Predictive analytics in the operational risk framework has been saved
Predictive analytics in the operational risk framework
Integrating new data to optimize risk identification methods
Data’s vast availability makes it possible to evaluate more types of operational risk than ever before. The challenge for organizations is to evolve their data architecture and models to support forward-looking risk management and more accurately determine their risk exposure.
The evolution to predictive risk identification methods
Globally, more banks are trying to make their operational risk management programs more forward-looking. Banks should seize the opportunities today’s advanced tools and vast data pools make possible. Predictive analytics techniques, machine learning, and artificial intelligence can help efficiently build and mine large and complex data sets that combine traditional Basel operational risk loss data with other data sources, including transaction data, non-transaction data, and external data.
These aggregated data sets provide billions of data combinations that can drive vastly improved risk identification methods through analytical results and insights. The data combinations can also greatly increase the likelihood of uncovering patterns and correlations that previously weren’t noticed until it was too late—if ever. This can help an organization prevent unpredictable outcomes and reduce operational losses and capital impacts.
This paper highlights one of the implementation challenges to actualizing a more predictive operational risk management program: the need for the evolution of the data architecture and models.
The phantom menace
The initial constraint for the design of the historical operational risk data model was the singular objective of facilitating the estimation of conservative capital so the organization could absorb the impact of loss events. By design, this made it backward looking.
Historical data models did not comprehensively contain information related to all types of operational risk exposures, such as conduct risks, sales practices, and market manipulation—or the subsequent losses that could occur. Therein may lay the “phantom menace”: risks that are already materializing but with losses that haven’t been recognized yet, and thus have not been captured in the data model or in the quantification of operational risk
If the operational risk data model captures only losses that have arisen in the past, the model does not reflect the current risk exposure of the institution and potential future loss. In this age of rapid technological and business disruption, few organizations can confidently and credibly claim to capture that view.
A new hope
One better way might be to learn from techniques derived from outside risk management, such as customer marketing and sales. These disciplines have well-grounded techniques to help understand customer behavior to generate additional sales and further build customer loyalty.
In operational risk management, we should emulate similar successes and begin to collect wide-ranging data through systems, applications, and processes—and through human interactions. Then we should derive meaningful patterns and behaviors in line with the unique risk challenges of individual organizations and lines of business.
Only through the collection of this data at the broadest level can we identify patterns and behaviors and thus determine which data is truly risk-sensitive. We should look beyond losses if we hope to accurately determine the operational risk exposure of a firm.
The data awakens
Organizations now have the opportunity to expand the traditional operational risk data model. As organizations undergo digital transformations, the availability and range of data become easier to access and more readily available for consideration and potential inclusion in the newly defined operational risk data model.
Moving toward a broader and more dynamic data model can open the door to more effective use of predictive analytics techniques and allow data science techniques to assist organizations in understanding risk drivers, themes, and behaviors. The defining effect of these dynamic operational risk models can permit greater predictability and probability for organizations to determine their current level of risk.
Stated differently, this new operational risk data model should be developed with defined rule sets that fuel deeper behavioral analysis, trend identification, and predictive analysis. Each organization will need to develop a unique set of characteristics and a bespoke implementation plan for a dynamic operational risk data model in line with its system and application architectures.
For more details, including illustrative examples, download the full report.