Modelling catastrophe claims with left-truncated severity distributions (Q2463663)

The authors analyze losses resulting from natural catastrophic events in the United States. They study the effects of data misspecification, under which the loss data available from a loss data base is truncated from below at a predetermined threshold level. The authors emphasize that such thresholds are often ignored in practice and this can result in biased estimates of the model parameters. Treating the available data as complete also results in under-estimating the intensity of the events which can lead to wrong (under-estimated) ruin probabilities for the compound risk processes. The authors present a method for treating the loss data samples with non-randomly missing observations where the number of missing data points is unknown. The theoretical aspects of the effects of necessary adjustments to severity and frequency distributions are examined. Also, an extensive empirical study for the 1990-1999 U.S. natural catastrophe data is given. The choice of most adequate models is considered via goodness-of-fit methods.