The Rise of Trading Bots

The promise & perils of AI-powered trading algorithms

An interesting relationship between tech and trading has developed over the last couple of decades. Electronic trading platforms such as E-Trade allowed people to trade financial assets from the comfort of their own homes. Then as computing power grew, it allowed for high-frequency trading in fractions of a second. Mobile apps such as Robinhood put that ability in the palm of people’s hands. Naturally, with the rise of AI, trading bots came along for the ride. Now, traders don’t even have to conduct the trades themselves, they can create a program that does it for them. Since these trading bots are automated, they can make transactions faster and without human error, setting up the potential for big profits. While that’s an exciting development, there are some things to watch out for. Should your trading bot do something illegal, with or without you telling it to, it won’t be the program that gets arrested.

Going Rogue

There haven’t been any real-world instances of trading bots going rogue yet, but regulators are certainly wary.

SEC chair Gary Gensler has warned of “the possibility of AI destabilizing the global financial market if big tech-based trading companies monopolize AI development and applications within the financial sector.” Regulators “have repeatedly highlighted the potential for AI to inadvertently amplify biases that could lurk in their designers, further jeopardizing competition and market efficiency.”

In a test by Apollo Research, an AI safety watchdog, a GPT-4 bot illegally used inside information to make a trade that would benefit the fictitious company it was acting on behalf of. Having been told the company was in dire financial straits, the trading bot decided to use information about a potential merger to make a trade, even though it had previously acknowledged it should not do so. When asked, the bot denied doing insider trading.

“This is a demonstration of a real AI model deceiving its users, on its own, without being instructed to do so,” Apollo Research said in its report to the UK government’s Frontier AI Taskforce. “Increasingly autonomous and capable AIs that deceive human overseers could lead to loss of human control.”

The firm’s CEO, Marius Hobbhahn, said that an AI model acting this way in the real world was unlikely, but that the test shows it is easier to train trading bots to be helpful than it is to train them to be honest.

“In most situations, models wouldn’t act this way. But the fact that it exists in the first place shows that it is really hard to get these kinds of things right,” he told the BBC.

Artificial Stupidity

With Nasdaq getting SEC approval to debut an AI-driven order type and top investment firms such as Blackrock and J.P. Morgan using AI, market watchers are on the lookout for potential unintended consequences. AI algorithms could learn to collude via programming that instructs them to avoid outlier behavior or through homogenized learning biases, worry Wharton finance professors Winston Wei and Itay Goldstein. Along with Yan Ji of the Hong Kong University of Science and Technology, they authored a paper called “AI-Powered Trading, Algorithmic Collusion, and Price Efficiency.”

The trio conducted a series of tests to study the potential effects of collusion between autonomous trading algorithms. Their key findings were that by manipulating excessively low order flows, “informed AI speculators” could “achieve supracompetitive profits” and that tacit collusion could be sustained “through the use of price-trigger strategies.”

One way to combat this is to avoid homogenized learning biases, which create what the authors term “artificial stupidity.” That is, trading bots should not be designed by the same people or using the same basic strategies.

“Collusion through punishment threat (artificial intelligence) only exists when price efficiency and information asymmetry are not very high. However, collusion through homogenized learning biases (artificial stupidity) exists even when efficient prices prevail or when information asymmetry is severe,” they wrote.

Crunching Numbers

These concerns are important because of the widespread effect financial markets have on the economy and society, and responsible development of AI systems is paramount. Fortunately, none of these worst-case scenarios has occurred, and trading bots are capable of yielding massive benefits to users. They can crunch huge amounts of data very quickly, spotting patterns a human brain might not be able to pick up on. They can use this analysis to make predictions on where certain stocks or commodities might be headed.

With their ability to better quantify risk, they can help manage it, and allow users to diversify their portfolios in a way that maximizes returns. They can forecast anomalies in markets and spot bubbles forming, allowing users to exit before those bubbles burst. Because they can process so much data so quickly, they might even help regulators spot evidence of market manipulation or insider trading before bad actors can cause too much damage.

Leading trading bots such as Trade Ideas provide simulated training for beginners before they make live trades, and suggested Entry Signals and Exit Signals alert users to when stocks are hot and not. TrendSpider’s Strategy Tester and StockHero’s simulated exchange let users play around with various strategies, homing in on the right amount of risk and reward before going live.

With all the advantages they offer, AI trading bots are set to transform financial markets. As with any transaction, let the buyer beware.