意外、健康保险欺诈概率识别

2014-10-30 19:51周建涛秦炳山等
河北经贸大学学报 2014年6期
关键词:保险金额回归系数欺诈

周建涛 秦炳山等

摘要:保险理赔过程中时常发生投保方欺诈现象,通过对实证调研的2000—2011年度258个意外、健康保险理赔诉讼样本进行描述性统计和Logistic回归,发现年龄41~60岁欺诈概率甚于其他年龄段,年龄41~50岁欺诈频次最高,年龄51~60岁欺诈概率最大,工人欺诈概率甚于其他职业,北京欺诈概率甚于外地,北京郊区县欺诈概率甚于北京市区;保险金额与欺诈概率正相关,疾病身故保险欺诈金额最大,保险期间与欺诈概率负相关,意外伤害住院医疗保险欺诈频次最高,健康险欺诈概率甚于意外险;摔扭伤欺诈概率甚于其他伤害,交通事故机动车司机欺诈概率甚于交通事故非机动车伤害,被保险人索赔欺诈概率甚于投保人索赔,非律师参与欺诈概率甚于律师参与。

关键词:欺诈概率;索赔人特征;保单信息;案件情况;健康保险;统计分析;保险金额;律师

中图分类号:D90 文献标识码:A 文章编号:1007-2101(2014)06-0138-05

一、引言

自保险业诞生之日起,保险欺诈(Insurance Fraud)随之而来。美国雪城大学(Syracuse University)“政府档案交流中心”(Transactional Records Access Clearinghouse,TRAC)统计,2011年1-8月美国发生903起健康险欺诈诉讼,相比2010年、2006年、1991年分别增加85%、157%和822%;美国保险研究委员会(Insurance Research Council,IRC)发现,2007年美国汽车人身伤害保险欺诈造成保险公司多支出48亿美元~68亿美元;英国保险人协会(Association of British Insurers,ABI)估计,保险欺诈每年给保险业带来20亿英镑损失,未能识别的欺诈金额每年高达19亿英镑,每个投保人每年额外支付约44英镑。广东省是我国经济、保险发达地区,索赔欺诈相对严重:东莞,2010年财产险赔付支出21.79亿元,其中约4亿元属于欺诈索赔;深圳,2008年各保险机构车险赔案,疑似骗赔案件976起,涉案金额超过2 000万元,某些险种欺诈导致的赔款支出最高达到保费收入的5倍。

保险欺诈是否属实,法院判决具有客观公信力。实证调研部分保险理赔诉讼案件,深入研究欺诈动因,提炼欺诈识别因子,对于有效遏制欺诈、促进保险业可持续发展,具有重要的理论意义和实践价值。

二、研究假设、模型设计与样本选择

(一)研究假设

假设1:保险欺诈受索赔人特征(性别、年龄、职业、婚姻状况、所在地区等)影响。

假设2:保单信息(保险金额、保险费、险别、保险期间等)对索赔欺诈也有重要影响。

假设3;索赔欺诈还受案件情况(保险事故类型、索赔人类型、是否律师参与、是否存在第三方赔付等)影响。

(二)模型设计

三、描述性统计分析

对258个案件数据进行描述性统计,结果见表2。

表2显示,男性欺诈概率(0.659)甚于女性(0.341),年龄41~50岁欺诈概率(0.437)甚于其他年龄段,工人欺诈概率(0.333)甚于其他职业,已婚欺诈概率(0.881)远甚于其他婚姻状况,北京欺诈概率(0.570)高于外地(0.430),北京郊区县欺诈概率(0.281)高于北京市区(0.148),外地地市欺诈概率(0.333)高于外地区县(0.237)。

高保险金额欺诈概率(0.532)甚于低保险金额(0.491),高保险费欺诈概率(0.548)甚于低保险费(0.458),短期保险欺诈概率(0.800)甚于长期保险(0.200),健康险欺诈概率(0.993)高于意外险(0.733)。

