Aktüerya Bilimleri Bölümü Tez Koleksiyonuhttp://hdl.handle.net/11655/2922019-10-23T07:59:34Z2019-10-23T07:59:34ZKasko Sigortasında Makine Öğrenmesi Yöntemleri İle Hasarlı/Hasarsız Durum TahminiKilisli, Sedahttp://hdl.handle.net/11655/94592019-10-21T12:45:25Z0019-09-19T00:00:00ZKasko Sigortasında Makine Öğrenmesi Yöntemleri İle Hasarlı/Hasarsız Durum Tahmini
Kilisli, Seda
The aim of this study is to estimate whether new insured who will join insurance company have claim by using machine learning methods. The increase in the number of vehicles in traffic each year brings with it an increase in the risk of accidents. Increasing number of accidents increases the costs of insurance companies and this increase is also reflected in the insurance premiums. However, due to price competition in insurance companies, sales below the optimal premium cause companies to fall behind their profitability targets. In order for insurance companies to maintain their profitability, it is very important to include the profitable insured profile in the portfolio. In order to include the profitable insured in the portfolio, models that simulate individual behaviors are needed. Therefore, by using machine learning algorithms used in many sectors recently, the model that makes the best estimate of claimed policies in the portfolio is determined and compared with the logistic regression model commonly used in the insurance sector.
0019-09-19T00:00:00ZHEDEF PROGRAMLAMA VE EN KÜÇÜK KAR FARKI YAKLAŞIMLARI İLE OPTİMAL REASÜRANSKaragül, Betül Zehrahttp://hdl.handle.net/11655/93372019-10-21T12:22:43Z2019-01-01T00:00:00ZHEDEF PROGRAMLAMA VE EN KÜÇÜK KAR FARKI YAKLAŞIMLARI İLE OPTİMAL REASÜRANS
Karagül, Betül Zehra
The aim of this study is to contribute to the optimal reinsurance studies, which have a considerable role in the actuarial literature, by considering the situation from a perspective that takes into account the insurer and the reinsurer together. For this purpose, firstly, the solution of the analytical model which minimizes “the VaR of the absolute value of the difference between the profits of the insurer and the reinsurer” is obtained. Secondly, the models which both taking into account both sides and multi objective / constrained, are examined by using goal programming.
In the literature review, optimal reinsurance studies are discussed under 4 main headings; studies on optimal reinsurance for insurer, studies on optimal reinsurance for reinsurer, studies on optimal reinsurance for insurer and reinsurer, studies on optimal reinsurance with multiple criteria. In addition, basic information about reinsurance, types of reinsurance, principles of premiums and risk measurements are given.
In order to illustrate the necessity of taking into account both the insurer and the reinsurer company as the parties of a reinsurance contract when calculating the optimal retention, this thesis examines a prior study of optimal reinsurance which only considers the insurer’s point of view. Subsequently discussed is a model which determines the optimal retention from both points of view of insurer and reinsurer with the corresponding simulations used to justify this model. This forms the preliminary work this thesis builds upon. This model assumes that the claim numbers are Poisson distributed and the claim sizes are exponential, lognormal and Pareto distributed. The results are obtained separately for both the stop-loss and the excess-of-loss reinsurance. As the premium principle, both standard deviation premium principle and expected value premium principle are used. The results are compared with the tables and figures for these two studies.
This thesis seeks to contribute to the existing literature by taking into account both the insurer and the reinsurer. An analytical model is set up and makes use of “the VaR of the absolute value of the difference between the profits of the insurer and the reinsurer” as a risk measure, the solution of which is obtained and presented in this thesis. The results of this analytical model are compared with the results of the preliminary work of this thesis and with the optimal reinsurance study which is examined in this thesis and has only considered the insurer point of view by considering the real world examples and using the numerical examples. We assume the aggregate loss is exponential and Pareto distributed and the premiums of both the insurer and the reinsurer are calculated using the expected value premium principle with stop-loss reinsurance.
The second aim of this thesis, is to contribute to the literature by addressing multi objective/constrained models using goal programming. First a general definition of the goal programming method used, subsequently the terminology and the variants of goal programming are summarized. For this application, 11 different multi objective/constrainted optimal reinsurance models have been constructed and their solutions have been investigated making use of goal programming. The constraints in the models are as follows: the value at risk of the absolute value of the difference between the insurer's profit and the reinsurer's profit; the absolute value of the difference between the standard deviation of insurer’s expected profit; the standard deviation of reinsurer’s expected profit; the expected utility function of insurer (with exponential utility and logaritmic utility); the expected utility function of reinsurer; the expected profit of reinsurer; the expected profit of insurer; the value at risk of the total cost of insurer; the value at risk of the total cost of reinsurer.
