11.19.2019
By Shanny Basar

Nasdaq To Expand AI for Surveillance With ‘Transfer Learning’

Nasdaq has introduced artificial intelligence for surveillance in its US equities market and will use transfer learning to allow the technology to be utilised in other markets and asset classes with far less data.

Tony Sio, vice president and head of marketplace regulatory technology at Nasdaq, told Markets Media: “The same technology could be used in other areas of the exchange or to do surveillance of innovative new markets. Nasdaq is looking to use its technology in other types of markets such as digital assets or sports betting.”

The technology took just over one year to develop in a collaboration between Nasdaq’s market technology business, its machine intelligence lab in Boston and the company’s US market surveillance team in Washington DC. The surveillance team currently reviews more than 750,000 alerts each year that flag unusual price movements, trading errors and potential manipulation.

Nasdaq has a patent pending in the US for the new technology which provides unique artificial intelligence capabilities for surveillance including deep learning; transfer learning and human-in-the-loop learning. Deep learning allows computers to understand complex patterns and hidden relationships in large amounts of data.

Tony Sio, Nasdaq

“There are many scenarios which are very hard to code for, with lots of subjective parameters, but the analysts will often immediately recognize the scenarios when they see it,” Sio added. “We feel the module will work particularly well for these cases.”

The new technology has been launched in US equities with the intention of migrating it to other Nasdaq exchanges including its Nordic markets; and then to other exchanges and regulators via the Market Technology business. The range of scenarios the system detects will also be increased.

However some smaller clients and exchanges will not have the volume of data or examples to train an algorithm to recognise anomalous patterns.

Sio explained that the concept of transfer learning is that the shape of the suspicious behavior has some consistencies and those learnings can be transferred to other markets and asset classes such as non-US equities, commodities, foreign exchange and fixed income.

“In some of our test applications of transfer learning we managed to achieve usable results in new markets with 10 times fewer training inputs, so savings are significant,” he said. “For example, if a new model had needed 1,000 inputs, then only 100 cases will be sufficient.”

Michael O’Rourke, senior vice president, head of machine intelligence, at Nasdaq, said in a statement: “Artificial intelligence and machine learning have broad application across our company – from predicting market trends with Nasdaq’s proprietary data or creating more sophisticated market surveillance capabilities.”

In Europe regulations such as MiFID II and the Market Abuse Regulations have led to an increased emphasis on having adequate controls in place to investigate potential suspicious behaviour.

Sio said: “Regulation in Europe has led to everyone looking for innovative ways to make surveillance more efficient, and the application of machine learning and artificial intelligence is a natural progression that regulators need to be on top of.”

Nasdaq’s latest annual study of market surveillance in global capital markets found that nearly half, 42% of respondents, had recently invested in AI/machine learning. In addition two thirds, 65%, plan to invest over the next 12 to 24 months.

Owen Lau, an analyst at Oppenheimer, said he had a positive view on Nasdaq due to the exchange’s involvement in emerging technologies including AI and blockchain according to the Seeking Alpha blog.

“We are specifically impressed by the level of focus that Nasdaq is allocating towards the monetization of these technologies,” said Lau.

Related articles

  1. Market surveillance gets a boost from disruptive technology.

  2. Structured data will push the platforms beyond their pre-investment tasks.

  3. Legislators need to pay attention to results instead of the methods.

  4. The new offering supports reconciliation, matching and exception management applications.

  5. A strong data foundation, dedicated AI teams, and C-suite buy-in are needed for success.