疾病欺诈概率(0.321)高于伤害,摔扭伤欺诈概率(0.207)远大于其它伤害,交通事故机动车司机伤害欺诈概率(0.163)略高于交通事故非机动车伤害(0.156),被保险人索赔欺诈概率(0.540)远大于投保人索赔(0.193),非律师参与欺诈概率(0.319)大于律师参与(0.170),无第三方赔付欺诈概率(0.919)大于第三方赔付(0.081),无证人证言欺诈概率(0.852)远甚于证人证言(0.148)。

对每个索赔案件欺诈总额(索赔总额-判决总额)及各子项欺诈金额(索赔金额-判决金额)进行描述性统计,结果如表3。

欺诈总额,从79.48元~399 138.7元不等,均值29 121.08元,不同案件欺诈金额差异很大。135例欺诈案件中,意外伤害住院医疗保险(63例)欺诈数量最多,其次为疾病住院医疗保险(35例)和伤残保险(33例)。从子项欺诈金额看,疾病身故保险(1例)与意外伤害身故保险(25例)欺诈均值最高,分别为80 000元和68 267.09元;其次是伤残保险(33例)和重大疾病保险(1例),分别为30 901.44元和10 000元;疾病医疗保险(1例)最低,均值499.5元。

四、Logistic回归分析

为避免多重共线性,本文筛选出的各变量容忍度(Tolerance)均大于0.2,方差膨胀因子(Variance inflation factor,VIF)均小于5,Logistic回归结果见表4。

表4显示,极大似然估计值为195.624,Nagelkerke R2为0.615,说明模型拟合度较好;从P-value看,年龄41~50岁、年龄51~60岁、工人、北京市区、北京郊区县、高保险金额、健康险、长期保险、摔扭伤、交通事故机动车司机伤害、被保险人索赔、非律师参与的回归系数在5%置信度水平下显著不为零,年龄41~50岁、年龄51~60岁、工人、北京市区、北京郊区县、高保险金额、健康险、长期保险、摔扭伤、交通事故机动车司机伤害、被保险人索赔、非律师参与是欺诈概率识别的关键要素。

年龄51-60回归系数为2.002,年龄51~60岁比其它年龄组欺诈对数发生比平均高2.002倍,可能是年龄51~60岁已进入老年,但尚未退休,对未来预期较悲观,为应对未来不确定性而欺诈。年龄41~50岁回归系数为0.314,年龄41~50岁比其他年龄组欺诈对数发生比平均高0.314倍,可能是年龄41~50岁上要赡养父母,下要抚养子女,经济压力较大,为缓解经济压力而欺诈。从回归系数看,年龄41~50岁欺诈明显弱于年龄51~60岁,但结合表2描述性统计I,年龄41~50岁欺诈案件最多(59例),是年龄51~60岁(22例)的2.68倍,这两个年龄段欺诈特点不同,应采取不同的应对措施。工人回归系数为1.68,工人比其他职业欺诈对数发生比平均高1.68倍,可能是工人收入偏低,通过欺诈缓解自身经济压力。北京郊区县回归系数为2.422、北京市区回归系数为2.200,北京郊区县比外地欺诈对数发生比平均高2.422倍、北京市区比外地欺诈对数发生比平均高2.200倍,北京欺诈甚于外地,可能源于北京高房价、高工作强度、高消费等压力;北京郊区县欺诈甚于北京市区,可能是北京郊区县聚集了相当数量外来务工人员,文化素质较低,对保险条款理解肤浅,出险后不管事故严重程度如何,往往按保险金额上限索要。

高保险金额回归系数为1.352,高保险金额比低保险金额欺诈对数发生比平均高1.352倍,欺诈和保险金额正相关,高保险金额更会诱致欺诈,例如疾病身故保险欺诈金额80 000元,意外伤害身故保险欺诈金额68 267.09元。长期保险回归系数为-1.198,长保险期间比短保险期间欺诈对数发生比平均低1.198倍,欺诈与保险期间负相关,短期保险更易导致欺诈,例如意外伤害住院医疗保险欺诈63例,可能为弥补保费和免赔额而欺诈。健康险回归系数为2.811,健康险比意外险欺诈对数发生比平均高2.811倍,可能是健康险承保常免于体检,便于被保险人隐瞒所患疾病。

