財務報導與財經新聞資訊內涵之差異分析研究 - 政大學術集成

全文

(1)國立政治大學會計學系碩士論文. 財務報導與財經新聞資訊內涵之差異分析研究政治. 大. 立 Financial Reporting and Financial News －. ‧ 國. 學. An Information Content Gap Analysis. ‧. n. er. io. sit. y. Nat. al. 中. 華. Ch. engchi. i n U. v. 指導教授：. 諶家蘭博士. 研究生. 黃冠穎. 民. 國. ：. 104. 年. 撰. 6. 月.

(2) 摘要新聞報導在現今社會中為最便利與即時的資訊來源，也是公司與管理階層用來發布年度表現與其他重要消息的管道。本研究想要去找出新聞報導的內容與股票市場的反應之間的關連性，以及財務報表附註提供比新聞報導更多的其他資訊。本研究以 2013 年度臺灣 90 家上市公司為樣本去探討股票市場與新聞報導及財報附註之間的關係。藉由內容分析法的方式去評估新聞報導與財報附註中. 政治大兩種不同的資訊來源，找出股票市場對於此兩者是否有不同的反應，並從中判立. 的正向、負向情緒以及兩者之間的資訊差異以及兩者對股票市場的影響。藉由. ‧ 國. 學. 斷兩種資訊內容的差異。. 本研究發現新聞報導中的正向情緒與股票市場有正向的關連性，然而無法. ‧. 顯著證明在市場反應方面，財報附註能提供比新聞報導更豐富的資訊。. n. er. io. sit. y. Nat. al. i n U. v. 關鍵字：累積異常報酬、新聞報導、內容分析法、財報附註。. Ch. engchi.

(3) Abstract Financial news articles are the most convenient and timely information in today’s world. Companies and managers can announce current year performance and other concurrent disclosures to investors and stakeholders by financial news articles. This study hopes to find the relationships between financial news articles and the stock market and the information content gap between financial news articles and footnotes to financial statements.. 治政大articles and footnotes to responses to the information content in the financial news 立 financial statements. This study implements content-analysis technique, which count This study uses 90 listed companies in TSE of 2013 to test stock market. ‧ 國. 學. words for characterizing as optimistic and negative tone and helps to decide the. ‧. abundancy of information content elaborated, to find the information content to stock market reactions and compare the information content gap between financial news. y. Nat. n. er. io. al. sit. articles and footnotes to financial statements.. i n U. v. This study finds that optimistic sentiment expressed in financial news articles. Ch. engchi. positively relates to the stock price movement. However, this study is unable to reach a conclusion that information disclosed in the footnotes to financial statements is significantly enough to represent the existence of information content gap as compared to financial news articles.. Keywords: cumulative abnormal return, financial news articles, content analysis, footnotes to financial statements..

(4) Table of Contents 1. INTRODUCTION ................................................................................................... 1 1.1 RESEARCH PURPOSE AND MOTIVATION ............................................... 1 1.2 RESEARCH ISSUE......................................................................................... 5 1.3 RESEARCH PROCESS .................................................................................. 6 2. LITERATURE REVIEW........................................................................................ 8 2.1 EFFICIENT MARKET HYPOTHESIS .......................................................... 8 2.2 SOURCES OF FINANCIAL INFORMATION ............................................ 11. 政治大 2.4 DISSCUSSION .............................................................................................. 16 立 2.3 CONTENT ANALYSIS ................................................................................. 13. 3. REASERCH METHOD ........................................................................................ 25. ‧ 國. 學. 3.1 HYPOTHESIS DEVELOPMENT................................................................. 25. ‧. 3.2 DATA AND SAMPLE SELECTION ............................................................ 30. sit. y. Nat. 3.3 CONTENT ANALYSIS METHOD ............................................................... 35. io. er. 3.4 REGRESSION MODEL................................................................................ 39 4. RESEARCH RESULTS AND ANALYSIS .......................................................... 42. al. n. v i n C h ......................................................................... 4.1 DESCRIPTIVE EVIDENCE 42 engchi U 4.2 EMPIRICAL RESULTS ................................................................................ 47. 5. CONCLUSION AND DISCUSSION ................................................................... 54 5.1 RESEARCH DISCUSSION AND CONTRIBUTION .................................. 54 5.2 LIMITATION AND FUTURE RESEARCH WORK .................................... 56 REFERENCES........................................................................................................... 58 APPENDIX ................................................................................................................. 62.

(5) Lists of Tables Table 2-1. Summary of Related Research.................................................................... 19 Table 3-1. Results of Selection of Samples ................................................................. 33 Table 3-2. Industry Breakdown.................................................................................... 33 Table 3-3. Variable Definition...................................................................................... 40 Table 4-1. Descriptive Statistics .................................................................................. 43 Table 4-2. Correlation Statistics................................................................................... 45 Table 4-3. Results for the Test of Relationships between Footnotes Scoring and Stock. 政治大 Table 4-4. Results for the Test of Relationships between Footnotes Scoring and Stock 立. Price Movement ........................................................................................ 48. Price Movement by Adding the Information Content of Financial News. ‧ 國. 學. Articles ...................................................................................................... 51. ‧. n. er. io. sit. y. Nat. al. Ch. engchi. i n U. v.

(6) Lists of Figures Figure 1-1. Research Structure ...................................................................................... 6 Figure 3-1. Research Model and Hypotheses .............................................................. 29. 立. 政治大. ‧. ‧ 國. 學. n. er. io. sit. y. Nat. al. Ch. engchi. i n U. v.

(7) List of Appendixes Appendix 1. Footnotes Scoring Example .................................................................... 62 Panel A. Rule 1 ...................................................................................... 62 Panel B. Rule 2 ...................................................................................... 62 Panel C. Rule 3 ...................................................................................... 63 Panel D. Rule 4 ...................................................................................... 63 Panel E. Rule 5 ....................................................................................... 63 Panel F. Rule 6 ....................................................................................... 64. 政治大 Appendix 2. DET_FS Word Lists ................................................................................ 65 立. Panel G. Rule 7 ...................................................................................... 64. Panel A. BS_D Word Lists ..................................................................... 65. ‧ 國. 學. Panel B. SCF_D Word Lists .................................................................. 68. ‧. Appendix 3. Word Lists of Sentiment .......................................................................... 71. sit. y. Nat. Panel A. Positive Words ......................................................................... 71. io. er. Panel B. Negative Words ....................................................................... 73 Appendix 4. News Scoring Criteria ............................................................................. 74. al. n. v i n Ch Panel A. Rule 1 ...................................................................................... 74 engchi U Panel B. Rule 2 ...................................................................................... 76 Panel C. Rule 3 ...................................................................................... 77 Panel D. Rule 4 ...................................................................................... 78 Panel E. Rule 5 ....................................................................................... 78.

(8) 1. INTRODUCTION 1.1 RESEARCH PURPOSE AND MOTIVATION Managers and companies convey performance to recent and prospective investors by accounting numbers. It is the most important and convenient way for investors to understand a company in today’s world. By referring to standardized format of financial statements required by laws and exchanges, investors can utilize and implement useful firm-specific information to help them evaluate and make appropriate decisions in time. High quality financial reporting can not only reduce the information asymmetry. 政治大. because of principal–agent problem but also facilitate sound capital markets and security market efficiency.. 立. ‧ 國. 學. Financial reports are sources of information companies must provide to the public. ‧. quarterly and annually. Major content included in the financial reports are Statement of. sit. y. Nat. Financial Position, Statement of Comprehensive Income, Statement of Stockholders’. io. er. equity, Statement of Retained Earnings and footnotes to financial statements.. al. v i n C h mainly provide content of the data provided, the former e n g c h i U numerical information about n. Differences between these four statements and footnotes to financial statements are the. the reporting period and the latter mostly provide explanatory and supplementary information to help financial report users better understand the calculation and explanation of an account or figure. With fundamentals and supplementary information required by laws and exchanges, one can acquire related company-specific information easily. By just referring to a company or exchange’s websites, when an investor is interested in a public company, he can find financial, operational and other readily available information easily accessible. However, besides financial reports and other formal announcement issued by 1.

