Catastrophe risk has long been recognized to pose a serious threat to the insurance sector. Catastrophe risk pooling offers an effective way to diversify losses arising from catastrophic events. In this paper, we investigate a structure of catastrophe risk pool and optimize it so that participants can attain the maximum diversification benefit from joining the pool. Determining the practical optimal pool entails solving a high-dimensional optimization problem, for which analytical solutions are typically unavailable and numerical methods can be computationally intensive and potentially unreliable. To address this challenge, we evaluate the diversification benefit in the limit and use it to derive an asymptotically optimal pool which approximates the practical optimal pool. Through simulation studies, we show that the asymptotically optimal pool provides an accurate and reliable approximation to the practical optimal pool. We also conduct an empirical analysis using data from the U.S. National Flood Insurance Program to illustrate how the framework can be applied in practice.

This paper proposes PyFi, a novel framework for pyramid-like financial image understanding that enables vision language models (VLMs) to reason through question chains in a progressive, simple-to-complex manner. At the core of PyFi is PyFi-600K, a dataset comprising 600K financial question-answer pairs organized into a reasoning pyramid: questions at the base require only basic perception, while those toward the apex demand increasing levels of capability in financial visual understanding and expertise. This data is scalable because it is synthesized without human annotations, using PyFi-adv, a multi-agent adversarial mechanism under the Monte Carlo Tree Search (MCTS) paradigm, in which, for each image, a challenger agent competes with a solver agent by generating question chains that progressively probe deeper capability levels in financial visual reasoning. Leveraging this dataset, we present fine-grained, hierarchical, and comprehensive evaluations of advanced VLMs in the financial domain. Moreover, fine-tuning Qwen2.5-VL-3B and Qwen2.5-VL-7B on the pyramid-structured question chains enables these models to answer complex financial questions by decomposing them into sub-questions with gradually increasing reasoning demands, yielding average accuracy improvements of 19.52% and 8.06%, respectively, on the dataset. All resources of code, dataset and models are available at: https://github.com/AgenticFinLab/PyFi .

Despite the empirical success of the rough Bergomi (rBergomi) model in modeling volatility dynamics, its practical use remains challenging due to high computational complexity in both pricing and calibration arising from its non-Markovian structure. To address these difficulties, we develop an efficient computational framework. First, we propose a modified-sum-of-exponentials (mSOE) Monte Carlo scheme within the class of hybrid multifactor approximations. The method combines an exact treatment of the singular kernel over the first time step with a sum-of-exponentials approximation over the remaining time interval, and exact Gaussian simulation of the resulting multifactor components. For a fixed number of exponential terms, the method maintains linear online complexity with respect to the number of time steps. It achieves high pricing accuracy in numerical experiments, particularly for out-of-the-money options. Second, building on this pricing engine, we formulate a calibration approach based on distributional matching of the terminal underlying asset via the Wasserstein-1 distance. Instead of fitting option prices only at selected strikes, this method compares model-generated and market-implied terminal distributions through the Kantorovich-Rubinstein dual representation. Numerical experiments indicate that the mSOE scheme exhibits stable convergence, and the Wasserstein-based calibration scheme improves parameter recovery, optimization stability, and out-of-sample performance relative to conventional MSE-based fitting in the rBergomi setting considered in this paper.

Extracting specific items from 10-K reports is challenging due to variations in document formats and item presentation. To improve over traditional rule-based approaches, this study introduces and compares two advanced item segmentation methods: (1) GPT4ItemSeg, using a novel line-ID-based prompting mechanism to utilize a large language model, ChatGPT-4o, for item segmentation, and (2) BERT4ItemSeg, combining a pre-trained language model, BERT, with a Bi-LSTM model in a hierarchical structure to overcome context window constraints. Trained and evaluated on 3,737 annotated 10-K reports, BERT4ItemSeg achieves a macro-F1 of 0.9825, surpassing GPT4ItemSeg (0.9567), conditional random field (0.9818), and rule-based methods (0.9048) for core items (1, 1A, 3, and 7). These approaches enhance item segmentation performance, improving text analytics in accounting and finance. BERT4ItemSeg offers satisfactory item segmentation performance, while GPT4ItemSeg can easily adapt to regulatory changes. Together, they provide an extensible framework for 10-K item segmentation that supports reliable and reproducible results.

Collective decision-making requires aggregating multiple noisy information channels about an unknown state of the world. Classical epistemic justifications of majority rule rely on homogeneity assumptions often violated when individual competences are heterogeneous. This paper studies endogenous epistemic weighting in binary collective decisions. It introduces the Epistemic Shared-Choice Mechanism (ESCM), a lightweight and auditable procedure that generates bounded, issue-specific voting weights from short informational assessments. Unlike likelihood-optimal rules, ESCM does not require ex ante knowledge of individual competences, but infers them endogenously while bounding individual influence. Using a central limit approximation under general regularity conditions, the paper establishes analytically that bounded competence-sensitive monotone weighting strictly increases the mean quality of the aggregate signal whenever competence is heterogeneous. Numerical comparisons under Beta-distributed and segmented mixture competence environments show that these mean gains are associated with higher signal-to-noise ratios and large-sample accuracy relative to unweighted majority rule.