VIX與財務預警 – 數據分析觀點 - 政大學術集成

全文

(1)國立政治大學會計學系碩士班碩士學位論文. VIX 與財務預警 – 數據分析觀點 VIX and Financial 政 Warning 治 – A Data Analytics. 立 Perspective. 大. ‧. ‧ 國. 學. n. er. io. sit. y. Nat. al. Ch. engchi. i n U. v. 指導教授：諶家蘭教授. 研究生：呂樂憫. 中. 華. 民. 國. 一〇七. 撰. 年. 七. 月. DOI:10.6814/THE.NCCU.ACCT.025.2018.F07.

(2) 致謝辭感謝上帝的帶領，讓我可以順利的完成畢業論文。回顧過去兩年碩士生活，我在學業、研究和生活上得到了許多老師、同學和朋友們的幫助和支持。在此，我要向他們表示我誠摯的謝意。論文得以順利完成，首先要感謝的是指導教授諶家蘭教授，老師在學術上給了我相當多的協助，每次遇到難題，老師都會抽空與我討論，沒有您悉心的指導，便沒有這篇論文的誕生；同時，也要感謝口試委員廖珮真教授及張仲岳教授提供許多精闢的見解，使畢業論文內容更趨完善。在此，再次向老師們表達謝意。另外，感謝資管所賴士詮提供與文字探勘技術相關的協助，幫助我們從上百萬篇新聞中，取得論文相關資訊。也感謝吳昱萱同學及陳羽謙同學在論文撰. 政治大寫期間的熱情幫助。感謝曾經在各方面給予幫助的夥伴們，在此，我再一次真立誠地向幫助過我的老師們和同學們表示感謝！. ‧ 國. 學. ‧. 最後，感謝我的父母及教會的弟兄姊妹無私的付出，在生活上給予很多的支持及鼓勵，使我能順利取得學位。. n. er. io. sit. y. Nat. al. Ch. engchi. i n U. v. DOI:10.6814/THE.NCCU.ACCT.025.2018.F07.

(3) 摘要新聞資訊能幫助投資人快速瞭解金融市場及總體經濟環境所發生之事情。從新聞內容中，投資人可以判斷整體市場的走勢。因此，新聞內容將影響投資人的投資決策，台灣投資人之情緒主要受本地新聞媒體所撰寫之報導所影響。本研究之樣本期間為 2007 年至 2017 年，新聞來源為全曜財經資訊股份有限公司 (CMoney)資料庫。本研究使用文字探勘技術，研究財務預警新聞與台灣投資人情緒之關聯性。本研究使用台灣恐慌指數(VIXTWN)作為衡量整體台灣投資人情緒之變數，觀察本地產業新聞及國際主要股市新聞與市場恐慌指數之關聯性。本研究之結果顯示，台灣投資人之整體情緒受本地產業新聞及全球股市新聞內容所影響。投資人情緒波動將反映在當日及明後兩日之恐慌指數上。新聞中所使用的字詞及語調，將影響投資人之情緒及對市場未來走勢之看法，並進一步影響投資人之投資決策。. 立. 政治大. 關鍵詞：恐慌指數、新聞、文字探勘、情緒分析、財務預警. ‧. ‧ 國. 學. n. er. io. sit. y. Nat. al. Ch. engchi. i n U. v. DOI:10.6814/THE.NCCU.ACCT.025.2018.F07.

(4) Abstract Mass media communicates with readers, investors can understand issues of the financial market through reading news articles. Information provided in the news articles plays an important role in affecting investors’ perspective on the future trend and opportunities of the financial market. Financial news are extracted from CMoney and the research period is 2007 to 2017. In this study, we the text mining technique to analyze the association between financial warning news and investors’ sentiment. The market volatility index (VIXTWN) will be used to quantify Taiwanese investors’ sentiment, models are established to observe how local industrial news and global stock market news affect market volatility. The empirical result of this study proves the relationship between local industrial and global stock market news and market volatility. Wordings and tone of news affect. 政治大 investors’ sentiment and their perspective on future market return. Therefore, changes 立 in investors’ sentiment affect their investment decision and further affect market ‧. ‧ 國. 學. volatility. Moreover, the study proves that market volatility reaction consist of two parts, immediate reaction and delayed reaction.. n. al. er. io. sit. y. Nat. Keywords: Volatility Index, Financial News, Text Mining, Sentiment Analysis, Financial Warning. Ch. engchi. i n U. v. DOI:10.6814/THE.NCCU.ACCT.025.2018.F07.

(5) Table of Contents 1. Introduction…………………………...………………………......….1 1.1 Research Purpose and Motivation…….…………………….…..1 1.2 Research Problem………………………………………….……6 1.3 Research Process………………………………………………..7 2. Literature Review…………………………………………………….8 2.1 Volatility Index………………………………………………….8 2.2 Macroeconomic Events………………………………………..13. 政治大. 2.3 Financial Warning……………………………………….……..16. 立. 2.4 Text Mining and Sentiment Analysis……………………….….18. ‧ 國. 學. 3. Research Method……………………………………………………23 3.1 Hypothesis Development………………………………………23. ‧. 3.1.1 The Relationship between Mood of News and VIXTWN...24. y. sit. Nat. 3.1.2 The Relationship between News Tone and VIXTWN…….26. er. io. 3.2 Data Collection………………………………………………..27. n. a 3.3 Text Analytic Approach on News………………………………30 v 3.4. i l C n hengchi U Regression Model………………………………….…………..34. 3.4.1 Dependent Variables…………………………………..…36 3.4.2 Independent Variables……………………………………36 3.4.3 Control Variables…………………………………………38 4. Empirical Result………………………………………………….…41 4.1 Descriptive Statistics…………………………………………..41 4.2 Correlation Analysis…………………………………….…..…44 4.3 Regression Analysis………………………………………...….49 4.3.1 Mood of Local Industrial News and VIXTWN……….…..49. DOI:10.6814/THE.NCCU.ACCT.025.2018.F07.

(6) 4.3.2 Mood of Global Stock Market News and VIXTWN……...52 4.3.3 Tone of Local Industrial News and VIXTWN…………...54 4.3.4 Tone of Global Stock Market News and VIXTWN……….56 4.3.5 Regression Result with Fixed Effect……………………..58 5. Conclusion and Discussion………………………………………….63 5.1 Research Discussion and Contribution………………………..63 5.2 Limitation and Future Research Work…………………………65 Appendix………………………………………………………………..67 References……………………………………………………………....78. 立. 政治大. ‧. ‧ 國. 學. n. er. io. sit. y. Nat. al. Ch. engchi. i n U. v. DOI:10.6814/THE.NCCU.ACCT.025.2018.F07.

