法人說明會與股票價格波動性之關聯性研究– 以內容分析為例 - 政大學術集成

全文

(1)國立政治大學會計學系碩士論文. 法人說明會與股票價格波動性之關聯性研究– 以內容分析為例 Conference Calls and Stock Price Volatility – A Content Analysis Approach. 中. 指導教授：. 諶家蘭博士. 研究生. 吳怡慧. 華. 民. ：. 國. 104. 年. 撰. 6. 月.

(2) 謝辭還清晰記得錄取政大會研時的興奮與喜悅，一眨眼就到了論文完成的時候，會研的兩年，充滿了挑戰與學習，很多感受與想法，是大學時不曾有的。這兩年要感謝的人很多，謝謝我的指導老師－諶家蘭老師，從研究案的指導到論文，老師是一位很認真、事必躬親的老師，不時的討論並指點我正確的方向，使我在這些年中獲益匪淺。老師對學問的嚴謹更是我輩學習的典範。謝謝我的麻吉幽默好夥伴們－小璧、老大、駿廷，和你們在一起可以討論課業、生活、工作、買東西，你們是我的好榜樣，謝謝老大讓我知道縱容有一百件忙碌的事也可以從容優雅的完成，謝謝小璧讓我知道有天才可以如此謙讓，兩年裡的日子，共同的生活點滴，學術上的討論、言不及義的閒扯、讓人又愛又怕的宵夜、趕作業的革命情感........；謝謝研究案子計畫的助理們，和一群不同系所的人相處，總是可以激盪不同的火花，帶給我不同的想法，參加研究案讓我得到了 10 幾個好朋友；謝謝祐子在論文資料上的蒐集，在這個資訊爆炸的時代，擁有處理資料的能力真的很厲害，謝謝你的協助也謝謝你給我的素食體驗！謝謝筱芳學姐的幫忙！謝謝秉軒、駿朋，謝謝你們！謝謝庭宇，在我壓力大、撰寫論文時給予的鼓勵，謝謝你的陪伴，很感謝有你，讓我覺得很踏實。謝謝我的媽媽無私的奉獻與支持，謝謝我的爸爸，謝謝家人們、姐妹、好友的支持。謝謝會研 102 級的同學們，很開心認識你們每一個人！本論文的完成另外亦得感謝台大會計系的李艷榕教授大力協助，及台大會計系廖芝嫻教授的支持。因為有你們的體諒及幫忙，使得本論文能夠更完整而嚴謹。這兩年的學習會讓我受益一輩子的，我會非常懷念指南山下的空氣、雨水和我在這裡遇到的每個人！. 2.

(3) 摘要本研究主要探討公司召開法人說明會(Conference Calls)時，經由媒體報導後，其新聞內容與股價波動性(Stock Price Volatility)之關聯性探討。近年來，法人說明會已逐漸成為公司傳達資訊的重要管道，藉由法人說明會企業可以完整說明公司資訊，投資人、分析師等也可直接與管理階層互動；文獻指出公司召開法人說明會可以減少管理階層與投資人之資訊不對稱，進而影響投資人於股票市場之交易之決策。本研究以國內上市櫃公司有召開法人說明會為研究對象，樣本期間為 2010 至 2014 年每年的三月份。本研究參考過去文獻及書籍，建立判讀公司法人說明會新聞之分數規則(Rules of Scoring)，衡量法人說明會新聞之內容，並以客觀量化之方式計算新聞內正向字詞與負向字詞。本研究假設法人說明會新聞之分數(SCORE)越正向，股價波動性越小；法人說明會新聞報導帶給投資人之情緒(TONE)越正向，股價波動性越小；法人說明會新聞數量(QUANTITY)越多，股價波動性越大，並以迴歸分析檢測假說。研究結果顯示，法人說明會之新聞內容分數(SCORE)和股價波動性呈現顯著負相關，投資人之情緒 (TONE) 與股價波動性具負向關係，法人說明會新聞之數量 (QUANTITY)與股價波動性具有顯著正相關。研究結果顯示公司召開法人說明會之媒體報導的新聞內容與股價波動性之關聯性。. 關鍵字：法人說明會、股價波動性、內容分析法、新聞. 3.

(4) Abstract This study focuses on the relationship between news content of conference calls and stock price volatility for TWSE corporations and GTSM corporations. Recently, conference calls have been a useful tool for companies to transmit information to the public, and management can communicate with analysts and investors face to face so as to reduce information asymmetry. Past literatures have proved that companies held conference calls can reduce information asymmetry, and have an impact on stock market. Our sample contains companies including listed in a Taiwan Stock Exchange companies and listed in Over-the-Counter companies that holding their conference calls on March during five-year period 2010-2014. Referring to past literature, this study builds rules of scoring to read the news of conference calls and use dictionary to objectively count the positive words and negative words in the news. This study’s hypothesis are that (1) the less of the SCORE and TONE the more stock price volatility will produce (2) the more QUANTITY of news, the more stock price volatility will produce. We use ordinary least squares regression to test hypothesis. The empirical results show that there is a significantly negative relation between scores (SCORE) of news content and stock price volatility and there is a negative relation between negative tone (TONE) of news and stock price volatility. In addition，there is a significantly positive relation between quantity (QUANTITY) of news and stock price volatility.. Keywords：Conference call, Stock volatility, Content Analysis, News. 4.

(5) TABLE OF CONTENTS 1.. INTRODUCTION ............................................................................................... 10. 1.1 RESEARCH MOTIVATION ......................................................................... 10 1.2 RESEARCH PURPOSE ................................................................................ 11 1.3 RESEARCH ISSUE ...................................................................................... 20 1.4 RESEARCH PROCESS ................................................................................ 21. 2.. LITERATURE REVIEW .................................................................................... 23. 2.1 CONFERENCE CALLS ............................................................................... 23 2.2 CONTENT ANALYSIS ................................................................................ 24 2.3 STOCK PRICE VOLATILITY ..................................................................... 27 2.4 DISCUSSION ................................................................................................ 29. 3.. RESEARCH METHOD ...................................................................................... 40. 3.1 HYPOTHESIS DEVELOPMENT ................................................................ 40 3.2 DATA COLLECTION ................................................................................... 42 3.3 RESEARCH DESIGN................................................................................... 48 3.4 REGRESSION MODLE ............................................................................... 48 3.4.1 DEPENDENT VARIABLE .................................................................. 49 3.4.2 INDEPENDENT VARIABLE .............................................................. 50 3.4.3 CONTROL VARIABLE ....................................................................... 52. 4.. RESEARCH RESULTS AND ANALYSIS ......................................................... 53. 4.1 DESCRIPTIVE STATISTICS ....................................................................... 53 5.

(6) 4.2 EMPIRICAL RESULTS ................................................................................ 57. 5.. CONCLUSION AND DISCUSSION ................................................................. 60. 5.1RESEARCH DISCUSSION AND CONTRIBUTION .................................. 60 5.2LIMITATION AND FUTURE RESEARCH WORK .................................... 61. APPENDIX ................................................................................................................. 67. 6.

(7) LIST OF TABLES TABLE 1. DISCUSSION OF LITERATURES .................................................. 31 TABLE 2. SAMPLE SELECTION ..................................................................... 45 TABLE 3. SAMPLE DISTRIBUTION ............................................................... 46 TABLE 4. DESCRIPTIVE STATISTICS............................................................ 54 TABLE 5. PEARSON CORRELATION............................................................. 56 TABLE 6. REGRESSION RESULT ................................................................... 59. 7.

(8) LIST OF FIGURES FIGURE 1. RESEARCH STRUCTURE ............................................................ 22 FIGURE 2. HYPOTHESIS DEVELOPMENT ................................................... 42. 8.

(9) LIST OF APPENDIXES APPENDIX 1. DEFINITION OF VARIABLE ................................................... 67 APPENDIX 2. RULES OF SCORING THE NEWS CONTENT OF COMPANY’S CONFERENCE CALL ....................................... 69 APPENDIX 3.THE RESULTS OF THE CONTENT ANALYSIS ABOUT NEWS OF CONFERENCE CALLS FROM 198 COMPANIES 72 APPENDIX 4.POSITIVE WORD LIST ............................................................. 87 APPENDIX 5.NEGATIVE WORDS LIST ......................................................... 91. 9.

