《保险研究》20250703-《消费贷款保证保险对消费者还款影响研究——基于大语言模型处理裁判文书数据集的实证》（许荣、刘灿阳）

［摘   要］本文基于2005~2023年裁判文书网99136篇消费贷款保证保险判决书，运用大语言模型（GPT-4o-mini）与自然语言处理技术（NLP）抽取合同信息，构建借款人数据集，实证检验贷款综合费率（贷款利率+保费率）对还款行为的影响。结果显示，综合费率与还款比例呈显著倒U型：适度上调费率可提升还款比例，过高或过低反而抑制还款。机制分析发现，综合费率主要通过调节主观还款意愿发挥作用，而非单纯削弱客观还款能力。异质性分析显示：倒U型效应在中等贷款金额区间最明显；消费者在2020年一季度经济下行周期对综合费率更敏感；高风控银行或保险总公司适度提高综合费率能促进还款，但两者皆强时促进作用被部分抵消。本研究拓宽了消费贷款保证保险的实证边界，为银保审慎定价提供借鉴，研究方法上展示了基于大语言模型的文本挖掘在保险、法律文书和经济学交叉研究领域的价值。

［关键词］贷款保证保险；综合费率；大语言模型

［基金项目］本文为中国保险学会2025年度课题成果（ISCKT2025-N-1-10）。

［作者简介］许荣，中国人民大学财政金融学院、中国财政金融政策研究中心、中国人民大学保险研究所教授、博士生导师；刘灿阳（通讯作者），中国人民大学财政金融学院博士研究生。

The Impact of Consumer Loan Guarantee Insurance on Borrower Repayment：An Empirical Study Using a Large Language Model to Analyze Court Judgments

XU Rong，LIU Can-yang

Abstract：By leveraging a large language model（GPT-4o-mini） and natural language processing（NLP） techniques，we extract the contract information，establish the borrower dataset，and conduct an analysis of 99，136 court judgments disclosed on China Judgments Online from 2005 to 2023 in order to empirically examine the effect of the "combined rate"（loan interest rate plus insurance premium rate） on consumers’ repayment behaviors.The results reveal a significant inverted U-shaped relationship between the combined rate and the repayment ratio.A moderate increase in the combined rate enhances the repayment ratio，whereas excessively high or low rates markedly reduce it.The mechanism analysis suggests that the combined rate influences borrowers’ willingness to repay rather than merely exerting pressure on their objective repayment capacity.Heterogeneity tests show that：the inverted U-shaped effect of the combined rate on the repayment ratio is most pronounced when the loan amount is in the medium range;during the economic downturn of the first quarter of 2020，consumers were more sensitive to the combined rate;banks and insurance companies with relatively stronger risk control capabilities or market positions can promote higher repayment rates by moderately raising either the loan interest rate or the premium rate.However，when both are strong，this enhancing effect is partially offset.By employing the large-scale text’s data mining value in the interdisciplinary research field among insurance and legal documents and economics，this study not only expands the research boundaries of consumer loan guarantee insurance but also provides useful references for prudent pricing by insurance companies and banks.At the same time，it offers policy references for controlling credit risks，enhancing consumer confidence，and facilitating the formation of a safe and sustainable credit consumption environment.

Key words：loan guarantee insurance;combined rate;large language model