6.3 3 Key Assumptions of Model

1. Incremental losses \(iid\)

   • Test this using residual analysis

2. \(\frac{Variance}{Mean}\) scale parameter \(\sigma^2\) is fixed and known

   • Technically this should be estimated with the other parameters but will makes things intractable

3. Variance estimates are based on the approximation to the Rao-Cramer lower bound

   • Variance based on information matrix \(I\) (6.6)

   • \(I\) is exact only when using linear functions

   • In our case this is simply a lower bound

   • We are using approximated parameters

The above temper the volatility in the model, actual results can be more variable

Model only works for positive expected incremental losses, but a negative loss here or there is fine as well