actuarial revenue modeling

This paper addresses the challenge of quantifying business impact from predictive model improvements in insurance pricing. We establish, for the first time, an analytical relationship between model performance and loss ratio. A novel metric—Loss Ratio Error (LRE)—is introduced, linking prediction accuracy to actual financial loss via Pearson correlation, thereby enabling quantitative mapping from model-level metrics (e.g., RMSE) to business-level KPIs. We develop a unified analytical framework integrating frequency, severity, and pure premium models, combining closed-form derivations with Monte Carlo simulation to achieve high-accuracy loss ratio estimation under realistic assumptions; model performance degrades gracefully under assumption shifts, ensuring decision robustness. Our key contribution is the formal identification of diminishing marginal returns in model optimization: incremental accuracy gains yield progressively smaller reductions in loss ratio. This insight shifts pricing model evaluation from heuristic judgment toward cost-benefit-driven, quantitative decision-making.

This study addresses the dual challenge of actuarial fairness and group equity in non-life insurance pricing, which is essential to mitigate legal and reputational risks. It introduces the multi-calibration framework—originally developed in algorithmic fairness—to the insurance domain, unifying self-calibration and conditional mean independence as coherent fairness criteria. The proposed approach ensures that, within each premium group, average claims match expected revenues while simultaneously satisfying formal fairness guarantees. To operationalize this framework, the authors develop a practical pricing model incorporating local regression, within-group bias correction, and credibility adjustments to enforce multi-calibration constraints. Empirical analysis using real-world auto insurance data demonstrates that the method effectively balances financial soundness with equitable treatment across demographic groups, offering a viable pathway toward fair and sustainable insurance pricing.

This study addresses the challenge that existing fair pricing methods, primarily designed for short-term insurance, are ill-suited for long-term insurance products governed by multi-state transition models. To bridge this gap, the authors reformulate the estimation of multi-state transition rates as a series of Poisson regression problems, thereby establishing the first unified framework for fair pricing in long-term insurance. This framework seamlessly integrates with mainstream fair machine learning techniques and accommodates preprocessing, in-processing, and post-processing fairness strategies. Using data from the University of Michigan’s Health and Retirement Study, the approach is validated in the context of long-term care insurance pricing, demonstrating both effectiveness and flexibility while filling a critical methodological void between multi-state modeling and fair machine learning.

This study addresses the limitations of traditional actuarial pricing, which relies heavily on manual feature engineering and struggles to leverage unstructured textual data. It proposes a novel approach that integrates pretrained large language models (LLMs) into the insurance pricing pipeline by extracting deterministic embeddings from policyholders’ natural language descriptions via controlled prompting. These embeddings are then incorporated as input features in a generalized linear model for Poisson regression of claim frequency. The method significantly reduces dependence on handcrafted features while preserving model governability. Empirical evaluation on French motor third-party liability insurance data demonstrates that the embedding-driven model substantially outperforms conventional approaches in low-data regimes. In high-data settings, performance is influenced by model architecture and embedding dimensionality, with further gains achieved through domain-specific fine-tuning.

Traditional revenue forecasting approaches struggle to uncover the underlying customer behavioral drivers—such as customer acquisition, repeat purchase rates, and average transaction value—that influence revenue dynamics. To address this limitation, this work proposes the Customer-Based Multi-Task Transformer (CBMT), which uniquely integrates multi-task learning with a Transformer architecture to jointly model customer behavioral metrics and total revenue through shared representations. Furthermore, CBMT incorporates a downstream alignment mechanism to enhance both interpretability and predictive accuracy. Empirical evaluation on real-world customer transaction panel data demonstrates that CBMT outperforms existing methods across 23 out of 24 evaluation metrics, achieving a 30% reduction in total sales prediction error compared to the strongest baseline and significantly surpassing single-task models employed by 74.3% of firms.

This work addresses the challenge that existing calibration tests for conditional quantile predictors struggle to handle distributional shifts and discrepancies in information sets, lacking feature-aware, continuous monitoring capabilities. The authors propose a distribution-free, game-theoretic sequential auditing framework that formally defines conditional quantile calibration under varying feature information sets—a notion not previously established—and provides finite-time detection guarantees without requiring independent and identically distributed data. By integrating contextual linear betting strategies with nonparametric e-processes, the method enables interpretable, feature-level calibration audits. Empirical evaluations demonstrate that the framework effectively detects significant miscalibration in state-of-the-art time series models, such as Chronos-2, across critical features.

This study addresses the high computational cost of asset-liability management (ALM) models in the insurance industry, particularly in large-scale sensitivity analyses and stress testing for solvency capital assessment and asset allocation optimization. It introduces path signature theory into ALM modeling for the first time, approximating key outputs—such as embedded value and best estimate—as linear combinations of path signatures derived from economic scenario trajectories. A regularized linear regression framework is then employed to construct a surrogate model. This approach substantially reduces computational overhead while maintaining high predictive accuracy and demonstrating robustness to shifts in the underlying economic scenario distribution, thereby enabling efficient large-scale balance sheet evaluation and rapid decision support.

This work addresses a critical limitation in existing algorithmic attribution methods, which focus solely on flipping model predictions while neglecting how recommendations can genuinely enhance individuals’ underlying qualifications. Such oversight often triggers strategic gaming and necessitates frequent model retraining. To overcome this, the study introduces structural causal models into algorithmic attribution for the first time, employing a causal intervention framework to characterize feature interactions and their effects on true outcomes. By integrating iterative dynamics to solve the resulting non-convex optimization problem, the authors propose a causal attribution strategy that achieves stable equilibrium. Experiments on both semi-synthetic and real-world credit datasets demonstrate that the method substantially outperforms empirical risk minimization, effectively mitigating distributional shifts caused by strategic behavior and significantly reducing the need for model retraining.

This study addresses the challenges of unstructured documents, heterogeneous data, and automated compliance-aware decision-making in insurance underwriting by proposing an “Agentic RAG” multi-agent framework for straight-through underwriting of small commercial policies. The approach integrates multi-agent planning, reflection mechanisms, and retrieval-augmented generation (RAG) to enable transparent, auditable, and human-in-the-loop decision-making through structured retrieval, third-party data validation, and explicit multi-step rule evaluation. Experimental results demonstrate that, compared to single large language models and naive RAG baselines, the proposed framework significantly enhances decision reliability and regulatory compliance—particularly in scenarios involving missing information or complex reasoning—while effectively preventing unjustified straight-through underwriting approvals.