11.1 Introduction & Synopsis
Applying different estimate methods to 200 triangles and comparing the actual outcomes to the predicted distribution to see if the models accurately estimates the distribution of outcomes
Models examined:
Estimate from Mack and ODP do not have enough variability
Incurred losses with variable row parameters and AY correlation \(\rho\) is sufficient
Paid losses requires variable row parameters and change in settlement rate parameter \(\gamma\)
Three reasons model doesn’t predict well:
Insurance loss environment has experienced changes that are not observable at the valuation date
i.e. There would be different “black swan” events that invalidate any attempt to model loss reserves
There could be other models that better fit the existing data
Data used to calibrate the model is missing crucial information needed to make a reliable prediction
e.g. Changes in the way the underlying business is conducted, like changes in claim processes of changes in the direct/ceded/assumed reinsurance composition of the claims triangle
If we can find better model and/or better data we can rule out 1)
- If we review many models and none of them validate it gives 1) credence but does not confirm