(7) List of Figures Figure 1-1 Fraud Triangle………………………………………………...3 Figure 1-2 Trend Graph of VIXTWN………………………………….....4 Figure 1-3 Flow of Research Process…………………………………....7 Figure 2-1 Occurrence of Fraud Schemes: Account and Evidence Scheme Combination…………………………………………………………….16 Figure 3-1 Factors Affecting VIXTWN…………………………………23. 立. 政治大. ‧. ‧ 國. 學. n. er. io. sit. y. Nat. al. Ch. engchi. i n U. v. DOI:10.6814/THE.NCCU.ACCT.025.2018.F07.

(8) List of Tables Table 2-1 Summary of Key Volatility Index Literatures……………..…11 Table 2-2 Summary of Key Macroeconomic Literatures…………….…15 Table 2-3 Summary of Key Text and Sentiment Analysis Literature…...21 Table 3-1 Standard Industrial Classification of R.O.C………………….28 Table 3-2 Result of News Search……………………………………….29 Table 3-3 Summary of Data Sources……………………………………33 Table 3-4 Description of Variables……………………………………...40. 政治大. Table 4-1 Descriptive Statistics…………………………………………43. 立. Table 4-2 Pearson Correlation (Local Industrial News)………………...45. ‧ 國. 學. Table 4-3 Pearson Correlation (Major Stock Market News)……………46 Table 4-4 Tolerance and VIF of Variables (Model 1)…………………..47. ‧. Table 4-5 Tolerance and VIF of Variables (Model 2)…………………..47. y. sit. Nat. Table 4-6 Tolerance and VIF of Variables (Model 3)…………………...48. er. io. Table 4-7 Tolerance and VIF of Variables (Model 4)…………………..48. n. a of Model 1………………………………..51 Table 4-8 Regression Result v i l C n h Model Table 4-9 Regression Result of e n g c2………………………………..53 hi U Table 4-10 Regression Result of Model 3………………………………55 Table 4-11 Regression Result of Model 4………………………………57 Table 4-12 Regression Result of Model 1 with Fixed Effect…………..…59 Table 4-13 Regression Result of Model 2 with Fixed Effect…………..…60 Table 4-14 Regression Result of Model 3 with Fixed Effect……………..61 Table 4-15 Regression Result of Model 4 with Fixed Effect……………..62. DOI:10.6814/THE.NCCU.ACCT.025.2018.F07.

(9) List of Appendix APPENDIX 1 Taiwan Major Industries and Their Abbreviations..……...67 APPENDIX 2 Major Stock Markets and Their Abbreviations…………68 APPENDIX 3 Quantity of Industries News Collected………………….69 APPENDIX 4 Quantity of Stock Market News Collected……………...70 APPENDIX 5 Partial Sentiment Dictionary – Positive….………………71 APPENDIX 6 Partial Sentiment Dictionary – Negative…………………73 APPENDIX 7 Weighting of Different Industries of TAIEX……………..75. 政治大. APPENDIX 8 Sentiment Variables Calculation Illustration……………..76. 立. ‧. ‧ 國. 學. n. er. io. sit. y. Nat. al. Ch. engchi. i n U. v. DOI:10.6814/THE.NCCU.ACCT.025.2018.F07.

(10) 1. Introduction 1.1 Research Purpose and Motivation We live in an ever-changing world, we face many uncertainties in our daily life. The financial world is even more complex and constantly changing. In order to understand the important events that have occurred or will occur all over the world, investors gather information from the mass media. Information available on public media may show signs of firm-specific, industrial or macroeconomic financial warning. Contents of financial warning related news play an important role in affecting investors’ perspective on the future trend and opportunities of the financial market. Rational investors would adjust their investment decisions according to the information that they have collected. These adjustments may help investors to benefit more from the financial market.. 政治大 Text mining and sentiment analysis are techniques that scholars usually used to 立 analyze the relationship between investors’ sentiment and the financial market. ‧. ‧ 國. 學. Solomon (2012) suggested that the mass media help investors to process public information and decide which news event are economically important. Investor would expect greater profitability for corporations with more positive news, leading to a shortterm price increases on that particular stock. Ferguson et al. (2013) suggested that news. n. al. er. io. sit. y. Nat. content would affect investors’ sentiment and content of news has strong predictive power on future stock market returns. There is a strong relationship between news content and investors’ trading decision in short-term. However, news articles do not impact the market in long run.. Ch. engchi. i n U. v. Most recent researches developed models to investigate the impact of firm-specific news on that particular stock. Li et al. (2014) quantified investors’ sentiment related to firm-specific news and examined its predictive power on future CSI 100 stock movements. Ferguson et al. (2013) examined the association between news tone (percentage of positive and negative words), future stock return and trade volume in the UK stock market. Both studies have proved that investors’ sentiment is associated with tone and volume of news available publicly. Investors’ sentiment affect investors’ investing behaviors and further enhance market volatility. From recent scholars’ researches, we learn that firm-specific stock return and volume are affected by firm-specific investors’ sentiment. Investors’ sentiment would influence investor’s financial decision and affect market volatility. The Volatility Index can act as an indicator of overall investors’ sentiment. It represents the overall investors’ 1. DOI:10.6814/THE.NCCU.ACCT.025.2018.F07.

