When Reputation Meets Disaster: Quantifying Idiosyncratic Risk in Primary Cat Bond Spreads by Florian Horky, Brian M. Lucey, Sovan Mitra, Boru Ren :: SSRN

Climate-finance debates increasingly point to catastrophe (CAT) bonds as a way to shift the costs of extreme weather from public budgets to capital markets. That promise rests on two linked perspectives: a macro perspective, where CAT bonds can broaden society’s ability to absorb disaster losses, and a micro perspective, where their primary-market spreads must be efficiently predictable to investors, supervisors and sponsors. We address both aspects by analysing how spreads are formed, drawing on a hand-collected dataset of 702 CAT bonds issued by 101 sponsors in 1997–2020. We compare 10 machine-learning (ML) models, including the recent CatBoost and XGBoost models, and the standard multiple regression approach used in most earlier work. Our models capture both traditional catastrophe metrics such as expected loss and idiosyncratic information on sponsor and underwriter reputation. Three main findings emerge. First, idiosyncratic factors do matter. Bonds brought to market by large, experienced players consistently price at lower spreads. Second, most ML methods tested predict spreads more accurately than linear regression. XGBoost with idiosyncratic variables can lead to a substantial reduction in error compared to the Linear benchmark. Third, different risk categories show different dynamics, with idiosyncratic factors being mostly relevant in lower risk categories. Together, these findings suggest that better modern data tools can improve transparency in CAT-bond pricing at the deal level and help the market better fulfil its broader risk-management role.