Application of data mining methods to solving the problems of actuarial modeling and estimation of financial risks
DOI:
https://doi.org/10.20535/SRIT.2308-8893.2017.1.04Keywords:
actuarial processes, insurance risks, generalized linear models, Bayesian approach, data mining, group method for data handling, fuzzy GMDHAbstract
Results of application of the data mining to solving the problem of actuarial processes modeling and risk estimation for insurance companies are presented. As a mathematical modeling tool the following approaches were used: generalized linear models, Bayesian networks, the group method for data handling, fuzzy GMDH, and Bayesian parameter estimation techniques. Using actual statistical data from the insurance industry, new generalized linear models were constructed that were used for estimation of a possible loss by an insurance company. Also, a model in the form of a Bayesian network was constructed that was applied to estimate the bankruptcy risk in a case of insurance losses. The best model constructed in this case turned out to be the gamma distribution based model and logarithmic link function whose parameters were estimated within four iterations of the estimation algorithm. A substantial computed value of the insurance company bankruptcy risk reflects the fact that the company under consideration does not possess an effective mechanism for managing its own capital and the payments from clients. Thus, an application of data mining is an effective approach to solving the problems of short-term forecasting financial processes and estimation of actuarial risks.References
Jones M.T. Artificial Intelligence: a systems approach / M.T. Jones. — Hingham, Massachusetts: Infinity Science Press LLC, 2008. — 516 p.
Munakata T. Fundamentals of the New Artificial Intelligence / T. Munakata. — London: Springer, 2008. — 266 p.
Koski T. Bayesian Networks / T. Koski, J.M. Noble. — Chichester: John Wiley & Sons, Ltd, 2009. — 368 p.
Darwiche A. Modeling and Reasoning with Bayesian Networks / A. Darwiche. — Cambridge: Cambridge University Press, 2009. — 562 p.
Dubinina S.V. Metodyka analizu ekstremal'nykh danykh ta yiyi vykorystannja pry otsinjuvanni parametriv uzahal'nenykh linijnykh modelej / S.V. Dubinina, P.I. Bidjuk // Radioelektronika, informatyka, upravlinnja. — 2016. —No. 1.
McCullagh P. Generalized Linear Models / P. McCullagh, J.A. Nelder. — New York: Chapman & Hall, 1990. — 526 p.
Tripp M.H. Quantifying operational risk in general insurance companies / M.H. Tripp // Working Paper [Presented to the Institute of Actuaries] (22 March 2004). —137 p.
Besag J. Markov Chain Monte Carlo for Statistical Inference / J. Besag // Working Paper, Center for Statistics and the Social Sciences. — 2001. — N 9. — 25 p.
Dubinina S.V. Zastosuvannja merezh Bajyesa do pobudovy modelej otsinjuvannja ryzyku aktuarnykh protsesiv / S.V. Dubinina, P.I. Bidjuk // Scientific Journal ScienceRise. — 2016. — 8, No. 2 (25).
Dovhyj S.O. Systemy pidtrymky pryjnjattja rishen' na osnovi jmovirnisno-statystychnykh metodiv / S.O. Dovhyj, P.I. Bidjuk, O.M. Trofymchuk. — K.: Lohos, 2014. — 419 s.
Bidyuk P.I. Forecasting based on Bayesian type models / P.I. Bidyuk, O.M. Trofymchuk, O.P. Gozhyj // International Journal of Computers and Technology. — 2015. — Vol. 15, N 3. —P. 6570–6584.
Murphy K. A Brief Introduction to Graphical Models and Bayesian Networks [Digital source] / K. Murphy // USA. — 1998. Digital source: URL: http://www.ai.mit.edu/~murphyk/Bayes/bnintro.html.
Negnevitsky M. A guide to intelligent systems / M. Negnevitsky. — London: Addison Wesley, 2005. — 435 p.