(10) 1. 1.1. INTRODUCTION. RESEARCH MOTIVATION In Taiwan, stock market’s fund is often from individual and institutional investors.. Institutional investors are professional investors, having professional capability, and their investment decisions are more stable than individual investors. Also they have large fund that can be invested in the market. Institutional organization definitely own a lot of professional analysts to analyze public companies’ operating conditions, corporate’s prospect and corporate’s financial statements etc. As a result, conference calls are a useful tool that business can use to convey their information to update the newest information to the professional analysts. It can helpfully reduce the information asymmetry. Listed in Taiwan Stock Exchange Corporation ("TWSE") and the Gre Tai Securities Market ("GTSM”) listed Companies are usually holding their conference calls every season or half a year. In their conference calls, companies will announce their achievement and operating forecast and so forth. In the question and answer time, chairman of the board, chief manager, or chief finance officer will directly face investors to explain their questions. In short conference calls is indeed a good opportunity for TSE companies and OTC companies to enhance the notability and pass on their complete information to professional analysts and media which make them make investment choice, so conference calls is a very important activities between every TSE company and OTC company and their investors. Mass media will transmit the information from conference calls in the financial news. Financial news has an impact on Retail investors and Institutional Investors, they are usually displayed in the form of text, and people read text information will have opinions. Opinions are central to almost all human activities and are key influencers of our behaviors. 10.

(11) Our beliefs and perceptions of reality, and the choices we make, are, to a considerable degree, conditioned upon how others see and evaluate the world. For this reason, when we need to make a decision we often seek out the opinions of others. Sentiment analysis and opinion mining mainly focuses on opinions which express or imply positive or negative sentiments. Volatility is the relative rate at which the price of a security moves up and down. Volatility is found by calculating the annualized standard deviation of daily change in price. If the price of a stock moves up and down rapidly over short time periods, it has high volatility. If the price almost never changes, it has low volatility. When it comes to volatility, it goes back to Oct 19, 1987, when the Dow Jones Industrial Average fell from 2246.7 to 1738.4, over 508 points. This was the largest one-day drop since Dow Jones began computing index numbers in 1885. It was also the largest percentage drop-about 22.5 per cent. Nevertheless, most public attention focused on the absolute size of the drop. The 190-point drop on October 13, 1989 also caused a large public reaction, although it represented only 6.9 per cent drop in value. Finance academicians widely agree that volatility should be measured in percentage changes in prices, or rate of return. If you invest $1,000 in a portfolio of common stocks its rate of return tells you the proportional change in the value of your investment over the period. A 10 per cent rate of return would mean an increase in value of $100, whether the Dow Jones Industrial Average was at 100, 1000 or 2500. We will focus on the how the financial news report conference calls and use our content analysis techniques to extract information from our data and analyze its relationship between stock market volatility and conference calls. 1.2. RESEARCH PURPOSE 11.

(12) Conference calls typically begin with prepared statements by management, which usually reiterate the press release, and are then opened up to questions from analysts. The dialogue between management and analysts relative to firm performance is a potentially rich information source. From the company's perspective, conference calls save time and mitigate selective disclosure problems, because management can talk to hundreds of analysts and money managers simultaneously, and the investor relations staff receives fewer calls. From the analyst's perspective, the benefits of conference calls include the opportunity to hear the questions of others (55%); the saving of time and money vis-a-vis traveling to meetings (45%); the receipt of timely information (13%); and the receipt of information at the same time as other investors (9%).. 1. Frankel, Johnson, & Skinner (1999) state that conference. calls provide benefits to the company and its analysts. Conference calls are often used to supplement mandated disclosures, in particular, quarterly earnings releases. It seems likely that conference calls will be used when such supplementation is most beneficial; for example, in helping analysts ascertain the extent to which earnings changes are permanent or transitory. Although Conference calls are deem as voluntary disclosure, there are also laws to regulate conference calls. As the law of Corporate Governance Best Practice Principles for Taiwan Stock Exchange Corporation ("TWSE") and the Gre Tai Securities Market ("GTSM”) listed Companies goes, the article 58 states way of holding institutional investor meeting: A TWSE/GTSM listed company shall hold an investor conference in compliance with the regulations of the TWSE and GTSM, and shall keep an audio or video record of. 1. Percentages taken from a survey of 122 analysts and portfolio managers by Christensen and. Associates [1992]. 12.

(13) the meeting. The financial and business information disclosed in the investor conference shall be disclosed on the designated Internet information posting system and provided for inquiry through the website established by the company, or through other channels, in accordance with the TWSE or GTSM rules. Also the law of Taiwan Stock Exchange Corporation Procedures for Verification and Disclosure of Material Information of Companies with Listed Securities goes, the article 8 states : Listed company holding conference calls should comply with the following matters: A TWSE listed company shall be invited to hold or hold a conference call at least one time every three year in Taiwan. Primary listed companies shall be invited to hold or hold a conference call at least one time every year in Taiwan. We can see government and capital market pay more attention to the conference calls. Recent literature argues that earnings conference calls have become an increasingly important medium through which firms convey value relevant information to the market. Because there is information asymmetry between corporate management and investors so recent literature argues that earnings conference calls have become an increasingly important medium through which firms convey value relevant information to the market. Frankel et al. (1999) studied the role of conference calls as a medium of voluntary disclosure and found that companies that used this tool more frequently tended to be larger, more profitable, faster growing and more interactive with the market. They empirically examine conference calls as a voluntary disclosure medium by analyzing stock volatility and trading volume levels at the time of the conference calls. They find that return variances and trading volume are both elevated during conference calls, suggesting that conference calls provide information to the market beyond that which is found in the press release alone. In other words, the calls contain material information and investors’ trade on this information in real time. 13.

(14) Kimbrough (2005) contends that conference calls are a voluntary disclosure mechanism of increasing importance that provide corporate managers with a forum in which they can emphasize specific aspects of recent performance and highlight their implications for future financial performance. Following opening remarks by management, analysts are allowed to make comments and pose questions. Kimbrough notes that during this somewhat informal exchange, details not contained in the earnings press release are often disclosed. The disclosures during conference calls help mitigate potential information asymmetry between managers and investors. During the presentation portion of the call, managers provide their interpretation of the firm’s performance during the quarter and provide any additional, voluntary disclosures they wish to communicate. In addition to possibly providing new disclosures during the presentation, managers also provide the information verbally, which is potentially informative to the market because of the information content of verbal cues Mayew & Venkatachalam (2012). There are at least two reasons that conference calls might be incrementally informative over a press release. First, managers are able to provide information in a less constrained fashion relative to financial statements and written press releases. Second, analysts can play a direct role in uncovering information during the question-and-answer session asking follow-up questions, requesting more detail, and perhaps questioning management’s interpretation of events. If the ability to disclose information in a less constrained fashion results in greater disclosure, then they expect the presentation portion of the call to be incrementally informative over the accompanying press release (Matsumoto, Pronk, & Roelofsen, 2011). Recent proposed changes by the Public Company Accounting Oversight Board (PCAOB) suggest that these calls can also be informative for assessing auditing risk (PCAOB Release 2009-007). 14.

(15) Because conference calls can mitigate potential information asymmetry, investors can get more information to make decision in the stock market, and conference calls will have an impact on stock market. Brown, Hillegeist, & Lo (2004) argued that the information provided in conference calls influenced analysts, enabling them to better assess their outlook on companies performance. Their results showed that analysts’ involvement in conference calls improved their forecasting ability, thus suggesting the use of this form of communication increased the quantity and quality of the information available on the company. Their study also pointed out, that conference calls could generate an information gap between different classes of investors and analysts with worst estimates in the past benefited most by participating in conference calls sessions. Moreover, they found that conference calls decreased dispersion among analysts, and consequently reduced the level of information asymmetry. Recent empirical studies such as Henry (2007), Davis et al.(2007), Li (2008), Tetlock (2007), and Tetlock et al. (2008), Sadique (2008) indicate that language used in firms’ earnings disclosures or in news media reports significantly affects stock returns. They extract the 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 relevant information is conveyed by managers in their word choices. Formal reports such as balance sheet, income statement, statement of cash flows have its official format to comply with, so managers face more constraints when communicating with investors through formal reports and announcements (e.g. annual reports, earnings announcements, etc.) and, consequently, suggests that conference calls may provide a better setting in which to explore the relation between linguistic content and firm performance. In other words, quarterly earnings conference calls provide a forum in 15.

