Ortec Finance: The evolution of insurance asset portfolio optimization
By Ashish Doshi, Head of UK Insurance at Ortec Finance
How machine learning is transforming investment strategies for insurance asset managers
The investment landscape is becoming ever more unpredictable driven by economic uncertainty, geopolitical risks and evolving regulations. This is putting a strain on traditional asset portfolio optimization techniques.
Traditional methods are becoming less effective in addressing the rapidly evolving financial environment and insurers are facing the challenge of struggling to balance complex regulatory and financial objectives using tools that were designed for a simpler, more stable era.
Shortcomings of traditional portfolio optimization
For decades, investors have relied on techniques rooted in linear relationships such as mean-variance optimization, which seeks to balance expected return against risk. These closed-form approaches offer clear frameworks for decision-making but require simplified approximations of insurer-specific objectives.
Insurance companies face objectives far more complex than simply maximizing return for a given level of risk. They must also account for objectives such as solvency capital requirements, regulatory compliance and liquidity management. Traditional optimization approaches struggle to accommodate these objectives, particularly when constraints are non-linear and when conflicting goals must be considered simultaneously.
To overcome this challenge in managing insurance portfolios, asset managers have resorted to trial-and-error or brute-force methods, manually generating portfolios until one fits the desired criteria. While this approach can work, it is inefficient and offers no assurance of optimality. The time and resources expended in this process can be considerable and the resulting portfolios may still fall short of meeting the required objectives.
Scenario-based machine learning - a new approach
Scenario-Based Machine Learning (SBML) represents a paradigm shift in portfolio optimization, enabling users to evaluate any combination of objectives within a stochastic scenario framework. Unlike traditional methods, SBML embraces the full complexity of the real world, allowing for non-linear objectives and the simultaneous optimization of multiple competing goals.
The key to SBML is its ability to learn from vast data sets of generated balance sheet projections driven by a stochastic real-world scenario generator. Machine learning algorithms train on these projections, identifying patterns and relationships between the complex objectives and constraints. This learning process identifies asset portfolios that best meet the objectives and constraints defined in the optimization exercise creating an efficient frontier of suitable portfolios.
Targeting balance sheet metrics
One of the defining features of using SBML tools for strategic asset allocation (SAA) optimization is the capacity to target the complex balance sheet metrics that matter most to insurers, leading to a targeted SAA approach.
Let’s take solvency capital as an example. By and large, for all insurance regulatory frameworks globally, the amount of capital held is directly influenced by the risk profile of the investments held. Regulatory frameworks, such as Solvency II in Europe, impose strict standards on insurers, requiring them to maintain sufficient capital to cover the risks of running asset portfolios. SBML enables insurance asset managers to directly incorporate these considerations into the optimization process maximizing returns or surplus while minimizing solvency capital and imposing a constraint on the amount of capital required.
Another example is where insurers want to consider the impact or the level of dividends payable from their surplus assets. A reliable dividend stream is important in the context of shareholders but there are several complexities that means implementing a suitable investment strategy backing this dividend stream is crucial. This is where SBML can really help, as it allows investors to specifically target the dividend as an objective metric of interest. The flexibility of the tool allows investors to either maximize or target the level of dividends whilst measuring the performance and risk profile of candidate portfolios through a stochastic lens. Adding a complex constraint such as solvency capital into the optimization can provide further insight into how to structure a suitable asset allocation.
Asset managers that embrace tools that use machine learning for portfolio optimization will be best positioned to achieve their goals, adapt to new challenges, and secure their place in the evolving landscape of global finance.
