Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

CATNet: A geometric deep learning approach for CAT bond spread prediction in the primary market

Dixon Domfeh, Saeid Safarveisi

Abstract

Traditional models for pricing catastrophe (CAT) bonds struggle to capture the complex, relational data inherent in these instruments. This paper introduces CATNet, a novel framework that applies a geometric deep learning architecture, the Relational Graph Convolutional Network (R-GCN), to model the CAT bond primary market as a graph, leveraging its underlying network structure for spread prediction. Our analysis reveals that the CAT bond market exhibits the characteristics of a scale-free network, a structure dominated by a few highly connected and influential hubs. CATNet demonstrates higher predictive performance, significantly outperforming strong Random Forest and XGBoost benchmarks. Interpretability analysis confirms that the network's topological properties are not mere statistical artifacts; they are quantitative proxies for long-held industry intuition regarding issuer reputation, underwriter influence, and peril concentration. This research provides evidence that network connectivity is a key determinant of price, offering a new paradigm for risk assessment and proving that graph-based models can deliver both state-of-the-art accuracy and deeper, quantifiable market insights.

CATNet: A geometric deep learning approach for CAT bond spread prediction in the primary market

Abstract

Traditional models for pricing catastrophe (CAT) bonds struggle to capture the complex, relational data inherent in these instruments. This paper introduces CATNet, a novel framework that applies a geometric deep learning architecture, the Relational Graph Convolutional Network (R-GCN), to model the CAT bond primary market as a graph, leveraging its underlying network structure for spread prediction. Our analysis reveals that the CAT bond market exhibits the characteristics of a scale-free network, a structure dominated by a few highly connected and influential hubs. CATNet demonstrates higher predictive performance, significantly outperforming strong Random Forest and XGBoost benchmarks. Interpretability analysis confirms that the network's topological properties are not mere statistical artifacts; they are quantitative proxies for long-held industry intuition regarding issuer reputation, underwriter influence, and peril concentration. This research provides evidence that network connectivity is a key determinant of price, offering a new paradigm for risk assessment and proving that graph-based models can deliver both state-of-the-art accuracy and deeper, quantifiable market insights.

Paper Structure

This paper contains 30 sections, 22 equations, 13 figures, 12 tables.

Table of Contents

  1. Introduction
  2. A brief literature review
  3. Data
  4. Graph representation of CAT data
  5. Multi-relational graph
  6. Graph construction
  7. CAT bond network properties
  8. Geometric Deep Learning for node prediction task
  9. Graph neural networks
  10. Relational graph convolutional networks (R-GCNs)
  11. R-GCN application to CAT bond risk premium prediction
  12. R-GCN Implementation
  13. Data transformation
  14. Quantifying structural influence via topological diagnostics
  15. Training: inductive versus inductive learning paradigms on graphs
  16. ...and 15 more sections

Figures (13)

  • Figure 1: Relational graph representation of a typical CAT bond contract.
  • Figure 2: CAT bond network visualization for bonds issued in 2021. Larger nodes indicate higher degrees. Different edge colors signify unique relationships between the entities.
  • Figure 3: Histogram plot of degree frequency
  • Figure 4: Centrality measures by node (top 5 per measure)
  • Figure 5: Fitness of different node types.
  • ...and 8 more figures