swarm XVI to a larger pool of investors is tricky, however. Upping the number of agents in the programme requires reams of processing power: doubling the number of agents in the programme would require the computing power of Nasa. Trimming the number of agents in the programme also erodes the benefits of sharing information. “Sixteen [agents] is pretty much the sweet spot for us,” says Drew.

Instead, Swarm is mulling the possibility of expanding the programme ‘sideways’, by setting up a separate programme that does not communicate with the original. A new programme would use the same structure as its predecessor: 16 agents trading across four markets.

Editing by Alex Krohn