Schedule P triangle data is a rich resource for building and testing development models. However, challenges arise when fitting tail models, since the data is right-censored after 10 years for long-tailed lines. Building on prior research, we propose a new method to use Schedule P prior year data to directly estimate tail models, rather than just using it as a validation tool. This new method relies on Bayesian inference and is implemented in Stan. By expanding the scope of broadly available data that is amenable to loss reserve modelling, we can improve the accuracy of tail modelling, and gather empirical evidence to evaluate long-held assumptions about tail development.
In this session, we discuss the theory behind the method, the assumptions it makes, and the practical implementation details associated with the method. We provide concrete guidance and resources for working actuaries looking to implement this method into their professional practice.
Learning Objectives:
think more broadly about the best way to fit and validate tail development models using Schedule P data.
use the proposed method and code provided to fit and validate their own tail models using Schedule P prior year data.
understand how Bayesian inference and Stan can be used to solve complex modelling problems
