Cat bonds & Artificial Neural Networks | An example of reinsurance products’ pricing using machine learning methods (2016) by Mikaël BENIZRI, supported by Ziad FARES

Over the last fifty years, numbers and costs of natural disasters have not ceased to multiply. Given this phenomenon, insurers and reinsurers struggle to cover the associated losses. Consequently, they turned to financial markets in order to obtain new hedging capabilities, by using various types of products, suchas excess of loss contracts (named XL) and cat bonds.

This paper presents a mathematic model allowing to predict the number and the cost of incoming catastrophes. Data used include wind catastrophes affecting the southeast area of the United States and whose damages are worth more than a billion dollars. This model helps to price insurance risk transfer products, such as XL contracts or cat bonds. First a regression relying on neural network methodology is implemented in order to predict the global annual cost of future catastrophes. Then, based on the same methodology, a classification is done in order to allocate these costs to the various catastrophes.

Our models are used to price several contracts included in the reinsurance program of “Heritage Insurance,” so our results help estimate the share of premiums received by the reinsurer. Two calculations methods are applied: the exposure curve and the “burning cost” method.

Ours models are also validated on cat bond products, but this time using a financial method. This method allows to estimate the share of the Expected Excess Return (EER) depending of the Probability of First Loss (PFL) and the Conditional Expected Loss (CEL).


Follow us