A tale of man versus algo in Norway

A court in Oslo has ruled it was market manipulation. For others, though, it is a tale of how two day traders outwitted the rapid-fire machines that have come to dominate financial markets.

Peder Veiby was trying to earn some money in the stock market to support his studies at the Norwegian School of Management when he was hit by a criminal charge for alleged market manipulation. Now he has found himself at the centre of a landmark legal case involving computer algorithms, or programmes, at the heart of automated trading systems.

Mr Veiby and another Norwegian day trader were handed suspended prison sentences and heavy fines on Wednesday after the court found them guilty of exploiting flaws in the electronic trading platform of a US broker to send “false and misleading signals” to the market.

The two men each worked out how to make money by predicting how the computer algorithm of Timber Hill, a unit of US-based Interactive Brokers, would respond to certain trades. They denied that this amounted to market manipulation but the court disagreed.

The case comes amid growing scrutiny of automated trading systems after the so-called “flash crash” in May, when a single algorithm triggered a plunge in US stocks.

Algorithms are computer programmes that have emerged in recent years as trading has become fragmented across many different types of trading venues.

The trading landscape has been transformed beyond recognition in the US, where as little as a decade ago, most stock trading was executed manually, either on the floor of the New York Stock Exchange or on traders’ desks.

The algorithms, “algos” in market jargon, are used by brokers and asset managers to help navigate this complex trading landscape. They are also used by firms using their own money to trade to submit thousands of orders in the blink of an eye. One type of algorithm, known as “efficiency” or “scheduling” algos, takes a large order, splits it into smaller pieces and sends it out to find a match periodically, finding the best possible price at the time.

Anders Brosveet, lawyer for Mr Veiby, says his client noticed a pattern in how the algorithm behaved while he monitored a long-term investment he had in a lightly traded Norwegian stock.

Mr Veiby found that the bid and ask prices moved up and down in tandem after each trade, making it easy to predict the spread between them.

He also noticed that the algorithm would respond in the same way to a small trade as it did to a larger one. This allowed him to buy a large number of shares at a low price and then make several smaller trades to bid up the price before selling out at a profit.

Svend Egil Larsen, the other defendant and a full-time day-trader for the past seven years, made the same discovery separately.

The two men did not know each other and it emerged in court that they had sometimes inadvertently undermined each other’s strategies as they each made similar trades in low-volume Norwegian stocks whose prices could be moved easily.

“I did not set out to trick the robot,” Mr Larsen told Norway’s Dagens Naeringsliv newspaper. “But after acting against it a few times, you see how it behaves. The computer was fairly obvious.”

The reaction among Norway’s amateur trading community has been largely sympathetic towards the men, even though prosecutors claimed they made tens of thousands of dollars in profits at the expense of other investors. “How can we be able to make money if we are not smarter than the robots?” asked one commentator on an online forum.

That the two men made their discoveries separately has raised questions over whether other traders around the world could be doing the same. “Anyone who is observant and trading in a stock with low liquidity and a stupid computer [algorithm] can do this,” said Mr Brosveet.

Algorithm experts in London said the two Norwegian traders seemed to have been using a “market-making” algo that post bids and offers out into the market, with the trader hoping to make money by capturing the bid-ask spread.

Niels Buhl, senior partner at Arctic Lake, which builds algorithms, said he had no knowledge of the Norwegian case but that it would be possible for someone to work out what an algo was doing by looking at market data, prices and any trading patterns that the algo had produced and working back from there.

“You can measure statistically what is happening in the market and do some forecasting,” he said, but added: “It’s not always easy to guess what an algo is doing otherwise a lot of people would be doing it.”