Financial Media Information And Investors' Over-trading

With the rapid development of Internet and new media,online financial media information has been greatly leveraging investors' decision making,but also aggravating their limited rationality.The online dissemination of financial media information affects the allocation of investor attention,changes investors' cognition regarding financial assets.The bullish and bearish signals conveyed in media information further drive investors' expectation of the future return of financial assets,leading to investors' over-trading of the attention-and sentiment-salient assets,and the deviation and biases of asset prices from rational prices in financial markets.Due to the lack of technical processing,in order to understand the mechanisms of the limited attention and biased sentimental beliefs of investors and their impact on financial asset prices,the majority of prior studies in behavioral finance resorts to simple counting and calculation of sentimental words from sentiment dictionaries or traditional classification models,leading to a shallow portray of investor sentiment from media information.Meanwhile,research in information sciences has not shed much light on financial asset pricing even in recent years.In response to the above problems,this study seeks to unravel the impact of financial media information on investors over-trading behavior and financial market asset prices in nowadays Internet and new media environments.The study attempts to combine the knowledge from information sciences and finance,on the basis of a systematic literature review on the theories of investor bounded rationality(particularly,in terms of investor attention and sentiment),overtrading and media effects,as well as the methods regarding investor sentiment measurement in finance and textual sentimental analysis,deep learning,etc.,in information sciences.As two emphases of this research,technical/method exploration and empirical investigation are conducted in this study by following the logic path from financial media information,investors' sentimental beliefs of asset future returns,investors' limited attention to those information-or sentiment-salient assets,investor overtrading,to asset mispricing.(1)With regard to the technical approach,the research focuses on obtaining and designing the method of fine-grained and target-dependent sentiment analysis of financial media information.The word embedding space trained by basic Word2 Vec has inherently limited ability in effectively distinguishing and tracking different asset entities/targets,and effectively identifying the valance of different sentimental words.In order to solve this issue,this research proposes a model of task-driven Word2 Vec.The model takes manfully constructed keyword dictionaries of both stock entities and sentimental words as the input of extra domain-driven supervising information to a joint training of Word2 Vec and the word embedding of stock entities and sentimental words.Vectors modelled in this way are fed to further sentiment analysis.Then,in light of aspect-dependent,a model of target-dependent sentiment classification is developed on the basis of dynamic entity tracking chain,which modifies the Long Short-Term Memory model(LSTM)by incorporating the recursive entity network(Ent Net).A series of experiments are conducted to compare the performance between the proposed target-driven Word2 Vec and the basic Word2 Vec,and between the proposed target-dependent sentiment classification method and the existing deep learning methods(e.g.,CNN and LSTM).Parameters for the proposed sentiment classification model are optimized by further experiments.The experiments demonstrate that the proposed task-driven joint word vector modelling is better than basic Word2 Vec,and the proposed target-dependent sentiment classification overwhelms the basic deep learning methods such as CNN or LSTM,in tracking the asset targets and capturing the sentiments specific to different assets.(2)On the empirical level,the research focuses on unravelling the impacts of financial media information on investor behavior and asset prices in China A-share market.Financial media information is automatically collected from Internet and individual stock asset prices are collected from financial databases.Indicator of individual stock sentiment is extracted from the above-mentioned target-dependent sentiment classification,which is further adjusted by a multiplier of social interaction as the explanatory variable of the empirical modelling.Grounded on the insights from behavioral finance,the changes of asset prices and returns,the instant over-trading indicator(measured by abnormal turnover rate)as well as the continuing over-reaction indicator are established as the explained variable.Autoregressive distributed lag models are established in terms of the hypotheses of the relationships between the explanatory and explained variables,which is further tested by Grange causality testing,correlation analysis and regressive analysis.The empirical results lead to the following conclusions:(i)The price changes,the rates of yield and the excess benchmark rates of return of individual stocks are positively related to the sentiments conveyed in or induced by financial media information.In particular,the negative sentiments show short-term effects,and in a longer term there is a reversal effect in the market.(ii)The instant over-trading behavior of investors(abnormal turnover rate of individual stocks)is positively related to the sentiments conveyed in or induced by financial media information.(iii)Compared to good information,the sentiment in bad information is more greatly spread and exaggerated.Investors are more vulnerable to the impact of negative information and continually overreact to negative information.(iv)The impact of investors' sentiment learned from financial media information on their trading behavior and asset prices contained is regulated by investors' limited attention to information,as evidenced by the increased correlation between investors' over-trading and the sentiment indicator adjusted by the multiplier of social interaction.The study adds to literature regarding investors' over-trading and its biasing impact on asset prices in financial markets from the perspective of media effects.It also provides implications for financial market regulation and market stability controlling from the perspectives of market participants and the effective diffusion of information.