(9) managers and companies, other sources of information provide not only accounting figures but also information with different perspective to the public. For example, analysts’ research reports, government publications, periodicals and, most handy and even free of charge sources of instantaneous information, financial news articles. Compared to periodic financial reporting, which normally has time lag before being available to the public, these alternatives can provide information on a more timely basis and help investors act instantly on receiving each pieces of information. It is said that because of the time lag when financial reports are announced, financial reports. 政治大 with current information to help them predict the future outcome of a company. 立. provide past performances of a company and therefore investors usually combine them. Therefore, current and prospective investors may combine financial reports provided. ‧ 國. 學. by companies with more current financial news articles together to perform better. ‧. decision-making.. Nat. sit. y. Ever since the invention of the World Wide Web (WWW), people are able to keep. n. al. er. io. crucial and valuable information highly related to the financial markets at their. i n U. v. fingertips. Various types of financial information are free of charge and they are just a. Ch. engchi. click away. For example, Wall Street Journal (WSJ), Bloomberg, CNN, and UDNdata, a Taiwanese-based news database with searchable feature of full-text newspapers. All these make textual information attractive resources to discover. Hundreds of millions of news on the Internet are updated daily. Compared to TV news and newspapers, online news is the easiest way one can capture what is happening and what might happen in the future in no time. For example, financial news articles may discuss possible reasons to the current economic condition and make prediction to the coming months. Efficient Markets Hypothesis (EMH) states that the market reacts instantaneously according to new incoming information (Fama 1965). News releases may influence 2.

(10) investor’s behavior and expectations and in turn affect stock market fluctuations. In the financial markets, changes in stock prices are the consequences of many unpredictable actions taken by investors and companies and even economic condition of a country. The wording throughout the financial reports and financial news articles may signal or imply managers’ or market expectations about the future performance. Managers’ earnings release language varies significantly across firms and ranges from straightforward to promotional (Mahoney et al. 2004). It is well established that presenting information in positive terms results in more favorable evaluations than does. 政治大 positive and negative language has a substantial influence on how information is 立. presenting information in negative terms (Levin et al. 1998). At the most basic level,. processed. Language also influences how information is both perceived and understood. ‧ 國. 學. (Katz 2001; Morris et al. 2005). As the standardized financial reports are required for. ‧. presentation, there is comparatively little room for companies and managers to express. sit. y. Nat. information in his favorable way.. n. al. er. io. Most of researches on data analytics during the past several decades aimed at. i n U. v. mining information from numerical data. There are also advanced techniques. Ch. engchi. developed to help analyze textual information. In recent years, many studies are focus on analyzing semi-structured and unstructured data, such as textual data. The major difficulty in analyzing textual data is that it has to be preprocessed first to make it possible for existing data-mining and related statistical software to analyze. Text analytics are processes by computer software, which automatically extract new and previously unclear information with mathematical and statistical methods. Steps in text analytics includes collecting textual data, processing the data into computer-readable data, organizing large amount of data into categorized data, and analyzing features. This study focuses on the sentiment and the information abundancy disclosed in financial 3.

(11) news articles and annual financial statements, especially footnotes to financial statements. That is, this study expects to observe market responses to words and information presented in both news articles and financial statements and further discusses the information content gap between financial news articles and footnotes to financial statements.. 立. 政治大. ‧. ‧ 國. 學. n. er. io. sit. y. Nat. al. Ch. engchi. 4. i n U. v.

(12) 1.2 RESEARCH ISSUE By nature, qualitative disclosures and news articles provide opportunities for companies and the press to imply or even try to influence market participants’ perceptions of firm’s future performance. New York Stock Exchange (NYSE) rules require that press releases place news in the ‘‘proper perspective’’ and that managers should avoid ‘‘overly optimistic forecasts, exaggerated claims, and unwarranted promises’’ (NYSE Manual), which bring out the inherent flexibility of qualitative disclosures in general, but also highlight the importance of language used about releases. 政治大. to the market as well as to regulators.. 立. Not only analyzing the terms used in the financial reports and news articles, but. ‧ 國. 學. also analyzing the sentiment used by the author, this study plans to implement a dictionary-based content analysis method that is capable of analyzing frequency of. ‧. words used and build an overall sentiment of each article after proper evaluation. Our. y. Nat. sit. motivation is to build and test such artifact with an emphasis on the sentiment and. n. al. er. io. information incorporated within financial news articles and financial statements of a. i n U. v. company. By evaluating the intricacies of article sentiment and words used, we seek to. Ch. engchi. find out whether the role sentiment and textual information can play in stock market. According to the research purpose and motivation, the research issues of this study are as follows: 1.. Does the language used throughout financial news articles provide a signal to the market response?. 2.. With the aid of detail information contained in the financial reports, does the market reaction to new information is more associated to it than merely news article alone?. 5.

(13) 1.3 RESEARCH PROCESS Research process is presented as follows: Figure 1-1. Research Structure Introduction. Literature Review. 學. ‧ 國. 立. 政治大. Research Method. ‧. n. al. er. io. sit. y. Nat Empirical Results and Analysis. Ch. engchi. i n U. v. Conclusion and Discussion. This study begins with introduction, as providing background information and research motivation to express the reasons and purposes of this study. Next, in literature review section, this study collects relevant ideas and methods of former research. In research method section, when and how samples are collected and processed are elaborated as well as research hypotheses and research models. Next, in 6.

(14) research results and analysis section, empirical results of regressions and association between dependent variable and independent variables are discussed. In conclusion and discussion section, conclusion and contribution of this study are shown as well as limitations and suggestions to future research. In appendix, variable definition, word lists used to perform word counts and scoring and related examples are presented. The remainder of this study is organized as follows. Section 2 reviews relevant literature. Section 3 presents research method. Section 4 discusses results. This study concludes in section 5.. 立. 政治大. ‧. ‧ 國. 學. n. er. io. sit. y. Nat. al. Ch. engchi. 7. i n U. v.

(15) 2. LITERATURE REVIEW In literature review section, this study first introduces Efficient Market Hypothesis to express the idea that an efficient market will react to new-coming information. Next, talks about two main financial information sources, financial reporting and financial news articles. Third, elaborates content analysis method. Last, gathers related works about related research issues and how this study is implemented based on prior research. 2.1 EFFICIENT MARKET HYPOTHESIS. 治政大 is introduced, the market of complete market information and when new information 立 instantly adjusts the stock price to reflect it. From EMH theory, it is believed that strong Efficient Market Hypothesis (EMH) states that the price of a security is a reflection. ‧ 國. 學. markets are efficient and that price corrections occur instantly, making price prediction. ‧. from market data impossible. The notion of market efficiency has been defined in many ways. Fama (1965) presents a thorough discussion of both theoretical issues and. y. Nat. er. io. sit. empirical tests of this proposition. After a discussion of the theory, empirical work concerned with the adjustment of security prices to three relevant information subsets. n. al. Ch. i n U. v. is considered. First, weak form tests, in which the information set is just historical prices,. engchi. are discussed. Then semi-strong form tests, in which the concern is whether prices efficiently adjust to other information that is obviously publicly available, are considered. Finally, strong form tests concerned with whether given investors or groups have monopolistic access to any information relevant for price formation are reviewed (Fama 1970). That is, market efficiency can be divided into three categories: (1) weakform tests (How well do past returns predict future returns?), (2) semi-strong-form tests (How quickly do security prices reflect public information announcements?), and (3) strong-form tests (Do any investors have private information that is not fully reflected in market prices?) Later, Fama (1991) changes the categories. Instead of weak-form 8.