(11) level of confidence on the macroeconomic financial market. In this study, we would examine the impact of local industrial news and global stock market news on overall market volatility. In this study, we will focus on analyzing the relationship between macroeconomic news sentiment and local market volatility. The financial market is globalized, firms expand their business to other countries and investors can invest in stock markets all over the world. Movements in major global equity market would affect local Taiwanese investors’ sentiment, as the movements may affect investors’ return and the businesses that the investors invested in. Global political issues and political environment changes in the Republic of China (R.O.C.) are issues that would affect the financial market, investors’ sentiment and their sense of security. Agarwal et al. (2017) suggested that during financial crisis uncertainty is high and high level of uncertainty would lead to high volatility in the financial market. We would investigate whether the mood and tone. 政治大 of news regarding issues of major local industries and global stock market would 立 increase the overall market volatility. We believe that global political movements, ‧. ‧ 國. 學. natural disasters and some other financial events are signs of financial warning. Therefore, we would further analyze the relationship between global stock market news, local industrial news and stock market volatility.. n. al. er. io. sit. y. Nat. Financial distress is a condition that an enterprise experience difficulties in meeting or paying off the financial obligations to creditors. Financial warning is the sign that company are under financial distress or may face financial difficulties in the future. According to the fraud triangle suggested by Donald R. Cressey in the 1950s, there are three major factors that would persuade people to commit fraud. Those three factors are pressure, opportunity and rationalization. Pressure is what motivates an individual to commit fraud, when company are facing financial difficulties, the corporate management may have more incentive to alter financial records. Signs of financial warning may be discovered by the mass media and disclosed in news articles. Information about non-firm-specific financial warning would affect investors’. Ch. engchi. i n U. v. confidence in a particular industry or the overall financial market. We would include elements and signs of financial warning in our sentiment dictionary to observe the association between macroeconomic financial warning and market volatility. The fraud triangle is shown in Figure 1-1.. 2. DOI:10.6814/THE.NCCU.ACCT.025.2018.F07.

(12) Figure 1-1 Fraud Triangle We believe that macroeconomic and industrial news play an important role in impacting the financial market, global equity market and local industrial market news sentiment would be considered in our research model. News articles describing events. 政治大 in the macroeconomic world 立would be classified into local industrial news and global ‧. ‧ 國. 學. equity market news. The news tone and mood of both types of news will be quantified and scored. In this study, we would examine the effect of different types of external environment news on the overall market volatility in Taiwan. We believe that news articles would reveal signs of industrial-wide or market-wide financial warnings.. n. al. er. io. sit. y. Nat. The impact of external environment news on the financial market can be measured by using the Volatility Index (VIX). Market volatility reflect investors’ sentiment, as investors adjust their investing behavior according to the information that they gathered. VIX is introduced by the Chicago Board Options Exchange (CBOE) in 1993. It can be used to estimate the market expected volatility on Standard & Poor's (S&P) options over the next 30 days. VIX was originally designed to measure the implied volatility of S&P 100 index options. In 2003, the methodology of calculating VIX was revised, it expanded its range of calculation and measure the volatility of S&P 500 index options. The revision enhances the accuracy of prediction on investor trading intention. VIX is calculated by averaging the weighted prices of S&P puts and calls over a wide range of. Ch. engchi. i n U. v. strike prices. High volatility level reflects investors’ pessimism in the future and expects high trading level of options in the near future, and vice versa (Fernandes et al, 2014). VIX can also be known as the investor fear index or the fear gauge, it can measure the level of fear and sense of security of investors on the financial market. VIX is a useful tool in investigating investors’ sentiment on the macroeconomic market.. 3. DOI:10.6814/THE.NCCU.ACCT.025.2018.F07.

(13) Our study focuses on quantifying the news sentiment of macroeconomic news, VIX calculated by CBOE is not a useful indicator of Taiwanese market volatility. The volatility index of Taiwan is calculated and published by the Taiwan Futures Exchange (TAIFEX). TAIFEX was established in 1998 under the Futures Trading Act, it offers futures and options on major Taiwanese stock indices, equity option etc. In 2006, the TAIFEX introduced the Taiwanese options volatility index (VIXTWN). VIXTWN can be used to measure the level of volatility on the Taiwanese option market. VIXTWN is calculated by using the VIX-formula developed by the CBOE, the application of VIXformula by the TAIFEX is used under the authorization of S&P. In this study, we will use VIXTWN to measure the volatility level of the Taiwanese stock market. Figure 12 demonstrate the trend of VIXTWN between January 2007 and December 2017.. 70. 立. 60. io. y. sit. 0. Nat. 10. ‧. 20. er. 30. ‧ 國. 40. 學. 50. VIXTWN 治政大. al. n. v i n CTrend Figure 1-2 of VIXTWN h e nGraph h gc i U. As we mentioned above, VIXTWN will be used to observe the changes in investors’ sentiment. In this study, we will focus on researching the impact of macroeconomic news on investors’ sentiment. Carretta et al. (2011) used the text mining and sentiment analysis techniques to quantify investors’ sentiment. In the study, news tone is used to measure investors’ sentiment, news tone can be determined by using the formula (P − N)/(P + N), where N and P represent the number of positive and negative words appeared in the financial news respectively. Negative words and positive word are classified according to the definition provided by the Harvard IV Psycho Social Dictionary. In this study, we further develop and enrich the dictionary built by the Innovative and Mobile Financial Service Technologies, Modeling and Applications project of the Ministry of Science and Technology. Words related to financial warning and risk are added to the dictionary, in order to consider the 4. DOI:10.6814/THE.NCCU.ACCT.025.2018.F07.

(14) interaction between VIX and financial warning. Those financial warning related words are extracted from the dictionaries built by Chang (2009) and Lin (2013). We will use the integrated sentiment dictionary to analyze news tone and other sentiment variables of news. We believe that this research can contribute to the society by providing reasonable prediction about future market volatility trend when different macroeconomic events occur. In this study, we would analyze the financial news article announced by CMoney, an Institutional Investors Investment Decision Supporting System, the research period is January 2007 to December 2017. In order to understand the relationship between macroeconomic events and market volatility, quantity of related news collected, their news tone and other sentiment variables will be examined.. 立. 政治大. ‧. ‧ 國. 學. n. er. io. sit. y. Nat. al. Ch. engchi. i n U. v. 5. DOI:10.6814/THE.NCCU.ACCT.025.2018.F07.

(15) 1.2 Research Problem Based on the research’s purpose and motivation stated in Section 1, this study examines the relationship between macroeconomic news, macroeconomic and industrial financial warning and market volatility (VIXTWN). Research questions of this study are listed as follows: How does the mood of local industrial news affect VIXTWN? How does the mood of global stock market news affect VIXTWN? Does the tone of local industrial market news affect VIXTWN? Does the tone of global equity market news affect VIXTWN?. 立. 政治大. 學 ‧. ‧ 國 io. sit. y. Nat. n. al. er. 1. 2. 3. 4.. Ch. engchi. i n U. v. 6. DOI:10.6814/THE.NCCU.ACCT.025.2018.F07.

