The NASDAQ, the second-largest U.S. stock market, if we address the market capitalization of listed companies, has so far been operating in an automated manner, but without safeguards in the face of attempts at market manipulation, a deficiency that it now seeks to remedy by implementing mechanisms based on artificial intelligence.
Today, for example, an investor may carry out an excessive amount of orders to sell a company’s shares, cause the value of shares to fall, cancel those orders and then start buying more shares of those companies, whose prices are would be at a low by their action.
This type of fraud is known as ‘spoofing’, and it is difficult to avoid because traders can resort to algorithms to issue orders and cancel them almost instantly, over and over again. That’s why NASDAQ officials are looking to address these kinds of threats by using the same weapon: AI.
The tool they ended up developing, based on deep learning technology, was launched a few weeks ago, a year after the project began: the goal is to be able to analyze the movement of index actions (billions of actions on a normal day) to detect the patterns characteristic of a fraud operation.
In the words of Michael O’Rourke (head of the NASDAQ AI Team), the good thing about deep learning is that it is “useful for finding things that are very difficult to describe.” O’Rourke built a dataset based on the historical trading data of Nasdaq. Given the large volume of data handled daily by the index, the quality of this training data when it comes to identifying patterns is very high.
The team then built an AI model that would examine trade data and identify all the activities that deviated from normal market trends. Such activity would then be analyzed by humans to determine whether it is benign or whether, on the contrary, it would be worthy of further investigation.
The future of algorithms in the financial field
One of the fears about the use of this AI was that it generated an avalanche of false positives, which would burden the personnel in charge of reviewing suspicious operations. However, O’Rourke believes that for now the false positive rate is “acceptable”.
The company had previously experimented the use of AI for monitoring purposes at other stock exchanges operating in Northern Europe following its acquisition of the Baltic-Scandinavian operator OMX; however, this is the first time NASDAQ has applied artificial intelligence in its Us market.
The plan is, if the experiment is successful, to turn what is known as ‘learning transfer’: apply what we have learned from this model to monitor other markets where less historical information is available. Ultimately, they hope to be able to implement this technology in the software they sell to other operators, such as the Hong Kong Stock Exchange.
Martina Rejsjo, head of Nadsaq’s market surveillance team, says that – regardless of the outcome of this experiment – AI will play a key role in the supervision of stock markets in the future, owing to the large volume of commercial information that needs to be analyzed on a daily basis. For Rejsjo, the role of humans in this task will be relegated to checking the warnings of the algorithms.
However, as the NASDAQ begins to bet on automation, the largest U.S. stock market, the New York Stock Exchange, is betting on going the wrong way: leaving human-de-being-only.
Thus, since August, quoted funds based on bonds, commodities and currencies, which operated on their own platform (the NYSE Arca Exchange, based on automated algorithms), have gone on to do so in the NYSE parquet, where the ‘designated market makers’ (DMM), can monitor operations related to these funds, in order to reduce their volatility.
A special reporting from Aarti Dhar and A Vaidyanathan.