New Models Uncover Hidden Risks in Loss Data Analysis
The article discusses using severity models to study loss data. Instead of the normal distribution, a thicker-tailed loss distribution is needed. Data samples show a right-skewed shape common in loss data. Two heavy-tailed distributions, GEV and GPD, are used to model severity data. Extreme value theory deals with limiting probability distributions for record values. Extreme stress losses can be seen as contaminations of basic loss patterns.