(16) 1.3 Research Process Our study begins with an Introduction. The introduction in Section 1 suggested the primary research purposes, the research motivation and briefly introduced the VIX index, investors’ sentiment, their backgrounds and relationships. Recent proposed literatures and related works are reviewed and discussed in Section 2, the literature review provide theoretical background and foundation for this study. In Section 3, detail discussion of hypothesis development, data collection, choice of variables, text analytic skills and the regression model of this study is provided. The empirical result of the study will be demonstrated in Section 4, including the descriptive statistics result, result of the regression model and the correlation analysis. The findings, limitations of the study and future research directions will be concluded in Section 5. The research process is shown as follows:. 立. 政治大. ‧. ‧ 國. 學. Introduction. Literature Review. n. er. io. sit. y. Nat. al. i n U. v. Research Method and Model Design. Ch. engchi. Empirical Result. Conclusion and Contribution. Figure 1-3 Flow of Research Process. 7. DOI:10.6814/THE.NCCU.ACCT.025.2018.F07.

(17) 2. Literature Review 2.1 Volatility Index VIX index was introduced by the CBOE in 1993. There were considerable amount of recent researches investigating the relationship between VIX, financial return and other macroeconomic events. VIX index is also regarded as the fear index or the fear gauge. VIX is highly correlated with investors’ sentiment and affect their financial decisions. In this section, we will discuss the findings of VIX-related literatures. Fleming et al. (1995) developed a model to investigate the relationship between the fluctuation of VIX and S&P 100 return. The research found that there is a strong short-term relationship between changes in VIX and market return. Negative movements and events in the financial market would impact on the volatility more significantly than positive events. Result shown that there is an inverse relationship. 政治大 between volatility and stock prices. 立. ‧. ‧ 國. 學. Simon and Wiggins (2001) identified volatility index (VIX), the put-call ratio and the trading index (TRIN) for Standard & Poor’s (S&P) 500 futures returns as marketbased indicators of investors’ sentiment. Research found that VIX, put-call ratio and TRIN are useful indicators in predicting S&P futures return statistically and. al. er. io. sit. y. Nat. economically. Excellent buying opportunities are created when the level of fear in the stock market rise.. n. Corrado and Miller (2005) examined the forecast quality of VIX index based on S&P 100 and S&P 500 stock returns. Scholars found that the forecast quality of CBOE implied volatilities for the S&P 100 (VXO) and S&P 500 (VIX) has improved since 1995. However, VXO and VIX yield upwardly biased volatility forecasts but they were more efficient than historical volatility. The writers confirmed the predictive power of VXO and VIX on future returns, forecast errors had been significantly improved after 1995.. Ch. engchi. i n U. v. Becker et al. (2009) claimed that recent researches had identified the relationship between implied volatilities and future forecast of volatility. They considered the information content provided by the S&P 500 VIX index. The scholars believed that VIX index indicates market confidence and plays an important role in market forecast. Result shown that VIX index subsume information relating historical jump contributions to total volatility of S&P 500 and reflects incremental information relating to future market jump activities and market returns. 8. DOI:10.6814/THE.NCCU.ACCT.025.2018.F07.

(18) Brière and Drut (2009) said that VIX can be used to represent investors’ expectations on future equity volatility and there is a strong relationship between currency carry trade and the equity market. Scholars used VIX to anticipate future currency market movements, VIX can help investors to determine whether they should use the carry trade strategy or the purchasing power parity strategy in the currency market. Baba and Sakurai (2011) investigated on whether macroeconomic variables can predict regime changes in VIX, regime-switching approach was applied. They identified three stages of regime changes, namely tranquil regime, turmoil regime and crisis regime. Tranquil regime with low volatility, turmoil regime with high volatility and crisis regime with extremely high volatility. They found that term spread are related to regime switch in VIX.. 政治大 Hammer and Russo (2012) suggested that VIX index provide evidence of 立 continued economic uncertainty, VIX can also be referred to as the “fear gauge”. Daily ‧. ‧ 國. 學. VIX could help investor to understand the current market condition and status. Investors and their advisers would bring market conditions into consideration, and develop suitable investing plans and related tax strategies. The paper introduced seven strategies that may mitigate investment risk from market uncertainty and enhance after-tax return.. y. Nat. sit. n. al. er. io. Goodell and Vähämaa (2013) examined the effects of political uncertainty and the political process on implied stock market volatility during US presidential election cycles. Scholars used VIX to measure market uncertainty. The findings of this research suggested that presidential election process engendered market anxiety and the market volatility increased. Investors would revise their future market expectations due to changes in macroeconomic policies. The paper provided information about the interactions between stock markets, public policies, and the potential role of political uncertainty in the financial markets.. Ch. engchi. i n U. v. Chung and Chuwonganant (2014) suggested that market uncertainty would affect the market volatility seriously and lead to a co-movement of individual assets liquidity. The study found that the impact of VIX on market liquidity is a great determinants of stock liquidity and surpass the determinants suggested by other researches. VIX could explain risk that are not captured by asset liquidity and asset liquidity is correlated with VIX.. 9. DOI:10.6814/THE.NCCU.ACCT.025.2018.F07.

(19) From the research done by Kim et al. (2016), we learn that the dispersion in GDP forecast and VIX can be used as indicators measuring macroeconomic uncertainties. Dispersion in GDP forecast and VIX can both capture future expectations about uncertainties of the economic world. VIX could act as an indicator of changes in investors mind due to macroeconomic changes. From researches mentioned above, we know that VIX index is a hot subject matter in global academic researches. In this study, we would focus on examining the relationship between changes in investors’ sentiment due to macroeconomic issues and VIX.. 立. 政治大. ‧. ‧ 國. 學. n. er. io. sit. y. Nat. al. Ch. engchi. i n U. v. 10. DOI:10.6814/THE.NCCU.ACCT.025.2018.F07.

