Open Conference Systems, 50th Scientific meeting of the Italian Statistical Society

Font Size: 
Modelling insurance losses via contaminated unimodal distributions
Salvatore Daniele Tomarchio

Last modified: 2018-05-18

Abstract


Forecast the loss associated with a claim is crucial in insurance industry. These types of payments are generally highly positively skewed and with heavytails, highlighting the necessity of flexible models. Contaminated models are a profitable way to accommodate situations in which some of the probability masses areshifted to the tails of the distribution, and in this work a general approach to contaminate unimodal hump-shaped distributions defined on a positive support is introduced. The proposed models are hence fitted to a real insurance loss dataset, alongwith several standard distributions used in the actuarial literature. Comparison between the models is made using information criteria and risk measures such as VaRand TVaR.

References


1. Ahn, Soohan, Joseph HT Kim, and Vaidyanathan Ramaswami. â€A new class of models forheavy tailed distributions in finance and insurance risk.†Insurance: Mathematics and Eco-nomics 51.1 (2012): 43-52.

2.Aitkin, Murray, and Granville Tunnicliffe Wilson. â€Mixture models, outliers, and the EMalgorithm.†Technometrics 22.3 (1980): 325-331.

3. Bagnato, Luca, and Antonio Punzo. â€Finite mixtures of unimodal beta and gamma densitiesand the k -bumps algorithm.†Computational Statistics 28.4 (2013): 1571-1597.

4. Bakar, SA Abu, et al. â€Modeling loss data using composite models.†Insurance: Mathematicsand Economics 61 (2015): 146-154.

5. Berkane, Maia, and Peter M. Bentler. â€Estimation of contamination parameters and identifi-cation of outliers in multivariate data.†Sociological Methods & Research 17.1 (1988): 55-64.

6. Bickerstaff, David R. â€Automobile Collision Deductibles and Repair Cost Groups: The Log-normal Model.†PCAS LIX (1972): 68.

7. Burnecki, Krzysztof, Adam Misiorek, and Rafal Weron. â€Loss distributions.†Statistical Toolsfor Finance and Insurance. Springer, Berlin, Heidelberg, 2005. 289-317.

8. Chen, Song Xi. â€Probability density function estimation using gamma kernels.†Annals of theInstitute of Statistical Mathematics 52.3 (2000): 471-480.

9. Cooray, Kahadawala, and Malwane MA Ananda. â€Modeling actuarial data with a compositelognormal-Pareto model.†Scandinavian Actuarial Journal 2005.5 (2005): 321-334.

10. Eling, Martin. â€Fitting insurance claims to skewed distributions: Are the skew-normal andskew-student good models?.†Insurance: Mathematics and Economics 51.2 (2012): 239-248.

11. Jeon, Yongho, and Joseph HT Kim. â€A gamma kernel density estimation for insurance lossdata.†Insurance: Mathematics and Economics 53.3 (2013): 569-579.

12. Kazemi, Ramin, and Monireh Noorizadeh. â€A Comparison between Skew-logistic and Skew-normal Distributions.†Matematika 31.1 (2015): 15-24.

13. Punzo, Antonio, and Paul D. McNicholas. â€Parsimonious mixtures of multivariate contami-nated normal distributions.†Biometrical Journal 58.6 (2016): 1506-1537


Full Text: PDF