Since investor sentiment broke the "rational man assumption" of asset pricing theory,the influence of investor sentiment on the stock market has become a fact that cannot be ignored.In particular,in modern behavioral finance,scholars have further confirmed that investor sentiment affects the limited rationality of investors,and that investor sentiment has become an important factor in the risk of stock market fluctuation.Investors' sentiment can synthetically influence changes in stock market development and plays an important role in the direction of stock market resource allocation development.Although the academic research on investor sentiment has formed more unified measurement indicators and matured research methods regarding the overall stock market sentiment measurement,the research on investor sentiment at the level of individual stocks,which is of concern to investors,has been challenged by a single source of data collection information,inconsistent measurement methods for directly measuring individual stock investor sentiment,and inconsistent research on the impact of investor sentiment.However,investor sentiment research on the individual stock level has been challenged by the single source of data collection information,the inconsistent measurement methods for directly measuring individual stock investor sentiment,and the inconsistency of investor sentiment impact studies.At the same time,fundamental issues such as whether investor sentiment can be used to invest in quantitative trading strategies are worthy of indepth analysis.Against this background,this thesis firstly constructs a measurement index system for investor sentiment under multi-channel information by taking individual stocks in the stock market as the research object and using multidimensional and large-scale data collection,processing and analysis techniques.Secondly,multichannel investor sentiment factors are extracted based on the measurement index system and verified the validity of their effects on the stock market.Finally,the multi-channel investor sentiment factor is introduced into quantitative investment analysis,and a deep learning model-based trading strategy is constructed to explore the optimal stock selection pattern for the multi-channel investor sentiment factor.Therefore,this study has three implications and effects:(1)This study enriches and expands the measurement method of investor sentiment based on the perspective of big data;(2)It verifies the validity of the measurement method of investor sentiment based on multi-channel information,explores the impact of multi-channel investor sentiment on the stock market,and explores new methods and ideas for investor sentiment research;(3)It provides new strategic models for the multichannel investor sentiment in quantitative investment research,and provides important insights and practical directions for using multi-channel investor sentiment for market analysis.The main research contents of this thesis are as follows.(1)A measurement system of multi-channel investor sentiment is constructed based on investor information sources.Considering that the stock market is a very complex system with rich dimensions of information sources for investors' decision making,this thesis proposes the idea of using multi-channel information to build a measurement system for investor sentiment,combining information from financial data on stock market transactions,online news information reported by professional news media,and social media information representing the views of the majority of "grassroots" investors.This can avoid information omissions and provide a comprehensive picture of investor sentiment.In measuring the sentiment of investors in the trading market,we construct a measure of investor sentiment in the trading market based on factors such as stock market volatility,stock trading volume and capital changes,and use the PCA(Principal Component Analysis)method to measure sentiment in the trading market of a sample of stocks.The measurement of investor sentiment based on online news and social media channels mainly utilizes crawling techniques for social media news and forum text data,and carries out the collection of online news and social media text data with 10 million data volume,and proposes a convolutional neural network based on CSSCNN(Chinese stock sentenceConvolutional We propose a convolutional neural network(CSSCNN)based investor sentiment mining method,train the model by using professional financial sentiment dictionary and manual marking,extract the text information of each stock's news and stock forums posting text,and realize the measurement of online news and social media investor sentiment.In particular,this thesis measures not only the sentiment from the stock exchange market channel but also the investor sentiment from online news and social media,and analyzes the investor sentiment characteristics of stocks with different activity levels,seasons and industries based on the measurement results of investor sentiment,revealing the multi-channel investor sentiment characteristics of individual stocks with different activity levels,industries and seasons in the stock market,providing a The results reveal the characteristics of multi-channel investor sentiment of stocks with different activity levels,different industries and different seasons in the stock market,and lay the foundation for the subsequent analysis of multi-channel investor sentiment.