Mathematical representations of the models are given by using the goal programming model and optimal retention levels and deviation variables are obtained using the stop-loss reinsurance are presented in the tables. The tables are prepared for both the expected value and the standard deviation premium principle and assumptions that the losses are distributed by Pareto, exponential and lognormal distributions. Comparisons between the models are made. In addition, how the results change with different premium loading coefficients and different initial wealth is shown in the tables.
As a result of this study, these models are interpreted from the joint perspective of insurer and reinsurer and examined in terms of their contributions and innovations to the literature. The difference between unilateral optimal reinsurance studies and optimal reinsurance studies taking both sides into account is demonstrated in terms of their real-world suitability and acceptability for both sides. In addition, the results of the multi objective reinsurance study and the advantages of using goal programming method in the solution of this study are also revealed.
2019-01-01T00:00:00ZHayat Dışı Sigortalarda Doğrusal Olmayan Bağımlılığın Kopulalar ile Dinamik Finansal AnaliziKaragül, Betül Zehrahttp://hdl.handle.net/11655/56772019-05-15T07:10:31Z2013-01-01T00:00:00ZHayat Dışı Sigortalarda Doğrusal Olmayan Bağımlılığın Kopulalar ile Dinamik Finansal Analizi
Karagül, Betül Zehra
In this study, with Dynamic Financial Analysis model approach that includes basic components for a non-life insurance company, two different simulation studies have been done. Non linear dependencies have been integrated into the model using the copulas. In line with the simulation study, these dependencies effects on the insurer s risk and return profile, the default risk and the ruin probability have been evaluated.Basic Information about Dynamic Financial Analaysis, dependence measures and copulas, that are main structure of the simulation studies which we done, have been given. The Copula families and DFA model framework which are used in the simulation study are detailed. The financial ratios, using for evaluating risk, return and performance of company, are mentioned.In the simulation studies for the same DFA model there are different dependence structures.
2013-01-01T00:00:00ZMultivariate Stochastic Prioritization Of Dependent Actuarial Risks Under UncertaintyNevruz, Ezgihttp://hdl.handle.net/11655/55652019-05-15T07:10:26Z2018-07-01T00:00:00ZMultivariate Stochastic Prioritization Of Dependent Actuarial Risks Under Uncertainty
Nevruz, Ezgi
The main prompting factor behind decision making is comparing or ordering risks. Risk management strategies should be based on the dynamics of stochastic ordering relations and influences of decision makers' tendencies on risk prioritization. The objective of this thesis is to construct a concept for stochastic risk prioritization of multivariate aggregate claims.
The definition of risk from perspectives of individuals, companies or governments may vary according to their risk perceptions, as risk is indicated not only by objective measures but also by subjective characteristics. In order to describe the risk accurately, the theoretical background of multivariate stochastic prioritization of dependent actuarial risks should be understood. For this aim, we familiarize ourselves with order theory that allows comparing and ordering objects characterized by multiple indicators.
Being an important issue of human behaviour, this area falls within the boundaries of several fields, one of which - public
health - is our specific interest. We intend to apply the order theory to a chosen risk area such as foodborne or agricultural risks, since they are rather vulnerable aspects of public health. Analytic tools may not always be sufficient for prioritization especially when we work on environmental risks. Hence, geographic information system is a useful tool for risk prioritization in such cases.
In this thesis, we aim to prioritize aggregate claim vectors of different risk clusters in agricultural insurance under the assumption that the individual claims exposed to similar environmental risks are dependent. For this purpose, first we obtain risk clusters for a crop-hail insurance portfolio considering spatial and temporal features of hazard regions. We propose an extended approach for differential evolution optimization which determines the optimal sample set used in inverse distance weighting with reduction technique. Second, we prioritize the aggregate claims taken as actuarial risks by using various stochastic ordering relations that are studied within the framework of partial order theory. These relations are stochastic dominance, stochastic majorization and stop-loss dominance. Having discussed the concept of risk itself, we also investigate the risk measures which could be sufficient and accurate criteria for determining the riskiness of a portfolio.
The classical first-order stochastic dominance is useful to design the risk prioritization context. We also suggest stochastic majorization relation according to multivariate representation of actuarial risks. This relation is very beneficial for our study since it enables us to order aggregate claim vectors partially using Schur-convex risk measures.
On the other hand, we consider the impacts of risk perception on prioritization of risks. Working within this context and attempting to contribute to it, we seek for a behavioral approach which could enhance and facilitate the description of the choices individuals make in risky situations. An example of such approaches could be cumulative prospect theory (CPT), as a more accurate alternative to expected utility theory. In the stop-loss dominance context, we adapt the zero-utility premium principle in order to obtain solutions for stop-loss premiums and propose stop-loss dominance relation under CPT.
2018-07-01T00:00:00Z