AI 中文总结
本文针对公私部门信贷项目的委托监督问题，构建机制设计模型分析其投资不足与过度投资扭曲，结合SBA SBIC数据与模拟验证模型，为补贴贷款设计提供依据。
AI 中文摘要
本文研究将信贷准入项目委托给私募股权和风险投资中介的公私合作伙伴关系。公共部门旨在缓解信贷配给，扩大对具有社会价值的企业的放贷；而委托监督者则负责筛选申请人、分配补贴贷款，并承担代理成本。本文构建了一个机制设计模型，表明同一委托中介结构可能同时产生斯蒂格利茨-韦斯（Stiglitz-Weiss）式投资不足和德梅扎-韦伯（De Meza-Webb）式投资过度扭曲。当筛选不完善时，利率排序可能会排除信用良好的目标企业；当补贴削弱监督和还款激励时，高风险企业可能获得过多信贷。实际发生的扭曲取决于监督曲率和补贴强度。该模型进一步预测，随着预期风险上升，贷款合约间可行利差会收窄。一个序贯扩展模型纳入了信用评分和贝叶斯更新，表明当监督资源依据借款人特征对后验风险分类的边际效应进行分配时，分类错误带来的福利损失最小。由于无法获得详细的借款人层面数据，实证部分提供了来自美国小企业局（SBA）小企业投资公司（SBIC）2018-2025年报告的州-年度描述性证据，以及一个校准模拟，说明该模型的比较静态可从经验中恢复；实证分析被作为设计验证，而非因果检验。
英文摘要
This paper studies public-private partnerships that delegate access-to-credit programs to private equity and venture-capital intermediaries. The public sector seeks to relax credit rationing and expand lending to socially valuable firms, while delegated monitors screen applicants, allocate subsidized loans, and bear agency costs. The paper develops a mechanism-design model showing that the same delegated intermediation structure can generate both Stiglitz-Weiss underinvestment and De Meza-Webb overinvestment distortions. When screening is imperfect, interest-rate sorting may exclude creditworthy target firms. When subsidies weaken monitoring and repayment incentives, high-risk firms may obtain excessive credit. The operative distortion is determined by monitoring curvature and subsidy intensity. The model further predicts that the feasible spread between loan contracts narrows as expected risk increases. A sequential extension incorporates credit scoring and Bayesian updating, showing that welfare loss from misclassification is minimized when monitoring resources are allocated according to the marginal effect of borrower characteristics on posterior risk classification. Because granular borrower-level data are unavailable, the empirical component provides descriptive state-year evidence from SBA SBIC reports (2018-2025) and a calibrated simulation illustrating that the model's comparative statics are empirically recoverable. The empirical analysis is presented as design validation rather than as a causal test.
CommentsThis paper has an attached Internet Appendix
