Seminar by Dr. Glen Barnett
- Topic: Boostrapping predictions in loss reserving with a power-variance function - a comparison of transformation and GLM approaches - Part I
- Speaker: Dr Glen Barnett, Visiting Fellow, Department of Applied Finance and Actuarial Studies, Macquarie University
- When: 1st June, 2012, (Fri)
When modelling data with the variance proportional to a power of the mean (Var(Y) ∝ μp), common approaches include the use of linear models after applying a power transformation, such as the Box-Cox transformation and generalized linear models using the Tweedie distribution.
In particular, when the power index, p, lies between 1 and 2 - as sometimes appears to be the case with reserving data - the choice may not be an obvious one.
This part examines the case where prediction intervals are generated by the bootstrap and where the power-index is treated as known, even where it is estimated; this would be the situation most likely to be found in practice.
Both approaches are quite feasible to use with the bootstrap; we compare them on several practical criteria. In the examples we compare the bootstrapped mean and standard deviation for the total outstanding, and upper one-sided prediction intervals both for moderate and extreme quantiles.
To view the full paper click - Boostrapping predictions in loss reserving with a power-variance function - a comparison of transformation and GLM approaches - Part I