以重大訊息文本數據為基礎之上市公司風險預警模型之研究

RESEARCH ON EARLY RISK WARNING MODEL OF LISTED COMPANIES BASED ON MATERIAL TEXT INFORMATION

郭螢璇(Ying-Hsuan Kuo) ； 李御璽(Yue-Shi Lee) ； 陳明輝(Ming-Huei Chen) ； 林基玄(Chi-Hsuan Lin) ； 徐靜婷(Jing-Ting Hsu)

《Journal of Data Analysis》 18卷1期 (2023/04) Pp. 43-74

https://doi.org/10.6338/JDA.202304_18(1).0003

摘要

股票市場作為上市公司籌集資金的主要途徑之一，同時亦為資金門檻相對低且操作方式相對容易的標的物之一，其變化影響著整個市場經濟。為確保市場資訊之公平與透明，各國政府與投資人皆尤其重視資訊揭露，臺灣則於證券交易所之「公開資訊觀測站」平台提供所有公司之財務、營運及公司治理等資訊。由股票市場分析方法之消息面角度切入，公開資訊觀測站揭露之重大訊息雖相對其餘市場消息更具正確性，然而大量的資訊是否得以被投資者消化與接受？是否真的能幫助投資人做出正確的判斷？為達到資訊揭露制度之效益，投資者能有效地利用這些資訊，並制定正確的投資決策應為首要條件。本研究主要以公開資訊觀測站公布之上市公司各式重大訊息為文本輸入資料，以及各上市公司之TCRI信用評等作為目標變數，利用文字探勘技術與分類機器學習進行文本分類，自然語言處理部分以Jieba斷詞系統進行斷詞，TF-IDF（Term Frequency - Inverse Document Frequency）計算字詞重要性，詞袋模型（Bag-of-Words Model，BoW）將字詞轉換為結構化向量，進一步建構風險預警模型，並評估、比較模型效果，以期為投資人提供正確的投資決策參考，同時驗證重大訊息公告之效益。

關鍵字

重大訊息 ； TCRI ； 文本分類 ； 文字探勘

並列摘要

As one of the main method for listed companies to raise funds, the stock market is also one of the objects with relatively low capital level and relatively easy operation. Its changes affect the entire market economy. In order to ensure the fairness and transparency of the market information, governments and investors of all countries focus on information disclosure. The Market Observation Post System of the Taiwan Stock Exchange provides the financial, operational and corporate governance information of all companies. From the point of view of the stock market analysis method, although the information disclosed by the Market Observation Post System is more accurate than other market information, however, can there be a large amount of information be digested and accepted by investors? Can it really help investors make correct judgments? In order to achieve the benefits of the information disclosure system, investors should be able to effectively use these information and make correct investment decisions. This research uses material information of listed companies in the Market Observation Post System as text input data, and TCRI credit ratings of listed companies as target variables. Text classification technology in text mining is used to classify text. In natural language processing, Jieba word segmentation system, TF-IDF method and Bag-of-Words (BoW) Model are used to segment words, calculate word importance and convert words into structured vectors, respectively. Based on the structured vectors, this research constructs an early risk warning model, which evaluates and compares the effects of the model. The purposes of this research are to provide investors with correct reference for investment decisions and verify the effectiveness of material information.

並列關鍵字

Material Information ； TCRI ； Text Classification ； Text Mining

參考文獻

TCRI台灣企業信用風險指標（無日期）。台灣經濟新報。2022年8月30日，取自：https://www.tej.com.tw/solution/TCRI%E2%84%A2%20%E5%8F%B0%E7%81%A3%E4%BC%81%E6%A5%AD%E4%BF%A1%E7%94%A8%E9%A2%A8%E9%9A%AA%E6%8C%87%E6%A8%99。

TEJ信用風險觀測TCRI（2010）。台灣經濟新報社。2022年8月30日，取自：https://www.tej.com.tw/webtej/doc/crwatch.htm。

王美齡（2019）。國內外重大訊息案例研析。證券服務，637，23-40。

王詮富（2011）。信用評等與違約率之回顧 - 以TCRI為例。國立台灣大學財務金融研究所碩士論文。

王慶助（2016）。信用評等TCRI調整對企業經營績效影響之探討與分析。國立臺北大學國際財務金融研究所碩士論文。