Seismic loss modelling in insurance industry : towards a new model for better claims management.

Summary Several scientific reports have highlighted that, both on the history of losses related to natural disasters, and on the average cost expected for the years to come, the seismic risk is the one for which the uninsured loss is the highest among all natural disasters. In this context, the subject chosen for my thesis is: "Modeling seismic risk in insurance: study of a new model for better insurance management". The work is organized in three parts: 1° presentation of the different insurance systems against seismic risk throughout the world. 2° identification of the main areas for improvement in risk modelling . 3° proposals for a new insurance model.The review of insurance systems focused on the following two countries: France and California. By combining several variables related to the level of risk put in perspective with different economic indicators, we have created a specific maturity scale for earthquake insurance. This tool allows us to measure the evolution, both upward and downward, of an insurance market. By applying it to the two countries studied, it appears that neither of them is equipped with a sustainable insurance system. In France, the risk of earthquakes is covered by the CAT-NAT regime. A probabilistic model for the recognition of natural disasters was developed by studying all the municipalities recognized as natural disasters following an earthquake. By applying it to two scenarios representative of a major earthquake in metropolitan France, the results show that the State could be in difficulty in paying the cost of the claims that it is responsible for under the CAT-NAT regime. In California, despite the fact that the risk is significant, only 15% of the population is insured. We conducted a study differentiating the effect of Californians' perception of seismic risk from that of the price of insurance. The results show that Californians do not buy insurance because of the price, not because they underestimate the risk. The model also shows that the majority of Californians would buy earthquake insurance if the price of the insurance was reduced by a factor of three.Based on the previously developed maturity scale, the evolution of this insurance system requires better risk modeling. For this purpose, we have developed a new method of comparison between probabilistic hazard maps and hazard fingerprint models. Refinement of stochastic earthquake loss models also requires strengthening the relationship between the damage scale used and the associated repair costs. For this purpose, a database of economic consequences of past earthquakes has been developed. In addition, an economic model has been developed to test the existing damage-cost models with the historical data previously collected.Finally, the last part of my thesis work concerns the study of a new insurance model in which the amount of premiums is allocated to each building. As long as an earthquake does not damage an insured building, the premium amount is invested to increase the available resources. When an earthquake occurs and damages an insured building, the insurance company pays for the repair or reconstruction work. If the accumulated resources are large enough before an earthquake occurs, they are used to finance seismic strengthening work. The associated cost is then compensated by the gain in resistance of the building. This reduces the premium paid by the insured and creates a virtuous loop of risk prevention.