Home Insurance Affordability and Socioeconomic Equity in a Changing Climate

2 CPD Points

Deglobalization - Modelling the impacts of the insurance industry

The insurance industry has long excelled in quantifying and managing uncertainty. Today, this expertise faces a new test with the growing unpredictability in the structure and operation of the global economy. As we shift from an interconnected, diversified global system toward more fragmented and regionalised markets, insurers confront emerging risks and opportunities including impacts on financial performance, business interruption, and the structure of operations. In this presentation, we will: * Trace the trajectory of globalisation and deglobalisation, focusing on major economic pairings and their relevance to the insurance sector * Introduce a robust modelling framework designed to quantify the impacts of deglobalisation across key insurance dimensions. * Apply this framework to a recent major economic event, illustrating actionable insights and strategic considerations for industry leaders. Participants will gain a practical understanding of how deglobalisation reshapes risk landscapes, operational structures, and strategic decision-making within insurance.

1 CPD Point

Deglobalisation - Modelling the impacts of the insurance industry

The insurance industry has long excelled in quantifying and managing uncertainty. Today, this expertise faces a new test with the growing unpredictability in the structure and operation of the global economy. As we shift from an interconnected, diversified global system toward more fragmented and regionalised markets, insurers confront emerging risks and opportunities including impacts on financial performance, business interruption, and the structure of operations. In this presentation, we will: * Trace the trajectory of globalisation and deglobalisation, focusing on major economic pairings and their relevance to the insurance sector * Introduce a robust modelling framework designed to quantify the impacts of deglobalisation across key insurance dimensions. * Apply this framework to a recent major economic event, illustrating actionable insights and strategic considerations for industry leaders. Participants will gain a practical understanding of how deglobalisation reshapes risk landscapes, operational structures, and strategic decision-making within insurance.

3 CPD Points

Summit 2026: Reimagining Health Economics: Actuarial Pathways to Sustainable Care

This paper introduces the Actuarial Risk Architecture for Comprehensive Health (ARCH) as a comparative actuarial diagnostic that distinguishes value-based care maturity from longrun fiscal sustainability. ARCH assesses system sustainability across four dimensions — Adequacy, Resilience, Capacity, and Horizon — and is applied to Australia, the United States, and India alongside a companion Actuarial VBC Maturity Model that maps each system’s progression from fee-for-service to population-based payment. Existing healthsystem frameworks describe components; ARCH is designed to diagnose which financing constraint is binding for the next feasible reform step. The framework reframes healthsystem diagnostics in terms of capital adequacy, shock-absorption capacity, and riskadjustment integrity rather than clinical performance metrics—complementary to the WHO building-blocks and OECD resilience frameworks, but built for actuaries advising on fiscal sustainability and risk transfer rather than on clinical performance alone. The central finding is a paradox of payment sophistication. The United States has built one of the most extensive VBC architectures globally yet scores weakest on Horizon; Australia shows strong coverage foundations but a structural prevention gap and binding workforce constraints; India’s digital infrastructure creates long-run opportunity, but Resilience and riskadjustment capacity remain binding. Three applications illustrate ARCH in operation: a Behavioural Adjustment Factor (BAF) framework for prevention economics, a GLP-1 stress test of cross-system resilience, and a concrete Australian policy implication — a PHI-routed prevention fund designed as the first stage of a longer-horizon financing reform. ARCH is offered as a tool to help the actuarial profession move from measuring health system performance to designing it.

0 CPD Points

Summit 2026: Beyond our means - Appendix A - Worked Example

4 CPD Points

Summit 2026: Beyond our means? A distributional approach to assessing actuarial reserving in general insurance

