Choosing the Right Frame for Analyzing Data

The leaders of financial firms are beginning to realize the value of data analytics and the necessity of having the right data governance framework to direct the use of analytics, according to operations, analytics, marketing and business executives at firms, service providers and a government-funded research organization.

Ways to define data analytics issues and map data accordingly are beginning to emerge, these executives also say.

The first step involves C-level executives recognizing data analytics problems, says Leif Hanlen, a business development executive at the Commonwealth Scientific Industrial Research Organization (CSIRO), which is funded by Australia's national and state governments to develop digital operations for the country's domestic industries, and Data61, a unit of CSIRO focused on industries' data requirements.

"What's changed in the analytics space is that the hardware's exponential growth has become doubly exponential in the software world," says Hanlen. "So we're seeing exponential on exponential growth. In the analytics world, that means the intelligence you can expect today is almost unpredictable, given what you might see in a few months from now.

"The nature of analysis means the problems that are solved tend to be solved much faster and more broadly," he adds. "The question tends to be how do you make all of this smart so it works with your enterprise system, which is obviously not evolving at that same speed?"

In the absence of proper governance and proper provenance… then what we get at the end can be a case of ‘garbage in, garbage out
Leif Hanlen, Commonwealth Scientific Industrial Research Organization

Data analytics used to be called "decision support systems," observes Darie Urbanky, vice president of investment and wealth management services IT at CI Investments in Toronto. "We depend so much on data," he says. "The groups I support, such as trade order management systems, are incredibly sensitive to data. We're in the midst of adding risk and performance attribution systems in the groups we're managing. It's incredible how much data is required and the data quality that you need."

Mapping data using analytics can provide value to the front office in the form of business intelligence, says Urbanky. "Business questions never come up just a single time," he says. "For unstructured data, [mapping] is still valuable and relevant."

Mapping can include machine learning techniques, says Brian Sentance, CEO of Xenomorph, an enterprise data management (EDM) services provider. "There are rules-based mappings where people prescriptively define what the mapping is. Then there is ‘fuzzy logic,' which actively extends what you have, using defined rules, to do matching and reconciliation," he says.

Flexible Frameworks

Data analytics systems built to work with firms' existing operations systems, such as those created by business intelligence and visualization tools provider Information Builders, might not need data governance plans that are as strict as those required for operating data warehouses, says Jake Freivald, the company's New York-based vice president of product marketing.

"Not everything has to be governed the same way," he says. "Compared to a strict, rigid and top-down approach, that's a dramatic shift by the analysts, architects and others who are now trying to achieve a balance in governance. It's a big change of mindset."

Still, some governance is necessary, as Hanlen explains. "In the absence of proper governance and proper provenance—as in where the data came from and who did something to it—then what we get at the end can be a case of ‘garbage in, garbage out,'" he says. "It's always a risk. There are ways of mitigating those risks, but if you don't know how the data was put into the system, you're never entirely confident that what comes out will be based on reality, rather than someone's version of it."

If a governance framework cannot be found for data being received, firms' ability to best use that data depends on providing "appropriate incentives and governance so we don't really ever throw data away," says Hanlen. "We can always convert it, change it or refine it, but we need to build in approaches that treat data as sacrosanct. People will always want to go back to the source."

Iterations and Sharing

Defining solutions through data governance plans is more challenging because the "iterative nature" by which data scientists work is changing, according to Baiju Devani, director of analytics at the Investment Industry Regulatory Organization of Canada (IIROC).

"Traditionally, it's been more along the lines of ‘you tell me why you need the data and what you need,' then we figure out a solution for you," he says. "The cultural change for us when analytics are involved is being honest with our stakeholders and saying that, more than half the time, we don't quite know what the solution is ... There's no one fixed definition of the data most of the time. We need to go back to the data over and over again because we're going through this iterative process.

"Having fixed mandates around your requirement, where we go get the data and you [a vendor or provider] come back with the solution—it's not quite like that anymore," Devani adds.

Not everything has to be governed the same way. Compared to a strict, rigid and top-down approach, that’s a dramatic shift by the analysts, architects and others who are now trying to achieve a balance in governance. It’s a big change of mindset
Jake Freivald, Information Builders

Data governance and analysis can potentially be improved by building mechanisms to share data with other organizations, explains Hanlen. "When you have too much data to deal with, the question should be how to build analytics that don't throw material away," he says. "Keep a good sample of the data that lets you infer, even if you can't see the original anymore."

Within EDM, however, data analytics is seen as "downstream," says Xenomorph's Sentance. "The analytics applied to data and the derived data they produce should be treated as an asset to be managed, audited, reported on and governed, just as any raw piece of data would be."

Rules set by data vendors and by exchanges can also present a challenge to sharing data, however advanced and innovative a data sharing system may be, says John Denheen, data team lead at London-based proprietary trading firm Tyler Capital. "We try to work with academics who have some good ideas about strategies but don't have the means to access these larger datasets," he says.

Machine learning can also prove useful for implementing analytics under a governance plan, including following rules for data sharing, as Denheen explains. "One of the big issues with getting good reference data is that you need to be able to tack it on to other datasets," he says. "We might buy market data from one provider and reference data from a completely different provider, then try to join those two sets together. That's a major challenge."