The AI-driven trading firms are worried about the risks AI poses to profits

After a fake photo of an explosion near Pentagon caused a short sell-off on US stocks, hedge funds and other computerized trading firms are increasingly worried about the threat artificial intelligence poses to their profits.

The S&P 500 index dropped 0.3 percent within 30 minutes last month, after a tweet that went viral from a Twitter account with a blue check showed an image of an explosion which never occurred. Bellingcat, an investigative website, and others speculated that the image spread quickly on social media, and was soon shown to be a fake.

The incident highlights how AI-generated images and news could be a problem for Hedge Funds, and ultrafast proprietary trading companies that use complex algorithms in order to scan vast amounts of social media and news for market-moving indicators that they can quickly trade.

Executives at quantitative trading firms say that while their computers are more adept at identifying false news articles and social media posts than ever before, machine-generated misinformation represents a new frontier.

Doug Greenig is the founder of the hedge fund Florin Court Capital. The fund focuses on long-term trends on alternative markets, rather than very short-term movements.

AI’s ability to create highly convincing stories and images in large quantities is of particular concern to traders.

This could present a number of problems for hedge funds and proprietary trading firms that have invested heavily for years in algorithms to parse and analyze critical information. They also assess the language used and the sentiment expressed within a source, and then use this data as a trigger for an automated trade.

“We see quants face two hurdles: fake pictures that could fool a reporter, and reports of false images that might fool the algorithm,” said Peter Hafez. Chief data scientist at RavenPack software, which uses AI for reading large amounts of data from banks, hedge funds, and other firms.

Hafez said that powerful algorithms mimic the brain by learning pattern recognition and natural languages. However, they may still have difficulty with a real news report on fake news – for example, a reliable news provider reporting a fake Pentagon explosion. “So [they] could treat them as true events and produce the corresponding analytics”, he added.

Yin Luo is the head of quantitative analysis at New York-based data group Wolfe Research. She predicts a “cat-and-mouse” game between those who spread fake news to move markets and traders trying to stay ahead.

He said that for the moment, investors will likely rely on reputable news and information sources. Algorithms are already being developed to check multiple news sources in order to ensure data accuracy.

A London-based quant funds executive stated that the rise of AI would likely push traders to use data firms that aggregate multiple sources into sentiment scores.

Investors’ concerns over market issues at the time, like the US debt ceiling impasse and the impact of higher interest rates, may have also contributed to the sharp decline in the S&P.

Charles-Henry Monchau said that these factors have led to an increase in the popularity for tight stop-loss order on deals. These orders require that demand positions be sold at a specific price level to protect investors from further losses.

Monchau said that there was a lot of uncertainty in the market right now. “There’s a big tug-of-war going on between bulls versus bears on an intraday level,” he added. “Any sharp movement that is not explained by macro numbers, which [algorithms] recognize, they will react to it and force some sales, accelerating the trend.”

This issue is not likely to affect all quant firms. Quant strategists at leading investment groups said that firms use checks and balances to make sure “dangerous data” does not cause quants to sell in a way that will push prices even lower, which would then trigger more selling. Quants often trade market patterns, rather than social media or news. They also tend to look at trends that span longer periods of time and ignore short-term price movements.

The majority of computer-driven traders make many small bets to minimise potential losses resulting from price movements originating from unreliable sources. The strategist stated that “bad data of any type is a major concern, but has been for a long time.”

Kit Juckes is a macro strategist with Societe Generale. He said: “In a way, we’re back in the past when there wasn’t accurate and fast news.” This is another step on the road of easy disinformation. It’s made possible, in part, by technology, but also by laziness and regulation. “But yes, a very important step.”

But those who built their business on marrying trading with technology know it will be a difficult road. They are also in the front line.

Mike Zigmont is the head of trading for Harvest Volatility Management, a US-based firm. He said: “Whether the fake story was exploited for profit by its creators is unknown. But there will be many more stories like this for a very long time, and the perpetrators of the crimes will try to extract value from markets.”