| Market efficiency is an essetial topic in financial studies.Traditionally,due to the difficulties in processing unstructured data,the research has been focusing on structured data.However,in the big data era,various machine learning methods,specifically,the natural language processing(NLP)methods,offer the possibility to explore voluminous information in unstructured data,such as analyst reports.Analyst reports contain ample information associated with stock market and investors,which is very important to reduce the information asymmetry and increase the market efficiency.Although analyst reports contain potentially valuable information,due to the limitations in methodologies and accessibility,there are relatively little research studing the textual information and most of the research focus on quantitative information.Analysts collect information through field research,participation in conference calls,paying attention to company announcements and data analysis,and many other ways to carry out information mining.Anlyasts have forecast and investment advice about the future performance of listed companies,stock prices and development and other aspects of the analysis and finally summarized their opions in analyst research reports.Therefore,the text information of analyst research reports contains a wealth of information resources that can not be included in quantitative information.If the information role of analyst reports is only studied from the dimension of quantitative information,and the qualitative information in analyst reports is ignored,it is undoubtedly a great loss for fully understanding and understanding the information function of analysts and the information validity of the market.Only by combining the quantitative and textual information of analyst reports can we more comprehensively analyze and measure the information role played by analysts in the financial market to improve the pricing efficiency of the capital market.How to collect and process the big textual data of the securities analyst report is the premise and key step to analyze the analysts’ textual information.In this paper,using the network data crawling method,more than 400,000 analyst reports related to A-share listed companies are crawled from Oriental Fortune Network from January 2010 to December 2020,and after sample screening,239,834 valid listed company research reports are finally used.Considering that the methods commonly used in the existing literature(such as the dictionary method and the Naive Bayes method)cannot deal with the problem of word order,it is not possible to effectively extract textual information at the sentence level.In view of the complexity of the Chinese language itself and the importance of word order problems,this paper mainly uses LSTM(Long Short-Term Memory),a deep learning method that can be more effectively analyzed at the sentence level,to extract the analyst tone.The empirical research in this paper mainly revolves around the information role of analysts textual tone and the market response to information.First,this paper uses event research methods to analyze the role of analyst text intonation information.Secondly,from the perspective of industry analysis,this paper analyzes the information role of analysts’ industry-related tones and construct trading strategies at the industry level.Furthermore,from the perspective of topic studies(Topic Analyses),this study explores the investor reaction caused by analysts’ attention to a certain topic.Specifically,taking the analyst’s attention to the important theme of green environmental protection as an example to study the impact of analysts’ topics on the market.Finanly,this study extracts analyst sentiment from analyst texts,and further indepth thinks about the informational role of analysts’ texts.Studying the informative role of the analyst tone,this paper finds that the accuracy performance of LSTM is superior to the Naive Bayes method in both insample and out-of-sample validity tests.In a two-way sorting test,the samples are sorted independently by analyst tone and analyst investment recommendation rating(recommended rating correction).This study finds that under the condition of controlling the analyst recommendation rating(recommended rating correction),different combinations of analyst tones can produce significantly different market reactions,indicating that the analyst tone provides the market with a new source of information in addition to the analyst rating information.Regression analyses finds that the more positive the tone of the text in the analyst report,the more positive the response to stock yields.Furthermore,we divide the analyst tone into positive and negative tones and analyze the market’s reaction to positive and negative tones.Regression analysis results show that the market reacts more strongly to the negative tone of the analyst.The results of the heterogeneity analysis show that the market reacts more strongly to the tone of star analysts,and the tone of analysts who rank high in market capitalization is more intense.At the same time,this paper finds that the information environment is better for stocks thatc can carry out margin financing business and join the list of Shanghai-Hong Kong Stock Connect and ShenzhenHong Kong Stock Connect,and because the information role of analysts in this type of stock is more significant,the market response to the text tone of analysts in this type of stock is more intense.Studing the role of analyst information from the industry perspective,this paper finds that based on the analysts’industry related tone,the arbitrage portfolio can produce excess returns that cannot be explained by systemic risk.Furthermore,the results of the placebo test support the analysts’ textual tone about the industry.The results of the regression analysis are consistent with those of the portfolio analysis,demonstrating the significant predictive role of the analysts’textual tone about the industry in predicting industry returns.At the same time,trading strategy based on quantitative information constructed at the industry level cannot generate significant returs.In the mechanism analysis section,the article argues that informatione value of the analyst’s analysis related to industry is from the analyst’s expectation of the growth and valuation of the industry.The analysis shows that the analyst industry analysis tone is positively correlated with the fundamentals of the industry(standardized unexpected revenue and standardized unexpected earnnigns),indicating that the analysts’information role at the industry level stems from analysts’understanding of the industry fundamentals.Taking the analyst’s attention to the theme of environmental protection as an example,the main findings are as follows:First of all,the article constructs a dictionary of environmental protection words,and constructs the analyst’s concern for environmental protection.The analysis finds that analysts’attention to environmental protection has a short-term positive effect on green stocks,increasing the short-term excess cumulative yield of stocks.This effect becomes more pronounced as analyst tone increases positively.Analysts’attention to environmental themes has a short-term positive effect on green stocks,and the role of analysts’ price discovery becomes more pronounced when there are star analysts.Analysts’ environmental concern receives more attention from investors as the national environmental regulations are stricter and the public’s attention to environmental protection increases,and the positive effect on the short-term market value of green stocks becomes more significant.In the robustness test,this paper finds that small market capitalization companies in the green industry usually receive less attention than companies with large market capitalization.Therefore,when market attention increases,the market response to small market value is stronger due to sudden increased attention in environmental protection.At the same time,the company with a high book market value ratio in the green industry,the market valuation is relatively low,when the market attention increases,for the sudden increase in concern,the market response of the company with high book market value ratio is stronger.The main findings about the market forecasting part of analyst sentiment are as follows:the paper uses the partial least squares method(PLS)to sum up the intonation information at the industry level,and this chapter finds that the analyst sentiment index constructed using the PLS method can significantly negatively predict future excess market returns in the sample,and its forecasting ability can last for more than one year.At the same time,in the study of the sub-sample,this paper finds that the analyst sentiment index had stronger predictive power during the high sentiment cycle.By comparing the predictive power of macroeconomic variables for future market yields,this paper finds that the analyst sentiment index remains significantly predictable after controlling for the impact of macroeconomic variables.In order to avoid the problem of small sample distortion caused by the in-sample test and the problem of forward-looking error caused by the PLS method itself,this paper also performs the out-of-sample predictive ability test.The results show that the analyst sentiment index is still highly predictive and has significant economic significance in the out-of-sample test.In an asset allocation test,investors can obtain positive certainty equivalent returns and a sharp ratio that is higher relative to market yields when using the predicted value of the analyst sentiment index for asset allocation. |
PDF Full Text Request