关键词: 年度报告, 信息披露, 文本分析, Black-Litterman投资组合模型

Abstract: The annual reports of listed companies, which are characterized by reliability, equality and rich content, increase the information supply of listed companies to the market, meet the information needs of investors, and also affect investors’ investment decisions and perspectives. Previous studies have found that annual report information disclosure has a significant impact on investors’ investment decisions and viewpoints. However, currently there is little research in China that uses annual report text information to improve investor viewpoint parameters in the Black-Litterman model (BL model). The BL portfolio model is proposed based on the Markowitz mean variance model, which introduces investor perspectives and modifies expected returns on the basis of equilibrium returns. This innovative method that combines prior returns and investor perspectives makes the calculation of expected returns in investment portfolios more reasonable. Investors can combine their personal expected returns, risk preferences and other perspectives to make more accurate asset allocation decisions.
The annual reports of listed companies contain rich and complex information, including a large number of textual descriptions, financial data and other related information. Traditional reading and comprehension methods often have low efficiency in processing these large-scale data. With the development of computer technology, various text analysis and mining tools have emerged one after another, making it possible to efficiently interpret and analyze the text content disclosed in annual reports. Furthermore, it can better assist investors in making investment decisions and market analysis. This article first combines computer technologies such as machine learning and deep learning to construct an annual report information disclosure attribute indicator that includes three dimensions: readability, similarity and risk factors. These indicators can measure the disclosure attributes of annual reports of listed companies. Then, this article uses a random forest regression model to predict the rise and fall of stock prices by adding annual report information disclosure attribute indicators as input features on the basis of traditional indicator prediction. Finally, we apply the predicted stock price fluctuations as an investor perspective to the traditional BL investment portfolio model and construct a new BL investment portfolio model that considers the disclosure of annual report information by listed companies. Our paper optimizes the traditional BL investment portfolio model and improves the measurement level of annual report information disclosure.
This article conducts an empirical analysis based on real data from the domestic A-share market. The results indicate that the BL investment portfolio model constructed in this article, which considers the disclosure of annual report information of listed companies, performs well in indicators such as Sharpe ratio. And our model can achieve returns exceeding market indices under market conditions. The research results can provide certain standard references for listed companies, relevant regulatory departments, and investors in the formulation, supervision, and analysis of annual report information disclosure content.
In future research, further exploration can be conducted from the following two aspects. One is to broaden the scope of information disclosure content collection, including disclosure information from self-media platforms (such as financial reports, Weibo, and WeChat) to achieve more dimensional measurement of information disclosure attribute indicators. The second is to optimize and improve more types of investment portfolio models based on the annual report information disclosure attribute indicators, further expanding the application scenarios of information disclosure attribute indicators.

Key words: annual report, information disclosure, text analysis, Black-Litterman portfolio model