(20) Table 2-1 Summary of Key Volatility Index Literatures Author. Research Period. Algorithm. Hypothesis. Variables. Conclusion. Fleming et al. (1995). 1986 - 1992. Regression. VIX can predict future market movements and returns.. VIX and Stock Index. VIX exhibits a strong temporal relationship with stock market returns and there is an inverse relationship. 立. ‧ 國. Volatility index, put-call ratio and trading index can predict returns on the Standard & Poor’s (S&P) 500 futures contract.. contribute to price volatility.. Volatility,. VIX reflects past jump activities and provide information to explain future. VIX reflect incremental information pertaining to future jump activity.. VIX. jump activity.. sit. y. CBOE implied volatilities appear to contain significant forecast errors before 1995, little indication of significant forecast errors after 1995.. al. iv n C Historical h ejump n gactivity c h i U Realized. n. 2 January 1990 – 17 October 2003. Regression. VIX, put-call ratio and trading Index had both statistically and economically significant predictive power for S&P returns.. VXN, VXO, VIX can VXN, VXO, predict return of Nasdaq 100 VIX stock, S&P 100 and S&P 500 stock respectively.. er. Regression. io. Becker et al. (2009). VIX, Put-call Ratio, Trading Index (TRIN). ‧. 1995 - 2003. Regression. Nat. Corrado and Miller (2005). January 1989 June 1999. between expected volatility and stock market prices.. 學. Simon and Wiggins (2001). 政治大. 11. DOI:10.6814/THE.NCCU.ACCT.025.2018.F07.

(21) Table 2-1 Summary of Key Volatility Index Literatures (cont.) Author. Research Period. Algorithm. Hypothesis. Baba and Sakurai (2011). January 1990 June 2010. Regression Macroeconomic variables VIX, Consumer can predict the regime shifts Price Index, in the VIX index. Producer Price. 政治 Index, 大 Index of. Regression Market uncertainty would affect the market volatility seriously and lead to a comovement of individual assets liquidity. Regression Macroeconomic uncertainty. Nat. io. n. Kim et al. (2016). 1996-2008. Ch. engchi. plays an important role in managers’ decision to issue management earnings forecasts.. VIX. y. January 2007 – December 2009. sit. Chung and Chuwonganant (2014). ‧ 國. Regression There is a negative association between stock market volatility and uncertainty about the winner of the election.. There are three stages of regime changes (tranquil, turmoil and crisis). Term spread are related to regime switch in VIX. Political uncertainty around US presidential elections affects stock market volatility.. ‧. 1992 – 2008. Industrial Production VIX, Inflation rate, Unemployment rate, Consumer Confidence Index. 學. Goodell and Vähämaa (2013). al. Conclusion. er. 立. Variables. i n U. VIX could explain risk that are not captured by asset liquidity and asset liquidity is correlated with VIX.. v. Dispersion in GDP,. Dispersion in GDP forecast and the. VIX. volatility index can be used as indicators in measuring macroeconomic uncertainties. 12. DOI:10.6814/THE.NCCU.ACCT.025.2018.F07.

(22) 2.2 Macroeconomic Events Recent studies investigated on economic factors and non-economic factors, both types of factor would impact the financial market and affect the macroeconomic world. In this study, we would examine the impulse that macroeconomic news assert to the financial market. Kurihara (2006) analyzed the relationship between stock market and macroeconomic variables based on information available in the Japanese market. Scholar suggested that there are many different types of factors affecting the financial market, these factors include enterprise performance, interest rates, dividends, gross domestic product (GDP), exchange rates, stock prices of other equity markets, money supply, employment rate etc. and factors that influence the stock market would change over time. However, Kurihara (2006) found that domestic interest rate does not influence the Japanese stock market.. 立. 政治大. ‧. ‧ 國. 學. Savor and Wilson (2011) researched on the relationship between macroeconomic news announcement and stock return. Important macroeconomic announcement include inflation rate, interest rate and unemployment rate announcements. The Sharpe ratios and the stock market average returns are usually significantly higher on the scheduled-announcement-day. Scholars found that investors would trade-off between. al. er. io. sit. y. Nat. macroeconomic risk and asset returns, the research estimated investors demand to bear macroeconomic risk.. n. Gilbert (2011) suggested that important macroeconomic announcement may undergo significant revision after a period of time, investors should filter out noises from news announcement. Result stated that revisions do matter and investors care about the true final value of a macroeconomic release. The economic significance of the result highlighted the influence of macroeconomic events on investors’ sentiment and decisions.. Ch. engchi. i n U. v. Pastor and Veronesi (2012) suggested that governmental policy changes would affect the financial market, they analyzed the impact of changes in government policies on stock prices. Government’s decision usually have economic or non-economic motives, the study found that both type of policy change would affect the financial market. In general, they found that uncertainty in government policies is negatively correlated with stock prices, policies change would increase market volatility. Examples of policy change include reforms in health care policies, subsidizing criterion and financial regulations. 13. DOI:10.6814/THE.NCCU.ACCT.025.2018.F07.

(23) Bansal et al. (2014) investigated the relationship between cash flow, discount rate, volatility and equity return. Cash flow risk, discount rate risk, and volatility risks, carry separate risk premium and must be taken into consideration. Scholars emphasized the importance of macroeconomic volatility movements in determining assets price. Result suggested that increase in market volatility is related to the increase in discount rate and the decline in consumption. Kelly et al. (2016) analyzed the price of political uncertainty, based on the impact of different government policy decisions. Scholars exploited the variation of stock price around the national elections and global summits. Result suggested that the political uncertainty is priced in the equity option market, and the effects of political uncertainty are larger when the economy is weaker. However, the effects of political uncertainty are not larger when the uncertainty is higher. We can conclude that the political environment would impact the stock and option market.. 立. 政治大. ‧. ‧ 國. 學. Liu et al. (2017) examined the impact of political uncertainty on the equity market. Scholars used the Chinese political event, Bo Xilai political scandal in 2012, to identify the impact of political issue on stock prices. They found that the event caused a significant drop in stock prices, and the impact were more significant on politically sensitive firms. The financial impact on the equity market was mainly driven by the. al. er. io. sit. y. Nat. changes in discount rate. Political issue would significantly affect stock prices and investors’ sentiment.. n. Based on the research stated above, we know that market behaviors are affected by macroeconomic activities. In this study, we would further analyze the impact of macroeconomic events on investors’ sentiment and market volatility.. Ch. engchi. i n U. v. 14. DOI:10.6814/THE.NCCU.ACCT.025.2018.F07.