(16) tests, which are only concerned with the forecast power of past returns, the first category now covers the more general area of tests for return predictability, which also includes the burgeoning work on forecasting returns with variables like dividend yields and interest rates. Instead of semi-strong-form tests of the adjustment of prices to public announcements, he uses the now common title, event studies. Instead of strong-form tests of whether specific investors have information not in market prices, he suggests the more descriptive title, tests for private information. Accounting research has long been interested in the valuation implications of. 政治大. corporate earnings. Traditionally, these studies have been couched in an efficient. 立. markets framework that postulates that accounting information is efficiently impounded. ‧ 國. 學. into stock prices by rational agents in well-functioning capital markets (Holthausen et al. 2001; Kothari 2001). The information content of accounting numbers can be. ‧. measured by the extent to which security prices change in response to the. Nat. sit. y. announcement of financial statements. One of the most commonly used measures is the. al. n. following model:. er. io. Earnings Response Coefficient, ERC. Specifically, an ERC is estimated using the. CARit  a  bUEit  eit. Ch. engchi. i n U. v. where eit is white noise and CARit (Cumulated Abnormal Return; CAR) is the measure of risk-adjusted return for security i cumulated over an event window around the earnings announcement at time t. The ERC literature often adopts Fama and French three-factor model (Fama and French 1993), which incorporates excess market returns (market returns minus risk-free interest rates), small minus large firm returns (SML), and high minus low book-to-market firm returns (HML) as main risk factors. Expected returns can be computed after the loadings of risk factors are estimated using historical 9.

(17) data. Abnormal returns are the difference between actual stock returns and their expected returns computed with the three-factor model. It is a common practice to accumulate the abnormal returns over a period to capture the market reaction to earnings events. UEit is the unexpected earnings divided by security price at time t. Unexpected earnings capture the earnings variation that has not yet been incorporated into the investors' belief. There are two approaches commonly used to compute unexpected earnings. The first approach estimates expected earnings based on a time series model.. 政治大. The second approach uses analysts' forecasts as a proxy for market expectation.. 立. Empirical evidence suggests that analysts' forecasts are a better proxy for market. ‧ 國. 學. expectation of earnings (Livnat and Mendenhall 2006).. ‧. However, developments in the field of behavioral finance, however, are increasingly spurring accounting researchers to question this view (Lee 2001; Daniel. y. Nat. er. io. sit. 2004; Hirshliefer et al. 2003; Hirshliefer et al. 2009). Lee (2001), for instance, challenges the traditional assumption that the price adjustment to new information is. n. al. Ch. i n U. v. always rational and instantaneous, and calls for more work that incorporates the. engchi. findings of behavioral finance into capital market research in accounting. Therefore, better understand the process by which stock prices incorporate accounting information.. 10.

(18) 2.2 SOURCES OF FINANCIAL INFORMATION Markets thrive on new information and this data is constantly streaming. There are many possible types of information can move a stock price, such as rumors, insider information, and eavesdropping. Amongst all, financial news articles are considered to be a more stable and trustworthy source. Many studies rooted in the Information Retrieval (IR) methodologies have examined the use of textual news articles. While most are interested in studying the relationships between articles in a pure IR point of view, few focus on the domain of financial text as it relates to quantifiable stock market data.. 立. 政治大. The information content of accounting numbers can be measured by which. ‧ 國. 學. security prices change in response to the announcement of financial statements. Cecchini (2010) aims to create a new method that can be used by accounting and. ‧. finance researchers who want to define a dictionary of key terms from financial text. Nat. sit. y. that are associated with an event or a state of nature. His method would help to automate. n. al. er. io. the task of finding a dictionary of many key words and rank these words in order of. i n U. v. discriminatory power. The dictionaries can also be quantified into token counts, which. Ch. engchi. can be used in analyses for the benefit of augmenting prior research models that utilized only quantitative data. He creates dictionaries of keywords that can help predict fraud and bankruptcy, based on text from the Management Discussion and Analysis Section (MD&A) of the 10-K reports. The results show that textual information is competitive with quantitative information. His research also shows that text and quantitative information, when combined together, give the best results, showing that complementarities exist between text and quantitative data. Financial reporting measurement is mostly focused on transaction-based recognition. Therefore, large amount of information embedded is historical in nature. 11.

(19) However, security prices reflect both current and future earnings information that are available to the market (Kothari et al. 1992). In that case, value-relevant information may have been mirrored on the prices before related financial reports are issued. However, most of these studies failed to explore other information sources that may have contained future earnings information. Financial news is one such information source (Ma et al. 2009; Muntermann 2009). Before being officially announced, information related to a firm's earnings is sometimes disclosed through various news articles. Investors could incorporate the information into their. 政治大. determination of reasonable security prices. The return-earnings relation can be better. 立. modeled if relevant information from financial news could be captured (Chen et al.. ‧ 國. 學. 2011).. ‧. In a different type of study, by examining the textual role that financial news articles may play in determining levels of company cooperation and connectedness. y. Nat. er. io. sit. among one another. Ma et al. (2009) constructs an intercompany network, which employs social network analysis techniques to identify a set of attributes from the. n. al. Ch. i n U. v. network structure, and feeds the attributes to machine learning methods to predict the. engchi. company revenue relation (CRR) based on two companies' relative quantitative financial data. The strength of the links can provide an idea of the amount of connectedness between companies. The act of correlating financial news articles to a quantifiable price movement is a difficult task, even when the information within a financial news article can have a visible impact on price (Gidófalvi et al. 2001; Lavrenko et al. 2000; Mittermayer 2004; Wuthrich et al. 1998).. 12.

(20) 2.3 CONTENT ANALYSIS Prior research documents a significant increase in the number of words used by managers in earnings press releases between 1980 and 1999 (Francis et al. 2002b). Davis’s (2012) research finds out by the end of 2003, press release length has an increase of 90 percent from 1998. This increase in earnings press release length is due, in part, to the provision of additional qualitative disclosures as examined in prior research (Hoskin et al. 1986; Francis et al. 2002b). Yet, prior studies are limited to laborintensive, judgmental measures of both the thematic content of officers’ comments and. 政治大. other qualitative disclosures as well as the related future-earnings implications of those disclosures.. 立. ‧ 國. 學. Content analysis, on the other hand, has been employed to provide insights into how specific wording in statements may affect stock prices. Content analysis enables. ‧. quantification of text based information. A recent stream of literature utilizes tools. Nat. sit. y. relatively new to finance, but well establishes in a wide variety of other disciplines,. n. al. er. io. allows a better understanding of the information content of words.. Ch. i n U. v. Tetlock (2008) incorporates text-based content analysis. He quantifies news. engchi. coverage about S&P 500 companies in the Wall Street Journal (WSJ) before earnings announcements and models its interaction with the return-earnings relation with content analysis method. His results show that news coverage decreases the information content of unexpected earnings and thus leads to a lower earnings response coefficient (ERC) for firms with higher news coverage. The results suggest that financial news articles may play an important role in conveying value-related information to the markets. Engelberg (2008) applies the method outlined in Tetlock (2008) to measure the qualitative content of Dow Jones News Service stories about firms’ earnings 13.

