基本医疗保险的逆向选择与道德风险研究——基于CHARLS数据实证检验

Study on Adverse Selection and Moral Hazard of Basic Medical Insurance —An Empirical Test Based on CHARLS Data

摘要: 目的：探究城乡居民医疗保险、新型农村合作医疗、城镇居民医疗保险三类基本医疗保险的逆向选择与道德风险，为提升医疗保险的覆盖广度、优化医疗保险制度设计提供参考。方法：基于中国健康与养老追踪调查(CHARLS)数据，采用Probit模型进行因果推断，并辅以边际效应分析。进一步地，构建基于LightGBM的道德风险预测模型，以捕捉医保参保后健康行为的复杂变化模式。结果：自评健康状况对城乡居民医疗保险参保无显著影响；新型农业合作医疗存在逆向选择，即健康状况较差者参保概率更高。城镇居民医疗保险中则存在正向选择。三类险种单独检验时均未发现显著的道德风险，但三类险种合并考察显示，参保后个体体力活动显著减少，表明存在道德风险问题。此外，分析发现低收入、低教育水平的农村老年男性群体的参保决策对健康状况的变化最为敏感，是逆向选择的“高发区”。结论：优化政府补贴与个人缴费的比例结构，推进医保缴费标准统一化；强化健康知识普及与行为引导，并探索建立基于精准数据的医保精算与风险预警机制，推动医保管理从“粗放式”向“精准化”转变。

Abstract: Objective: To investigate adverse selection and moral hazard in three basic medical insurance schemes—Urban and Rural Resident Basic Medical Insurance, New Rural Cooperative Medical Scheme, and Urban Resident Basic Medical Insurance—and to provide evidence for expanding insurance coverage and optimizing the design of the medical insurance system. Methods: Utilizing data from the China Health and Retirement Longitudinal Study (CHARLS), we employed a Probit model for causal inference, supplemented by marginal effect analysis. Furthermore, a LightGBM-based prediction model was constructed to capture complex patterns of health behavior changes following insurance enrollment. Results: Self-rated health status showed no significant effect on enrollment in Urban and Rural Resident Basic Medical Insurance. The New Rural Cooperative Medical Scheme exhibited adverse selection, with individuals in poorer health being more likely to enroll. In contrast, Urban Resident Basic Medical Insurance demonstrated favorable selection. No significant moral hazard was detected when the three insurance types were examined separately; however, pooled analysis revealed a significant reduction in physical activity after enrollment, indicating the presence of moral hazard. Additionally, analysis identified that elderly, low-income, less-educated rural males were most sensitive to changes in health status when making enrollment decisions, representing a group prone to adverse selection. Conclusions: It is recommended to optimize the structure of government subsidies and individual premium contributions, promote the standardization of insurance premium criteria, and strengthen health education and behavioral guidance. Furthermore, establishing a precision data-based actuarial and risk early-warning mechanism is essential to transition medical insurance management from a “one-size-fits-all” approach to a more tailored and precise model.