Fundamentally, actuaries are asked to put a number to uncertain future events. We spend a lot of effort quantifying this uncertainty to provide a range of reasonable outcomes. As part of standard valuation work, actuaries will present a central estimate alongside a metric for uncertainty, for example, a quantile of the loss distribution in the form of a risk margin or a risk adjustment. A core part of the actuarial control cycle is the monitoring and assessment of the models against experience. In practice, this is focussed on the central estimate, with less emphasis placed on the distribution. “Actual vs Expected” analysis assesses the percentage variation against the central estimate, with a layer of actuarial judgement applied to justify the materiality of any differences (e.g. based on portfolio maturity, size of book, known events). This is likely sufficient in many cases but in situations where the modelling decisions/results are less clear cut, we wanted to explore the ability to leverage the distributional thinking already performed at past valuations to provide a better data-driven assessment to support judgement. It turns out that this isn't as simple as just applying the overall risk margin to the central estimate and backing out some quantiles. This is because: • The central estimate models are almost always different from the risk margin models • Risk margins include consideration of systemic risk which may not be in the historical data • Adjustments in central estimate models (e.g. blending or accounting for data quality/scarcity) means analytical expressions can be hard to derive explicitly • Allocation - how do you allocate a risk margin calculated for the full projection period, to just the next reporting period? In this paper, we describe an approach to calculate a distribution for each cell in the projection (or the combination of cells you are interested in). Most importantly, the additional time and effort required to do this analysis is minimal; much of it would already be part of your standard valuation pipeline. Having access to these distributions opens up a number of areas where "distributional hindsight analysis" can be useful to help support actuarial decision making by adding statistical rigour (because you can't do statistical tests on X% above the mean). Examples of such decisions include: 1. When should we change our roll-forward modelling assumptions? 2. When should we revise our risk margins? 3. How outlying was the new event and does it require a post-balance date adjustment? 4. Should I analyse my new portfolio separately or should I just combine it with my current book?

8 CPD Points

Summit 2026: Navigating the Electric Vehicle Environment: E Bikes and E-Scooters

The increasing adoption of Personal Mobility Devices (PMDs) across Australia is leading to a diverse and everchanging landscape. PMDs are small, electrically powered devices, designed for a singular user to travel short distances. They typically include devices such as electric scooters (e-scooters), electric bikes (e-bikes) and electric skateboards (e skateboards). More broadly, their increased adoption represents a sustainable, affordable and efficient alternative to traditional transportation methods; promoting healthier lifestyles whilst creating efficient solutions to common infrastructure issues including traffic congestion and the ‘last mile’ problem. However, the rapid adoption of PDMs has also introduced various new and distinct challenges for insurers, regulators, policymakers and urban planners. Different licensing, road and vehicle design laws have created significant different experiences between Australian states and territories. PMDs can exhibit vastly different risk characteristics to other forms of conventional transport due to their unconventional design, rider behaviour and vehicle usage. Road infrastructure and urban environments, state-based legislation, interactions with pedestrians and other road users and the implementation of telematics all have the propensity to further influence risk profiles, thereby creating further complexities for insurers. With a focus on e-bikes and e-scooters; this paper considers the academic literature to look at their different risks characteristics and contributing factors - which includes the regulation on licensing, PMD designs, road use (helmets and speed limits) and road infrastructure. A review of Australian injury data provides a better understanding of their risks for physical injury and fire damage. Lastly, the paper illustrates the insurance gap that presently exists for Australian PMD users given the current insurance regulations and design of the Compulsory Third Party and Workers Compensation schemes.

0 CPD Points

Treasury Consultation: Education Reform for Financial Advisers

We strongly support the initiative to make it easier for financial services professionals to transition into financial planning through a more flexible and streamlined approach to recognising existing educational attainment while maintaining high standards.

7 CPD Points

Summit 2026: Beyond pairwise correlation: capturing nonlinear and higher-order dependence with distance statistics

Measuring and modelling dependence between risks is crucial in many actuarial applica tions, such as when assessing diversification benefits or setting capital requirements. Current industry practice relies heavily on Pearson’s correlation coefficient, despite well-known limita tions. In particular, it captures only linear association, is restricted to pairwise relationships, and does not naturally extend to multivariate settings involving multiple random variables or random vectors (such as when one-hot encoding categorical variables). Sole reliance on Pearson’s lin ear correlation may fail to detect important nonlinear, higher-order, and mutual dependence structures. In this paper, we discuss and illustrate several distance-based dependence statistics which do not suffer from the same limitations as correlation, and discuss how they can be used in actuarial applications. In the bivariate setting, we consider the Hellinger correlation (Geenens &LafayedeMicheaux, 2022)asatoolfor measuring dependencebetweentwocontinuousuni variate random variables, and distance covariance (Székely et al., 2007) as a tool for detecting andtesting dependence, especially when randomvectorsareinvolved. Wethendiscussexten sions to higher-dimensional settings, including joint distance covariance for assessing mutual dependence across multiple random variables (or vectors), and the auto-distance correlation function for time series applications such as forecasting mortality rates over time. Throughout the paper, we illustrate the use of these tools in actuarial contexts and we direct the reader to available software implementations. Overall, the paper aims to provide actuaries with a practical introduction to distance-based dependence statistics and to show how they can complement classical correlation-based tools in actuarial workflows.

