Scoring Models for Insurance Risk Sharing Pool Opimization

We introduce a flexible scoring model that can be used by property and casualty insurers that have access to a risk-sharing pool to better select the insureds to transfer to the pool. The model discriminates between insureds whose transfer is likely to be profitable under the pool regulations against those paying a fair premium. This model makes use of feature selection methods to automatically discover the most relevant model inputs, yet is robust to overfitting due to the use of a rank averaging technique. By analogy to the knapsack problem, we show what should be the most suitable sorting criterion depending on the pool regulations. We illustrate the performance of the approach by testing against the historical data of a mid-sized Canadian insurer.