主管:中国科学院
主办:中国优选法统筹法与经济数学研究会
   中国科学院科技战略咨询研究院

中国管理科学 ›› 2020, Vol. 28 ›› Issue (3): 191-200.doi: 10.16381/j.cnki.issn1003-207x.2020.03.020

• 论文 • 上一篇    下一篇

个体投资者情绪与股票价格行为的互动关系研究

黄创霞1, 温石刚1, 杨鑫1, 文凤华2, 杨晓光3   

  1. 1. 长沙理工大学数学与统计学院, 湖南 长沙 410114;
    2. 中南大学商学院, 湖南 长沙 410081;
    3. 中国科学院数学与系统科学研究院管理、决策与信息系统重点实验室, 北京 100190
  • 收稿日期:2018-05-25 修回日期:2019-06-24 出版日期:2020-03-20 发布日期:2020-04-08
  • 通讯作者: 杨鑫(1988-),男(汉族),湖南桃源人,长沙理工大学数学与统计学院,讲师,研究方向:金融风险管理,E-mail:yangxintaoyuan@163.com. E-mail:yangxintaoyuan@163.com
  • 基金资助:
    国家自然科学基金资助项目(71471020,71873146,71850008);湖南省自科基金资助项目(2016JJ1001,2019JJ50650);湖南省教育厅重点项目(15A003,18C0221);湖南省研究生科研创新项目(CX2018B571)

The Interactive Relationship between Individual Investor Sentiment and Stock Price Behaviors

HUANG Chuang-xia1, WEN Shi-gang1, YANG Xin1, WEN Feng-hua2, YANG Xiao-guang3   

  1. 1. School of Mathematics and Statistics, Changsha University of Science and Technology, Changsha 410114, China;
    2. Business School, Central South University, Changsha 410081, China;
    3. Academy of Mathematics and Systems Science, Chinese Academy of Science, Beijing 100190, China
  • Received:2018-05-25 Revised:2019-06-24 Online:2020-03-20 Published:2020-04-08

摘要: 本文运用情感分析技术,在情感倾向点互信息(SO-PMI)算法的基础上,引入"拉普拉斯修正"和"情绪分类阈值",提出了一种改进的个体投资者情绪度量的SO-LNPMI算法;基于上证指数股吧的31万条论坛信息,运用格兰杰因果检验方法研究了个体投资者情绪与市场收益率和成交量的互动关系。研究表明:(1)与经典的SO-PMI算法相比,本文提出的SO-LNPMI算法的情感识别精度更高;(2)积极情绪是股票收益率的格兰杰原因,消极情绪对其影响不显著;(3)投资者情绪与成交量存在双向的格兰杰因果关系;(4)当投资者处于积极状态时,会热衷于使用表情符号表达情绪。本文的研究为投资者情绪度量提供了一种新的有效算法,有助于投资者更好的利用网络论坛信息进行投资决策。

关键词: 投资者情绪, SO-LNPMI算法, 格兰杰因果检验

Abstract: The measure of individual investor sentiment in Chinese stock message board Guba Eastmony and its interactive relation to the market returns and trading volume is investigated. In order to measure sentiment, the construction of sentiment lexicon is a key procedure. Traditional methods for lexicon acquisition are commonly based on Semantic Orientation from Pointwise Mutual Information(SO-PMI) algorithm. A novel algorithm Semantic Orientation from Laplace Smoothed Normalized Pointwise Mutual Information(SO-LNPMI) is proposed, which has the higher accuracy for sentiment classification. Empirical analyses on the interactive relationship between individual investor sentiment and market returns and trading volume show that: (i) positive sentiment is the cause of market return while passive sentiment does not cause it; (ii) investor sentiment and trading volume present two-side Granger causality. In addition, an interesting phenomenon is that individual investors are enthusiastic about the use of emoticons when individual investors are positive.

Key words: investor sentiment, SO-LNPMI algorithm, Granger causality analysis

中图分类号: