算法黑箱:网络传播误导性旧信息操纵市场的研究  

The Black Box Algorithm:A Study on Market Manipulation by Spreading Misleading Outdated Information Online

作　　者：王春[1] WANG Chun(Nanjing Audit University,211815)

机构地区：[1]南京审计大学金融学院,211815

出　　处：《财贸经济》2021年第11期85-100,共16页Finance & Trade Economics

基　　金：江苏省社会科学基金项目“市场操纵视角下网络传播误导性信息识别及风险防范研究”(19EYD004);江苏高校优势学科建设工程三期项目“南京审计大学应用经济学”(苏政办发[2018]87号)。

摘　　要：智能互联网时代,新闻以机器学习为基础生产和推荐,公众难以了解其中间环节,从而形成算法黑箱。算法黑箱以看似技术中立输出的与客观事实不符的信息,误导投资者决策,破坏市场定价功能,易引发系统性风险。本文以财经新闻文本相似性为网络传播误导性旧信息的代理变量,采用向量空间模型和词频-逆向文档频率法进行度量。基于数行者(深圳)科技有限公司开发的报刊新闻量化舆情数据平台中1046620条样本新闻数据,发现网络传播误导性旧信息会操纵资本市场,表现为:网络传播误导性旧信息发布一周后,公司股票超额收益反转,且财经新闻文本相似性越高,收益反转越明显;网络传播误导性旧信息还操纵了公司股票交易活动,即降低了公司股票收益波动性和超额交易量。将样本公司按机构投资者持股比例分组后,发现机构投资者持股比例较低的公司受网络传播误导性旧信息操纵的影响更大。网络传播误导性旧信息操纵市场的机制在于个人投资者行为偏见,即个人投资者混淆了财经新闻中包含的新信息与旧信息,并且按照旧信息进行交易。本文为互联网平台算法生产的财经新闻纳入监管提供了政策依据。In the era of intelligent Internet,news is produced and recommended based on machine learning,and the general public does not understand its inner workings,hence the term“black box algorithm.”The black box algorithm,though seemingly technology-neutral,produces information that is inconsistent with facts,misleading for investment-decision making,destroys the market pricing function,and easily leads to systemic risks.In this paper,the text similarity of financial news is used as the proxy variable of misleading information spread on the Internet,and the vector space model and the TF-IDF(term frequency and inverse document frequency)method are used to measure the misleading information.Based on 1,046,620 sample entries of news data in the quantified public opinion platform of newspapers and periodicals developed by Datago Technology Co.,Ltd.,it is found that misleading information spread on the Internet will manipulate the capital market.The results are as follows.One week after the release of misleading information,the excess return of stocks reversed,and the greater the text similarity,the more obvious the return reversal was.The misleading information also reduced the stock trading activity,i.e.the volatility of the stock earnings and excess trading volume.After grouping the sample companies by the proportion of institutional investors,it is found that the companies with a lower proportion of institutional investors are more vulnerable to the manipulation of misleading information.The misleading information manipulates the market by taking advantage of the behavior bias of individual investors who could not tell outdated information from new information in the financial news and traded based on the outdated information.This study provides a policy basis for the regulation of algorithm-generated financial news produced by Internet platforms.

关 键 词：网络信息 文本分析法 财经新闻 市场操纵 

分 类 号：F830.9[经济管理—金融学]