2020年4月,金融学顶刊《Journal of Finance》(简称JF)刊发论文“Risk Management in Financial Institutions”,该文使用对冲利率和外汇风险的数据研究金融机构的风险管理问题。作者是来自杜克大学的Adriano A. Rampini 和S. Viswanathan ,以及法国巴黎高等商学院的Guillaume Vuillemey。
该文于2017年4月投稿,2019年7月被接收,2021年1月获得2020年JF杰出论文奖(Brattle Group Distinguished Paper Prize)。
但是,在2021年7月5日,该文章被撤稿,主要原因是该研究的核心实证结果无法复制,研究结果可靠性不足。随后,该文获得的JF杰出论文奖和相应奖金也一并被撤回。
值得说明的是,这是JF自创刊以来首次公开撤稿,也是金融学国际三大顶刊(Journal of Finance, Review of Financial Studies, Journal of Financial Economics)首次公开撤稿。撤稿论文速览
摘要:我们使用对冲利率和外汇交易风险的数据研究金融机构的风险管理问题。本文提供了强有力的证据表明,如果在横截面的不同机构之间对比,或是对同一机构在时间序列上的对比,无论风险头寸如何,当金融机构净值较高时对冲力度更强。随后,我们基于房价下跌导致贷款损失从而造成的净值冲击,来进行实证上的因果识别。结果表明,与其他类似的机构相比,承受此类净值冲击的金融机构会显著减少对冲交易。在房地产风险敞口较高的金融机构中,对冲交易减少的幅度更大。这些结果与金融约束会阻碍融资和对冲的理论是一致的。
论文核心结果难以复制
英国伦敦国王商学院的金融学者Paul Guest尝试复现这项研究时,他发现核心结果无法被复制。经反复验证后,Paul Guest发现,原论文补充信息(Supporting Information)部分中提供的复制代码不能完全重现论文发布的结果,并无法提供修订代码。
因此,Paul Guest认为原来的研究结果是不可靠的。该复现研究已经于2021年6月发表于《Journal of Finance》。
摘要:Rampini、Viswanathan 和 Vuillemey (RVV) 的实证表明,机构的净资产水平会推动对冲交易水平。我发现的一些差异性证据表明 RVV 的核心结果并不可靠:机构净值和对冲交易之间的正相关性与机构规模并不是独立的;房价下跌对净值的冲击(RVV 用于识别因果)对于对冲交易的影响是混杂的,且在不同设定下不稳健;在实验组机构中,当机构的房地产风险头寸更大时,上述冲击对机构净值和对冲的影响并没有增大,这与RVV的因果解释是不一致的。总的来说,我的分析结果不支持 RVV 的结论——更高的机构净值会导致更多的对冲交易。
3
《Journal of Finance》撤稿公告
作者在此撤回文章“Risk Management in Financial Institutions”,该文章发表在 2020 年 4 月的《Journal of Finance》上。一项复现研究发现,文章补充信息部分中提供的重复代码并不能复制出文章中报告的一些核心发现。在重新检查工作后,论文作者确认复现代码并不能完全复制已发表的结果,并且无法提供可以复现的修正代码。因此,作者得出结论,发表的结果不可靠,负责任的做法是撤回文章并退还文章获得的 Brattle Group 杰出论文奖。作者对这给期刊和学术界造成的损害深感遗憾。
三位作者对文章的具体贡献如下:
第一作者和第二作者提供了理论假设;
三位作者共同设计了实证方法和因果识别策略;
第三作者负责构建和处理数据,开展实证分析,并提供实证结果以及复现数据和代码。
第三作者表示,产生已发表结果的原始数据和代码已丢失。
第一作者和第二作者在原始数据和代码丢失时并未收到通知,事先不知道向期刊提供复现数据和代码的问题。
发布于《Journal of Finance》的撤稿公告
《Journal of Finance》编辑回复
Q: Why the retraction of this article when retractions have been virtually non-existent in finance and economics in the past? 为什么在金融和经济领域几乎没有撤稿的情况下撤回这篇文章?
A: The reason why retractions have almost never occurred in the past is unlikely due to the absence of errors in past published research. More likely, errors have not been exposed in the past because replication was not attempted or failures to reproduce or replicate results have not reached the public. The Journal of Finance introduced two policies a few years ago that encourage and facilitate replication: (i) The requirement that authors of accepted papers share program code; (ii) the introduction of a special category for “Replication and Corrigenda” papers in the Journal (the replication paper that uncovered errors in this case is forthcoming in this section of the Journal). When reproduction and replication is facilitated and taken seriously, it is to be expected that some errors will be found. If the uncovered errors are sufficiently serious, retraction is a necessary consequence. 此前几乎从未发生过撤稿的原因,不太可能是因为过去发表的研究成果没有错误。过去发表的研究没有发现问题,更有可能是因为以前没有尝试复现结果,或者结果无法复现的情况未公开。《Journal of Finance》几年前推出了两项鼓励和促进复现结果的政策:(i) 要求被接收论文的作者共享程序代码;(ii) 在期刊中加入了一个特殊类别论文(例如这次发现原论文无法复现的Paul Guest论文),即“复制和更正”。当复制和验证研究结果被认真对待时,可能会发现一些错误。如果发现的错误足够严重,撤回论文是必然的结果。
Q: How does the data and code used in the replication study differ from the “original data code” that has reportedly been lost according to the authors’ statement in the retraction notice? 复现研究中使用的数据和代码,与原作者在撤回声明中提到的已丢失的“原始数据代码”有何不同?
