CLS Looks for Value in Discarded Data

The information services division of foreign exchange settlement specialist and data provider CLS is sifting through years of “discarded data” to develop a new alternative data product that can be commercialized.

While the program is still in its infancy, the hope is to find new and more useful ways to wrap and combine existing CLS datasets to create new insights, says Masami Johnstone, head of information services at CLS.

“By really interrogating CLS’s data, we might identify interesting behaviors or relationships between counterparties, groups, or certain geographic locations, which is not something we have investigated previously,” she says.

Using an automation tool from start-up causaLens that sifts through data using machine learning to find patterns, CLS plans to look deep inside its systems to mine data from as far back as 2011. Johnstone says the key is to locate well-populated datasets that can be used to create new data packages.

“Using AI or machine learning to mine existing datasets is an important process to find valuable insights. AI can bring together a combination of different datasets and examines the intricate relationship between them,” she says

CLS receives vast amounts of data from the settlement of FX transactions that flows through its platform, and that data contains different tags, metrics, and messages. Johnstone says her team has already taken steps to clean its current dataset and has aggregated it at a high level. “During that process, we discarded certain information deemed unnecessary. Now, our data science team is re-reviewing that discarded data to search for any untapped insights,” she says.

Last year, even before the coronavirus greatly disrupted the FX market, Johnstone assembled a data science team in an effort to decode volatility in the market. She put together the three-member unit after she noticed an increased need for information on FX market dynamics from investors. Each of the data scientists on the team is a physicist or scientist. All were previously part of the broader CLS data analytics team but were recruited by Johnstone for this specific project.

Causal AI

Darko Matovski, CEO of causaLens, says that his company’s technology looks to imagine scenarios and answer ‘what if?’ questions by recognizing causal links between data to understand what the drivers of a market could be.

He says current machine learning techniques based on correlation are unable to distinguish between “spurious” correlations and true causal drivers. CausaLens licenses a product that clients can use with their own raw data. When implemented, the tool sits on top of a platform and runs through the data in the background at all hours of the day. It then uses AI functionality to correlate and decode relationships between multiple data points quickly.

“The machine can go day in [and] day out trying to find value in the datasets without the company having to put additional resources into it,” Matovski says.

The product can be distributed through a web interface that is accessed on a browser. It can also live behind the organizational firewall and connect to clients via an API. “Users can interface with our technology in multiple ways. They could get an email, look at a dashboard, and include it in their codebase,” Matovski says.

“Normally, traditional machine learning techniques have a limit to how much data they can process because they attempt to fit a curve through the data,” Matovski says. The causaLens technology can go beyond curve-fitting and process data in a streaming fashion, down to the tick level.

The product is aimed at traditional analysts. They can access the information through a user interface to view a diagram that displays the relationship between variables. The product has also been licensed to quant funds who use the API to access information programmatically through code.

The process for CLS involves pulling metrics from what is essentially a raw data table and applying the causaLens AI to determine whether there are any significant or sustainable signals that can predict FX volatility or currency pair movements. 

Matovski says the insights gained from applying the causaLens technology could be used to inform trading strategies, or by chief financial officers and treasurers to review hedging blueprints in the light of changing market dynamics as a result of the coronavirus. 

“As data is growing exponentially, and the data science talent is simply growing linearly, there’s a huge gap [for technology],” he says, adding that this kind of data-mining process is exploratory and unpredictable, as well as expensive. “A lot of organizations just don’t do it because it’s a huge investment upfront with an unknown return.”