(16) which to more fully examine corporate disclosure as executives interact with call participants through the unscripted question and answer sessions. In the context of earnings announcements, the new financial information usually studied is the amount of unexpected earnings announced (Li, 2008). The language used in earnings press release and conference calls will affect information receiver’s sentiments then affect their decisions in stock market. Using 24,000 quarterly firm issued earnings press releases from PR Newswire between 1983 and 2003, Davis et al. (2007) find contemporaneous relationship between returns and optimistic and pessimistic language usage in earnings press releases after controlling for other factors known to influence the market response to earnings announcements. Investors are also found to be affected by the use of language in media reports. Content analysis enables quantification of text based information. A recent stream of literature utilizing tools relatively new to finance, but well established in a wide variety of other disciplines, allows a better understanding of the information content of words. There are literatures employs psychosocial dictionaries to count words that reflect particular characteristics of the text such as General Inquirer (GI) or Linguistic Inquiry and Word Count (LIWC). Engelberg (2008) applies the method outlined in Tetlock (2007) to measure the qualitative content of Dow Jones News Service stories about firms ’ earnings announcements to analyze the link between “ soft ” information and equity prices. He finds support for Tetlock (2007) and demonstrates a relation between linguistic media content and the stock market. Companies that announce earnings will have an impact on stock price, Ball and Brown (1968) find evidence that stock prices continue to drift upward (downward) after initial positive (negative) earnings announcements, rendering the initial 16.

(17) stock price reaction to the earnings news incomplete and raising questions of market efficiency. Tetlock, Saar-Tsechansky, & Macskassy (2008) examine whether language used in news reports about a firm’s earnings predicts its earnings and returns for S&P 500 firms over the sample period of 1984 to 2004. They find that negative words in firm-specific news stories predict low firm earnings, that stock prices under react to negative words used in news stories and that the impact of negative words on earnings and returns is highest for stories about firm fundamentals. Investors that have seldom opportunity to fully understand the operating conditions usually read the financial news to know the conditions of stock market, and their opinions are often influenced by news. Mullainathan & Shleifer (2005) document that the origin of media spin bias (or demand-side bias) derives from the literature on bubbles and panics, which shows that people read and believe only those reports that satisfy their preconceived notions. Consequently, media shape news to be attractive to readers by highlighting content that reinforces their preconceived notions. Henry (2007) examines whether the writing style, including tone, of earnings press releases influence investors. Henry shows that (a) stylistic attributes of earnings press releases in addition to actual financial performance affect market reaction to earnings announcements and (b) earnings surprise has less impact on abnormal returns, but this depends on press releases’ length, numerical intensity, and textual complexity. Past research that do content analysis usually use computer-based content analysis techniques; in addition to use computer-based content analysis techniques which use this research’ specified dictionary (shown in Appendix 4 and 5 ) to count the words, this study establishes the rules to read the news and give scores based on manual reading.. 17.

(18) The specified dictionary is established by three graduate students including me, and are triple checked by three graduate students major in Accounting and our instructor professor Seng who expertizes in Data mining and Information and Technology. First, we select the financial news which are categorized as Economics and Trade, Finance, Industry, Technology and Financial Investment in Knowledge Management Winner (KMW) news database. Second, we select the positive words and negative words line by line such as good, bad, excellent etc. Third, the positive words and the negative words are triple checked by three graduate students and our instructor professor Seng to make sure the words we select are appropriate. Finally, the dictionary are finished. Appendix 2 shows that the rules to score the conference calls’ news and appendix 3 shows that the outcome of word counts and scores. This research focus on the how the financial news report conference calls and use our data analytics techniques to extract information from our data and analyze its relationship between stock market volatility and conference calls. Our measures of content are based on two methods, one is word counts of financially of each conference calls news, and the other is the rules to score the conference calls’ news. This research develop a customized word list of financially oriented words based on an analysis of commonly used words in conference calls. The other process of dealing with news follow the rules of scoring conference calls’ news and the news are from Knowledge Management Winner (KMW) news database. After counting and scoring the news, this research conduct a regression analysis to provide evidence that the financial news content have an impact on stock price volatility. The results show that conference calls’ content conveys information to market participants and affect stock price volatility. In addition, SCORE has significant explanatory power for stock price volatility, as measured by the five day event window (days zero through day four), after controlling the variables that would affect stock price 18.

(19) volatility. However, the TONE measured on a customized, domain relevant dictionary does not have significance for stock price volatility by the five day event window (days zero through day four).. 19.

(20) 1.3RESEARCH ISSUE Based on the prior research purpose and motivation, the research questions of this study are listed as follows: . How is relationship between conference calls news and stock market volatility using a content analysis approach?. The remainder of this paper is organized as follows. Section 2 presents relevant literatures. Research method is described in Section3. In Section 4, the study presents the empirical results. Finally, conclusion is provided in Section 5.. 20.

(21) 1.4 RESEARCH PROCESS The research process in this study is shown as following Figure1. In the above, this study has discussed research background and research problems identification, it demonstrates the importance of conference calls, the impact of conference calls, and content analysis techniques to analyze the content of conference calls. In the section 2, it shows literature review about our research, and it divides 4 parts to demonstrate literature review: 1. Conference calls 2. Content analysis 3. Stock price volatility and 4. Discussion Research method will be shown in the section 3, and there are 4 parts in the section: Hypothesis development, data collection, research design, and regression model. This section will show that how this paper builds hypothesis, where it collect conference calls’ data and news data , how this paper build the rules of scoring the news of company’s conference call, and how it use content analysis to calculate the positive words and negative words. Research results and analysis will be discussed in the section 4. Descriptive statistics show sample descriptive statistics included mean, standard deviation, maximum, and minimum. Empirical results shows regression result of this research’ regression model. In the section 5, there are conclusions and discussions of this study to conclude this research and some advice for future research.. 21.

(22) FIGURE 1 RESEARCH STRUCTURE. Research Background and Research Problems Identification. Literature Review. Research Method. Empirical Results and Analysis. Discussions and Conclusion. 22.

(23) 2.. LITERATURE REVIEW. In the section 2, it shows literature review about our research, and it divides 4 parts to demonstrate literature review: 1. Conference calls 2. Content analysis 3. Stock price volatility and 4. Discussion. 2.1 CONFERENCE CALLS. Frankel et al. (1999) study the role of conference calls as a medium of voluntary disclosure and find that companies that use this tool more frequently tended to be larger, more profitable, faster growing and more interactive with the market. In addition, they find that firms in high tech industries and with higher-than-average market-to-book ratios and sales growth rates are more likely to hold conference calls. Brown et al. (2004) hypothesize that conference calls are a voluntary disclosure mechanism that leads to long-term reductions in information asymmetry among equity investors. They show that information asymmetry, measured as the probability of informed trade (PIN), is negatively associated with conference call activity. Firms initiating a policy of regularly holding conference calls experience significant and sustained reductions in information asymmetry, in contrast to one-time callers who experience no significant decline in asymmetry. Based on prior work documenting that the cost of equity capital is increasing in the level of information asymmetry, Brown et al. (2004) interpret their results as suggesting that firms holding conference calls frequently have lower costs of capital. Kimbrough (2005) examines the effect of conference calls on post-earnings announcement drift in order to provide insight into whether conference calls improve the efficiency of the market reaction to earnings announcements. He finds that the initiation of conference calls is associated with a significant reduction in the post-earnings 23.

