| 摘要 |
A method of predictive modeling is for purposes of estimating frequencies of future loss and loss distributions for individual risks in an insurance portfolio. To forecast future losses for each individual risk, historical data relating to the risk is obtained. Data is also obtained for other risks similar to the individual risk. Expert opinion relating to the risk is also utilized for improving the accuracy of calculations when little or no historical data is available. The historical data, any current data, and expert opinion are combined using a Bayesian procedure. The effect of the Bayesian procedure is to forecast future losses for the individual risk based on the past losses and other historical data for that risk and similar risks. Probability distributions for predicted losses and historical data for use in the Bayesian procedure are obtained using a compound Poisson process model.
|
