Technical Issues in General Insurance
Backtesting is one of the more valuable tests performed when validating internal models: it tells us something about model output relative to results observed historically (hopefully that the two are in agreement!)
However, validating dependencies in this way is often challenging. For any pair of risks, we can calculate the historical correlation, but this only provides a single data point to rely on. There is also no guarantee that this figure will bear much resemblance to the selected correlation being modelled.
Our overall aim is to discuss a framework for validating this aspect of capital models in a useful and structured way, based on a recent case study.
The main areas of the talk will be as follows:
- A quick review of the main dependencies backtesting methods currently available (implicit and explicit)
- An extension of the explicit method, based on assessing the extent to which ranges of inputs/outputs fall within confidence intervals around historical values
- A further extension of the explicit method, in which we examine how well backtesting works for various segments of a typical insurance correlation structure
We will also discuss some potential other uses for the methodology, focusing on enhancements to a typical ESG validation exercise.