A: The authors of the retracted article provided replication code and data in the supplementary information section of their published article (provision of code is required by the JF’s code sharing policy). This is the data and code that was the starting point for the analysis in Guest’s replication paper. One problem Guest’s analysis uncovers is that some key results of the retracted article are not reproduced by these data and code files that the authors had provided to the JF upon acceptance (see Tables IV and V in the replication paper). The replication paper uncovers other problems as well, but the fact that the posted replication data and code do not reproduce key results reported in the article is a major one. What has reportedly been lost according to the authors’ retraction statement is some other version of data and code that the authors’ retraction statement refers to as the “original data and code that produced the published results.” This is all the editors know at this point. 撤回文章的原作者在其刊发论文的补充信息部分提供了复制代码和数据(JF 的代码共享政策要求提供代码)。这些数据和代码是为Paul Guest复现和分析原论文提供的理论依据。Paul Guest分析发现的一个问题是,原作者提供给 JF 的这些数据和代码无法复制被撤回文章的一些关键结果(参见复现论文中的表 IV 和 V)。复现论文也揭示了其他问题,但主要问题是原论文提供的复制数据和代码没有重现原文中报告的关键结果。根据原论文作者的撤回声明,丢失的是一些其他版本的数据和代码,作者的撤回声明将其称为“产生已发表结果的原始数据和代码”。这是目前编辑们知道的所有情况。
Q: Why were the authors allowed to describe in the retraction notice the division of tasks in their project? 为什么允许作者在撤稿公告中描述他们在论文中的任务分工?
A: The retraction guidelines of the Committee of Publication Ethics explicitly encourage Journals to mention the division of tasks in retracted research. At the same time, the guidelines also stress that “authorship entails some degree of joint responsibility” and that such a statement about the division of tasks does not allow authors to dissociate themselves from a retracted publication. 出版道德委员会的撤稿指南明确鼓励期刊提供撤稿研究的任务分工。同时,指南还强调“作者身份需要承担一定程度的共同责任”,任务分工声明不允许作者与被撤回的出版物脱节。
Q: Should referees have caught the errors that lead to the retraction? 审稿人是否应该发现这些导致撤回的错误?
A: They could not. Under our current system of refereeing, referees are not tasked with paper replication. Our code-sharing policy requires sharing of code only after the paper has been accepted. Hence, the code that the authors of the retracted paper posted in the supplementary information section of their article was not available to referees when they reviewed this paper. Publication of a paper does not guarantee with certainty that results are correct. The task of reproduction and replication of research results is left to the academic community. But if errors are found in this process, it is important that the scientific records gets corrected. This has been done in this case. By catching the errors in published work, the system worked as intended. 他们不能。在我们目前的审稿制度下,审稿人的审稿任务不是进行论文复制。我们的代码共享政策要求只有在论文被接收后才能共享代码。因此,被撤回论文的作者在其文章的补充信息部分发布的代码对审稿人在审阅这篇论文时是不可用的。发表论文并不能肯定地保证结果是正确的,研究成果的复制和论证任务留给了学术界。但如果在这个过程中发现错误,重要的是要进行纠正并记录。当前已经在捕捉已发表作品中的错误了。
Q: Should the Journal consider policy changes?《Journal of Finance》是否应该考虑改变政策?
A: Introduction of the Replications and Corrigenda section in the Journal and the code sharing requirement are not necessarily the end point of efforts to increase reproducibility and replicability. Further steps have been, and still are under discussion. 本期刊提供了复制和更正部分的介绍并要求作者共享,这些不一定是努力提升论文的可重复性和可复制性的终点。下一步的步骤仍在讨论中。
Reference:
1. Rampini, A. A., Viswanathan, S., & Vuillemey, G. (2020). Risk management in financial institutions. The Journal of Finance, 75(2), 591-637.2. Paul M. Guest (2021). Risk Management in Financial Institutions: A Replication. The Journal of Finance.3. Retracted: Risk Management in Financial Institutionshttps://onlinelibrary.wiley.com/doi/10.1111/jofi.130644. Answers to FAQ about the recent retraction of an article in the JFhttps://afajof.org/2021/07/answers-to-faq-about-the-recent-retraction-of-an-article-in-the-jf/