- Date: 26 Jul. (Thu.)
- Place: W.W. 6th-floor, Colloquium Room
- Time: 16:30-18:00
- Speaker: Xiaoshan Su (EMLYON Business School)
- Title: Some interesting insurance work: Distribution Choice in Non-life Insurance Risk Model with Statistical Learning Method & Optimal Insurance under Third Degree Risk
(Xiaoshan Su emlyon business school)
- Abstract:
The former work treats a distribution selection problem as a classification problem for claim frequency and claim severity in non-life insurance risk model, and trains machine learning classifier to predict most likely distribution for real data. The training of classifier uses a simulation training sample that is generated under a two-level hierarchical structure. The first level is to generate enough pieces of parameter sets for all competitive distributions, and for each piece of parameter set in each distribution, the second level is to simulate a large size of sample and compute the respective values of descriptive statistic variables, which forms a record of training sample by combination with the corresponding distribution label. Then, the cross-validation method compares the performance of commonly used classifiers, including decision tree, k-nearest neighbour classifier, neural network, support vector classifier, bagging, boosting and random forest, etc. Both of numerical experiments and empirical studies show decision tree and bagging, boosting and random forest that use decision tree as weak learners, perform better than other classifiers and also than traditional fitting measures. The latter work investigates the optimal insurance design of considering a wide coverage of insured including risk-averter and risk-lovers, by assuming that the insured is third degree risk averse. Under expected value premium principle, we show that the optimal insurance form is a change-loss insurance or a dual change-loss insurance, which depends on the coefficient-variance of the ceded loss. The insurer with a mean-variance preference has a strong motive to issue these two kinds of insurance contracts.
日本語
English