(24) Table 2-2 Summary of Key Macroeconomic Literatures Author. Research Period. Algorithm. Hypothesis. Kurihara (2006). 19 March 2001 – 30 September. Regression Japanese stock price is affected by. 2005. Variables. Conclusion. Exchange Rate, Interest Rate,. Macroeconomic variables influence stock prices but domestic interest rate. Rate 政治 Employment 大. macroeconomic variables.. 立. does not affect stock price.. Regression Stock return is affected by macroeconomic news announcement.. Inflation Rate, Interest Rate and Unemployment Rate. Sharpe ratios and stock market returns are usually higher on the scheduled-announcement-day and investors would trade-off between macroeconomic risk and asset returns.. Gilbert (2011). February 1985 – June 2005. Regression Macroeconomic announcements and their revisions affect the stock market.. Unemployment Rate, Industrial Production, Nonfarm payroll. Investors care about the true value of a macroeconomic release.. Discount rate, Consumption. Volatility news is an important source of risk that affects the measurement. n. 1930 – 2010. engchi. Regression Asset price fluctuations is associated with the macroeconomic world.. sit. io. Bansal et al. (2014). Ch. er. Nat. al. y. ‧. ‧ 國. 1958 – 2009. 學. Savor and Wilson (2011). i n U. v. and interpretation of underlying risks in the economy and financial markets.. 15. DOI:10.6814/THE.NCCU.ACCT.025.2018.F07.

(25) 2.3 Financial Warning Financial warning are often regard as an early alert of financial trouble in a specific firm. Financial warning affects investors’ sentiment and their intention to further invest in the stock market. Apart from firm-specific financial warning, we believe that industrial or regional financial warning will also affect investors’ sentiment. In this part, we will look into recent researches that have discussed different types of financial warning, fraud and misrepresentation. Financial warning related words are translated to Chinese and included in our sentiment dictionary. Gray and Debreceny (2014) create a taxonomy that combines research on patterns of observed fraud with an appreciation of areas that benefit from data mining. Scholars recognized major fraudulent behaviors and their relative frequency. Major fraudulent behaviors include fictitious revenue, premature revenue recognition, fictitious assets and overstated assets, understated liabilities and omitted disclosure. These fraudulent. 政治大 behaviors are often covered by using fake, altered or hidden documents, collusion with 立 third parties, fake products, client misrepresentation etc. The paper suggested auditors ‧. ‧ 國. 學. to widely incorporate data mining techniques in analyzing the reliability of information available publicly. The accuracy of information could affect investors’ financial decision. Popular fraud and their frequency of occurrence suggested in the literature is shown in Figure 2-1.. n. er. io. sit. y. Nat. al. Ch. engchi. i n U. v. Figure 2-1 Occurrence of Fraud Schemes: Account and Evidence Scheme Combinations 16. DOI:10.6814/THE.NCCU.ACCT.025.2018.F07.

(26) Chen et al. (2017) stated that annual reports are valuable reference for creditors and investors to understand the operational performance, financial status and social contributions. However, many annual reports exaggerated the activities that the enterprise have done to raise investors' capital and other financial supports. Effectively detecting fraud in the annual report of a company is thus a priority concern during an audit. From this research, we learn that information available in the market would affect investors’ financial decision. Therefore, fraudulent information would mislead investors. Yin et al. (2018) researched on the role of time-varying betas in estimating abnormal returns around important news announcements. The study analyzed the stock price reaction to financial profit warnings on stocks listed in the Hong Kong Stock Exchange. The result of this study suggest that abnormal returns are positive (negative) on the days on which positive (negative) warnings are released and negative (positive). 政治大 on the subsequent days. This study shows that post-warning abnormal returns patterns 立 are largely consistent with the predictions of the Efficient-market hypothesis.. ‧ 國. 學. ‧. From the above researches, we understand the importance of having accurate information about enterprise performance. Exaggerated information shown in financial reports may reflect financial warning in firm’s operation. We believe that industrial-. n. al. er. io. sit. y. Nat. wide and market-wide financial warning will influence investors’ sentiment and their investment decisions. Signs of financial warning are introduced in the above literatures, wordings related to financial warning are captured in our sentiment dictionary and financial warning related news tone will be calculated. In this study, we would analyze how macroeconomic news tone and their wordings affect investors’ sentiment and market volatility.. Ch. engchi. i n U. v. 17. DOI:10.6814/THE.NCCU.ACCT.025.2018.F07.

(27) 2.4 Text Mining and Sentiment Analysis Text announcement and media content would affect investors’ sentiment, investors’ sentiment would induce investors’ decision. News provide investors with information related to the financial market and future market trend, therefore, it is important to analyze information generated by the mass media and the impulse raised by the media content. Tetlock (2007) measured the interaction between stock market activity and daily media contents. Scholar analyzed news pieces available on Wall Street Journal (WSJ) column. Research studied on the linkage between measure of media pessimism and the stock market using basic vector autoregressions (VAR). Result suggested that high level of pessimistic and uncertainty quotes predict downward movement on stock price. Extremely high or low level of pessimism anticipate rise in trading volume. We can conclude that daily media content and stock market return and volume are highly. 立. correlated.. 政治大. ‧. ‧ 國. 學. Kaplanski and Levy (2010) suggested that investors’ mood and anxiety would negatively affect their investment decision and hence affect the entire stock market. The study investigated the impulse caused by aviation disaster on investors’ sentiment, scholar believed that aviation disaster would increase investors’ anxiety and reduce the. n. al. er. io. sit. y. Nat. market demand on risky investments and affect stock prices. Result suggested that aviation disaster are followed by negative rates of return in the stock market accompanied by a reversal effect two days later. VIX and VXO would increase after aviation disasters.. Ch. engchi. i n U. v. Akhtar et al. (2011) examined the impact of monthly released Australian consumer sentiment news on the equity market. Scholar believed that announcement of negative (positive) consumer sentiment news will induce a negative (zero) stock market reaction. Result reflected that consumer sentiment contain valuable information content and it is useful in predicting variation in stock market activities. Scholars suggested that the monthly Australian Consumer Sentiment Index (CSI) is a good proxy of for investors’ sentiment in the Australian equity market. If the recently released CSI is lower than that of the previous CSI, the Australian market will suffer from a significant announcement negative day effect. However, there is no significant impact if CSI is higher than that of the previous month.. 18. DOI:10.6814/THE.NCCU.ACCT.025.2018.F07.

