Book Description
Accurate loss reserves are essential for insurers to maintain adequate capital and to efficiently price their insurance products. Loss reserving for Property & Casualty insurance is usually based on macro-level models with aggregate data in a run-off triangle. The macro-level models may generate material errors in the reserve estimates when assumptions underlying the estimates evolve over time in an unanticipated way. In recent years, a small set of literature has proposed reserving models that use underlying individual claims data to estimate outstanding liabilities based on individual claim level information, analogous to approaches used in the life insurance industry. These models are referred to as "micro-level models". In this dissertation, I specify a micro-level model with a hierarchical structure to model the individual claim development that has the flexibility to accommodate assumptions that evolve dynamically over time. The dissertation consists of a simulation study and an empirical study. In the simulation study, I simulate claims data under different environmental changes and use both the macro- and micro-level models to estimate the outstanding liabilities. The results demonstrate that there are many scenarios in which the micro-level model outperforms the macro-level model by generating reserve estimates with smaller reserve errors and higher precision. For actuaries responsible for setting reserves, this study highlights scenarios in which micro-level models outperform traditional macro-level models and so can provide a new tool to provide insights when establishing accurate loss reserves. In the empirical study, I demonstrate the application of a micro-level model in a large portfolio of workers compensation insurance provided by a major P&C insurer. The model is estimated with historic data, validated with a hold-out sample, and compared with commonly-used macro-level models. I show that the micro-level model provides a more realistic reserve estimate than that given by the macro-level models, and the estimation error is largely reduced through the use of individual claims data. The micro-level model is more likely to capture the downside potential in reserves and to provide adequate allowance when extreme scenarios occur. I conclude that micro-level models provide valuable alternatives to traditional models for loss reserving.