(24) announcement drift. This reduction is concentrated in small firms, the companies for which prior literature has determined the drift to be most severe. Price, Doran, Peterson, & Bliss (2012) find earnings-specific dictionary is much more powerful in detecting relevant conference call tone; conference call discussion tone has highly significant explanatory power for initial reaction window abnormal returns as well as the post-earnings-announcement drift; conference call ‘question and answer’ tone matters more when firms do not pay dividends. Larker and Zakolyukina (2012) detect deceptive discussions in conference calls. The linguistic features of CEOs and CFOs in conference call narratives can be used to identify financial misreporting; deceptive CEOs use significantly more extremely positive emotion words and fewer anxiety words; deceptive CFOs use significantly. 2.2 CONTENT ANALYSIS. Content analysis is a wide and heterogeneous set of manual or computer-assisted techniques for contextualized interpretations of documents produced by communication processes in the strict sense of that phrase (any kind of text, written, iconic, multimedia, etc.) or signification processes (traces and artifacts), having as ultimate goal the production of valid and trustworthy inferences. Content analysis, 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 utilizing tools relatively new to finance, but well established in a wide variety of other disciplines, allows a better understanding of the information content of words.. 24.

(25) Henry (2008) quantifies tone using computer-based analytical tools to measure the frequency of positive and negative words found in earnings press releases. She finds that the tone of earnings press releases influences investors’ reaction to earnings in a sample of technology firms. Davis et al. (2007) confirm Henry ’ s findings across a broad sample of firms, with a significant positive association between levels of optimistic tone in 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. Studies have only recently begun to incorporate the richness of the linguistic information content of conference calls. In their investigation of investor relations costs, Mayew and Venkatachalam (2009) incorporate the linguistic content of conference calls as a control in their analysis of managerial affective states, utilizing audio files and vocal emotion analysis software, during earnings conference calls. The application of content analysis, in general, and textual tone analysis, in particular, has gained increased attention in disclosure research in recent years. Davis et al. (2008), Demers and Vega (2008), Henry (2008), and Sadique (2008) extract the 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 relevant information is conveyed by managers in their word choices. Using 24,000 quarterly firm issued earnings press releases from PR Newswire between 1983 and 2003, Davis et al. (2007) find contemporaneous relationship between returns and optimistic and pessimistic language usage in earnings press releases after controlling for other factors known to influence the market response to earnings 25.

(26) announcements. Investors also found to be affected by the use of language in media reports. Tetlock (2007) examines investor sentiment extracted from the “Abreast of the Market” column in the Wall Street Journal by measuring the pessimism index that is composed of mostly negative and weak words from the GI dictionary. Tetlock et al. (2008) examine whether language used in news reports about a firm’s earnings predicts its earnings and returns for S&P 500 firms over the sample period of 1984 to 2004. They find that negative words in firm-specific news stories predict low firm earnings, that stock prices under react to negative words used in news stories and that the impact of negative words on earnings and returns is highest for stories about firm fundamentals. Liu (2012) discusses that sentiment classification is perhaps the most extensively studied topic (also see the survey (Pang and Lee, 2008)). It aims to classify an opinion document as expressing a positive or negative opinion or sentiment. The task is also commonly known as the document-level sentiment classification because it considers the whole document as a basic information unit. The problem can be defined that given an opinion document d evaluating an entity e, determine the overall sentiment s of the opinion holder h about the entity, and time of opinion t are assumed known or irrelevant (do not care). There are two formulations based on the type of value that s takes. If s takes categorical values e.g., positive and negative, then it is a classification problem. If s takes numeric values or ordinal scores within a given range, e.g., 1 to 5, the problem becomes regression. Since online reviews have rating scores assigned by their reviewers, e.g., 1-5 stars, the positive and negative classes are determined using the ratings. For example, a review with 4 or 5 stars is considered a positive review, and a review with 1 to 2 stars is 26.

(27) considered a negative review. Most research papers do not use the neutral class, which makes the classification problem considerably easier, but it is possible to use the neutral class, e.g., assigning all 3-star reviews the neutral class.. 2.3 STOCK PRICE VOLATILITY. Schwert (1990) has computed sample standard deviations from intra-daily returns on the S&P 500 index. Frankel et al. (1999) empirically examine conference calls as a voluntary disclosure medium by analyzing stock volatility and trading volume levels at the time of the conference calls. They find that return variances and trading volume are both elevated during conference calls, suggesting that conference calls provide information to the market beyond that which is found in the press release alone. In other words, the calls contain material information and investor’s trade on this information in real time. Sadique et al. (2008) look at the impact of tone on volatility in addition to stock returns using earnings press releases and the financial news coverage of those releases with a sample of S&P 100 firms. They find that positive tone is directly related to returns, and negatively related to volatility. Antweiler and Frank (2004) do not find a statistically or economically significant effect of positive messages on stock returns; however, they find evidence of a relationship between message activity and both trading volume and return volatility. Similarly, Coval and Shumway (2001) establish that the ambient noise level created by traders in a futures pit is linked to volume and volatility, but not to returns. Dell' Acqua, Perrini, & Caselli (2010) find evidence that voluntary disclosure following the introduction of the Regulation Fair Disclosure, included in the Selective 27.

(28) Disclosure and Insider Trading Act issued by the SEC, reduces price volatility of high tech firms listed in the US market.. 28.

(29) 2.4 DISCUSSION The following table 1 discuss the above literature review and lists 5 papers’ author, hypothesis, model, variable to be adopted and whether can data be taken from TEJ database. Frankel et al. (1999) examines whether stock prices and trading volume increase during the time these calls take place. They study the role of conference calls as a medium of voluntary disclosure and find that companies that used this tool more frequently tended to be larger, more profitable, faster growing and more interactive with the market. In addition, they find that firms in high tech industries and with higher-than-average market-to-book ratios and sales growth rates are more likely to hold conference calls. Dell' Acquae et al. (2010) find evidence that voluntary disclosure following the introduction of the Regulation Fair Disclosure, included in the Selective Disclosure and Insider Trading Act issued by the SEC, reduces price volatility of high tech firms listed in the US market. Price et al. (2012) find earnings-specific dictionary is much more powerful in detecting relevant conference call tone; conference call discussion tone has highly significant explanatory power for initial reaction window abnormal returns as well as the post-earnings-announcement drift; conference call ‘question and answer’ tone matters more when firms do not pay dividends. Johnson et al. (2004) document a positive relationship between firm level news and firm level volatility contrary to the negative relationship between idiosyncratic news and firm volatility predicted by current theories. They confirm, as conjectured in prior literature, an increase in idiosyncratic IPO volatility over the period from 1973.

(30) through 2003. We find the increase in volatility is over twice as large for IPOs as for firms matched to the IPOs based on size and book-to-market ratio. Matsumoto et al. use a sample of more than 10,000 conference-call transcripts, they examine the information content of both segments of the call—the presentation and the discussion segment. They find that both segments have incremental information content over the accompanying press release. However, discussion periods are relatively more informative than presentation periods, and this greater information content is positively associated with analyst following. They also find that managers provide increased disclosures during the presentation segment when firm performance is poor, but relatively more information is released during discussion periods in these circumstances. Overall, their results are consistent with the notion that active analyst involvement in conference calls increases the information content of the calls, particularly when firm performance is poor.. 30.

(31) TABLE 1. DISCUSSION OF LITERATURES Author. Hypothesis. Model. Variable to TEJ Database be adopted. Pr(CCALL) = β0 + β1 𝐿𝑁𝐴𝑆𝑆𝐸ET + 𝛽2 MtB + 𝛽3 𝑁𝐴𝑁𝐴𝐿𝑌𝑆𝑇 + 𝛽4 𝑅𝑂𝐴. Empirical. Whether stock. Examination of. prices and. + 𝛽5 𝑆𝐺𝑅𝑂𝑊𝑇𝐻 + 𝛽6 𝑆𝑡𝑑𝑅𝑂𝐴 + 𝛽7 𝐷/𝐸 + 𝛽8 𝐼𝑁𝑇𝐶𝑂𝑉. taken from. conference calls. trading volume. + 𝛽9 𝑋𝐼𝑇𝐴 + 𝛽10 𝑆𝑃𝐼𝑇𝐴 + 𝛽11 𝑃𝑈𝐶𝑆 + 𝛽12 𝐷𝐸𝐵𝑇 + 𝛽13𝑂𝑇𝐻𝐸𝑅. TEJ database,. as a voluntary. increase during. +𝜀.. and. disclosure. the time these. medium. calls take. Variable definitions：. 1999 Frankel, R.. place.. CCALL is coded one of the sample firm had a conference call in the First Call. Johnson, M.. N/A. Data can be. calculated.. conference call database between February and November 1995, and zero otherwise. LNASSET is the log of total assets at the end of fiscal 1994. MtB is. Skinner, D. J. the market-to-book ratio computed as price divided by shareholders’ equity. NANALYST is the number of analysts following the company in the January.

