×

Spread pricing model of catastrophe bonds based on the machine learning. (Chinese. English summary) Zbl 1474.91206

Summary: The empirical research on the spread of catastrophe bonds is relatively mature, but there are still some controversies about the model form and variable selection. This thesis summarizes the pricing factors in historical literature and constructs new pricing factors. This thesis establishes a spread pricing model for catastrophe bonds based on the issued data of catastrophe bonds. First, this thesis constructs a generalized linear model (GLM) and finds that the risk load of logit performs well. Then, results of the GLM are embedded into deep neural network (DNN). This method improves the prediction ability of the GLM and improves the iteration efficiency of the neural network. Finally, this thesis compares the pricing effects of machine learning algorithms such as DNNs, DNNs embedded with GLMs, random forests, XGBoost, and support vector regression. Empirical results show that catastrophe bond pricing models based on machine learning algorithms are significantly better than traditional regression models. This thesis proposes to use Support Vector Regression to price spread of catastrophe bond.

MSC:

91G20 Derivative securities (option pricing, hedging, etc.)
68T07 Artificial neural networks and deep learning
PDFBibTeX XMLCite