Traditional portfolio optimisation techniques often fall short when dealing with multiple objectives and complex, non-linear constraints—especially in the insurance sector where risk capital requirements add an extra layer of complexity.
In a recent Insurance Asset Risk webinar, in partnership with Ortec Finance, Iain Ritchie, insurance solutions at M&G Investments and Ashish Doshi senior business specialist at Ortec Finance explored how Scenario-Based Machine Learning (SBML) can redefine the optimisation process by leveraging stochastic scenarios and advanced machine learning techniques.
Moving beyond traditional models
Conventional portfolio optimisation methods such as mean-variance analysis have long provided a framework for balancing risk and return. Yet, as Doshi pointed out, these methods struggle to cope with the realities of insurance investing.
While the methods have served a purpose, they are unrealistic for today's global investors, he said. "From an insurance perspective, they don't necessarily fully capture the risk return objectives that insurers really care about."
"Scenario-based machine learning (SBML) can be trained on thousands of scenarios... to learn the complex non-linear dynamics of how portfolios perform on investor-specific return objectives," Doshi explained.
This allows insurers to test strategies against forward-looking assumptions while capturing the complexity of regulatory capital and liability considerations, he said.
Phase one: Proof of concept
In the first case study with M&G Investments the aim was to optimise performance while holding the market risk solvency capital requirement (SCR) constant."The exercise focused on maximizing surplus mean and the 5% CVar of cumulative surplus, while maintaining the market risk SCR charge broadly the same level as the current portfolio," Doshi said.
The results were striking, he added, SBML produced an efficient frontier of viable portfolios, whereas traditional optimisation often yielded only a single usable solution.
For M&G Investments, the impact was immediate, Ritchie said.
"One of the main advantages of SBML modelling is the efficiency it brings to the optimisation process," he explained. "With the traditional process, you might only be left with one optimisation to consider. SBML allowed us to identify quite quickly these viable portfolios and focus much more on the portfolio construction and the strategic conversation with clients."
Phase two: Expanding the horizon
Phase two extended the approach by adding a third optimisation objective. Instead of a two-dimensional frontier, insurers could now visualise a three-dimensional plane of possibilities.
"In the second phase, we added a third objective: optimising the mean versus the 5% CVar and also versus the market risk SCR charge at the same time. Instead of an efficient frontier, it creates an efficient plane," Doshi explained.
For Ritchie, this development could not be timelier: "The insurance market is becoming increasingly competitive. The ability to optimise across these three objectives simultaneously - returns, risk, and capital - will be a significant enhancement. It should help us assess the trade-off more holistically and identify portfolios that strike the right balance."
A central theme of the discussion was the balance between technology and human expertise. While SBML offers powerful new ways to optimise portfolios, both speakers stressed that it is designed to support decision-making rather than replace it. Transparency and explainability are vital, with tools such as contribution analysis helping insurers understand what drives the results.
The audience Q&A highlighted practical issues such as the importance of robust assumptions, the integration of assets and liabilities, and the value of stress testing to build confidence in outcomes. Flexibility was shown to be key, with the ability to incorporate client-specific views alongside standard capital market assumptions.
Ultimately, the conversation returned to the need for a "human overlay". SBML can surface trade-offs and present richer sets of options, but insurers must interpret these results against their own strategy, regulatory context, and market realities.
The speakers also acknowledged challenges ahead, from embedding SBML into existing governance frameworks to ensuring that supporting tools and stress tests keep pace with its capabilities.
For a deeper dive into these issues, and to hear the full perspectives from Ortec's Doshi and M&G's Ritchie, the webinar is available on-demand here.