(2)Analyze the linkage and difference of multi-channel investor sentiment on stock returns,and verify the effectiveness of multi-channel investor sentiment.This thesis analyzes the impact of investor sentiment on the stock market based on the data of individual stocks in the A-share stock market,and then explores the linkage and difference of multi-channel investor sentiment on stock returns to verify the validity of multi-channel information source investor sentiment variables in the asset pricing model.This study finds a significant positive correlation between investor sentiment and stock returns,and examines the variability of the impact of stocks with different levels of activity and industries on returns by classifying indicators such as turnover rate,industry,and season.The impact of investor sentiment on returns is more sensitive in the sample of stocks in the active group.In the exploration of seasonal effects,although there is no significant seasonal effect on trading market investor sentiment,there is a negative seasonal sensitivity of financial news investor sentiment in summer and autumn,and investor sentiment based on financial news has a lower impact on returns than other seasons,which also indicates that a higher sensitivity to the sentiment of financial news assets in summer and fall will earn lower returns during periods of rising sentiment and higher returns during periods of falling sentiment,and stock forum news investor sentiment has a negative seasonal sensitivity in summer.In the heterogeneity analysis of industries,investor sentiment in more than 80% of industries can positively and significantly affect stock returns,and the higher the investor sentiment in that industry,the higher the future returns of its market.Robustness tests are completed by examining the lagged effect of multi-channel investor sentiment,U-shaped relationship,omitted variables and shrinkage tests.Finally,the effectiveness of multichannel investor sentiment on asset pricing is verified using the introduction of multichannel investor sentiment into the Fama-French five-factor model.This thesis provides a more detailed micro-econometric analysis of the impact of investor sentiment on the stock market,further demonstrating the effectiveness of the investor sentiment analysis method and providing a reference for in-depth research on the impact of multi-channel investor sentiment on the stock market.(3)Reveal the duration and intensity of the impact of investor sentiment from different channels on market return shocks.It has been found that the impact impact of investor sentiment on returns from different sources of information is different,the impact of investor sentiment on returns from trading market and stock bar forum is significant,and the impact of news media sentiment on returns is less significant.In the analysis of sustained impact,investor sentiment from trading market has the most significant impact on market returns,the strongest is one day,and the longest sustained period is 5 days,followed by stock forum and other social media investor sentiment,the impact of news media investor sentiment is less significant,and the impact of news media and social media peaks one day earlier and lasts for up to 3 days.The difference in the sustained influence of different information source channels on stock returns provides a reference basis for quantitative strategy design using multi-channel investor sentiment,and we focus on designing the moment for quantitative stock trading within the strongest sustained impact cycle,which can provide a reference for investors in making multi-channel information decisions,and also provide suggestions for information disclosure of listed companies.(4)Explore the optimal model of quantitative investment based on multichannel investor sentiment and verify the ability of stock selection based on multichannel investor sentiment factor.This thesis compares the traditional market sentiment MACD(Moving Average Convergence and Divergence)and investor sentiment by selecting multifactor stock selection indicators,including financial factors,technical characteristics and market multi-channel investor sentiment factors,after designing a quantitative strategy using the deep learning model LSTM(Long Short-Term Memory).Divergence)and investor sentiment,and confirmed the effectiveness of investor sentiment in multi-factor stock selection.As a benchmark for the selection of quantitative strategies studied,a benchmark strategy based on the ultra-short-term momentum effect is designed to screen out quantitative investor sentiment strategies that outperform the benchmark.This thesis further explores the optimization mechanism of trading strategies,including the overlay of financing and shorting mechanisms and trading signals and their diversity studies.The inclusion of the shorting mechanism of financing and financing provides a meaningful attempt to realize bilateral trading of stocks;the quantitative trading strategies are further optimized in the study of the overlay of trading signals and their rich diversity to achieve the growth of winning effects.Ultimately,the optimal model of a multichannel investor sentiment factor stock selection strategy that outperforms the benchmark model is identified.The quantitative investment study further demonstrates that multi-channel investor sentiment provides better stock selection than both traditional stock selection factors and single-channel investor sentiment.The research on quantitative stock selection strategies of investor sentiment in quantitative stock trading provides investors,listed companies and regulators with clear references for decision making,and the findings also enrich the results in the field by providing empirical evidence from China for the research on investment model selection based on investor sentiment.In summary,the main innovation points of this study are mainly reflected in the following four aspects.