网络在线投资者互动平台舆情数据对公司股票未来表现的影响研究

Research on the impact of public opinion data on the online investor interaction platform on the future performance of the company's stock

李翔

11930567

硕士

数学

周倜

2021-05-18

2021-06-10

南方科技大学

深圳

互动易与e互动分别是深交所与上交所于2010年与2013年上线的在线投资者——上市公司互动平台。投资者可以在平台上对公司进行提问，公司被要求及时且准确的回复相关提问。相比于其他投资者在线互动平台，互动易与e互动有着及时性高、回复质量受监管、更新及时等优势，数据的质量较高，有一定研究价值。本文利用了来自这两个平台的提问与回复的相关数据，从投资者关注度、投资者情感以及公司回复质量三个角度，依照定性与定量的方式对平台上的提问回复的数量及内容进行分析，构建了相关的特征并使用学术界常用的Fama-MacBeth回归法、信息系数IC法、单变量投资组合排序法和时序回归检测法，并结合统计学上的各种统计量对特征的有效性进行了检验。本文还把统计人工智能算法与金融实际问题相结合，使用了多元线性回归、隐变量回归和XGBOOST回归算法来构建投资策略模型，使用单变量投资组合排序法以及更能模拟市场环境的量化平台的回测框架进行回测来测试投资策略的表现，从而从实证的角度进一步检测特征对股票未来收益的预测能力。研究结果发现，基于提问数量的投资者关注度与公司回复质量的特征都能在一定程度上影响股票的下期收益率，且不同类型的提问对股票的下期收益率也有不同的影响；特征的预测能力在每年财报的披露期要好于其他时期，这与一些传统关注的公司特征相反，特征包含一些传统因子及传统定价模型未捕获到的信息；将特征与XGBOOST回归算法结合的模型对股票的下期收益率有一定的预测能力，且可以获得相对于Fama-French三因子模型的显著为正的超额收益；模型及特征对未来表现不好的股票的预测能力要好于对未来表现较好的股票，且模型的预测结果与可能导致股票暂停交易或者大幅波动的信息相关，导致模型在更贴近现实的测试环境中表现不佳。

Hudongyi and Ehudong are online interactive platforms for listed companies launched by Shenzhen Stock Exchange and Shanghai Stock Exchange respectively in 2010 and 2013.Investors can post questions about the company on the platform, and the company is required to respond to relevant questions in a timely and accurate manner. Compared with other online interaction platforms for investors, Hudongyi and Ehudong Interactive have advantages such as high timeliness, regulated response quality and timely update, and high data quality, which is of certain research value. This paper makes use of the relevant data of questions and replies from the two platforms to analyze the quantity and content of questions and replies on the platforms in a qualitative and quantitative way from three perspectives, namely investor attention, investor sentiment and company response quality. The relevant features are constructed and the validity of the features is tested by using the regression method, IC method, univariate portfolio ranking method and time-series regression detection method commonly used in academia. Financial combined with artificial intelligence, this article also used the linear regression, hidden variables regression and XGBOOST regression algorithm to build the investment strategy, the use of portfolio sorting method and better market environment quantitative simulation platform of returning the measurement framework for measurement to test the performance of the investment strategy, thus further testing from the perspective of empirical features the ability to predict future earnings for equities. The results show that the characteristics of investor attention and company response quality based on the number of questions can affect the next return rate of stock to a certain extent, and different types of questions also have different effects on the next return rate of stock. The predictive power of the characteristics is better in the annual financial disclosure period than in other periods, which is contrary to some traditionally focused corporate characteristics. The characteristics contain some information that are captured by the traditional factors and traditional pricing model. The model combining features with XGBOOST regression algorithm can predict the stock `s future return to a certain extent, and can obtain significantly positive excess return compared with the Fama-French three-factor model. The prediction ability of the model and its characteristics for stocks with poor future performance is better than that for stocks with good future performance, and the prediction results of the model are related to the information that may cause the stock to suspend trading or fluctuate sharply, which leads to the poor performance of the model in a test environment more close to reality.

投资者关注度 互动易 e互动 特征筛选 收益率可预测性 统计机器学习

investors’ attention Hudongyi Ehudong feature selection predictability of returns statistical machine learning

中文

独立培养

南方科技大学

李翔. 网络在线投资者互动平台舆情数据对公司股票未来表现的影响研究[D]. 深圳. 南方科技大学,2021.