Chapter 23 ERA 3.2 Modeling Parameter Uncertainty - Venter, Gluck

Parameter risk impact big companies more as their process risk is smaller (23.1)

Projection risk

• Different ways to model the severity trend

  • Model the trend independently or adjust the general inflation then model the superimposed inflation

  • Model with AR(1) (23.2)

Estimation risk in selecting the parameters and the downstream impact

• Use MLE and negative log liklihood

• Slope of the negative log likelihood determines our confidence in the selection

• Infomration matrix (23.3) and correlation matrix (23.4) use with joint lognormal to correlate parameters

Model risk

• Use HQIC to adjust the loglikelihood with a penalty for # of parameters

• Also use a pool of distribution of reasonable parameters

Projection models

• Form of the model we use to do the modeling e.g. expected parameters vs pool of distribution of parameters

• Should focus on the spread of potential outcomes instead of best estimate

• Add flexibility and let the parameters reflect the uncertainty in the assumptions