Impact of Climate Risk on Parameter Estimation and Pricing for Property Insurance Portfolios Using Catastrophe Modeling and Machine Learning

Project Abstract

This study investigates how climate risk influences parameter estimation and pricing in property insurance portfolios by integrating catastrophe modeling with advanced machine learning techniques. We address the growing exposure of property insurance to extreme weather events—such as tropical cyclones, heatwaves, floods, and wildfires—driven by climate change, and the resultant challenges in accurate risk quantification, capital adequacy, and premium determination. The research develops a hybrid framework that combines stochastic catastrophe models with data-driven machine learning algorithms to produce robust parameter estimates for loss distributions, tail risk measures, and dependencies across portfolio lines. We first synthesize climate risk drivers, historical loss data, and exposure information into a cohesive dataset, ensuring rigorous data cleaning, outlier treatment, and feature engineering that captures spatial-temporal heterogeneity and climate-driven volatility. The methodology integrates physical catastrophe modeling for event generation with statistical and probabilistic approaches to calibrate severity and frequency parameters, augmented by machine learning models that learn complex non-linear relationships and parameter uncertainty from high-dimensional data. Key contributions include (1) a calibrated multi-hazard catastrophe framework that links climate scenarios to portfolio-level loss distributions, (2) adaptive parameter estimation procedures that jointly fit severity, frequency, and correlation structures under climate stress, and (3) pricing innovations that incorporate climate-adjusted risk measures, such as conditional tail expectations and spectral risk metrics, into actuarial pricing kernels and reinsurance placement decisions. The study explores model risk and parameter uncertainty through Bayesian inference, ensemble learning, and scenario analysis across representative climate pathways. We implement cross-validation and backtesting protocols to evaluate predictive performance, focusing on out-of-sample accuracy, calibration, and stability of parameter estimates under regime shifts. The research further investigates the impact of granularity in exposure data—property-level versus portfolio-level—on pricing efficiency and capital requirements, identifying thresholds where machine learning augmentation yields meaningful gains. Empirical results are expected to show that integrating catastrophe modeling with machine learning improves estimation of tail risk and the accuracy of premium loadings under climate stress, reduces mispricing during extreme events, and enhances risk-based capital allocation. Comparative analyses across geographic regions, peril types, and policy features illuminate where climate-informed models outperform traditional methods and where data limitations constrain improvements. The study also discusses operational considerations, including data governance, computational requirements, model interpretability, and regulatory implications for model governance and disclosure. By demonstrating a rigorous, reproducible framework that blends physical risk assessment with data-driven inference, this work provides actionable insights for insurers seeking resilient pricing strategies, improved capital management, and enhanced resilience to evolving climate-driven insured losses.

Project Overview

What This Project Is About

---

A plain-language overview of how climate factors affect insurance decisions, focusing on how we estimate risk and set prices for property insurance using computer models and simple data analysis. The project looks at how weather and climate-related events influence damage costs and the accuracy of price estimates.

---

---

The Problem It Addresses

---

The current pricing methods often assume stable climate conditions, which is not true in many places. This gap can lead to underpricing or overpricing risk, expensive losses for insurers, and higher premiums for customers. The project explores how climate risk changes model estimates and pricing decisions.

---

---

Objectives of the Project

---

1. Explain how climate factors affect property risk and pricing in simple terms.
2. Introduce basic methods to estimate risk more accurately in changing climate conditions.
3. Show how catastrophe thinking helps explain large losses from extreme events.
4. Demonstrate how machine learning can improve parameter estimates and prices.
5. Provide a clear framework for evaluating model performance and fairness.

---

---

What You Will Do Step by Step

---

1. Gather publicly available data on property claims and weather/climate indicators. 2. Clean and prepare data for analysis. 3. Build simple baseline pricing models. 4. Introduce catastrophe modeling concepts to account for extreme events. 5. Apply basic machine learning to refine risk parameters. 6. Compare model outputs to real losses and check pricing accuracy. 7. Assess limitations and robustness of results. 8. Prepare a short report with findings and practical recommendations.

---

---

Expected Outcome

Clear, easy-to-understand insights on how climate risk should be reflected in pricing, with simple methods that can be used by students and practitioners. The project should show when more complex models help and how to communicate pricing changes to stakeholders.