(32) Author. Hypothesis. Model. Variable to TEJ Database be adopted. 1995 statistical period on the IIBIEIS consensus tape. ROA is the return on total assets for fiscal 1994. SGROWTH is the geometric mean of annual sales growth between fiscal years 1991 and 1994. StdROA is the variance in annual return on assets between fiscal 1990 and 1994. DIE is the debt-to-equity ratio at the end of fiscal 1994. INTCOV is the interest coverage ratio for fiscal 1994 computed by dividing cash flow from operations by interest paid. XJ/TA is the amount of extraordinary items for fiscal 1994 divided by total fiscal 1994 assets. SpuTA is the amount of special items for fiscal 1994 divided by total fiscal 1994 assets. PUCS is an indicator variable taking a value of one if the company had a public common stock issuance transaction in the Securities Data Corporation (SDC) database, and zero otherwise. DEBT is an indicator variable taking a value of one if the company had a debt issuance transaction in the SDC. 32.

(33) Author. Hypothesis. Model. Variable to TEJ Database be adopted. database, and zero otherwise. OTHER is an indicator variable taking a value of one if the company had a non-common- stock and non-debt-issuance transaction in the SDC database, and zero otherwise. The portion of the SDC database covering the period between January 1995 and May 1996 is used in this sample. Conference Calls. Whether a. and Stock Price. more intensive. Volatility in the. use of. StDevt = β0 + β1 CCALLTot + 𝛽2 OUTSH01 + 𝛽3 b3ROA01 + 𝛽4 b4LEV01. StDevt. Data can be. + +𝛽5 MKTCAPend01 + 𝛽6 BtoMend01 + 𝜀 .. OUTSH. taken from TEJ database,. Variable definitions： Post-Reg FD Era. conference. 2010. calls by high. Dell’Acqua, A.. tech firms has. Perrini, F.. been beneficial. BtoM and. StDevt is standard deviation to measure stock volatility, CCALLTot is annual number of conference calls, OUTSH01 is outstanding shares during 2001. ROA01 is Return on Asset Ebit (Earnings before interest and taxes) on total. 33. LEV. calculated..

(34) Author. Hypothesis. Model. Variable to TEJ Database be adopted. Caselli, S.. in terms of. assets, Book-to-market ratio (BtoMend01), Leverage (LEV01) is computed. reducing the. through the ratio of net financial debt on total assets , Market capitalization. stock price. (MKTCAPend01) at the end of FY 2001, calculated as the number of. volatility.. outstanding shares times the stock price of the last trading session of FY 2001. 𝜎𝑎𝑛𝑛𝑢𝑎𝑙= √∑𝑇1(𝑅𝑗− 𝑅𝑉𝑊𝑖𝑛𝑑𝑒𝑥 )2 , stock volatility measure (StDev), Ta is the number of trading days in the year, Rj is the one-day firm return and RVWindex is the one-day value-weighted index return.. Earnings. The research. conference calls. expect a. and stock returns. significant. The incremental. relation. 𝐶𝐴𝑅𝑗 = 𝛾0,𝑖 + 𝛾1,𝑖 𝑆𝑈𝑅𝑃𝑖,𝑗 + 𝛾2,𝑖 𝑇𝑂𝑁𝐸𝑖,𝑗 + 𝐶𝑂𝑁𝑇𝑅𝑂𝐿𝑆𝑖,𝑗 + 𝜀𝑗. H-tone. Calculating the variable. Variable definitions： 𝐶𝐴𝑅𝑗 represent the cumulative abnormal return measures for conference call j defined above; 𝑇𝑂𝑁𝐸𝑖,𝑗 is interchangeable with 𝐻 − 𝑇𝑂𝑁𝐸𝑖,𝑗 , both as. 34. of H-tone should use.

(35) Author. Hypothesis. Model. Variable to TEJ Database be adopted. informativeness. between the. defined above and where i = 1 or 2; and 𝐶𝑂𝑁𝑇𝑅𝑂𝐿𝑆𝑖,𝑗 includes measures of. the dictionary. of textual tone. tone of the. call length, firm size, book-to-market equity, profitability, leverage, trading. of this. 2012 Price, S. M.. question and. volume, returns volatility, and analyst coverage for firm j. Unexpected earnings,. research,. answer portion. SURPj, are calculated using a seasonal random walk model where the difference. shown in. of the. between the earnings per share and the earnings-per-share in the same quarter of. Appendix 4. conference. the prior year is scaled by the stock price at the close of the lagged quarter:. and 5.. Doran, J. S. Peterson, D. R. Bliss, B. A.. calls and stock. 𝑆𝑈𝑅𝑃𝑗 =. returns.. (𝐸𝑃𝑆𝐽 − 𝐸𝑃𝑆𝑗,𝑞−4 ) 𝑆𝑡𝑜𝑐𝑘𝑃𝑟𝑖𝑐𝑒𝑗,𝑞−4. 𝐻 − 𝑇𝑂𝑁𝐸1,𝑗 = 𝐻 − 𝑇𝑂𝑁𝐸2,𝑗 =. 35. 𝑃𝑜𝑠𝑖𝑡𝑖𝑣𝑒𝑗 𝑁𝑒𝑔𝑎𝑡𝑖𝑣𝑒𝑗. 𝑃𝑜𝑠𝑖𝑡𝑖𝑣𝑒𝑗 − 𝑁𝑒𝑔𝑎𝑡𝑖𝑣𝑒𝑗 𝑁𝑒𝑔𝑎𝑡𝑖𝑣𝑒𝑗 + 𝑁𝑒𝑔𝑎𝑡𝑖𝑣𝑒𝑗.

(36) Author. Hypothesis. Model. Variable to TEJ Database be adopted. The Effect of. Greater. News on. number of. Volatility: A. news. Study of IPOs. announcement. 2004. contributes to. William C.. greater return. Johnson. volatility. St. Dev = β0 + β1 𝐻𝑖𝑡𝑠𝑡 + 𝛽2 𝐼𝑃𝑂 𝐷𝑢𝑚 + 𝛽3 𝐻𝑖𝑡𝑠𝑡 ∗ 𝐼𝑃𝑂 𝐷𝑢𝑚. ROE. Data can be. + 𝛽4 𝐿𝑛(𝑀𝑘𝑡. 𝐶𝑎𝑝)𝑡−1 + 𝛽5 𝑆𝑡. 𝐷𝑒𝑣𝑡−1 + 𝛽6 𝑅𝑒𝑡𝑢𝑟𝑛𝑡. LEV. taken from. 𝑏 + 𝛽7 𝑅𝑂𝐸𝑡−1 + 𝛽8 𝐿𝑒𝑣𝑡−1 + 𝛽9 𝐿𝑛 ( ) + 𝛽10 𝑇𝑒𝑐ℎ 𝐷𝑢𝑚 𝑚 𝑡−1. B/M. + 𝛽11 1980𝑠 𝐷𝑢𝑚 + 𝛽12 1990𝑠 𝐷𝑢𝑚 + 𝜀. TEJ database, and calculated.. Variable definitions： 𝜎𝑚𝑜𝑛𝑡ℎ𝑙𝑦= √∑𝑇1(𝑅𝑗− 𝑅𝑉𝑊𝑖𝑛𝑑𝑒𝑥 )2 where T is the number of days in the month, R. Jennifer. firm is the one-day firm return and R Vwindex is the one-day value-weighted. Marietta-Westber. index return. 𝜎𝑚𝑜𝑛𝑡ℎ𝑙𝑦= √∑𝑇1(𝑅𝑖𝑛𝑑𝑒𝑥 )2 where T is the number of days in the. g month and R index. is the one-day return for either the value-weighted or. equally weighted CRSP stock index. The variables are formed as follows. We. 36.