摔扭伤回归系数为2.561,摔扭伤比其他伤害欺诈对数发生比平均高2.561倍,国际欺诈识别研究表明,摔扭伤系软组织伤害,最难界定,更易欺诈。交通事故机动车司机伤害回归系数为1.729,交通事故机动车司机伤害比交通事故非机动车伤害欺诈对数发生比平均高1.729倍,可能是交通事故机动车司机就是事故直接责任人,迫于息事宁人的赔偿压力而欺诈。被保险人索赔回归系数为2.608,被保险人索赔比其他人索赔欺诈对数发生比平均高2.608倍,因为被保险人就是保险事故当事人,比投保人、受益人等更了解意外事故真实情况和自身疾病详情,索赔时更能够避重就轻,掩盖事实真相或夸大受伤、疾病程度。非律师参与回归系数为2.239,非律师参与比律师参与欺诈对数发生比平均高2.239倍,非律师参与人多是索赔方的亲戚、朋友、同事等,对保险、法律也不甚了解,可能基于同情保险事故而包庇、纵容索赔人欺诈行为,亲亲相隐。

五、结论和建议

本文对实证调研的258个意外、健康保险理赔诉讼样本进行描述性统计和Logistic回归,发现索赔人年龄、职业、所在地区、保险金额、险别、保险期间、保险事故类型、索赔人类型、律师是否参与等是欺诈概率识别的关键因子。

1. 年龄41~50岁、年龄51~60岁欺诈概率甚于其他年龄组,其中年龄41~50岁欺诈频次最多,年龄51~60岁欺诈概率最大。工人欺诈概率甚于其他职业。北京欺诈概率甚于外地,北京郊区县欺诈概率甚于北京市区。建议保险公司理赔时,重点审查年龄41~50岁、年龄51~60岁、工人、北京、北京郊区县等投保方特征。

2. 保险金额与欺诈概率正相关,身故保险金额较高,欺诈概率远甚于其他保险。保险期间与欺诈概率负相关,意外伤害住院医疗保险期限较短,欺诈概率远甚于其他保险。健康险欺诈概率甚于意外险。建议保险公司理赔时,高度关注高保险金额、短期保险、健康保险等保单信息。

3. 摔扭伤欺诈概率甚于其他伤害,交通事故机动车司机伤害欺诈概率甚于交通事故非机动车伤害。被保险人索赔欺诈概率甚于投保人索赔,非律师参与欺诈概率甚于律师参与。建议保险公司理赔时,重视保险事故类型(摔扭伤、交通事故机动车司机伤害)、索赔人类型(被保险人索赔)、律师是否参与(非律师参与)等案件情况。

参考文献:

[1]M. Artís,M. Ayuso,M. Guillén. Detection of Automobile Insurance Fraud with Discrete Choice Models and Misclassified Claims. The Journal of Risk and Insurance,2002,(69):325-340.

[2]S.B. Caudill,M. Ayuso,M. Guillén. Fraud Detection Using a Multinomial Logit Model With Missing Information. The Journal of Risk and Insurance,2005,(72):539-550.

[3]J.Ai,P. Brockett,and L. Golden. Assessing Consumer Fraud Risk in Insurance Claims with Discrete and Continuous Data. North American Actuarial Journal,2009,(13):438-458.

[4]J.P. Boucher,M. Denuit,M. Guillen. Number of Accidents or Number of Claims? An Approach with Zero-Inflated Poisson Models for Panel Data,The Journal of Risk and Insurance,2009,(76):821-846.

[5]H.J. Smoluk. Long-Term Disability Claims Rates and the Consumption-to-Wealth Ratio. The Journal of Risk and Insurance,2009,(76):109-131.

[6]周建涛,巨珣,董楠.保险公司应对欺诈的诉讼研究[J].北京工商大学学报(社会科学版),2011,(4):117-121.

