编者按

在本月 MS 文章精选中，我们梳理了顶刊 Management Science 三月份发布有关OR/OM/FE以及相关应用的文章基本信息，旨在帮助读者快速洞察行业/学界最新动态。本文为第二部分（2/2）。

推荐文章1

● 题目：How Much Can Machines Learn Finance from Chinese Text Data?

机器能在多大程度上从中文文本数据中学习金融?

● 原文链接：https://pubsonline.informs.org/doi/abs/10.1287/mnsc.2022.01468

● 作者：Yang Zhou, Jianqing Fan, Lirong Xue

● 发布时间：2024.3.18

● 摘要：

This paper introduces a novel framework named FarmPredict, which aims to learn financial returns directly from Chinese text data through unsupervised learning and sparsity regularization. The FarmPredict model is capable of extracting information directly from articles without relying on predefined dictionaries or pre-trained models. The model converts articles into vectors of hidden components consisting of multiple factors and idiosyncratic residuals via Principal Component Analysis (PCA), and determines the number of hidden factors using the Adjusted Eigenvalue Thresholding method. FarmPredict also screens idiosyncratic variables correlated with the target stock returns, conditional on factors, to reduce dimensionality to a more manageable level. Finally, the model employs the LASSO method or other machine learning algorithms to predict asset prices using hidden factors and screened idiosyncratic components. The study demonstrates that FarmPredict performs well in forecasting stock market returns in China, with sentiment scores from news articles significantly predicting stock excess returns. Moreover, compared to other statistical and machine learning methods available in the market, FarmPredict significantly outperforms them in terms of model prediction and portfolio performance.

本文提出了一种名为FarmPredict的新框架，旨在通过无监督学习和稀疏正则化从中文文本数据中直接学习金融上的回报。FarmPredict模型能够直接从文章中提取信息，无需依赖预定义的词典或预训练模型。该模型通过主成分分析（PCA）将文章转换为包含多个因素和特异残差的隐藏成分向量，并通过调整特征值阈值法确定隐藏因素的数量。FarmPredict还通过条件相关性筛选来筛选与目标股票回报相关的特异变量，从而减少维度至更易管理的水平。最后，模型使用LASSO方法或其他机器学习算法，结合隐藏因素和筛选后的特异组成部分来预测资产价格。研究结果表明，FarmPredict在预测中国股市回报方面表现出色，其基于新闻的情感得分能够显著预测股票的超额回报。此外，与市场上其他统计和机器学习方法相比，FarmPredict在模型预测和投资组合表现上都有显著优势。

推荐文章2

● 题目：Designing Sparse Graphs for Stochastic Matching with an Application to Middle-Mile Transportation Management

随机匹配稀疏图设计及其在中程交通管理中的应用

● 原文链接：https://pubsonline.informs.org/doi/abs/10.1287/mnsc.2022.01588

● 作者：Yifan Feng, René Caldentey, Linwei Xin, Yuan Zhong, Bing Wang, Haoyuan Hu

● 发布时间：2024.3.19

● 摘要：

This paper addresses the problem of designing a sparse subgraph that supports a large matching when nodes are randomly removed. The authors consider four families of sparse graph designs (Clusters, Rings, Chains, and Erdős-Rényi graphs) and demonstrate both theoretically and numerically that their performance is close to the optimal one achieved by a complete graph. The motivation for studying the stochastic sparse graph design problem primarily stems from a collaboration with a leading e-commerce retailer in the context of its middle-mile delivery operations. By testing the theoretical results with real data from the industry partner, the authors conclude that adding a bit of flexibility to the routing network can significantly reduce transportation costs.

本文研究了在随机删除节点的情况下，如何设计一个稀疏子图以支持大规模匹配的问题。作者考虑了四种类型的稀疏图设计（集群、环、链和Erdős-Rényi图），并从理论和数值上证明了它们的性能接近于完整图所能达到的最优性能。研究的动机来自于与一家领先的电子商务零售商合作，旨在优化其中间英里的配送运营。通过使用行业合作伙伴的真实数据进行测试，作者得出结论，增加路由网络的灵活性可以显著降低运输成本。

推荐文章3

● 题目：The Impact of Exchange-Traded Fund Index Inclusion on Stock Prices

交易所交易基金指数纳入对股票价格的影响

● 原文链接：https://pubsonline.informs.org/doi/abs/10.1287/mnsc.2022.02125

● 作者：John Duffy, Daniel Friedman, Jean Paul Rabanal, Olga A. Rud

● 发布时间：2024.3.20

● 摘要：

This paper reports on a laboratory experiment examining the impact of ETF index products on the prices and trading volume of assets. It compares an environment where an ETF index includes all assets in a market with an environment where a redundant asset is excluded from the index. The findings are that (i) participants place significant value on the ETF index asset beyond the value of its constituent assets; (ii) there is a substantial index premium for included assets; and (iii) the index premium persists even when short-selling is permitted. These results suggest that ETF products can distort markets to some degree. The paper also explores potential mechanisms behind the ETF index premium and rules out other possible explanations, such as signaling effects and increased liquidity of included assets, through the laboratory setting.