(37) Author. Hypothesis. Model. Variable to TEJ Database be adopted. continue to use the monthly measure of idiosyncratic volatility (St. Dev t) formed from daily returns as in CLMX (2001). We also cumulate daily excess returns to form a monthly return measure (Return t). The monthly market capitalization variable (Mkt. Cap t-1) uses the previous month’s market value (price times shares outstanding). We cumulate the number of days with news citations each month to form a monthly hits variable (Hits t ). Leverage is defined as the ratio of last fiscal year’s long-term debt to last fiscal year’s total assets (Lev t-1). ROE is the prior year’s net income divided by the prior year’s book value (ROE t-1). Book-to-market uses the prior year’s book value and divides by the prior month’s market value (B/M t-1).. 37.

(38) Author. Hypothesis. Model. Variable to TEJ Database be adopted. What Makes. |𝐴𝐵𝑅𝐸𝑇 𝑃𝑅𝐸𝑆 | = β0 + β1 |𝑅𝐸𝑇 𝐷𝐴𝑌𝐵4 | + CONTROLs + 𝜀 .. Conference Calls. |𝐴𝐵𝑅𝐸𝑇 𝐷𝐼𝑆𝐶 | = β0 + β1 |𝑅𝐸𝑇 𝐷𝐴𝑌𝐵4 |+𝛽2 |𝐴𝐵𝑅𝐸𝑇 𝑃𝑅𝐸𝑆 | + CONTROLs + 𝜀. Useful? The Variable definitions： Information Content of Managers’ Presentations and Analysts’. |𝐴𝐵𝑅𝐸𝑇 𝑃𝑅𝐸𝑆 | is. the abnormal absolute returns during the presentation portion of. the conference call. |𝑅𝐸𝑇 𝐷𝐴𝑌𝐵4 | = quote midpoint at the start of the conference call (MIDstart ) less the quote midpoint at the same time one trading day before the conference call ( MIDdayprior ) divided by MIDdayprior. |𝐴𝐵𝑅𝐸𝑇 𝐷𝐼𝑆𝐶 | is the abnormal absolute returns during the discussion portion of the conference call.. Discussion Sessions 2011 Matsumoto, Pronk, and 38. N/A. N/A.

(39) Author. Hypothesis. Model. Variable to TEJ Database be adopted. Roelofsen. 39.

(40) 3.. RESEARCH METHOD. 3.1HYPOTHESIS DEVELOPMENT In this section, hypotheses will be discussed and provided as follows： The research question of whether the score and tone of the conference calls news content affect stock-price volatility stems from investors’ lack of perfect knowledge about the true state of the company’s condition. Evidence that readers are influenced by how information is written appears in analogous research in judgment and decision making. Veronesi (1999) documents that during times of greater uncertainty; investor’s expectations about future cash flows become more sensitive to new information, which causes greater stock price volatility. This uncertainty affects investor reaction to accounting information. In periods of uncertainty, investors are more likely to make expectation errors about asset fundamentals. Unreliable and biased framing of accounting information can prevent investors from processing available information in a purely rational fashion (Barber & Odean, 2008) (Hirshleifer & Teoh, 2003). Tetlock (2007) find that high media pessimism predicts downward pressure on market prices followed by a reversion to fundamentals, and unusually high or low pessimism predicts high market trading volume. These and similar results are consistent with theoretical models of noise and liquidity traders, and are inconsistent with theories of media content as a proxy for new information about fundamental asset values, as a proxy for market volatility, or as a sideshow with no relationship to asset markets. Hong, Lim, and Stein (2000) argue that the momentum effect documented by Jegadeesh and Titman (1993) is asymmetric and has a stronger negative effect on.

(41) declining stocks than a positive effect on rising stocks. A potential explanation behind the idiosyncratic volatility results is that stocks with very low returns have very high volatility. Of course, stocks that are past winners also have very high volatility, but loser stocks could be overrepresented in the high idiosyncratic volatility quintile. Abarbanell and Bernard (1992) state that the literature on both under reaction and overreaction provides evidence suggesting that stock price behavior around earnings announcements may be caused by the failure of market participants to fully appreciate the information content of current earnings. This implies that we need to further our understanding of contemporaneous stock price reactions to the information content associated with earnings announcements. Such information is not limited to the numerical content of an announcement alone, but includes statements that typically accompany the announcement. An excellent example is the conference calls that usually immediately follow the release of earnings. If formal earnings disclosures increase uncertainty about the disclosure and framing of information, thereby reducing investor ability to interpret information in a systematic way, then one may expect that the score and tone of the conference calls news content affects volatility around the day of announcement. Firm issued conference calls and their news coverage give investors some idea about the company’s past performance and future prospects. Accordingly, it is logical to expect that a report’s SCORE and TONE affects volatility. Since positive (negative) tone reduces (increases) uncertainty about a firm’s future performance, volatility is expected to fall for reports with positive tone and to rise for reports with negative tone. The quantity of news are deemed as whether the market have a lively discussion, so the QUANTITY of news are expected that the more quantity of conference calls news reported, the more stock price volatility will produce. Past research and above 41.

(42) discussion leads to our three hypotheses: Hypothesis 1. The score and tone of conference calls news affects stock price volatility. It is expected that negative score will increase stock price volatility. Hypothesis 2.The quantity of news has impact on stock price volatility. It is expected that the more quantity of news, the more stock price volatility will produce. The following Figure 2 demonstrate the relation between hypothesis1 and hypothesis 2. The score and tone of conference calls news affects volatility (H1), and the quantity (H2) of news affects stock price volatility.. FIGURE 2. HYPOTHESIS DEVELOPMENT. News’ TONE and SCORE. News’ QUANTITY. H1. Stock Price Volatility H2. 3.2 DATA COLLECTION This study focus on the relationship between news content of conference calls and stock price volatility for TWSE corporations and GTSM corporations. During the season of financial statements uploading to Taiwan’s Market Observation Post System, this paper observed that most enterprise uploaded their financial statements. 42.

(43) on March. Meanwhile, they frequently have their conference calls are also on March. So this research set our observation on March. Our sample contains companies including listed in Taiwan Stock Exchange companies and listed in Over-the-counter companies that holding their conference calls on March during five-year period 2010-2014. The event window used in this study is a 5 trading day (t-1 to t+3) window around the date of the conference calls as provided in the Taiwan’s Market Observation Post System, which is the date the conference calls are first publicly reported in various news media such as China Times. It uses a time horizon beginning the day prior to the event day (for the company’s conference calls) because that date can reflect the date of print media based on a press release issued on the previous day. Another reason for using a time horizon beginning the day prior to the conference calls date is that prior research has empirically shown that the market exhibits significant reaction to earnings announcements beginning on day t-1 when day 0 is the announcement date (Ball & Kothari, 1991; Patell & Wolfson, 1981). Related previous accounting research that has examined market response to financial and other information disclosed in earnings press releases use a 3-day event window from day t-1 to t + 1 , after our collection, this paper find many related news will miss ,so this paper extend our event window to 5 trading days. Conference calls date are collected from Taiwan’s Market Observation Post System and are double-checked against TEJ database. This paper hand collect the news talking about firm’s conference calls (no matter in the title or context) the sample from KMW news database. Table 2-1 shows that news selection process; after deducting the news memorandum are the observation sample news in this research. Using 435 conference calls news, this paper obtain 565 companies, shown in Table 2-2. This paper eliminate 43.

(44) banking and insurance business, not in event windows' company, foreign Company, repeated company, not holding conference calls' company, and missing value and get our final sample 198 companies. In table 3, there are our sample companies’ industry distribution, including computer. and. peripheral. equipment,. electronic. parts. and. components,. communications and internet, semiconductor et cetera. Most our sample companies are high-tech industry. The rules of categorizing industry follow Taiwan Stock Exchange Corporation.. 44.

