Generalised linear models

At Taylor Fry, GLM valuation models are used extensively - not because they are clever or fashionable, but rather because they make the modelling process much easier. GLMs are better at capturing the multivariate properties of the valuation data, and tend to require fewer model parameters, thereby increasing their predictive power.

That said, we do recognise a place for deterministic models in many circumstances, and are well equipped to use them when required.

The GLM is one example of a model with a fully stochastic specification. Dynamic GLMs (DGLMs) are GLMs with parameters that vary over time. A particular example of DGLMs in practice is in the area of adaptive reserving. Taylor Fry is at the leading edge of adaptive reserving research and its practical application.

One example of the power of the GLM approach to loss reserving is in the area of segmentation. That is, when outstanding liability estimates are required for various cohorts or segments of the whole. This is best described by way of an example:

Consider an insurer with four distribution channels. For profit sharing calculations the insurer requires the valuation of outstanding claims liability by channel. This problem is often addressed in one of two ways:

• A single valuation is conducted on the entire business, and the resultant liability is dissected by channel (say, on the basis of case estimates); or
• A separate valuation is conducted for each channel, and the results are combined for the insurer as a whole.