2 CPD Points

Summit 2026: From Data Taker to Data Generator: Introduction to Causal Inference and Randomised Control Trial Playbook

As actuaries, we rely on observational data to build models, understand historical patterns and tell stories. Those datasets show us what has happened, but rarely tell us why things happen. The why matters because it pinpoints to action and consequence, it enables decision makers to understand causal effects of changes rather than being blinded by correlations, and it provides robust policy evaluations. Understanding the why, or the cause and effect is challenging, there are techniques and methods to illuminate the hidden causal relationship, randomised controlled trial (RCT) is widely considered as the gold standard for uncovering causality, as it directly addresses the problem of endogeneity by ensuring treatment assignment is exogenous, while other causal methods such as differences-in-differences, regression discontinuity, or instrumental variables arguably suffer from the endogeneity problem to various degrees. RCT, as a method of field experiment, generates meaningful data, sharper insights and business guidance beyond what observational data can provide. It adds to the actuary's toolkit and redefines our understanding of why things happen. This paper will introduce causal inference, the endogeneity problem, an RCT playbook with practical reflections, and a case study, which hopefully will inspire more actuaries to explore and embrace the spirit of experimentation.

4 CPD Points

Summit 2026: An Interpretable Deep Learning Model for General Insurance Pricing

The rapid advancement of machine learning has provided an opportunity to transform the modeling techniques in actuarial analytics. Novel machine learning methods, especially deep learning, have demonstrated versatile modeling capability and superiorpredictive performancecomparedtotraditionalactuarialapproachessuchasGeneralizedLinear Models. However, the widespread adoption of deep learning techniques in the insurance industry is often hindered by the lack of model interpretability, as the intricacies of their inner workings remain obscured behind the complex model architecture. This lack of interpretability is further complicated by the absence of a generally accepted definition of what an interpretable model is. There are also various practical requirements, such as smoothness and monotonicity, that a pricing model in general insurance should possess in addition to being interpretable. This paper introduces the Actuarial Neural Additive Model, an inherently interpretable deep learning model for general insurance pricing that offers fully transparent and interpretable results while retaining the strong predictive power of neural networks. This model assigns a dedicated neural network (or subnetwork) to each individual covariate and pairwise interaction term to independently learn its impact on the modeled output while implementing various architectural constraints to allow for essential interpretability (e.g. sparsity) and practical requirements (e.g. smoothness, monotonicity) in insurance applications. The development of our model is grounded in a solid foundation, where we establish a concrete definition of interpretability within the insurance context, complemented by a rigorous mathematical framework. Comparisons in terms of prediction accuracy are made with traditional actuarial and state of-the-art machine learning methods using both synthetic and real insurance datasets. The results show that the proposed model outperforms other methods in most cases while offering complete transparency in its internal logic, underscoring the strong interpretability and predictive capability.

3 CPD Points

Summit 2026: All is Fair in Love and War….but not in general insurance

There is a well-known quote suggesting that love and war are so important and overwhelming that people should be excused for acting in their own best interest. But general insurance is neither a battlefield nor a romance - it's a relationship built on trust, where fairness must prevail over self interest. In his October 2024 dialogue paper, “Fairness in Insurance”, Ian Laughlin challenged insurance Boards to elevate customer fairness as a strategic priority. This paper shifts the lens to explore the critical role general insurance actuaries can—and should—play in delivering fairer outcomes. Laughlin recognised that Boards occupy a unique vantage point for ensuring fair customer treatment, and identified fairness as a "natural area of interest" for actuaries. But what are the obligations and rules of fairness which apply to general insurance actuaries? This paper examines what practitioners need to know about fairness and what they could, or should, be doing to achieve this objective in practice. While Laughlin noted that Boards benefit from sitting "above the fray" with greater objectivity than management, this paper argues that general insurance actuaries are equally well positioned—if not better placed—to balance competing interests. Working in the front lines where pricing, underwriting, and claims decisions are made, general insurance actuaries can serve as vital arbiters of fairness, translating principle into practice where it matters most.