本文通过实验室实验研究了ETF指数产品对资产价格和交易量的影响。研究比较了包含市场中所有资产的ETF指数环境与排除冗余资产的指数环境。

实验发现：

(i) 参与者对ETF指数资产的重视程度超过了其组成资产的价值；

(ii) 被纳入指数的资产存在显著的指数溢价；

(iii) 即使允许卖空，指数溢价也持续存在。这些发现表明ETF产品在一定程度上可以扭曲市场。

文章还探讨了ETF指数溢价的潜在机制，并通过实验室环境排除了信号效应和流动性增加等其他可能的解释。

推荐文章4

● 题目：The Financial Consequences of Online Review Aggregators: Evidence from Yelp Ratings and SBA Loans

在线评论聚合器的财务价值:来自Yelp评级和SBA贷款的证据

● 原文链接：https://pubsonline.informs.org/doi/abs/10.1287/mnsc.2020.03003

● 作者：Ruidi Huang

● 发布时间：2024.3.25

● 摘要：

This paper investigates the financial and real consequences of online review aggregators on small business loans. Utilizing a regression discontinuity design to address the potential endogeneity between Yelp reviews and Small Business Administration (SBA) loan outcomes, the study demonstrates that higher Yelp ratings lead to improved loan terms and performance. Specifically, a half-star increase in Yelp ratings corresponds to a 25-basis-point decrease in loan spread and a 6% reduction in collateral requirements. These effects are more pronounced when banks have less information about borrowers. Additionally, higher Yelp ratings contribute to increased consumer demand and the likelihood of future business openings. The findings indicate that online review aggregators influence both consumer choices and banks' financing decisions.

本文研究了在线评价聚合器对小企业贷款的财务和实际价值。利用回归不连续设计，研究者克服了Yelp评价与美国小企业管理局（SBA）贷款结果之间潜在的内生性问题。研究发现，Yelp评级的提高可以改善贷款条件和表现。具体来说，Yelp评级半星的提高对应于贷款利差下降25个基点，以及抵押要求降低6%。当银行对借款人信息了解较少时，这些效应更为显著。此外，较高的Yelp评级还有助于增加消费者需求和未来商业开张的可能性。这些发现表明，在线评价聚合器不仅影响消费者的选择，也影响银行的融资决策。

推荐文章5

● 题目：On Statistical Discrimination as a Failure of Social Learning: A Multiarmed Bandit Approach

统计歧视作为社会学习的失败:一种多臂老虎机方法

● 原文链接 ：https://pubsonline.informs.org/doi/abs/10.1287/mnsc.2022.00893

● 作者：Junpei Komiyama, Shunya Noda

● 发布时间：2024.3.29

● 摘要：

This paper analyzes statistical discrimination in hiring markets using a multi-armed bandit model. The authors assume that all firms (decision-makers) are fully rational and non-prejudiced, and all workers are ex ante symmetric. In such an environment, an unbiased decision policy—hiring workers with superior skills—satisfies numerous fairness notions articulated in scholarly literature, including equalized odds and demographic parity. However, the long-term persistence of biased beliefs could still occur. The paper emphasizes that temporary affirmative actions can effectively enhance both welfare and equality. Two policy solutions are proposed: a novel subsidy rule (the hybrid mechanism) and the Rooney Rule. The results indicate that temporary affirmative actions effectively alleviate discrimination stemming from insufficient data.

本文通过多臂老虎机模型分析了招聘市场中的统计歧视问题。作者假设所有公司（决策者）都是完全理性且无偏见的，而所有工人在事先是对称的。在这样的环境下，无偏见的决策政策才能够雇佣技能更高的工人，进而满足学术文献中提出的多种公平观念，包括等概率和人口统计奇偶性。然而，即使在这种情况下，偏见信念的长期存在仍然可能发生。文章强调，临时的平权行动可以有效地提高福利和平等。文章提出了两种政策解决方案：一种新的补贴规则（混合机制）和Rooney规则。研究结果表明，临时的平权行动可以有效缓解因数据不足而产生的歧视。