(21) announcements to analyze the link between “soft” information and equity prices. He finds support for Tetlock (2008) and demonstrates a relation between linguistic media content and the stock market. Davis et al. (2008), Henry (2008), and Sadique et al. (2008) analyze the information content of the wording of official earnings announcements. Henry (2008) quantifies tone using computer-based analytical tools to measure the frequency of positive and negative words found in earnings press releases. He finds that the tone of earnings press releases influences investors’ reaction to earnings in a sample of. 政治大. technology firms. Davis et al. (2008) confirm Henry’s findings across a broad sample. 立. of firms, with a significant positive association between levels of optimistic tone in. ‧ 國. 學. earnings press releases and both future return-on-assets and the initial market response. They suggest that managers use tone in their press releases to provide investors with. ‧. information about expected future firm performance.. y. Nat. er. io. sit. The application of content analysis, in general, and textual tone analysis, in particular, has gained increased attention in disclosure research in recent years. Davis. n. al. Ch. i n U. v. et al. (2008), Demers and Vega (2008), Henry (2008), and Sadique (2008) extract the. engchi. tone of the wording of quarterly earnings press releases and relate it to things such as stock returns, volatility, and firm performance. All of these textual analysis studies find statistical significance for the linguistic tone of disclosure documents, suggesting that managers in their word choices convey relevant information. However, none of these studies examines the additional disclosures embedded in the language of the financial reports combined with financial news articles. This study aims to find out the information content of both financial news articles and financial reports. By analyzing two main data sources for decision making in 14.

(22) today’s world, this study hopes to discover information presented in financial news articles and footnotes to financial statements do really connect to the stock market. In addition, decides whether footnotes to financial statements provide more information than financial news articles do.. 立. 政治大. ‧. ‧ 國. 學. n. er. io. sit. y. Nat. al. Ch. engchi. 15. i n U. v.

(23) 2.4 DISSCUSSION Davis et al. (2012) examine whether a measure of language used throughout an earnings press release is associated with future firm performance and generates a market response. By analyzing a sample of approximately 23,000 quarterly earnings press releases published on PR Newswire between 1998 and 2003, they construct a measure of net optimistic language using an established, textual-analysis software program (DICTION), which counts words characterized as optimistic and negative. They later find that levels of net optimistic language in earnings press releases are. 政治大 suggest that managers use language in earnings press releases, at least to some extent, 立 predictive of firm performance in future quarters. They also interpret the evidence to. to communicate information about expected future firm performance to the market.. ‧ 國. 學. These results suggest that market participants consider at least some level of. ‧. managers’ language to be a credible signal.. Nat. sit. y. Chen et al. (2011) investigate the influence of news coverage on the ERC, which. n. al. er. io. measures the information content of earnings. They collect news articles in the Wall. i n U. v. Street Journal from August 1999 through February 2007 to construct measures for. Ch. engchi. news coverage on S&P 500 companies. Combined with data from classical financial databases such as IBES, Compustat and CRSP, they are able to study the effect of news coverage on earnings surprise. Empirical results indicate that news coverage has a significantly negative effect on the ERC; higher news coverage decreases the information content of earnings and reduces market responses to unexpected earnings. Mian et al. (2012) examine cross-sectional differences in the stock price response to earnings. The study takes a time-series approach and helps answer the call for more work that incorporates the findings of behavioral finance in capital market research in accounting (Lee 2001). Using a recent proxy of investor sentiment constructed by 16.

(24) Baker and Wurgler (2006, 2007), they examine how the stock price reaction to earnings news varies with investor sentiment. The results show that the stock price response to good earnings news is higher during periods of high sentiment, whereas stock price reaction to bad earnings news is higher during periods of low sentiment. This is consistent with the notion that as sentiment increases, investors place increasingly optimistic valuations on the incremental cash flows embedded in earnings announcements. Davis et al. (2012) investigate whether the proportion of total negative language. 政治大. reported in earnings press releases is associated with the intensity of managers’. 立. strategic reporting incentives. They expect that managers who are more sensitive to. ‧ 國. 學. the stock price effects of information disclosed face greater incentives to report strategically, and predict that they will report a lower proportion of total negative. ‧. language in their earnings press releases relative to the MD&A. They find that. Nat. sit. y. managers of high-growth firms and firms that exactly meet or just beat earnings. n. al. er. io. benchmarks in the current quarter report a lower proportion of total negative language. i n U. v. in their earnings press releases. They also find evidence consistent with managers. Ch. engchi. who habitually meet or just beat analysts’ earnings forecasts reporting a lower proportion of total negative language in their earnings press releases. Tama-Sweet (2014) investigates the association between changes in the optimistic tone of earnings announcements and CEOs' equity sales. Using a sample of approximately 20,000 firm-quarters from 1998 to 2007, he finds a positive relation between changes in earnings announcement tone and CEO equity sales. The positive relation is mitigated by SOX and by litigation risk. Consistent with the literature, he also finds a positive relation between changes in optimistic tone in earnings announcements and abnormal returns. 17.

(25) This study implements the ideas of prior works and adds different sources of information and method to conduct content analysis. Some of the variables are chosen and some are not because of different databases used and availability of data. The summary of related works is in the following table.. 立. 政治大. ‧. ‧ 國. 學. n. er. io. sit. y. Nat. al. Ch. engchi. 18. i n U. v.

(26) The following table discusses the literature review mentioned in section 2.4, including lists of author, hypothesis, model, variable definition and adopted variable in this study. Table 2-1. Summary of Related Research. 政治大 CAR     SURP 立  BEAT   LOSS   DET _ FS. Author. Hypothesis. Model. Beyond the Numbers: Measuring the Information Content of Earnings Press. Ceteris paribus, the unexpected level of net optimistic language in earnings press. CAR, SURP,   5 DIV _ INCi   6 NONREC _ POSi   7 NONREC _ NEGi BEAT,  8 PM i   9 ATi  10 DAi  11BM i  12 ( NETOPTi  LAGNETOPTi ) LOSS,   13 j IDij   14kYEAR ik   i j k DET_FS, CAR is the size-adjusted cumulative abnormal return over the three-day window DIV_INC,. Release Language*. Davis, A. K., Piger, J. M., & Sedor, L. M. (2012).. releases is positively associated with market returns around the earnings announcement. centered on the earnings press release date. SURP is the difference between actual earnings and the most recent consensus analyst earnings forecast made prior to the earnings announcement, scaled by stock price measured at the beginning of the current quarter. BEAT is a dummy variable, to be 1 if announced earnings for the current quarter met or exceeded analysts' expectations and 0 otherwise. LOSS is a dummy variable, to be 1 if earnings are negative and 0 otherwise. DET_FS is a dummy variable, define BS_D and SCF_D to be 1 if these word counts are greater than two for balance sheet -. date.. related words and greater than one for statement of cash flow related words, respectively. If the sum of BS_D and SCF_D is greater than 0 and include it in our analyses as a control for the presence of detailed financial statements. DIV_INC is a. 0. 1. i. 2. i. 3. i. 4. i. 學. ‧. ‧ 國. i. Adopted Variable. n. er. io. sit. y. Nat. al. Ch. engchi. 19. i n U. v. PM, AT, DA, BM, NETOPT.

(27) Model. Adopted Variable. dummy variable, to be 1 if the dividend change is positive. NONREC_POS is a dummy variable, to be 1 if the net effect of these components on current period earnings is positive and zero otherwise. NONREC_NEG is a dummy variable, to be 1 if the net effect of these components on current period earnings is negative and zero otherwise.. 政治大. PM is profit margin. AT is asset turnover. DA is total liabilities scaled by total assets (both measured at the end of the current quarter). BM is book-to-market ratio as the book value of equity scaled by the market value of equity (both measured at the end of the current quarter). NETOPT is the difference between the percentage of optimistic words and the percentage of pessimistic words. ID is an indicator variable, taking the value 1 if the press release represented in observation i is for a firm in the jth two-digit SIC industry and 0 otherwise. YEAR is an indicator variable, taking the value 1 if the. 立. 學. ‧. io. sit. y. Nat. earnings press release represented in observation i was released in year k and 0 otherwise.. n. al. er. Hypothesis. ‧ 國. Author. Ch. engchi. 20. i n U. v.