(28) Carretta et al. (2011) investigated on how news communication affect the stock market, they extracted information available between 2003 and 2007 in a major Italian financial newspaper. From the result of the analysis, scholars concluded that the mass media have significant impact on the financial markets since news reports contribute to the formation of investor expectations. The paper first proved the possibility that investor behavior is influenced by the content of the news and by the tone of communication. Zhang and Skiena (2011) used the natural language processing techniques to analyze the text content of blogs and news. They studied the anticipating power of text appeared in blogs and the mass media on market trade volume and future returns. Based on the sentiment data available from blogs and news, scholars developed a marketneural strategy to generate information from text data. Result suggested that news and blogs polarity are correlated with market volume and return. The sentiment conveyed. 政治大 by blogs will last longer than that of news, as content of blogs include more personal 立 feelings if investors.. ‧ 國. 學. ‧. Baker et al. (2012) constructed the investor sentiment indices for six major equity market and investigated the impact of investor sentiment on both local and global markets. Result shown that relative sentiment is correlated with the stock prices of dual-. n. al. er. io. sit. y. Nat. listed companies. Global investors’ sentiment is a useful contrarian predictor of country-level stock returns and both global and local sentiment are contrarian predictors of future returns. Investors’ sentiment play a significant role in international market volatility and generates market predictability. From the study, we learn that both local and global investors’ sentiment are related to future stocks returns.. Ch. engchi. i n U. v. Liu (2012) provided an overview of sentiment analysis and opinion mining. Scholar defined sentiment analysis as a field of study that investigate people’s opinions, emotions and attitudes. Liu (2012) believed that opinion words are powerful indicators of sentiment, these opinion words reflect positive or negative feelings. We could use the number of positive and negative wordings appeared in news article as variables of study.. 19. DOI:10.6814/THE.NCCU.ACCT.025.2018.F07.

(29) Ferguson et al. (2013) suggested that news media provide valuable information for market participants. Both private and public information shape investors’ expectation on firm’s future performance. Scholars used the information from UK market and gather international evidence for the predictive power of news content on future stock returns, there is significant interaction between positive and negative content and stock returns. Market participants react more significantly for news with higher level of visibility. Li et al. (2014) proposed a media-aware quantitative model to quantify how investors’ sentiment predict future stock market movements. Scholars used the percentage of positive and negative words in news articles to measure the optimistic and pessimistic mood of investors. Result showed that new article sentiment would fluctuate investors’ sentiment and interfere with their decision making. However, news article from different sources would affect the market differently. Moreover, news. 政治大 concerning executive personnel change and new product release are more influential 立 comparing with other types of news.. ‧ 國. 學. ‧. Benhabib et al. (2016) investigated the impact of financial information on investors’ sentiment fluctuations in equity market. Small investor sentiment shock would be amplified in the financial market and affect the economic situations. The. al. er. io. sit. y. Nat. paper illustrated the possibility of sentiment driven fluctuations in stock prices and realtime output.. n. Yang et al. (2017) applied the text mining approach to evaluate firms’ risks using unstructured textual disclosure from firm’s financial reports. Scholars suggested there are many hidden information hided between the lines in corporate public documents. The study used the Natural Language Processing techniques to extract firms’ selfidentified risks including financial, strategic, operational, and hazard risks and evaluate the relationship between firm’s overall risks and its audit fee.. Ch. engchi. i n U. v. From research discussed above, we learn that publicly available information such as news and annual reports content would affect investors’ sentiment and decision. These findings suggested the importance of sentiment mining based on news content. In this study, we would analyze the interaction between of the external environment news and VIXTWN.. 20. DOI:10.6814/THE.NCCU.ACCT.025.2018.F07.

(30) Table 2-3 Summary of Key Text and Sentiment Analysis Literatures Author. Research Period. Algorithm. Hypothesis. Variables. Conclusion. Tetlock (2007). 1984 - 1999. Regression. Pessimism factor serve as proxies for investor sentiment, return and. Pessimism Media Factor,. High values of media pessimism induce downward pressure on. 學. market price. High or low values of pessimism lead to temporarily high trading volume. Aviation disasters are followed by negative rates of return in the stock market accompanied by a reversal effect two days later.. 政治大 Daily Volume. daily trading volume.. ‧ 國. 立. January 1950 – Regression December 2007. Increased anxiety after aviation disasters will lead to a short-term reduction in demand for risky assets and affects stock prices. Akhtar et al. (2011). June 1992 – December 2009. The announcement of negative Stock Return, (positive) consumer sentiment Consumer news will induce a negative (zero) Sentiment Index stock market reaction.. n. al. er. io. sit. y. Nat. Regression. Carretta et al. (2011). 2003 – 2007. Stock Returns. ‧. Kaplanski and Levy (2010). Regression. Ch. engchi. i n U. v. Corporate governance news. Abnormal. would impact the stock market and investor expectations.. Return, News Tone. CSI has valuable information content. When a lower (higher) than previous month CSI is announced, the stock market suffers a (no) significant negative announcement day effect. Investors are influenced by the content (positive or negative) and tone of communication (strong or weak) of the news publications.. 21. DOI:10.6814/THE.NCCU.ACCT.025.2018.F07.

(31) Table 2-3 Summary of Key Text and Sentiment Analysis Literatures (cont.) Author. Research Period. Algorithm. Hypothesis. Variables. Conclusion. Zhang and Skiena (2011). 2005 - 2009. Regression. Company related news variables can anticipates the. News volume, Media Polarity,. News and blogs sentiment can predict volume and return. The. Trade Volume. sentiment conveyed by blogs can last longer than news.. 政治大Stock Return,. company’s stock trading volumes and financial returns.. io. Semantic measures of news media content can predict future stock returns. n. al. Li et al. (2014). Ch. engchi. January 2011 – Regression December. News article and public mood would influence stock market. 2011. movements.. Positive Content, Negative Content, News Volume, Stock Return. y. Regression. Stock Return, Sentiment Indices. sit. ‧ 國. Investors’ sentiment play a significant role in international market volatility and generates market predictability.. ‧. 1981 – 2010. Regression. Nat. Ferguson et al. (2013). 1980 – 2005. 學. Baker et al. (2012). er. 立. i n U. v. Relative sentiment is correlated with the relative prices of duallisted companies. Global sentiment is a contrarian predictor of country-level returns. Visibility and tone are key factors in determining how investors respond to news.. Optimistic Mood, Pessimistic Mood,. Media influence of financial news on stocks exists and can be. Stock Return. quantified using natural language processing techniques.. 22. DOI:10.6814/THE.NCCU.ACCT.025.2018.F07.