4 CPD Points

Summit 2026: Fairness Testing for Insurance Pricing: A Statistical Inference Framework

Ensuring fairness in insurance pricing has become a central concern for regulators, insurers, and the public. Existing approaches often rely on descriptive disparity measures or significance testing without explicit reference to fairness concepts or acceptable tolerances, leading to fragmented practices and ambiguous conclusions. We develop a unified framework that grounds fairness testing in classical statistical inference, providing clear definitions, reproducible methods, and auditable protocols. Our framework advances fairness testing along four key dimensions. First, we formalise widely discussed fairness criteria as hypotheses on identifiable estimands, making fairness notions statistically precise. Second, we introduce decision rules that incorporate explicit tolerance thresholds, reflecting regulatory standards in practice. Third, we propose inference procedures that bridge actuarial practice with fairness guarantees. Fourth, we design a quote-audit protocol that specifies how to collect, test, and validate fairness claims in a manner that is transparent and replicable. By combining inferential rigour with regulatory practicality, our approach delivers a coherent methodology that both regulators and insurers can implement. The framework integrates uncertainty quantification and audit design into a single pipeline, ensuring that fairness tests are interpretable, reproducible, and statistically robust. Beyond pricing, the principles extend to approval outcomes and process measures, providing a flexible foundation for responsible insurance analytics. The result is a regulator- and insurer-ready framework that aligns fairness testing with statistical guarantees and supports the broader goal of accountable and transparent insurance practices.

4 CPD Points

Summit 2026: Making all the right noise - Using diffusion deep learning models to perform multivariate prediction for simulation and reserving

As AI adoption accelerates, there is continued interest in adapting ideas to new contexts to see how deep learning models can be applied. For actuaries, there is an ongoing opportunity for new models to handle multivariate prediction problems where there are important dependencies between outputs. Examples include: claims reserving (estimating a suite of accident and development period combinations), pricing (where different risks are estimated and combined into a risk price), microsimulation (where many different characteristics are simulated over time in a consistent way) and DFA models (where different parameters will evolve over time to determine capital positions, with strong correlations across measures). Diffusion models are a common deep learning approach, particularly in image generation. Starting with random variation, a trained model will successively ‘de-noise’ a dataset until it resembles something coherent. In many cases they outperform other model structures, including in a small number of research papers testing applicability to tabular data. Our paper applies diffusion models to two key actuarial problems. First, we apply diffusion to microsimulation, where interrelated outcomes must be jointly predicted. The diffusion setup significantly outperforms the GAN and VAE model approaches explored in our 2025 summit paper (Gains ratio on a discriminator model of 28%, compared to about 70% for prior models), at the expense of speed. Second, we reformulate reserving using actuarial triangles as a deep learning image completion problem and obtain promising results on a database of historical data. Results are reasonable, although some tail misfit is evident – comparable with results seen elsewhere in the literature. The models represent an important contribution, with the potential to substantially simplify the training of complex systems, as well as augmenting standard actuarial practice. To the authors’ knowledge, this is the first time such diffusion models have been applied in an actuarial context.

1 CPD Point

NSW Planning: Consultation on NSW’s Climate Change and Natural Hazards SEPP

The Institute supports the SEPP’s overarching objective to “future‑proof” communities against climate‑amplified natural hazard risks (including coastal hazards, flooding, bushfires and urban heat) and sees this reform as a critical complement to strengthening the National Construction Code. We strongly recommend the SEPP adopt an explicit principle of “no new residential or other highly vulnerable development in high‑risk hazard zones”. This is not intended to preclude lower‑consequence uses in high‑risk hazard zones, where events do not result in major disruption or loss.