Many startup founders cut their entrepreneurial teeth at an early age, but David Widerhorn’s first company was a far cry from your run-of-the-mill lemonade stand operation.
At age 12, he was was running a computer hardware testing business with seven employees alongside his high school coursework.
Widerhorn shut down that business at 15, when he went off to study at the Massachusetts Institute of Technology, from which he graduated at 19. Widerhorn then spent some time as a high-frequency trader before starting a consultancy that built trading software for more than 100 trading firms around the globe.
Widerhorn’s latest company, Neurensic, helps regulators and compliance departments navigate the new finance landscape. Born out of the consultancy business Widerhorn started after college, the startup uses AI to detect potentially fraudulent and disruptive trading.
“We’ve seen this incredible transformation of how trades are executed,” said Widerhorn. “If you look at markets like the U.S. stock market or markets that trade in derivatives, between 60 and 80 percent of the transactions in those markets are done automatically by high-frequency trading strategies and other computer programs.”
Algorithmic trading poses a number of challenges for regulators. Increasing trading volumes gives law enforcement agencies a bigger haystack to dig through in search of wrongdoers. But the ability to place large numbers of orders in a short period of time also gives bad actors a whole new set of tools that regulators need to keep up with.
“Back in the day, you could tell people not to trade certain kinds of stock, or not to make transactions over a certain size,” said Widerhorn. “But when you have competing strategies trying to deceive each other in a game theory-like competition, the regulations become a lot more complex.”
Rather than forbidding specific kinds of trades, these new regulations get at things like whether a trading algorithm works by deception, collusion with other traders or by disrupting the marketplace. Neurensic’s AI platform is trained to spot those patterns, and as it ingests more market data, it gets better and better at identifying potential issues.
In fact, Widerhorn said, Neurensic’s AI has grown sophisticated enough to uncover new forms of fraud that regulators weren’t aware of yet.
“In a dataset we recently reviewed, we plugged the AI in and we started wondering why it was flagging a particular cluster because we didn’t see anything wrong with it,” he said. “Then we started replaying it in slow motion, and we realized that the trader was spoofing in an entirely new way that we’d never encountered before.”
Spoofing is a classic example of the kinds of activity Neurensic is well-equipped to detect. The trading strategy relies on deliberately manipulating the price of an asset, usually by placing large quantities of trade orders with the intention of canceling them before they go through.
Because order cancellations are commonplace and often legitimate, looking for large cancelled orders will create a lot of false positives. Looking at patterns of cancellations, on the other hand, can give a regulator or compliance officer a far better picture of what’s going on.
For enterprise clients, Neurensic can generate an “integrity score” for each trader — an assessment of whether that trader’s algorithms rely on strategies deemed suspicious by regulators.
Widerhorn said these scores can help firms shut down malfunctioning algorithms, and lead to more constructive conversations about why certain strategies need to be abandoned, even if they are profitable.
Neurensic has 25 employees, of which 18 are based in its Chicago office. The company also has an office in Silicon Valley. Widerhorn said his team is currently researching other uses for its AI platform within the finance industry, including credit card fraud prevention and money laundering investigations.
Neurensic has raised $9 million in funding to date.
Image via Neurensic.