Research On Investor Information Learning And Its Effects On Stock Price Synchronicity

Based on the strategic height of national financial security and adapting to the development of Web 2.0 Internet technology and social media,this paper investigates the information learning mechanism of investors and its effects on asset pricing efficiency(i.e.stock price synchronicity)from the perspective of online social networks.This is the urgent requirement for related academic research that Xi Jinping emphasizes on preventing and defusing financial risks,as well as maintaining financial security with“China's overall security concept”.Around the core questions of“how about the information environment in which investors learn information”,“how do investors learn information via online social networks”,and“what are the economic consequences of investor information learning”,this paper forms a hierarchical and logical research framework with three core components:the research on social user role recognition,the research on information learning mechanism,and the research on the impact of information learning on stock price synchronicity.The Introduction introduces the background and the significance of our research,and reviews the existing studies at home and abroad.On this basis,we put forward our research contents and research ideas.Chapter I describes the theoretical methods and key technologies used in this study,including clustering methods for social role recognition,especially K-Means clustering algorithm,and evolutionary game theory for modeling the process of investor information learning,especially replicator dynamic equations on both regular networks and complex networks.Chapter II investigates the impacts of textual quality on sentiment classification performance of deep learning algorithms.IMDb is used for experimental dataset,length and readability are chosen to measure the quality of texts,and three basic but representative deep learning algorithms,i.e.SRN,LSTM,and CNN,are used to conduct sentiment experiments;independent t tests and multiple linear regressions are also used to conduct statistical analyses to capture the relation between textual quality and sentiment classification accuracy.In this chapter,we try to seek the ways to improve the performance of sentiment analysis from the perspective of textual quality,and the corresponding findings can be helpful for data screening and classification model selection in Chapter V.Chapter I and Chapter II serve as the research basis of this paper,which lays a good foundation for the development of the main body contents.Chapter III,Chapter IV,and Chapter V are the main body contents of this paper.Chapter III proposes a method for social role recognition based on network topology features.We take users'contribution to network connectivity as the standards of role division,and use betweenness centrality and closing centrality to measure global connectivity of network,and use degree centrality and clustering coefficient to measure local connectivity of network.K-Means clustering algorithm is employed to divide users into core users with high global connectivity and high local connectivity,intermediary users with high global connectivity and low local connectivity,active users with low global connectivity and high local connectivity,and general users with low global connectivity and low local connectivity.Then we propose a method to evaluate network information environment based on various edge types and the rating of each edge type with respect to information spreading size,peak,and speed.To verify the validity of the proposed methods,we conduct experiments on three online social networks,i.e.Xueqiu,Guba,and Zhihu.Chapter IV establishes a dynamic model of investor information learning based on evolutionary game theory(EGT),in which network information environment and self-enhancing transmission bias(SET)are taken into consideration to better describe the dynamic evolutionary mechanism of investor information learning.Firstly,the investor group participating in information learning is mapped to the population participating in the evolutionary game,and three learning strategies are defined based on EGT,that is,S_T strategy for learning true information,S_F strategy for learning false information,and S_N strategy for learning no information.Secondly,payoff matrix of information learning game is defined by considering social earnings,investment income,the cost of information acquisition and interpretation;combined with research findings of network information environment in Chapter III,and considering the increase in the probability of investors releasing information and disseminating their investment opinions,arising from SET,information environment rating and SET are introduced as two key factors that affect the cost of information acquisition.Then we derive the replicator dynamic equations of strategy games,and analyze the condition that S_T is evolutionary stable strategy(ESS).Finally,we conduct several simulation experiments to verify the validity of the model.Based on this,we use the proposed model to analyze the information learning mechanism in different scenarios,such as epidemic situation and non-epidemic situation,different online social networks,bull market situation and bear market situation,and different levels of SET.Chapter V takes peer engagement in online social networks as the macro performance of investor information learning,and examines the economic consequences of investor information learning by investigating the impact of peer engagement on stock price synchronicity.Based on the theory of Wisdom of crowds(Wo C),we use informativeness,information diffusion degree,diversity,and expert proportion to measure the performance of investor peer groups;we develop the research hypothesis that stock price synchronicity would be lower for firms with the peer group that exhibits higher level of wisdom.In particular,for the measurement of peer engagement informativeness,besides a simple index of monthly average number of posts,we construct a more refined index of consistency of posts by applying deep learning algorithms to conduct sentiment analysis on posts.Based on the research hypothesis,we establish the correspondingly theoretical econometric models,and use China's A-share listed firms and more than 30 million social textual data scraped from Guba as samples to conduct empirical tests.A series of robustness tests are also conducted to verify the reliability of our empirical results.A series of creative achievements have been made in this paper through the above research.Some of them contribute to practical policy suggestions for alleviating the high stock price synchronicity,improving the market efficiency,and firmly holding the line that no systemic financial risks occur in the current post-epidemic period.For the question“how about the information environment in which investors learn information”,we find that Xueqiu is an online social network dominated by general users;its basic information diffusion mode is mainly general-general,supplemented by intermediary-general and core-general;its information environment rating is low(1.43).Although Xueqiu is considered to be a gathering place for high-quality investors,there is little interaction between investor peers,so high-quality information cannot be fully disseminated,thus resulting in poor information environment.Guba is an online social network dominated by active users;its basic information diffusion mode is mainly active-active,supplemented by intermediary-active,core-active,and general-general;its information environment rating is higher than Xueqiu(2.19).Guba has always been considered as a place for investors to relieve feelings.However,due to a large number of local active users,even a small amount of high-quality information can be fully disseminated in at least a local scope,thus resulting in better information environment.Zhihu is an online social network dominated by intermediary users;its basic information diffusion mode is dominated by intermediary-intermediary;its information environment rating is the highest(2.95),which benefits from high-quality users and high degree of user participation and activity on the platform.For the question“how do investors learn information via online social networks”,we find that S_T strategy is ESS when the parameters meet certain conditions.Investors group can learn true information and exhibit the wisdom of crowds.We also find that online social networks creating good information environment and generously providing social rewards to investors can accelerate the speed of investors learning true information;especially in epidemic situation,the evolutionary consequences of investor information learning are more sensitive to information environment and social rewards.Hence,relevant departments should strengthen a guidance for establishment and improvement of management system of online social networks,encourage managers of online social networks to make a regulation for rewarding social users who actively participate in information activities and produce high-quality contents,and promote the optimization of platform information environment,thus combatting the adverse effects of epidemic outbreaks on investor information learning.For the question“what are the economic consequences of investor information learning”,we find that informativeness and diversity of peer engagement are negatively related with stock price synchronicity,and the degree of information diffusion and the proportion of experts strengthen the negative effects of them on stock price synchronicity,respectively.Therefore,China's financial regulatory authorities should strengthen effective guidance and regulation towards investor peer activities in online social networks,such as encouraging investors to publish more posts and comments,especially high-quality posts and comments that could provide an accurate analysis of fundamental values;encouraging more investors with different levels of expertise to participate in social activities,and so on,so as to alleviate high stock price synchronicity,improve operational efficiency of capital market.We also find that the outbreak of the epidemic leads to higher stock price synchronicity,and the epidemic weakens the lowering effects of peer engagement informativeness and information diffusion degree on stock price synchronicity,but strengthens the lowering effects of group diversity and expert proportion on stock price synchronicity.Therefore,during the epidemic period,China's financial regulatory authorities should pay full attention to the roles of group diversity and experts.Encouraging the diversity of peer engagement in online social networks,and giving play to the advantages and guidance effects of experts might be the possible measures to deal with higher stock price synchronicity in the epidemic situation.