Research On Asset Allocation Under The Black-Litterman Model Based On Industry Investor Sentiment Indicators

Behavioral finance theory has confirmed that investor sentiment plays a significant role in asset allocation.The main contribution of the Black-Litterman model is to combine traditional finance and behavioral finance,and update the investment portfolio based on the historical market data combined with the subjective sentiment of investors.As a result,it has been widely used in the field of asset allocation of modern investment portfolios to better quantify investors’ stock selection decisions.Regarding the issue of investors’ subjective opinions measurement in the Black-Litterman model,the literature researches mainly predicts expected returns based on historical return indicators.However,these researches ignored the impact of investors’ psychological bias on their returns,which will inevitably affect the results of asset allocation.Thanks to the rapid development of the Internet and the advancement of data collection technology,social network texts provide a new channel for the measurement of investor text sentiment.At present,most of the investor sentiment classification methods based on data mining mainly use written text as the information source for judging sentiment polarity,while ignoring the behavior data generated by users in the platform.And there are few researches on the integration of multiple texts and multiple users per unit time.The core questions that this paper hopes to answer are,Firstly,how to quickly extract key and valuable emotional information from the massive financial social networking site data?How to combine the characteristics of financial social networking site and use information such as release time,user expressions and social network structure to construct investor text sentiment indicators? Secondly,do investor sentiment metrics based on text mining algorithms better reflect the price trend of financial assets? Thirdly,does investor sentiment based on text mining algorithms affect financial asset allocation?Are allocation results significantly improved? The industry plays an important role in the allocation of industry weights by fund managers.At the same time,before selecting and allocating individual stocks,natural person investors need to pay attention to the market conditions of the industries in which individual stocks are located and grasp the strong industries.Therefore,it has certain practical significance to construct investor network information and use it for mapping and matching of investor sentiment,and to measure investor sentiment indicators at the industry level.This paper crawls industry bar of guba.eastmoney.com as the text database according to the first-level industry classification standard of Shenyin & Wanguo,and converts the text into a structured data matrix.Comparing and analyzing the effectiveness of three methods for measuring investor text sentiment,including the Word Lists,the Long Short-Term Memory network based on the attention mechanism of time,and the Long Short-Term Memory network based on the dual attention mechanism of time and user,and constructing the investor’s subjective sentiment indicator —— text sentiment indicator based on social network text mining algorithm,so as to achieve the purpose of reflecting investor sentiment more truly and accurately.Then,combined with the screened objective sentiment indicators and the constructed subjective sentiment indicators,the principal component analysis method is used to synthesize comprehensive investor sentiment indicators of various industries that reflect the stock price trend of the industry.And investor sentiment indicators are embedded into the Black-Litterman model to construct an investor opinion and determine the asset allocation ratio between industries.The main content and research conclusions of this paper are mainly divided into three parts:1.Based on the Long Short-Term Memory network(LSTM)model based on the dual attention mechanism of time and user,the investor text sentiment indicator is constructed.In this paper,multi-text information,user social information and emotional time series are combined,and data matrix information is extracted by unsupervised learning method — Word Lists and supervised learning method — Long Short-Term Memory network(LSTM)based on the attention mechanism of time and the dual attention mechanism of time and user.According to the stock market trading day,this paper classifies and predicts the industry investor text sentiment metrics which are constructed in time series,and improves the effectiveness of the investor text sentiment metrics by comparing and analyzing the results of the three methods.The empirical results show that the emerging deep learning method — LSTM has better predictive ability than the traditional text big data analysis method — Word Lists.The fusion of time and user attention mechanism improves the accuracy of information extraction and provides a more effective method for accurately measuring investor text sentiment.2.Based on the optimized screening of six objective sentiment proxy variables and one subjective sentiment proxy variable,the principal component analysis method is used to construct comprehensive investor sentiment indicators in various industries.As the government formulates and implements different industrial policies for various industries,there are significant differences in investor sentiment during the investment process.In addition,the investment value of each industry in each period is different,so there are obvious differences in investor sentiment indicators among industries.Firstly,the rationality test--Granger causality test is used to screen the seven sentiment proxy variables in various industries.Then,the cross-correlation analysis method is used to select the synchronous,leading or lagging terms of the proxy variables with high correlation coefficient with return rate.Finally,the principal component analysis method is used to synthesize the comprehensive investor sentiment indicators in various industries.The empirical results show that the constructed investor sentiment indicators of various industries can better reflect the stock price trend of the industry.3.The industry investor sentiment indicators based on text mining technology is quantified as the confidence level that affect the investor’s opinion error co-variance matrix,which is embedded in the Black-Litterman model to construct the investor view and determine the asset allocation ratio among industries.The constructed comprehensive investor sentiment index is embedded into the Black-Litterman model to construct investor view matrix,and the Black-Litterman model is improved to optimize the asset allocation ratio among industries.The empirical test confirms that the portfolio efficiency and allocation results of the Black-Litterman model based on industry investor sentiment are significantly better than those of the traditional Markowitz mean-variance model under the minimum risk and the highest Sharpe ratio on the efficiency frontier.The results effectively improves the average daily return rate and Sharpe ratio of asset allocation.The empirical results are still robust after adjusting from multiple angles,such as allowing short selling and considering transaction costs,first three quarters of out-of-sample testing,and the stock market boom and bust phase and the faster and slower rising/falling phase,which further confirms that investor sentiment has a significant impact on asset portfolio.The Black-Litterman model based on industry investor sentiment has certain competitive advantages.The main innovations of this paper is reflected in the following three aspects:1.The adopted or expanded methods have realized the fusion of multi-text information within each trading day.The total amount of texts crawled in various industries during the statistical period is at least 100,1,000 or even more than 10,000.The multiple text contents of each trading day are integrated and fused,so as to construct the investor sentiment indicator of time series and form the daily frequency data.2.After adding user social data,the integration of multi-text and multi-user information in the trading day is realized.The information of different users in the social platform is considered in the supervised learning model for constructing investor text sentiment metrics.The social data such as the number of fans,influence,and bar age are used as the attention mechanism of the Long Short-Term Memory(LSTM)to capture the continuous and dynamic emotions of investors more realistically.3.This paper combines the investor sentiment indicators with the Black-Litterman asset allocation model.The investors’ subjective view in Black-Litterman model is established based on industry investor sentiment indicators,and the optimal programming model is established to dynamically simulate the optimal asset allocation strategy.In this paper,investors’ cognitive bias and other factors are fully considered,and investor sentiment indicators are constructed based on social network text mining technology,which solves the limitations of investors’ psychological cognition and behavior that cannot be intuitively revealed by prediction models that only rely on expected returns or historical data.From a new perspective,the problem of investor opinion generation in Black-Litterman model is solved scientifically.Moreover,the application of the model is subdivided into the stock markets of various industries in China,and the empirical data that investor sentiment in emerging stock markets has an important role in asset allocation is given.The theoretical system of the Black-Litterman model is extended,and the application of behavioral finance theory in asset allocation is promoted.