(45) TABLE 2. SAMPLE SELECTION Panel A. News selection 2014. 2013. 2012. 2011. 2010. TOTAL. Conference calls news. 98. 122. 100. 90. 92. 502. News Memorandum. (20). (20). (4). (6). (17). (67). Observation sample news. 78. 102. 96. 84. 75. 435. News are from KMW database.. Panel B. Company selection 2010. 2011. 2012. 2013. 2014. Total. Initial sample. 100. 112. 110. 124. 119. 565. Banking and insurance business. (16). (20). (19). (19). (13). (87). Not in event windows' company. 0. 0. (3). (3). (4). (10). Foreign Company. (1). (5). (6). (5). (1). (18). Repeated company. (33). (33). (28). (47). (38). (179). (12). (15). (9). (13). (12). (61). Missing value. (4). (1). (5). 0. (2). (12). Final sample. 34. 49. 198. Not holding conference calls' company. 38. 45. 40. 37.

(46) TABLE 3. SAMPLE DISTRIBUTION Industry. 2010. 2011. 2012. 2013. 2014. Total. 7. 7. 7. 9. 9. 39. Electronic Parts & Components 5. 2. 3. 2. 4. 16. Cultural and Creative Industry 0. 0. 0. 0. 3. 3. Foods. 0. 0. 0. 0. 1. 1. Communications and Internet. 6. 7. 5. 5. 7. 30. Semiconductor. 11. 7. 8. 8. 8. 42. Optoelectronic. 1. 4. 3. 7. 2. 17. Plastics. 0. 0. 0. 0. 2. 2. Tourism. 0. 0. 0. 0. 1. 1. Others. 0. 0. 4. 1. 4. 9. Biotechnology & Medical Care 1. 0. 0. 1. 3. 5. Electrical and Cable. 0. 0. 0. 0. 1. 1. Electric Machinery. 0. 0. 1. 1. 3. 5. Trading and Consumers' Goods 0. 3. 4. 0. 1. 8. Other Electronic. 0. 1. 3. 2. 0. 6. Cement. 0. 1. 1. 1. 0. 3. Information Service. 2. 0. 1. 0. 0. 3. Textiles. 0. 1. 0. 0. 0. 1. 0. 3. 0. 0. 0. 3. Shipping and Transportation. 0. 1. 0. 0. 0. 1. Chemical. 0. 1. 0. 0. 0. 1. Computer and Peripheral Equipment. Building Material and Construction. 46.

(47) Electronic Products 1. 0. 0. 0. 0. 1. 34. 38. 41. 37. 50. 198. Distribution Total. Industries are categorized by Taiwan Stock Exchange Corporation.. 47.

(48) 3.3 RESEARCH DESIGN This study use Ordinary Least Squares (OLS) regression to test the relation between the news content of conference calls and stock price volatility. The empirical results of stock price volatility (STDEV) and the scores (SCORE) given by reading the news that refer to company’s conference calls, stock price volatility (STDEV) and the affect or feeling of news (TONE) that refer to conference calls of a company and stock price volatility (STDEV) and quantity of news (QUANTITY); scores and tone are presented in Appendix 3. 3.4 REGRESSION MODLE STDEV = 𝛃0 + 𝜷𝟏 𝐐𝐔𝐀𝐍𝐓𝐈𝐓𝐘 + 𝜷𝟐 𝐒𝐂𝐎𝐑𝐄 + 𝛽3 𝐓𝐎𝐍𝐄 + 𝛽4 𝑹𝑶𝑬 + 𝛽5 𝑳𝑬𝑽 + 𝛽6 𝑶𝑺𝑯𝑨𝑹𝑬𝑺 + 𝛽7 𝑩𝒕𝒐𝑴 + 𝜀 . Where: STDEV = the stock price volatility is calculated as standard deviation (STDEV) of stock price. QUANTITY = the measure of the quantity (QUANTITY) of news is based on the quantity of news that are reported company’s conference call by the mass media in our time window. SCORE = the Scores (SCORE) are given by reading the news that refer to company’s conference calls, and its rules appears on Appendix 2. TONE = the affect or feeling of news (TONE) that refer to conference calls of a company. TONE is calculated as (Positive word-Negative word) divide by (Positive word+ Negative word) and use this research’s dictionary to calculate. ROE = Return on Shareholders' Equity (ROE) is a company's net income divided by its average stockholder's equity. 48.

(49) LEV = The Leverage (LEV) is computed as net debt/total asset. OSHARE= Outstanding shares (OSHARE) are the tradable shares available on the market at the end of every our research year. BtoM= The Book to Market (BtoM) is the ratio between book value and market capitalization. 3.4.1 DEPENDENT VARIABLE The most commonly used measure of stock return volatility is standard deviation Schwert (1990). This statistic measures the dispersion of returns. Financial economists find the standard deviation to be useful because it summarizes the probability of seeing extreme values of return. If the price of a stock moves up and down rapidly over short time periods, it has high volatility. If the price almost never changes, it has low volatility. When the standard deviation is large, the chance of a large positive or negative return is large. The methodology used to test our hypothesis follows the work of Johnson and Westberg (2004). Their study examined the role of corporate news reported by mass media on the stock price volatility after IPOs. Specifically, Johnson and Westberg (2004)2 employed a cross sectional multivariate regression using the average monthly idiosyncratic stock price volatility as dependent variable, a number of dummies indicating the level of news on the stock as independent variables and various control variables. In this study because the aim is to analyze the stock volatility on daily basis, this use a daily average stock volatility measure (STDEV) expressed as follows:. 2. The average monthly idiosyncratic volatility in Johnson and Westberg (2004) is measured as the sum of the daily squared differences between firm returns and those on a value weighted-index. Thus, monthly firm idiosyncratic volatility is: 𝜎𝑚𝑜𝑛𝑡ℎ𝑙𝑦= √∑𝑇1 (𝑅𝑗− 𝑅𝑉𝑊𝑖𝑛𝑑𝑒𝑥 )2 Where T is the number of trading days in the month, R j is the one-day firm return and R index return. 49. VWindex is. the one-day value weighted.

(50) 𝑇. 2. 𝜎=√∑(𝑅𝑗− R̅ 𝑗 ) 1. Where T is the number of trading days in our time window, Rj is the one-day closing price and R̅ 𝑗 is the one-day average stock price. 3.4.2 INDEPENDENT VARIABLE Liu (2012) Sentiment classification is usually formulated as a two-class classification problem, positive and negative. Training and testing data used are normally product reviews. Since online reviews have rating scores assigned by their reviewers, e.g., 1-5 stars, the positive and negative classes are determined using the ratings. For example, a review with 4 or 5 stars is considered a positive review, and a review with 1 to 2 stars is considered a negative review. Most research papers do not use the neutral class, which makes the classification problem considerably easier, but it is possible to use the neutral class, e.g., assigning all 3-star reviews the neutral class. Follow Liu’s concept, I use numeric score expressing the strength/intensity scores. Because news provide abundant information, I expand score to positive 10 scores and negative 10 scores to reflect information content. If the news make mention of company’s stock price limit up, the company will get positive 8~10 scores, and accurate scores will depends on how the news describe company’s condition; if the news refer to the company’s net income is significantly increasing, earning per share is higher than company in the same industry or customers love their products very much so their future performance is worthy expectation, the company will get positive 4~7 scores and accurate scores will depends on how the news describe company’s condition; if the news refer to the company is during the off season, so its operating achievements is not good as expectation or any other condition of the company’s performance is at the same level 50.

(51) but still have room for improvement , the company will get positive 1~3 scores and accurate scores will depends on how the news describe company’s condition. If the news refer to a company’s operating performance is under expectation, inferior to the industry performance, their operating income is negative, earning per share is negative or any other condition of poor performance, the company will get negative 1~3 scores and accurate scores will depends on how the news describe company’s condition; if the news mention that a company’s revenue is significantly decreasing or face the condition of operating difficulty or any other severe problem or severely poor performance, the company will get negative 4~7 scores and accurate scores will depends on how the news describe company’s condition; if the news make mention of company’s stock price limit down, the company will get negative 8~10 scores, and accurate scores will depends on how the news describe company’s condition. The variable of SCORE will follow the above rules and the rules of scoring are presented in Appendix 2. The measure of the quantity (QUANTITY) of news is based on the quantity of news that are reported company’s conference call by the mass media in our time window. When quantity of news is increasing, it means market pay more attention to the company, and market are in the midst of a lively discussion of the company. This paper expected that the more quantity of news reported, the more stock price volatility will produce, so this research expect the positive sign. The measure of Tone (TONE) is based on a frequency count of the number of positive and negative words, using the dictionary of this research. The words counted as positive and negative within conference calls are shown in Appendix 5 and 6 and the computed results are shown in Appendix 2. TONE is calculated as the count of positive words minus the count of negative words, divided by the sum of positive and negative word counts so the maximum and minimum values of TONE are 1 and –1, 51.

