Thesis title: Modellizzazione del rischio terremoto in Italia
This study aims to provide a comprehensive and transparent framework for catastrophe modeling, with a focus on deconstructing and analyzing the individual components that contribute to seismic risk assessments. Catastrophe models are often presented as black boxes, where users cannot fully understand the internal mechanisms or the interaction between key variables. In this work, the goal is to deeply investigate these components, offering a detailed exploration of how they contribute to the overall model, while proposing specific
improvements in terms of hazard, vulnerability, and exposure assessment. The first key component examined is seismic hazard. The physical characteristics of earthquakes are reviewed, alongside a detailed analysis of probabilistic models used in the literature. To estimate the expected number of earthquakes in Italy, a two-step approach is proposed: first, a Geographically Weighted Regression (GWR) model is developed, followed by a Random Forest model to improve the prediction of earthquake occurrences based on macroseismic intensities (Modified Mercalli Scale, MCS). The use of MCS makes the results more interpretable for stakeholders by linking the hazard directly to observed earthquake effects rather than more abstract physical measures like ground acceleration. The second major component, vulnerability, is addressed through multiple approaches. Building on the ShakeMap data, fragility curves are constructed, representing the probability of different damage levels given an earthquake’s intensity. A Beta distribution is introduced to directly simulate damage levels across different building categories. Furthermore, an XGBoost model is utilized to predict damage outcomes, leveraging both structural data and seismic intensity measures. This hybrid approach provides a detailed estimation of damage
for each building type and allows for greater flexibility in the vulnerability modeling process. In terms of exposure, the study reassesses the distribution of wealth and building types across Italian municipalities, offering insights into how regional variations in building stock influence potential losses. Historical earthquake damage data is analyzed to calibrate the exposure model, ensuring that it reflects the reality of past events and current vulnerability levels. This granular approach to exposure allows for more precise loss estimations at the local level, which is critical in a region with complex socioeconomic and structural variability like Italy. The final phase of the research focuses on calculating key cat metrics, namely the Average Annual Loss (AAL) and Annual Exceedance Probability (AEP) curve. These metrics are essential for understanding the expected financial impact of seismic events and are fundamental for the pricing and underwriting of earthquake insurance. To offer an insurance-oriented perspective, the study also calculates risk-adjusted premiums and the Solvency Capital Requirement (SCR) for a typical insurance policy. These calculations are designed to align with regulatory frameworks such as Solvency II, providing insights into how catastrophe models can be directly applied within the insurance sector. By exploring the fundamental components of a catastrophe model in depth and proposing new methodologies for hazard, vulnerability, and exposure estimation, this study contributes to the ongoing efforts to make seismic risk models more transparent, interpretable, and applicable to both scientific and insurance industry audiences.