A Causal Analysis of Cat Bond Markets

Abstract: This work is a contribution to the causal analysis of the catastrophe bond market, which has generated high excess returns over the last two decades. Since these excess returns remain partially unexplainable and the interest in catastrophe bonds is increasing, the causal study of the factors affecting their premiums is of high relevance. Most studies about the catastrophe bond market have used only linear models. However, more complex models such as random forests may be better suited for modeling this market. Considerable progress has been made in developing methods for inference in the specific setting of random forests. Causal random forests, especially when combined with double machine learning and Shapley values, represent a sophisticated empirical toolbox. They provide unbiased prediction intervals and allow the analysis of heterogeneous effects, while Shapley values help interpret individual predictions. I apply these methods to quantify uncertainties and heterogeneities of effects in the secondary catastrophe bond market in part I. In part II, I apply these methods to quantify uncertainties and heterogeneities of effects in the primary catastrophe bond market with respect to the issue date of the bonds. In particular, I analyze whether the expected loss has a time-varying effect on the catastrophe bond premiums. My results confirm that the effect of the analyzed factors on the premiums is often heterogeneous. In addition to mean effects, I present median effects, and their whole distribution. Expected loss is by far the most decisive factor. To date, the interactions among predictors have not been extensively studied, and there may be non-linearities in the explanatory predictors. My additional heterogeneity analysis provides answers as to which factors cause the variance in the impact of the expected loss on the premiums.