(1)Based on the perspective of big data research,it takes the lead in constructing a multi-channel investor sentiment measurement system,which breaks through the limitations of traditional investor sentiment research that more often uses a single channel measure and enhances the innovative research on investor sentiment indicator measures.This thesis identifies data sources for multi-channel investor sentiment research based on the perspective of investor information sources.Since this study collects complex large-scale data sets,through the collection and in-depth analysis of big data about investment transactions in trading markets,online news and social media,it helps to break through the limitations of single-channel data and more comprehensively portray investor sentiment,providing a comprehensive picture of the field of finance in complex big data This thesis provides an innovation based on data sources for the study of classical problems in finance in complex big data.At the same time,this thesis proposes an analysis method for measuring the characteristics of investor sentiment under different channels in the stock market,constructs a feature extraction method for investor sentiment for information from different sources,and carries out the construction of indicators for multi-channel investor sentiment,realizing the innovation of investor sentiment indicator metrics for different channels at the micro individual stock level.(2)Revealing the correlation between multi-channel investor sentiment and stock returns,verifying the effectiveness of multi-channel investor sentiment on stock market impact,and extending and enriching the research results on investor sentiment.First,in the part of empirical research on the impact of investor sentiment on returns,it is concluded that investor sentiment has a positive impact on stock returns,by further analyzing the differences in the impact of investor sentiment on stock returns on different activity levels,industries and seasonal effects.Next,the multichannel investor sentiment is investigated in terms of lagged effects,U-shaped relationship and omitted variable issues through robustness analysis.Finally,multichannel investor sentiment is introduced into the Fama-French five-factor model to verify the validity of multi-channel investor sentiment in the asset pricing model.(3)The sustained influence of multi-channel investor sentiment on the stock market is deeply explored,and the differences in the duration and intensity of the influence of different channels of investor sentiment are demonstrated,which provides important academic support for multi-channel investor sentiment research.This thesis firstly finds that trading market investor sentiment has more impact on returns than news and financial media and social media forums etc.through variance decomposition study,and after impulse analysis to study the duration of impact impact of different channels,the study finds that the strongest impact of investor sentiment of different channels occurs one day earlier,which provides important academic support for understanding the trading time of quantitative trading strategies using multi-channel investor sentiment This provides important academic support for understanding the design of trading time points for quantitative trading strategies using multi-channel investor sentiment and also provides an important academic reference for the correct use of multi-channel information to guide investors.(4)Based on the deep learning model,we explore the role of multi-channel investor sentiment in quantitative investment,and provide suggestions and guidance for using multi-channel investor sentiment for investment.By introducing multi-channel investor sentiment into multi-factor quantitative stock selection strategies and using LSTM deep learning models,the study compares the effectiveness of adding multi-channel investor sentiment indicators with traditional quantitative indicators in predicting stock ups and downs and returns and confirms that multi-channel investor sentiment has better stock selection effects in quantitative trading models compared with traditional stock selection factors and single-channel investor sentiment.In order to further improve the quantitative effect,this thesis also studies optimization mechanisms such as margin short selling mechanism,strengthening trading signal mechanism and adopting sample mirror diversity mechanism,so as to achieve the purpose of optimizing the effect of the model.It is found that investor sentiment has a better stock selection effect in quantitative trading,which verifies the effectiveness of multi-channel investor sentiment in quantitative trading strategy and helps investors to enhance the return of investment trading. |