(52) respectively. Formula is as following： 𝑇𝑂𝑁𝐸𝑗 =. 𝑃𝑜𝑠𝑖𝑡𝑖𝑣𝑒𝑗 − 𝑁𝑒𝑔𝑎𝑡𝑖𝑣𝑒𝑗 𝑃𝑜𝑠𝑡𝑖𝑣𝑒𝑗 + 𝑁𝑒𝑔𝑎𝑡𝑖𝑣𝑒𝑗. Where： 𝑃𝑜𝑠𝑡𝑖𝑣𝑒𝑗 = Positive is the total positive words in the news. 𝑁𝑒𝑔𝑎𝑡𝑖𝑣𝑒𝑗 = Negative is the total negative words in the news. 3.4.3 CONTROL VARIABLE Wei and Zhang (2003) provide evidence of an inverse link between ROE and volatility, when the ROE of a company is high, investors are more willing to invest in the stock market, so this paper insert the variable of ROE, this paper expected a positive sign. This paper also use control variables as indicative of the size, value and level of indebtedness. The variables for size and value are included because they are factors that significantly affect the expected returns of stock prices in financial markets (see Fama and French, 1992). The size is measured by the number of outstanding shares (OSHARE) as another variable. Because a larger quantity of tradable outstanding shares should be correlated to greater stability in stock prices. They are less likely to be influenced by volatility resulting from a scarce availability of shares on the market. The coefficient of the variable is expected to be negative (Dell'Acqua et al., 2010). The value is measured by the Book-to-market ratio (BtoM), calculated as the ratio of the book value on the market capitalization of the firm, both taken at the end of every our research year (2010-2014). As for the previous variable even the coefficient of the Book-to-market ratio is expected to be negative: a higher Book-to-market ratio has to be related to a lower volatility Firm’s indebtedness is determined by the Leverage (LEV) and is computed through the ratio of net financial debt on total assets, using for both balance sheet data 52.

(53) at the end of every our research year (2010-2014). This research expect a positive coefficient of this variable since higher leverage results in higher financial risk and ultimately higher stock price volatility.. 4.. RESEARCH RESULTS AND ANALYSIS. 4.1 DESCRIPTIVE STATISTICS This study applies the descriptive statistics to analyze the data from sample companies including Stock Price Volatility (STDEV), total Scores (SCORE) which are given by reading the news that refer to company’s conference calls, the affect or feeling (TONE) of a news that make mention of conference calls of a company and control variables. Table 4 provides descriptive statistics on each variable, and Table 5 presents a correlation matrix. Table 4 shows that the mean value of TONE is 0.667. By construction, a value of 1.0 would indicate a completely positive. The means of SCORE is 4.38, which indicate that news are mostly report optimistic operating condition. The max of SCORE is 9 and the min of SCORE is -4, indicating when company are in a good operating condition, news will give enormous publicity to the company, but when company are in a bad operating condition, news will report the company with retention of the phrase. The means of quantity of news (QUANTITY) are 1.8, indicating that when company held conference calls, there are average of 2 news report the conference calls. The standard deviation for outstanding shares is quite high reflecting the different operating structure and dimension of the companies in different industry. Leverage is generally very low, with an average net debt/asset ratio close to zero. Table 5 contains a correlation analysis of these variables. The correlation above 53.

(54) 0.5 are for ROE and BtoM. Notwithstanding this correlation, the analysis does not signal any problem of mucollinearity. TABLE 4.DESCRIPTIVE STATISTICS Variable. Mean. Median. Min. Max. Std. Dev.. STED. 4.133864. 0.157469. 0.000225. 168.9253. 18.37467. QUANTITY. 1.823232. 1. 1. 7. 1.326828. SCORE. 4.383117. 5. -4. 9. 2.654733. TONE. 0.666778. 0.75. -0.5. 1. 0.325501. ROE. 13.62495. 14.06. -57.12. 70.34. 14.54918. LEV. 0.062872. 0.024005. 0. 0.432999. 0.084553. OSHARES. 2068193. 465713.2. 26189. 2.59E+07. 4612234. BtoM. 0.647709. 0.554844. 0.10183. 4.503539. 0.456989. The table presents summary statistics for the variables employed in the study. The stock price volatility is calculated as standard deviation (STDEV) of stock price. The measure of the quantity (QUANTITY) of news is based on the quantity of news that are reported company’s conference call by the mass media in our time window. The Scores (SCORE) are given by reading the news that refer to company’s conference calls, and its rules appears on Appendix 2. The affect or feeling of news (TONE) that refer to conference calls of a company. TONE is calculated as (Positive word-Negative word) divide by (Positive word+ Negative word). Return on Shareholders' Equity (ROE) is a company's net income divided by its average stockholder's equity. The Leverage (LEV) is computed as net debt/total asset at the end of every our research year (2010-2014). Outstanding shares (OSHARE) are the tradable shares available on the market 54.

(55) at the end of every our research year (2010-2014). The Book to Market (BtoM) is the ratio between book value and market capitalization, both taken at the end of every our research year (2010-2014). All the financial data were provided by the TEJ database.. 55.

(56) TABLE 5.PEARSON CORRELATION STED STED QUANTITY. QUANTITY. SCORE. TONE. ROE. LEV. OSHARES. BtoM. 1 0.1209. 1. SCORE. -0.1176. 0.1518*. 1. TONE. -0.0381. 0.0021. 0.2918***. 1. ROE. 0.2369***. 0.1552*. 0.0641. 0.1512*. 1. LEV. -0.1227. 0.0492. 0.0829. 0.0096. -0.3519***. 1. OSHARES. -0.1578*. 0.2717***. -0.0549. -0.209**. -0.1516*. 0.2578*. 1. BtoM. -0.1772*. -0.0896. -0.1452*. -0.2529***. -0.5754**. 0.3347*. 0.2501***. Note: 1. *** indicates significant at 1% level, ** indicates significant at 5% level and *indicates significant at 10% level. 2. Variable definitions are shown in Appendix 1.. 1.

(57) 4.2 EMPIRICAL RESULTS This research conduct a regression analysis to provide evidence that the financial news content have an impact on stock price volatility. Our measures of content are based on two methods, one is word counts of financially of each conference calls news, and the other is the rules to score the conference calls’ news. This research develop a customized word list of financially oriented words based on an analysis of commonly used words in conference calls. Appendix 2 shows that the rules to score the conference calls’ news and appendix 3 shows that the outcome of word counts and scores. The finding in Table 6 indicates that Stock price volatility (STDEV) has a significantly negative association with news content that are quantified by the Scores (SCORE) which are given by reading the news that refer to company’s conference calls; this result is consistent with the hypothesis that the score of conference calls news affects volatility; it is expected that positive score will reduce volatility and negative score will increase volatility. It is an evidence that conference calls’ news indeed influence the stock market in the time window. Investor read the news to understand company’ operating condition and future plans and influence their decision making in stock market; however, the tone (TONE) is not significant but the negative coefficient is as this research expected. The quantity (QUANTITY) of news that reported company’s conference call is significant that prove the stock market are indeed influenced not only by the news content but also the frequency of the news reported. Our control variable are also provide some interesting observations. As expected, the Leverage (LEV) coefficient is also positive, coherent with the hypothesis, higher leverage results in higher financial risk and ultimately higher stock price volatility, but 57.

(58) Leverage (LEV) is not statistically significant. The Book-to-market (BtoM) enters the regression is in our expectation with a negative coefficient, but it is not statistically significant. The negative coefficient of outstanding shares (OSHARE) variable is significant as our expectation that a larger quantity of tradable outstanding shares should be correlated to greater stability in stock prices. They are less likely to be influenced by volatility resulting from a scarce availability of shares on the market. The coefficient of return on equity (ROE) is positive as this research expect, and it is statistically significant.. 58.