责任编辑、校对:张增强

Abstract: The insurance claim process is often accompanied by the claimant exaggerated claims. This article regresses the 258 accident and health insurance sample data of claiming litigations between 2000 and 2011 with descriptive statistics and Logistic model, finding that the 41-50 and 51-60 year-old is worse than the other year-old, and the 41-50 with the highest fraud frequency and the 51-60 with the highest fraud probability; the worker' fraud is worse than the others'; Beijing is worse than other areas, and Beijing suburb worse than Beijing city; fraud is significantly positive with insurance amount, the death insurance from illness with the largest fraud amounts, fraud is significantly negative with insurance term, casualty hospitalization insurance with the highest fraud frequency, health insurance worse than accident; strain/sprain fraud is worse than trauma, troublemaker in a traffic accident worse than the injured, the insured fraud worse than the insurant, fraud with non-lawyer worse than that with lawyer.

Key words: fraud probability, the insured characteristics, policy information, the case circumstances, health insurance, statistical analysis, insurance amount, lawyer

[5]H.J. Smoluk. Long-Term Disability Claims Rates and the Consumption-to-Wealth Ratio. The Journal of Risk and Insurance,2009,(76):109-131.

[6]周建涛,巨珣,董楠.保险公司应对欺诈的诉讼研究[J].北京工商大学学报(社会科学版),2011,(4):117-121.

责任编辑、校对:张增强

Abstract: The insurance claim process is often accompanied by the claimant exaggerated claims. This article regresses the 258 accident and health insurance sample data of claiming litigations between 2000 and 2011 with descriptive statistics and Logistic model, finding that the 41-50 and 51-60 year-old is worse than the other year-old, and the 41-50 with the highest fraud frequency and the 51-60 with the highest fraud probability; the worker' fraud is worse than the others'; Beijing is worse than other areas, and Beijing suburb worse than Beijing city; fraud is significantly positive with insurance amount, the death insurance from illness with the largest fraud amounts, fraud is significantly negative with insurance term, casualty hospitalization insurance with the highest fraud frequency, health insurance worse than accident; strain/sprain fraud is worse than trauma, troublemaker in a traffic accident worse than the injured, the insured fraud worse than the insurant, fraud with non-lawyer worse than that with lawyer.

Key words: fraud probability, the insured characteristics, policy information, the case circumstances, health insurance, statistical analysis, insurance amount, lawyer

[5]H.J. Smoluk. Long-Term Disability Claims Rates and the Consumption-to-Wealth Ratio. The Journal of Risk and Insurance,2009,(76):109-131.

[6]周建涛,巨珣,董楠.保险公司应对欺诈的诉讼研究[J].北京工商大学学报(社会科学版),2011,(4):117-121.

责任编辑、校对:张增强

Abstract: The insurance claim process is often accompanied by the claimant exaggerated claims. This article regresses the 258 accident and health insurance sample data of claiming litigations between 2000 and 2011 with descriptive statistics and Logistic model, finding that the 41-50 and 51-60 year-old is worse than the other year-old, and the 41-50 with the highest fraud frequency and the 51-60 with the highest fraud probability; the worker' fraud is worse than the others'; Beijing is worse than other areas, and Beijing suburb worse than Beijing city; fraud is significantly positive with insurance amount, the death insurance from illness with the largest fraud amounts, fraud is significantly negative with insurance term, casualty hospitalization insurance with the highest fraud frequency, health insurance worse than accident; strain/sprain fraud is worse than trauma, troublemaker in a traffic accident worse than the injured, the insured fraud worse than the insurant, fraud with non-lawyer worse than that with lawyer.

Key words: fraud probability, the insured characteristics, policy information, the case circumstances, health insurance, statistical analysis, insurance amount, lawyer

猜你喜欢
保险金额回归系数欺诈
关于假冒网站及欺诈行为的识别
关于假冒网站及欺诈行为的识别
警惕国际贸易欺诈
农作物保险金额确定方法比较
多元线性回归的估值漂移及其判定方法
基于“超额保险”案例的法律思考
电导法协同Logistic方程进行6种苹果砧木抗寒性的比较
多元线性模型中回归系数矩阵的可估函数和协方差阵的同时Bayes估计及优良性
网购遭欺诈 维权有种法
车辆损失险中保险标的残值的归属——兼谈《保险法》第59条的理解与适用