(28) Author. Hypothesis. Model. Adopted Variable. Giving context to accounting numbers: The role of news. Firms with higher news coverage before their earnings. CARit ,0, 2   a  b1UEit  b2 DUEitUEit  g1FFalpha it  g 2 log( Sizeit )  g3 log( BM it ). coverage. Chen, K. T., Lu, H. M., Chen, T. J., Li, S. H., Lian, J. S., & Chen, H. (2011).. announcements are associated with lower ERCs.. CARit ,0, 2   a  b1UEit  b2 DUEitUEit  c1 NewsFrequencyit  c2 NewsFrequencyitUEit. CAR, UE,  g 4 log( STurnoverit )  h1FFalpha itUEit  h2 log( Sizeit )UEit  h3 log( BM it )UEit DUE,  h4 log( STurnoverit )UEit  eit Size,. 立. 政治大.  g1FFalpha it  g 2 log( Sizeit )  g 3 log( BM it )  g 4 log( STurnoverit ). ‧ 國. 學.  h1FFalpha itUEit  h2 log( Sizeit )UEit.  h3 log( BM it )UEit  h4 log( STurnoverit )UEit  eit. ‧. CAR is the cumulated abnormal return of firm i over the trading day window [0, 2] relative to an earnings announcement day t. UE is the unexpected earnings computed. n. al. er. io. sit. y. Nat. based on the difference between the realized earnings and analysts' forecasts, divided by the stock price of firm i at day t. DUE is a dummy variable, where is 1 if UE is positive and 0 if UE is negative. FFalpha is the estimated intercept in the Fama–French three-factor model used to compute abnormal returns. SIZE is firms' market value at time t. BM is book-to-market ratio at time t. STurnover is share turnover rate during the reporting quarter associated with time t. NewsFrequency is Number of news articles that have mentioned firm i during the trading day window [−20, −1] relative to the. Ch. engchi. earnings announcement date t.. 21. i n U. v. BM, STurnover, NewsFrequency.

(29) 治政   UEDown  Sent   NonlUp  大 UEUp  MktPE 立   UEDown  MktPE     NonlDown. RETit   0  1Downit   0UEUpit  1UEUpit  Sentt 1   0UEDownit 1. t 1. it. 2. it. 2. 3. it. it. 3. t 1. it. t 1. it. Earnings is the primary earnings per share of firm i for quarter q. RET is the cumulative abnormal return for firm i during the earnings announcement event window from day 1 to +1 centered on the earnings announcement date in calendar month t. Up is a dummy variable, equals 1 if the unexpected earnings is positive, and 0 otherwise. Earnings is. Nat. the primary earnings per share of firm i for quarter q. P is the price per share at the end of quarter q. Down is a dummy variable, equals 1 if the unexpected earnings is negative, and 0 otherwise. Sent is the sentiment index developed by Baker and Wurgler (2006, 2007), which is available on a monthly basis. NonlUp is the square of UEUp. NonlDown is the square of multiplied by -1. MktPE is market price-to-earnings ratio.. y. sentiment, whereas the ERC of bad news is higher during periods of low sentiment.. Piq. io. sit. earnings news. Mian, G. M., & Sankaraguruswamy, S. (2012).. UEiq . Earningsiq  Earningsiq  4. n. al. er. The ERC of good news is higher during periods of high. Adopted Variable. ‧ 國. Investor sentiment and stock market response to. Model. ‧. Hypothesis. 學. Author. Ch. engchi. 22. i n U. v. RET, Down, Up, MktPE.

(30) Author. Hypothesis. Model. Managers’ Use of Language Across Alternative. Whether the unexpected amount of pessimistic. CAR _ PRi   0  1SURPi   2 BEATi  3 LOSSi   4UNEXP _ PESS _ PRi. CAR, SURP,   5 INDij    6YEARik   i j k BEAT, CAR is the cumulative abnormal return relative to the firm’s size-decile portfolio over LOSS,. Disclosure Outlets: Earnings Press Releases versus MD&A*. Davis, A. K., & TAMA‐. language in a firm’s earnings press release and MD&A is negatively related to market returns around. the three-day window centered on the earnings press release date. SURP is the difference between actual earnings and the most recent consensus analyst earnings forecast made prior to the earnings announcement, scaled by stock price measured at the beginning of the current quarter. BEAT is a dummy variable, to be 1 if announced earnings for the current quarter met or exceeded analysts' expectations and 0 otherwise. LOSS is a dummy variable, to be 1 if earnings are negative and 0 otherwise. UNEXP_PESS_PR is unexpected pessimistic language in the press release from a. the disclosure date.. random walk model. IND and YEAR is year and industry.. 立. 政治大. ‧. ‧ 國. 學. n. al. er. io. sit. y. Nat. SWEET, I. S. H. O. (2012).. Adopted Variable. Ch. engchi. 23. i n U. v.

(31) Author. Hypothesis. Model. Adopted Variable. Changes in earnings announcement tone and. There is a positive association between. CAR_SALE, OPTIMISM_PR,   j b5 j QTRitj   k b6 kYEARitk   l b17l INDUSTRYitl  e SURP, CAR_SALE is the abnormal return from one day before the earnings announcement BEAT,. insider sales. Tama-Sweet, I. (2014).. changes in the optimistic tone of earnings announcements and CEOs' subsequent equity sales.. until the day of a CEO equity sale. OPTIMISM_PR use DICTION_PC as dependent LOSS variable in tests. SURP is the difference between actual earnings and the most recent consensus analyst earnings forecast made prior to the earnings announcement, scaled by stock price measured at the beginning of the current quarter. BEAT is a dummy variable, to be 1 if announced earnings for the current quarter met or exceeded analysts' expectations and 0 otherwise. LOSS is a dummy variable, to be 1 if earnings are negative and 0 otherwise. QTR, YEAR, and INDSTRY are indicators for quarter, year,. CAR _ SALEit  a  b1OPTIMISM _ PRit  b2 SURPit  b3 BEATit  b4 LOSSit. 立. 政治大. ‧. ‧ 國. 學. y. sit. io. n. al. er. Table source: this research. Nat. and industry.. Ch. engchi. 24. i n U. v.

(32) 3. REASERCH METHOD 3.1 HYPOTHESIS DEVELOPMENT In this section, there are two hypotheses discussed and provided as follows: 3.1.1 Financial Reporting and the Market Response Because of financial news articles mostly provide timely information, they are usually presented in a shorter and simplified format to convey the most important news to the public. But financial reporting are not presented in the same way, there are financial reporting standards, such as International Financial Reporting Standards. 政治大. (IFRS), require the information should be conveyed and format should be laid out. 立. clearly and formally. As a rule of thumb, most financial reports contained four. ‧ 國. 學. statements and footnotes to detailed textual and numerical explanation. By referring to. ‧. the footnotes to financial statements, people can better understand the calculation why and how the numbers occur.. y. Nat. io. sit. Asquith et al. (2005) incorporate the contents of analyst reports in their entirety. n. al. er. rather than just the individual summary elements such as the stock recommendation. Ch. i n U. v. and reach to the conclusion that analysts both provide new information and interpret. engchi. previously released information. “In the end, stock ratings and target prices are just the skin and bones of analysts’ research. The meat of such reports is in the analysis, detail, and tone’’ (‘‘When a stock’s rating and target collide’’, Business Week Online, April 25, 2002). Therefore, this study scores part of the footnotes to financial statements based on certain scoring criteria (explained in later chapter) and tries to find possible connections between annual financial reports and the stock market. In addition, this study expects a positive relationship between the information abundancy of the notes to the financial statements and stock price movement during the financial statements announcement period. 25.

