农业产量保险费率的联合分布估计:基于高斯过程回归方法  

Probabilistic Estimation of Crop Insurance Rates Based on Gaussian Process Regression

作　　者：文旷宇 郑始潮 王耀璟 Wen Kuangyu;Zheng Shichao;Wang Yaojing

机构地区：[1]华中科技大学经济学院 [2]北京大学经济学院

出　　处：《统计研究》2024年第11期142-151,共10页Statistical Research

基　　金：湖北省自然科学基金面上项目“空间面板数据联合概率密度函数的非参数估计”(2023AFB1113);中央高校基本科研业务费专项资金项目:基于期权的远期概率分布及其在识别和管理我国经济风险上的应用(HUST:2022WKFZZX023);北京大学经济学院种子基金资助。

摘　　要：科学精准地厘定政策性农业保险费率有利于促进我国农业保险行业高质量发展。与现有文献侧重保险费率的点估计不同,本文构建了全新计量方法估计各地费率的联合分布,以便更灵活地在不同损失函数下确定最优费率。为克服单一县产量数据稀少的问题,本文汇集多县产量数据,基于地理坐标空间平滑各县数据以提高估计效率,使用高斯过程回归平滑各县经验费率,并得出各地费率估计的联合分布。与现有的经验费率法、贝叶斯模型平均法以及密度比法相比,蒙特卡洛模拟实验表明新方法具有更高的统计精准度。将新方法用于厘定吉林2022玉米保险费率和河南2023年棉花保险费率,所得出的费率在空间上平滑变化特征符合农户“相邻县费率类似”的心理预期,具备一定应用价值。Accurately deciding crop insurance rates is crucial for fostering high-quality agricultural development in China.Unlike previous studies,which primarily focus on point estimates of insurance rates,this paper introduces a novel econometric approach to estimate the joint distribution of rates across different locations.This distributional estimation is useful for designing optimal rates tailored to various loss functions.To address the challenge of limited yield data at the county level,we pool data from multiple counties and apply spatial smoothing based on geographic coordinates,enhancing the overall estimation efficiency.Specifically,we use Gaussian process regression to smooth empirical rates of each county,providing a comprehensive distributional estimate of insurance rates.Monte Carlo simulations demonstrate that this method achieves greater statistical accuracy than the existing approaches,such as empirical rates,Bayesian model averaging and density ratio methods.We apply the proposed method to corn data from Jilin Province and cotton data from Henan Province,producing results that are spatially smooth and align with observed regional rate patterns.This method shows significant practical value for setting more precise and regionally consistentcropinsurancerates.

关 键 词：农业保险 保险费率 高斯过程 空间平滑 

分 类 号：F224[经济管理—国民经济] F842.6