| Abstract [eng] |
In insurance mathematics are often used the lognormal ant the Pareto distributions with two parameters to model loss data. The lognormal distribution is used to model small data with higher frequencies, while the Pareto distribution is used to model large data with low frequencies. In order to achieve both of these losses in one model, the truncated lognormal and the Pareto distributions mixture with three parameter is presented. This work is written with reference to Kahadawala Cooray ir Malwane M. A. Ananda article "Modeling actuarial data with a composite lognormal – Pareto model" ("Scandinavian Actuarial Journal", 2005, 5, 321 - 334 pages), where composite lognormal – Pareto model is researched. In this work the necessity of the lognormal and the Pareto distributions mixture is discussed, the derivation of the truncated lognormal and the Pareto distributions mixture model are presented and behaviour of density function in dependent of parameter variation is discussed by illustrating. For practical application three data sets are chosen: simulated from the composite lognormal – Pareto distribution, personal accident insurance loss of one Lithuania insurance company, Danish fire loss data. For the each of data sets the parameters are estimated using maximum likelihood function, resulted estimators are compared with maximum likelihood estimators of composite lognormal – Pareto, lognormal, Pareto, Gamma and Weibull distributions. In order to compare the models the following criterions are used: Kolmogorov – Smirnov, Anderson – Darling criterions and the p – value from the chi square goodness of it test. The result indicate that the truncated lognormal and the Pareto distributions mixture fits most for every data set of insurance loss. |