(33) Hypotheses 1: Footnotes to financial statements provide information and are associated with market returns around annual financial reports released dates.. 立. 政治大. ‧. ‧ 國. 學. n. er. io. sit. y. Nat. al. Ch. engchi. 26. i n U. v.

(34) 3.1.2 The Relationships between News Articles and Stock Price Movement Can stock market responses be better explained with the aid of financial news articles? This study aims to investigate the information content gap between financial news articles and footnotes to financial statements by referring to the stock market response. Financial news articles contain timely updates of information on firm value. Since annual financial reports have been audited to provide confidence of the past performance to the financial report users but usually have time lag because of time required to conduct the audit process, financial news articles can provide more timely. 政治大 financial reports and financial news articles aim to provide, the “price lead earnings” 立. information ahead of financial reporting to the market. Since different goals for annual. viewpoint suggests that investors would have incorporated the information from. ‧ 國. 學. financial reporting into stock prices before the financial reporting announcements. ‧. (Beaver et al. 1980). Because prices lead earnings, the specification using the earnings-. sit. y. Nat. level-deflated-by-price variable in a price-earnings regression is ‘better’, in terms of. io. er. bias in the estimated earnings response coefficient and explanatory power, than specifications using earnings-change-deflated-by-price and earnings-deflated-by-. al. n. v i n An C accurate proxy for U h e n g c h i unexpected. lagged-earnings variables.. earnings, however,. outperforms the earnings-level- and earnings-change-deflated-by-price specifications (Kothari 1992). In addition, compared to financial reporting, financial news articles are easily catchable and readable; takes financial news articles alone may explained some parts of the market reaction to a certain level. Odean (1998) suggests that the tendency of investors to hold losing investments too long and sell winning investments too soon. Investors demonstrate a strong preference for realizing winners rather than losers. That is, when there are chances of winning, investors will act on the information. On the other hand, when a bad information suggests chances of losing, investors may refuse to change his position. As a result, this study expects a positive relationship between 27.

(35) financial news reporting and market reaction especially with positive news sentiment.. Hypotheses 2: Positive sentiment and information content in financial news articles combined with financial reporting around annual financial reports released dates is more associated with market responses than footnotes to financial statements alone.. Figure 3-1 schematically shows the hypothesized relationships in the research model. This study uses event study approach. An Event study is a statistical method to. 政治大 a merger between two business entities can be analyzed to see whether investors believe 立 assess the impact of an event on the value of a firm. For example, the announcement of. the merger will create or destroy value. The basic idea is to find the abnormal return. ‧ 國. 學. attributable to the event being studied by adjusting for the return that stems from the. ‧. price fluctuation of the market as a whole. Measured by the cumulative abnormal return. sit. y. Nat. (CAR), the thick arrow indicates the relationship between the footnotes to financial. io. er. statements and cumulated abnormal returns and the upper arrow indicates the relationships between the footnotes to financial statements combined with information. al. n. v i n C h and cumulated U content in the financial news articles abnormal returns. Hypothesis 1 engchi asserts the relationships between information content in footnotes to financial statements and market reaction. Hypothesis 2 asserts that sentiment and information content in financial news articles strengthen the relationship with the stock price movement than merely footnotes to financial statements alone. That is, test for the information content gap between news articles and annual financial reports.. 28.

(36) Figure 3-1. Research Model and Hypotheses. Financial News H2 Financial Reporting. Hypotheses 1:. 立. H1. Cumulated Abnormal Return. 政治大. ‧ 國. ‧. Hypotheses 2:. 學. Footnotes to financial statements provide information and are associated with market returns around annual financial reports released dates.. sit. y. Nat. Positive sentiment and information content in financial news articles combined with financial reporting around annual financial reports released dates is. io. n. al. er. more associated with market responses than footnotes to financial statements alone.. Ch. engchi. 29. i n U. v.

(37) 3.2 DATA AND SAMPLE SELECTION This study focuses on the Taiwan capital market and investigates the differences between news and annual financial reports. In order to examine research questions, this study not only collects financial news articles of selected companies but also annual financial statements as well. The detailed processes of data collection are as follows. The firm list of listed companies of 2013 is taken from the website of Taiwan Stock Exchange (TSE), for an initial sample of 882 companies. Companies in the Finance and Insurance Sector are excluded because of different capital structures and industry. 政治大. characteristics. Annual financial reports and each released dates of listed companies of. 立. 2013 are taken from the website of Market Observation Post System (MOPS). The post-. ‧ 國. 學. release dates of annual financial reports published during a 6-day working-day periods are collected, that is, [0, +5] working days, as the event day being the release day,. ‧. suggesting that the week after each annual financial reports are released would have an. Nat. er. io. sit. y. impact on that company’s stock prices.. Most of the data sources used in this study are from Taiwan Economic Journal. al. n. v i n database (TEJ), as most of theCinformation of public h e n g c h i Ucompanies is included in the. database. For each [0, +5] working days window, which depend on the publishing date for each company, this study collects a number of accounting and financial market variables. As dependent variables for our tests of Hypothesis 1 and Hypothesis 2, some measures of market returns around the date of the release date of the annual financial reports are required. To measure market returns, this study defines CAR as the cumulated abnormal return over the six-day working day window starting on the financial reports published date (0 working day to +5 working day). For each annual financial reports announcement dates, the regression coefficients estimated over the trading day window [−200, −21] relative to an earnings announcement day were used 30.

(38) to compute abnormal returns during the [0, 5] window. Excluding 20 trading days before financial report announcements avoids the confounding of the estimated coefficient with annual financial report announcements. The sum of the abnormal returns during the [0, 5] event window is the dependent variable. For control variables implemented for regression, this study collects variables likely associated with the market response, including variables meant to proxy for quantifiable information in the news articles other than language. This study measures the current quarter earnings surprise, SURP as the difference between annual actual. 政治大. earnings and the most recent consensus analyst earnings forecast made prior to the. 立. earnings announcement, scaled by stock price measured at the beginning of the year.. ‧ 國. 學. The analyst earnings forecasts are calculated as taking average of all forecasts of annual after-tax earnings for 2013, which ranging from 1 to 50 estimates per company. This. ‧. study defines dummy variable BEAT to be 1 if announced earnings for the year of 2013. Nat. er. io. sit. y. meet or exceeded analysts’ expectations (i.e., when SURP ≥ 0) and 0 otherwise. This study constructs proxies for the concurrent disclosures made with earnings. n. al. Ch. i n U. v. press releases found to be significant in prior research, which indicates that the presence. engchi. of detailed financial statements, including the announcement of dividend increases and the reporting of nonrecurring earnings components relates to earnings (Hoskin et al. 1986; Francis et al. 2002b). Including the announcement of dividend increases as proxies for other quantifiable information in earnings press releases, this study believes that it is likely to be associated with market returns around the annual financial reporting announcement period. However, because of the adoption of IFRS since 2013, the reporting of nonrecurring earnings components are no longer required for disclosure. Therefore, we do not include the reporting of nonrecurring earnings components as prior research does. 31.

