Milliman has announced that an innovative new study examining multi-criteria decision-making using an iterative process of advanced computing and human input has shown superior results in risk management when compared to machine algorithms or humans alone.
Using an illustrative example from the life insurance industry, the study looked at how optimisation techniques can be used to develop insights into drivers of economic capital within an internal model framework, and how to then use these insights for risk management decisions. The findings illustrate that advanced computing, visualisation, and complex systems-mining techniques that include expert input can deliver superior optimisation results when faced with multiple objectives and multiple constraints which machine algorithms alone find challenging to resolve.
While not obvious at the outset, combining human input with advanced computer modeling allows domain experts to analyse results and elicit insights into features that subsequent iterations of a model should contain, thereby refining the process.
Milliman’s study employed the DACORD platform from DRTS, Ltd. to support its system-mining efforts. “Future states are unknown, involve human affairs and are therefore complex,” says Jeff Allan, CEO of DRTS, Ltd. “Augmenting experts with the appropriate tools and processes can aid the reasoning and evaluation of a range of solutions.”
Adds Milliman’s Corey Grigg, “Looking toward the future, this sort of optimisation technique can extend to big data, simulations, and enhanced visualisation, ensuring that even as the complexity of our data and problems increases, experts can continue to add value.”
The results suggest a number of practical applications for Insurance ERM, including finding patterns in key risks driving capital losses and understanding diversification in order to enable quick judgements about the similarities and differences in the risk profiles of different portfolio elements.
Milliman’s Optimisation study was conducted in conjunction with Dr. Lucy Allan of University of Sheffield. To read the entire study, click here.