Data Doubles Every 3 Years
The volume of data continues to double every three years as information pours in from digital platforms
The
Machine Economy
According to David Cox, assistant professor of molecular and cellular biology and computer science at Harvard, the best way to advance computers so they can perform and think more like humans—a trend that could eventually help businesses gain lucrative insight from vast amounts of data—is by tapping into brains. Literally.
Paid Post for
Forward-thinking enterprises like Nasdaq, Inc., are leveraging artificial intelligence, and investing in machine learning technologies that can analyze and interpret data in a human-like manner. These technologies can generate productivity gains and cost efficiencies, but the big payoff is being able to leverage the machine to process massive amounts of data to extract information beyond a human’s capacity. Finding these “unknown unknowns” is where Nasdaq intends to differentiate itself from the competition.
“Machine learning and artificial intelligence—is a capability that can be applied to financial services very broadly,” said Brad Peterson, Executive Vice President and CIO of Nasdaq. “Nasdaq was a pioneer in moving from the human-based trading pit venue to fully electronic; this is the next phase of evolution in making the machine-to-machine markets function with unprecedented agility, accuracy, and intelligence.”
The Machine Economy
Machine Learning 101
We are still in the early days of a widespread rollout of artificial intelligence in business applications. Along with the fairly recent explosion of data flowing from the Internet, it’s only been in the past few years that the hardware capacity for data storage has increased enough to allow data scientists reasonable computing power at lower costs, according to Aditya Kaul, research director at research firm Tractica. That said, as the volume of data continues to double every three years as information pours in from digital platforms—according to a December 2016 report from the McKinsey Global Institute—analysts and corporations are betting on AI as the next big technological revolution.
“Artificial intelligence is changing the way that we perceive and interact with technology,” said Kaul, whose firm predicts that the $600 million market will rise to $36 billion in revenue by 2022. “It will touch almost every industry, just like software transformed every known industry.”
Deep learning—a subset of machine learning, which enables computers to replicate human decision-making through analyzing multiple levels of representations of data—will be the leading AI technology, accounting for 44% of all the sector’s revenue between 2016 and 2025.
“In industries where integrating more and better data can dramatically improve performance—such as banking, insurance, retail, the public sector, and beyond—the organizations that master this capability can realize major advantages,” said the McKinsey Global Institute report, The Age of Analytics, Competing in a Data-Driven World.
For Nasdaq, this means leveraging machine learning to better analyze the growing influx of data that it harnesses from around the world in running its exchanges.
“Our customers are demanding more intelligent solutions that provide their trade surveillance analysts and supervisors with an holistic view of what their traders and customers are doing,” said Bill Nosal, Head of Business Strategy and Development for Nasdaq’s Market Technology group. “They want to coalesce information from trading systems, news, research, market events, and communications to derive the full context of the trading activity. Moreover, the art and practice of Surveillance is advancing from an exclusively event-driven, reactive model to include predictive mechanisms that identify high risk entities who are likely to engage in inappropriate trading behaviors.”
How Nasdaq is using artificial intelligence to make its SMARTS Surveillance technology smarter
Data scientists at Nasdaq are using machine learning to provide more insight into the behavior of traders for customers.
LAYER 1
Data Ingestion Stage
Here information is fed into the software, including: billions of market transactions (orders and trades), news that may impact a stock, bits of e-mail and online chats from traders.
LAYER 2
Relating Behavior to Context
Here is where SMARTS forms a tick by tick view of the market and determines key behavioral metrics and figures out what are normal trading behavior patterns.
LAYER 3
Finding the Red Flags
Here is where SMARTS starts to detect bad trader behavior. SMARTS clients add in their own elements/factors to weed out false positive alerts.
LAYER 4
Ranking Alerts
Here is where SMARTS determines which alerts need more immediate action/attention over others.
One More Thing
Feedback loop
The analyst’s classification is used to train the model for the ranking of future alerts.
Machine Learning
SMARTS Surveillance Technology
Collect Text in Traders’ Email, Chat Room Conversations, Tweets and Other Communications
Analyze This Unstructered Data to Determine the Veracity of Manipulative Behavior
Score and Rank Alerts to Be Sent to Compliance Officials To Improve Efficiency
Nasdaq is currently using machine learning to provide more insight into the behavior of traders for customers with its SMARTS Surveillance technology, which powers surveillance and compliance for:
45
Marketplaces
17
Regulators
140+
Market Participants
“It’s about tackling the data size problem—what information can we extract and interpret from the numerous data sets that we’re collecting,” said Andrew Franklin, SMARTS lead developer.
Nasdaq SMARTS merges the private trading data of market participants, exchanges or regulators with public market data, news analytics, and other reference and contextual data – providing more context to and insight into traders’ activity. Based on rules written into its technology infrastructure, SMARTS can provide alerts to compliance staffers when the surveillance platform senses manipulative behavior, such as spoofing, insider trading or price fixing. “Once an analyst takes a hold of an alert, they’re using prior knowledge to work out whether or not to escalate it,” said Franklin. “That human intuition—about what makes an alert interesting or not—is something that we’re attempting to learn.”
Machine learning is helping Nasdaq rank and score the alerts, giving it a newfound ability to let customers know which activities might need more immediate attention.
“Scoring is one mechanism we provide to help customers minimize time investigating false positives,” said Valerie Bannert-Thurner, Nasdaq’s head of risk and surveillance solutions. “Our objective is to help our customers and users to focus their time and attention on those alerts and outlier events that, given past experience, are highly likely to represent significant issues that are worth a detailed review. Pointing users to what is ultimately going to be most important quickly will result in a huge efficiency boost for the surveillance analyst,” she said.
Key in perfecting the alert rankings is Nasdaq’s partnership with cognitive computing company Digital Reasoning, whose natural language processing expertise is helping Nasdaq analyze text in traders’ e-mail, chat room and tweets. The analyses of this unstructured data are then applied to the alerts. “We can now commingle the analyses of these unstructured data sets with the behaviors identified from the trade and order data,” said Nosal.
“The key is linking the system, data, and alerts and map them into suspicious behaviors”
“Being able to sift through millions of different data elements to identify the communications that are definitively related to specific trading activity presents the contextual understanding that our customers have been asking for.”
While this first phase of the partnership with Digital Reasoning—linking trading data to any related contextual analysis—has already kicked off, Bannert-Thurner is looking ahead to phase two, which calls for creating behavior profiles for certain traders who have elicited unusual patterns of behavior in the past. The final phase, she said, will pull everything together onto one platform, giving Nasdaq and its customers a way to predict where and from whom bad behavior might arise before it actually does.
“The key is linking the system, data, and alerts and map them into suspicious behaviors,” said Bannert-Thurner. “Once you bring all the different data elements together, and apply our Machine Intelligence tools for clustering, outlier detection, ranking and scoring to it, you can really retrieve value from it.”