Shedding Light on Dark Data

The term "dark data" and what it represents is being raised more often lately by industry service providers and consultancies as an issue of concern. With this increased attention, some are uncertain about what exactly "dark data" means in the context of financial services industry data management.

Dark data shouldn't be understood literally, as though it were about using data for "nefarious purposes," says Charles Fiori, a consultant with long industry experience at firms and service providers, including Bear Stearns, BrokerTec, Goldman Sachs, JPMorgan, Lehman Brothers and Reuters.

Fiori defines dark data as data whose existence is either unknown to a firm, known but inaccessible, too costly to access or inaccessible because of compliance concerns.

Defining dark data also means accounting for issues other than accessibility, according to Tim Lind, global head of financial regulation solutions at Thomson Reuters. "It's about connection and drawing connections," he says. "It's data that we store anyway for compliance or business purposes, that's sitting there waiting for you to discover its true meaning-or it's looking for, as we do in the regular analytics world, the predictive value of data when it comes together."

Fragmentation is a major challenge for connecting data to collect and reconcile all dark data. "We see some institutions out there with 30 trading systems that grew up over time in different regions servicing different businesses, but still within the same family," says Lind. "So the data standards, the operational structure and, daresay, the governance of how data is managed and how data standards are propagated within an institution, are the core challenges for getting data together."

Risk Aggregation Aims

The true purpose of collecting dark data by aggregating data is to aggregate all available risk information, says Lind. "Any data not being utilized properly, is unable to be connected, or is outside the grid somehow because of structure, operations or governance perspectives, is dark data," he says. "That data tells an interesting story about the risk an enterprise has with a counterparty."

BCBS 239, the Basel Committee on Banking Supervision's risk data aggregation principles, are challenging firms to access and manage their dark data, according to Lind. These rules, which have a US equivalent in CCAR (Comprehensive Capital Analysis and Review), require firms to stress test the risk in their holdings. To meet the requirements of these principles, firms must collect and aggregate data in a similar way as they would to obtain dark data.

"A lot of these principles are still outside the grasp of most financial institutions," says Lind. "They're still not quite able to link transactions and positions across asset classes, lines of business and geographies, to get a bank-wide picture of their exposure to a counterparty, to an issuer or to a country. These core risk aggregation and management items are about the ability to connect data we store already, a.k.a. dark data."

In fact, BCBS 239 may be spurring firms to recognize dark data as an issue, according to Hubert Holmes, managing director, reference data at Interactive Data, a data services provider. "Dark data has not been much of an issue until big data, data management, data governance, regulatory and risk mandates have brought it more to the attention of firms," he says.

Potential Solutions

Solving the issues inherent in the nature of dark data would most likely require structuring, tagging or organizing the data in a database. If that has not been done, or is not possible, the solution may be doing an audit, according to Holmes. "There are many risks that arise for having such information but not really managing it," he says, adding that these include legal or regulatory risks, intelligence, reputation and operational risks. "Some dark data is worthless and some is highly valuable, and everything in between. You can't tell without an audit."

Devising a way to address dark data is necessary to have a foundation for advanced analytics, according to Lind. "Start a little closer to home," he says. "We have a ton of work to do in structure, data standards, aggregation and drawing those connections to the most common objects like instruments and entities."