(39) In order to identify the presence of detailed financial statements, this study counts words and phrases in financial news articles. This study uses word lists of terms of accounts and wording used in the annual financial reports, such as “total liabilities” or ‘‘from financing activities”, related to the presence of a detailed balance sheet or a statement of cash flows, respectively. To control for dividend increases, compared to prior year, this study uses DIV_INC representing for a firm that dividend declared in 2014 greater than prior year. Then calculates the amount and defines DIV_INC to be 1 if the dividend change is positive. DIV_INC is 0 if the dividend change is less or equal. 政治大 is different to prior research (Davis et al. 2012). In the paper, they define DIV_INC as 立 to zero or there was no dividend announcement made in 20141. This variable definition. dividend declaration made during the three-day window surrounding the earnings press. ‧ 國. 學. release date. This study adopts different meaning and calculation for that since all. ‧. dividend declaration made didn’t match our six-day window, mostly would be. y. sit. io. n. al. er. stock market.. Nat. announced later in 2014 but still thinks dividends can be a source of information to the. i n U. v. Finally, as additional measures of quantifiable information, this study collects. Ch. engchi. book-to-market ratio (BM) as the book value of equity scaled by the market value of equity (both measured at the end of the cur-rent year). This study also collects current year sales (REV) and uses its natural logarithm (LOGREV). This study constructs a variable, FILELAG, equal to the number of days between the annual financial reports release date of each company and average companies to control for the potential for managers to choose when to release skeptically. After deleting missing data, the numbers of remaining companies and news are 90 1. Though we discuss 2013 financial reports, related information should be announced before March 31, 2014. Therefore, we decide whether there is an increase in dividend by announcement made in 2014. 32.

(40) and 212, respectively. As described below, the calculation and summary of the samples are summarized in Table3-1 and the industry breakdown is summarized in Table 3-2.. Table 3-1. Results of Selection of Samples 2013 listed companies. 882. Companies with news. 161. Delete companies without sufficient data. 71. Total. 90. 立. 政治大. Table 3-2. Industry Breakdown. ‧ 國. Industry. Percentage. 12. Manufacture of Wearing Apparel and. 1. 1.11. 2. 2.22. 1. 1.11. 2. 2.22. sit. io. Clothing Accessories. n. al. ‧. Manufacture of Beverages. Nat. 09. y. of Firms. er. of SIC. Number. 學. Two-digit. v ni. 25. Manufacture of Fabricated Metal Products. 08. Manufacture of Food Products. 22. Manufacture of Plastics Products. 1. 1.11. 29. Manufacture of Machinery and Equipment. 4. 4.44. 38. Waste Collection, Treatment and Disposal. 1. 1.11. Ch. engchi U. Activities; Materials Recovery 18. Manufacture of Chemical Material. 7. 7.78. 20. Manufacture of Pharmaceuticals and. 1. 1.11. 1. 1.11. Medicinal Chemical Products 21. Manufacture of Rubber Products 33.

(41) Two-digit. Number Industry. Percentage. of SIC 23. of Firms Manufacture of Other Non-metallic Mineral. 2. 2.22. 27. 30.00. 16. 17.78. 政治大 Construction of Buildings 立. 1. 1.11. 1. 1.11. Retail Trade. 7. 7.78. 2. 2.22. 1. 1.11. 1. 1.11. 1. 1.11. 1. 1.11. 1. 1.11. Products 26. Manufacture of Electronic Parts and Components. 27. Manufacture of Computers, Electronic and Optical Products. 41. 學. 47-48. Manufacture of Motor Vehicles and Parts. ‧ 國. 30. Water Transportation. 51. Air Transport. 55. Accommodation. 56. Food and Beverage Service Activities. 61. Telecommunications. 62. Computer Systems Design Services. 64. Financial Intermediation. 1. 1.11. 67. Real Estate Development Activities. 6. 6.67. Wholesale Trade. 1. 1.11. 90. 100.00. n. er. io. al. Ch. sit. y. Nat. 45-46. ‧. 50. engchi U. Total. 34. v ni.

(42) 3.3 CONTENT ANALYSIS METHOD 3.3.1 Approaches to Measure Notes to Financial Reports and Scoring Notes to the annual financial reports are also scored by seven criteria pointed out by the Financial Supervisory Commission (FSC), as the competent authority responsible for development, supervision, regulation, and examination of financial markets and financial service enterprises in Taiwan. These points brought out by FSC are meant to provide guidance to help listed companies disclose more detailed notes to the annual financial reports after the adoption of IFRS since 2013. For test of. 政治大 notes to financial statements, ranging from 1 to 7. This scoring is constructed manually, 立. Hypothesis 2, this study defines a variable NOTESCORE to represent the scoring of. ‧ 國. 學. because the format and ways of presentation in the annual financial reports are not machine-readable without specific modifications. The manual scoring criteria of annual. ‧. financial reports are summarized as follows:. sit. y. Nat. a) Valuation of Investment Property.. io. al. er. b) Material Component of Property, Plant, and Equipment.. n. c) Benefits to the Management.. Ch. engchi. d) The Impact of IFRSs Implementation.. i n U. v. e) The Impact of the New IFRSs Standard. f) Three level of fair value measurement of financial instruments. g) Disclosure of Material Accounting Measurement. The examples of the scoring criteria are presented in the appendix 1.. 35.

(43) 3.3.2 Approaches to Measure News articles Language and Scoring This study collects news articles published by Knowledge Management Winner (KMW) database between [0, +5] working days window. KMW is marketed and maintained by the largest financial news group in Taiwan and is the largest news database composed of China Times, China Times Express and Commercial Times. This research searches financial news articles using company names and codes from TSE as two key words 2 and further screens out the news articles unrelated to the financial reporting. Only news articles related to financial information about target companies,. 政治大 promotional and customer-related and other non-financial-related announcements and 立 industry analysis or stock performances and forecasts are chosen. Those relate to. information are eliminated.. ‧ 國. 學. In the beginning of the news articles, the database name, the publish date and time,. ‧. the title of the article and the author are shown. Then the text body follows and this. Nat. sit. y. study focus only on the text body. This study defines dummy variables BS_D and. n. al. er. io. SCF_D to be 1 if these word counts are greater than one for balance sheet related words. i n U. v. and greater than one for statement of cash flow related words, respectively. Then, this. Ch. engchi. study constructs a dummy variable, DET_FS, equal to 1 if the sum of BS_D and SCF_D to be equal to or greater than 1. The word lists used are provided in appendix 2. This research employs computerized textual-analysis tools to measure language throughout each of the news articles in the sample. Two measurements of financial news article language are from prior research of the Ministry of Science and Technology project. In the project, the team of professors and three graduate students read the textual content of financial news articles and score each news report based on the amount of information disclosed in the financial news. They score each news report based on the. 2. Use 4-digit numbers and 2-character abbreviated names on TSE to search news. 36.

(44) amount of extra information disclosed in the financial news articles yet in financial reporting and construct word lists to count optimistic, negative and scoring criteriarelated words in the full text of each news articles. The building processes of the word lists are manually select positive, negative and scoring-related words; include the word lists into computer software to act as features to automatically analyze financial news articles; and compare the results of both manual and automatic analysis. To increase validity of the word lists, they repeat the building processes many times to reduce subjectivity.. 政治大. This study implements the word lists to analyze the sentiment and information. 立. content of financial news articles. The scoring of financial news articles is between 1. ‧ 國. 學. and 5. The word lists of positive, negative and scoring criteria are presented in the appendix 3 and 4, and scoring criteria are summarized as follows:. ‧. a) Financial news corresponds to the financial reports and extends the reports.. Nat. sit. y. b) Financial news corresponds to the financial reports and provides the cross-year. n. al. er. io. or cross-firm or cross-industry analysis.. i n U. v. c) Financial news corresponds to the financial reports and provides further. Ch. engchi. information to explain the notes in the reports.. d) Financial news corresponds to the financial reports and provides opinions on the effect stock trends. e) Financial news corresponds to the financial reports and provides further information about earnings, investments, merger and acquisition, management, headquarters changes, and accounting method changes. For tests of Hypothesis 2, this study first defines a dummy variable TONE, using the word lists presented in the appendix 3 to count both optimistic (P) and negative (N) words in the full text of each financial news articles. TONE is calculated as the 37.

