《保险研究》20250804-《LightGBM模型在提升冬小麦农业气象指数保险赔付精准度上的应用》（张连增、李浩男、罗来娟）

［摘   要］农业气象指数保险作为一种应对气候变化和气象风险的有效工具，凭借其理赔快捷、管理成本低等诸多优势，正逐渐被广泛应用于农业生产中。然而，在农险实务中，由于受到多种因素的影响，农业气象指数保险的赔付精准度仍然较低，存在较大的基差风险。针对这一问题，本文将机器学习和可解释性方法引入到冬小麦农业气象指数保险的定价模型中，以山东省、河北省和河南省的县级数据为例，分别使用分位数回归模型和LightGBM（Light Gradient Boosting Machine）模型研究了气象指标与冬小麦产量之间的关系。结果显示：LightGBM模型在预测相对气象产量方面具有显著优势，有效降低了冬小麦农业气象指数保险的基差风险；不同生育期影响冬小麦产量的气象指标存在差异，其中气温、降水和光照是更加重要的气象因子；利用单一气象指标预测冬小麦产量损失情况容易产生偏差，综合考虑冬小麦不同生育期的多个气象指标能够有效降低保险产品的设计风险。本文的研究结论完善了冬小麦农业气象指数保险的定价机制，能够切实降低基差风险，为提升赔付精准度提供了重要的理论基础和技术支持，有助于农业保险的高质量发展。

［关键词］农业气象指数保险；基差风险；分位数回归；LightGBM；可解释性

［作者简介］张连增，南开大学金融学院教授、博士生导师；李浩男（通讯作者），南开大学金融学院博士研究生；罗来娟，南开大学金融学院博士研究生。

Enhancing Compensation Accuracy in Winter Wheat Weather Index Insurance through the Application of the LightGBM Model

ZHANG Lian-zeng，LI Hao-nan，LUO Lai-juan

Abstract：As an effective tool to deal with climate change and meteorological risks，weather index insurance is gradually being widely used in agricultural production due to its many advantages such as faster claims settlement and lower management costs. However，in the practice of agricultural insurance，due to the influence of various factors，the compensation accuracy of weather index insurance is still low，and there exists a large basis risk. To address this problem，this paper introduced machine learning and interpretability methods into the pricing model of weather index insurance for winter wheat. With county-level data from Shandong，Hebei，and Henan Provinces，the quantile regression model and the Light Gradient Boosting Machine （LightGBM） model were used respectively to study the regression relationship between meteorological indicators and winter wheat yield. The results show that：firstly，the LightGBM model has significant advantages in predicting relative meteorological yield，effectively reducing the basis risk of weather index insurance for winter wheat.Moreover，different meteorological indicators affect winter wheat yield in different growth periods，among which temperature，precipitation and light have more important effects.Lastly，using a single meteorological indicator to predict winter wheat yield is prone to misjudgment，and comprehensive consideration of multiple relevant meteorological indicators tailored to the distinct growth stages of winter wheat can mitigate the design risk of insurance products. The research findings of this article can enhance the pricing mechanism of weather index insurance for winter wheat. This improvement can effectively mitigate basis risk，provide important theoretical basis and technical support for enhancing compensation accuracy，and contribute significantly to the high-quality development of agricultural insurance.

Key words：weather index insurance;basis risk;Quantile Regression;LightGBM;interpretability