【作者】 朱斌；

【导师】 李亚静；

【作者基本信息】 电子科技大学 ， 应用经济学， 2023， 硕士

【摘要】 随着金融企业监管制度的成熟与完善,企业的信息披露行为也愈发变得规范。但是,“变脸”事件依旧屡屡发生,且业绩预告披露信息质量良莠不齐。为此,学者曾从许多角度对业绩预告质量的影响因素展开研究,本文乘着技术发展的大船,发掘了一个新的影响因素:媒体情绪。基于深度学习技术,对媒体的新闻文本进行学习训练,获得一个可以量化的情绪指标。将媒体情绪指标作为研究对象,探究媒体情绪对业绩预告质量的影响。立足于信息操纵理论和情绪渲染理论,研究媒体情绪如何影响管理层行为。而本文中的管理层行为特指对于业绩预告质量的控制。为探索媒体的文本内容对企业发布业绩预告意愿的影响,本文先对是否有媒体报道、以及报道的数量进行回归检验。结果表明受到媒体关注的企业更倾向于发布业绩预告,而受到关注的企业样本中,关注度与发布业绩预告愿意呈正相关。实证结果表明,媒体情绪对业绩预告呈现显著负向影响,即消极的媒体情绪可以达到外部治理效应,提高业绩预告质量。本文业绩预告质量使用了精准度、偏离度、及时性三个指标。其中精准度和偏离度回归结果一致为显著的负向影响,而媒体情绪对业绩预告及时性却存在显著正向影响,即媒体情绪越消极,企业管理层越拖着发布业绩预告。在加入机构投资者占比和是否执行股权激励计划影响因素以后,发现机构投资者占比对该效应存在显著的负向调节,即企业中机构投资者占比越高,企业管理层的决策愈发理性,因此媒体情绪对管理层决策的影响越小。本文还将数据样本按照企业特征、行业特征、经济环境特征进行分组实验,实验结果表明不同板块下的企业受到媒体情绪的影响效应不同,而结果在沪市主板的企业样本回归中最为显著。以精准度指标来看,某些特定的行业如:制造业、建筑业、卫生和社会工作、批发和零售业、金融业的企业受到媒体情绪的影响最显著,而租赁和商业服务业的企业受到的影响显著但与全样本结果相反。在疫情后,由于疫情防控政策,企业生产的不确定性更高。因此,疫情后媒体情绪对业绩预告质量的影响显著性不如疫情前。此外,以往学者对媒体相关信息的采集主要依靠搜索指数和对新闻文本的词频统计,本文使用深度学习模型构造媒体情绪指标,使得情绪衡量更合理。本文边际地拓展了关于业绩预告影响因素的立意角度,同时丰富了媒体文本在金融领域的实证研究中的应用。更多还原

【Abstract】 With the maturity and improvement of financial enterprise supervision system,the information disclosure behavior of enterprises is becoming more and more standardized.However,"face-changing" events still occur frequently,and the quality of performance forecast disclosure information is not good.Therefore,scholars have studied the influencing factors of performance forecast quality from many aspects.Taking advantage of the development of Technique,this paper has explored a new influencing factor: media sentiment.Technology based on Deep Learning,the media news text is learned and trained to obtain a quantifiable sentiment index.This paper probes the effect of media sentiment on the quality of performance forecast,utilizing media sentiment index as the research object.Based on earnings manipulation theory and emotional rendering theory,this paper studies the influence of media sentiment on the behavior of management,specifically the control of performance forecast quality.In order to explore the influence of media text content on the willingness of corporate management to issue performance forecasts,this paper first regresses whether there is media coverage and the number of reports.The results show that enterprises not subject to media attention are more inclined to issue performance forecasts,and among those receiving attention,the higher the attention,the more willing the management is to issue performance forecasts.The empirical results show that media sentiment has a significant negative effect on performance forecast,that is,negative media sentiment can achieve external governance effect and improve performance forecast quality.This paper uses accuracy,deviation and timeliness as three indicators of performance forecast quality.The regression results of accuracy and deviation are consistent,while media sentiment has a significant positive effect on timeliness of performance forecast,that is,the more negative the media sentiment,the more reluctant the management is to issue performance forecasts.After adding the influence factors of institutional investor proportion and whether equity incentive plan is implemented,it is found that institutional investors have a significant negative regulation effect on this effect,that is,the higher the proportion of institutional investors in the enterprise,the more cautious the decision-making of the management,so the influence of public opinion on the decision-making of the management is smaller.This paper also divides the data sample according to corporate characteristics,industry and environmental characteristics,and the experimental results demonstrate the influence effect of media sentiment on enterprises in different platelets is different,and the results are the most significant in the regression of Shanghai main board enterprise sample.Enterprises in some specific industries,such as manufacturing,construction,wholesale,retail and financial industries,are most affected by media sentiment,while enterprises in information services,leasing and commercial services are significantly affected by media sentiment,but opposite to the results of the whole sample.After the epidemic,due to the epidemic prevention and control measures,the uncertainty of enterprise production is higher,so the influence of media sentiment on performance forecast quality is more significant before the epidemic.In addition,in the past,scholars mainly relied on search index and word frequency statistics of news text to collect relevant media information.This paper uses deep learning model to construct media sentiment index,which makes the sentiment measurement more reasonable.This paper marginally expands the perspective of influencing factors of performance forecast and enriches the application of media text in the empirical research of finance.更多还原