(45) differences between frequency counts of optimistic words and negative words separately and divided by the sum of frequency count of positive and negative words. That is, (P－N)/(P＋N). By dividing by the sum of frequency count of positive and negative words, TONE is the proxy for sentiment, which eliminates the common biases for financial news articles mostly containing positive words. Second, this research defines a variable TOTALCORE to represent the information abundancy by adding the scores of footnotes to financial statements and financial news articles. For all the content analysis variables, TONE and TOTALCORE, this study will take average of. 政治大. them if a company has news articles more than one.. 立. ‧. ‧ 國. 學. n. er. io. sit. y. Nat. al. Ch. engchi. 38. i n U. v.

(46) 3.4 REGRESSION MODEL 3.4.1 Test of Hypothesis 1 Hypothesis 1 predicts that notes to the financial statements are associated with market returns. This test estimates a multivariate regression model on which CAR (the cumulated abnormal return over the six-day window beginning from the annual financial reports announcement date) is regressed on NOTESCORE and other variables. NOTESCORE is built to evaluate whether the additional and supplementary information disclosed in the notes to the financial statements can provide explanation. 政治大 CAR     SURP 立   BEAT   DIV _ INC   BM    LOGREV. for the stock price movement. The formal specification model is as follows: i. 0. 1. i. 2. i. 3. i. 4. i. ‧ 國. 學.  6 FILELAGi  7 NOTESCOREi   i. 5. i. (1). ‧. where i indexes the company observation and all variables are defined in Section 3.2 and Table 3-3.. n. er. io. sit. y. Nat. al. Ch. engchi. 39. i n U. v.

(47) Table 3-3. Variable Definition Variable. Definition. CAR. The cumulated abnormal return over the six-day working day window starting on the financial reports published date. The difference between TEJ annual actual earnings and the most recent consensus analyst earnings forecast made prior to the earnings announcement, scaled by stock price measured at the beginning of the year. Equal to 1 if announced earnings for the year of 2013 meet or exceeded analysts’ expectations (i.e., when SURP ≥ 0) and 0 otherwise.. SURP. BEAT. DIV_INC. Equal to one if the dividend change is positive, compared to prior year, and is 0 if the dividend change is less or equal to zero or there was no dividend announcement made in 2014.. 政治大 Book-to-market 立 ratio as the book value of equity scaled by the. 學. y. sit. n. al. er. Equal to 1 if the sum of BS_D and SCF_D to be equal to or greater than 1 and is 0 otherwise (BS_D and SCF_D to be 1 words identified are greater than one for balance sheet related words and sentences greater than one for statement of cash flow related words and sentences, respectively). The differences between frequency counts of optimistic words and negative words separately and divided by the sum of frequency count of positive and negative words. That is, (P－N)/(P＋N).. io. DET_FS. Nat. NOTESCORE. ‧. LOGREV FILELAG. market value of equity (both measured at the end of the cur-rent year). Current year sales (REV) and use its natural logarithm. Number of days between the annual financial reports release date of each company’s and average companies’. The scoring of financial news articles, depending on the abundancy of information discussed in the news articles.. ‧ 國. BM. TONE. TOTALSCORE. Ch. engchi. i n U. v. The sum of the scores of footnotes to financial statements and financial news articles.. 40.

(48) 3.4.2 Test of Hypothesis 2 Hypothesis 2 predicts that positive sentiment in news after annual financial reports released is associated with market returns around the post-released dates. This test estimates a multivariate regression model on which CAR (the cumulated abnormal return over the six-day window beginning from the annual financial reports announcement date) is regressed on DET_FS, TONE, TOTALCORE and other variables. To measure the incremental market responses to financial news articles language, this study includes control variables in analysis that are known to have information content,. 政治大 articles, this study includes DIV_INC, BM and LOGREV in the model. The formal 立 namely SURP and BEAT. To proxy for other quantifiable information in financial news. specification model is as follows:. ‧ 國. 學. CARi  0  1SURPi   2 BEATi  3 DIV _ INCi   4 BM i   5 LOGREVi. ‧.  6 FILELAGi  7 NOTESCOREi  8 DET _ FS i  9TONEi  10TOTALSCOREi   i. Nat. y. (2). n. al. er. io. and Table 3-3.. sit. where i indexes the company observation and all variables are defined in Section 3.2. Ch. engchi. 41. i n U. v.

(49) 4. RESEARCH RESULTS AND ANALYSIS 4.1 DESCRIPTIVE EVIDENCE Table 4-1 represents descriptive statistics for all accounting, financial market and content analysis variables. The statistical analysis is conducted with Stata/Se 12.0. During the sample period, approximately 61 percent of the sample companies report earnings that meet or beat analysts’ expectations, which means that approximately 39 percent of the sample companies reported a loss, compared to analysts’ forecast. The mean of DET_FS is 0.67, representing that about 67 percent of the financial news. 政治大. articles include detailed financial statement terms or sentences. Table 4-2 presents the. 立. correlation matrix for all accounting, financial market and content analysis variables.. ‧. ‧ 國. 學. This study employs multivariate analysis for all of hypothesis tests.. n. er. io. sit. y. Nat. al. Ch. engchi. 42. i n U. v.

(50) Table 4-1. Descriptive Statistics Variable. Obs.. Mean. Median. Min. Max. Std. dev.. CAR. 90. 0.012. 0.012. -0.268. 0.148. 0.051. SURP. 90. 0.006. 0.001. -0.834. 0.991. 0.157. BEAT. 90. 0.611. 1.000. 0.000. 1.000. 0.490. DIV_INC. 90. 0.644. 1.000. 0.000. 1.000. 0.481. BM. 90. 0.670. 0.619. 0.094. 1.961. 0.373. LOGREV. 90. 17.036. 16.862. 13.827. 22.098. 1.678. FILELAG. 90. 4.169. 62.980. 10.377. DET_FS. 90. 2.000. 0.519. NOTESCORE. 90. 5.700. 6.000. 4.000. 7.000. 0.771. 90. 0.684. 0.834. -1.000. 1.000. 0.387. 90. 10.211. 10.000. 8.000. 12.000. 1.008. ‧ 國. ‧. TOTALSCORE. 學. TONE. 立. 治 -4.020 政 -0.020 大 0.667 1.000 0.000. sit. y. Nat. Notes: This table presents descriptive statistics for all accounting, financial market and con. n. al. er. io. tent-analysis variables. CAR is the cumulated abnormal return over the six-day working day window starting on the financial reports published date. SURP is the difference between annual actual earnings and the most recent consensus analyst earnings forecast made prior to the earnings announcement, scaled by stock price measured at the beginning of the year. BEAT is equal to 1 if announced earnings for the year of 2013 meet or exceeded analysts’ expectations (i.e., when SURP ≥ 0) and 0 otherwise. DIV_INC is equal to one if the dividend change is positive, compared to prior year, and is 0 if the dividend change is less or equal to zero or there was no dividend announcement made in 2014. BM is book-to-market ratio as the book value of equity scaled by the market value of equity (both measured at the end of the cur-rent year). LOGREV. Ch. engchi. i n U. v. is current year sales (REV) and use its natural logarithm. FILELAG is equal to the number of days between the annual financial reports release date of each company and average companies’. NOTESCORE represents the scoring of notes to financial statements based on the seven criteria brought out by FSC. DET_FS is equal to 1 if the sum of BS_D and SCF_D to be equal to or greater than 1 and is 0 otherwise. BS_D and SCF_D to be 1 words identified are greater than one 43.

(51) for balance sheet related words and sentences greater than one for statement of cash flow related words and sentences, respectively. TONE is the differences between frequency counts of optimistic words and negative words separately and divided by the sum of frequency count of positive and negative words. That is, (P－N)/(P＋N). TOTALCORE is sum of the scores of footnotes to financial statements and financial news articles.. 立. 政治大. ‧. ‧ 國. 學. n. er. io. sit. y. Nat. al. Ch. engchi. 44. i n U. v.