(32) 3. Research Method 3.1 Hypothesis Development In this section, we will discuss the hypothesis development of this study. Our study identifies various factors affecting VIXTWN. VIXTWN measures the overall volatility of the market, it would be affected by many different macroeconomic factors. In light of recent researches, we find that factors affecting VIXTWN can be classified into three different groups. News affecting investors’ sentiment can be categorized into local industrial news and global stock market news. We believe that wordings used in the news articles would affect investors’ sentiment, tone of local industries news and global stock market news will impact the Taiwanese equity market differently. Apart from the sentiment variables, many local macroeconomic indicators would affect investors’ confidence on future market movements. Macroeconomic indicators such as inflation rate, unemployment rate, consumer confidence index and index of industrial production. 政治大 are used as control variables. Figure 3-1 illustrates three groups of factors affecting 立 VIXTWN, these factors will be analyzed in this study.. ‧ 國. 學 ‧. Local Industries News. VIXTWN. y. Nat. n. er. io. al. sit. Macroeconomic Variables. Ch. Global Stock Market News. engchi. i n U. v. VIXTWN Macroeconomic Variables. Figure 3-1 Factors affecting VIXTWN. 23. DOI:10.6814/THE.NCCU.ACCT.025.2018.F07.

(33) 3.1.1 The Relationship between Mood of News and VIXTWN In this study, we want to examine the effect of mass media information on the financial market. Rational investors normally rely on information described in news articles. We would analyze how the mood of news affect investors’ sentiment in our first hypothesis. In Section 2, we have discussed various types of events that would affect investors’ sentiment and behavior. Savor and Wilson (2011) suggested that inflation rate, interest rate and unemployment rate announcements would impact on investors’ investment behavior. Pastor and Veronesi (2012) introduced the interaction between governmental policy changes and financial market, uncertainty in policy change would increase market volatility and reduce stock price. In light of the findings introduced by the above researches, we would classify non-firm-specific news into different industries’ news or. 政治大 different equity market’s news according to their content. 立. ‧. ‧ 國. 學. Recent researches proved the interaction between the mass media and the financial market. Tetlock (2007) stated that high level of pessimism in financial news exert a downward pressure on future stock price. Tetlock (2007) suggested that news wordings would affect investors’ perspective on future return, extremely high or low levels of. al. er. io. sit. y. Nat. negative words appeared in news articles would predict high trading volume and media pessimism would anticipate low future returns.. n. Ferguson et al. (2012) suggested that the percentage of positive (negative) words over total number of words appeared in publicly available information would affect investors’ future expectations. The percentage of positive (negative) words over total number of words can be known as the optimistic (pessimistic) mood of news. Market movements could be anticipated by analyzing the information content provided in the news articles. In light of this study, we learn that the content of financial news affect the market volatility. A sentiment dictionary will be constructed to quantify the. Ch. engchi. i n U. v. optimistic and pessimistic mood of news. The association between optimistic and pessimistic mood of news and market volatility will be examined in this study.. 24. DOI:10.6814/THE.NCCU.ACCT.025.2018.F07.

(34) Furthermore, most recent researches focused on investigating the relationship between the news sentiment variables of the current day and the stock market. DellaVigna and Pollet (2009) researched on the association between earnings announcements and stock returns. They found that stock price reactions consist of two parts, immediate reactions and delayed reactions. We believe that news sentiment would generate immediate and delayed fluctuations in market volatility. In order to observe the delayed effect of news on market volatility, we construct hypotheses analyzing the relationship between VIXTWN and the sentiment variables of the day before and two days before. H1a: When the mood of local industrial news of today is more optimistic, the VIXTWN decreases.. 政治大. H1b: When the mood of local industrial news of the day before is more optimistic, the VIXTWN decreases.. 立. ‧ 國. 學. H1c: When the mood of local industrial news of two days before is more optimistic, the VIXTWN decreases.. ‧. H2a: When the mood of global stock market news of today is more optimistic, the. y. Nat. VIXTWN decreases.. sit. H2b: When the mood of global stock market news of the day before is more optimistic,. al. er. io. the VIXTWN decreases.. v. n. H2c: When the mood of global stock market news of two days before is more. Ch. optimistic, the VIXTWN decreases.. engchi. i n U. 25. DOI:10.6814/THE.NCCU.ACCT.025.2018.F07.

(35) 3.1.2 The Relationship between News Tone and VIXTWN Carretta et al. (2011) mentioned that news tone is a useful indicator in predicting future return, news tone is calculated by the formula (P − N)/(P + N), where P and N are the number of positive and negative words appeared in the news respectively. Zhang and Skiena (2011) used the same formula to measure media polarity. They found that daily news tone are correlated with stocks return. Recent researches have proved the relationship between news tone and stock returns. Our study focuses on analyzing the effect of local industrial and global equity market news tone on market volatility. The sentiment dictionary integrated by this study will be used to quantify news tone. News tone represent the strength of information illustrated by the news articles. We would observe the immediate and delayed relationship between news tone and market volatility.. 政治大 H3a: When the local industrial news tone of today is more optimistic, the VIXTWN 立 decreases.. ‧ 國. 學. H3b: When the local industrial news tone of the day before is more optimistic, the VIXTWN decreases.. ‧. H3b: When the local industrial news tone of two days before is more optimistic, the. sit. y. Nat. VIXTWN decreases.. io. al. n. VIXTWN decreases.. er. H4a: When the global stock market news tone of today is more optimistic, the. Ch. i n U. v. H4a: When the global stock market news tone of the day before is more optimistic, the VIXTWN decreases.. engchi. H4a: When the global stock market news tone of two days before is more optimistic, the VIXTWN decreases.. 26. DOI:10.6814/THE.NCCU.ACCT.025.2018.F07.

(36) 3.2 Data Collection Many macroeconomic variables are considered in this study. We believe that macroeconomic variables affect market volatility. TAIEX, unemployment rate, inflation rate, consumer confidence index and the index of industrial production are control variables of this study. The data of our dependent variable, VIXTWN, and other control variables are extracted from the Taiwan Economic Journal (TEJ) database. This study focuses on investigating the relationship between macroeconomic and industrial news variables and market volatility. The research period of the study is January 2007 through December 2017. Market volatility measures the overall volatility of the external financial environment and we will examine how macroeconomic news affect investors’ sentiment. The macroeconomic news articles that we use in this study is extracted from CMoney, the Institutional Investors Investment Decision Supporting System. CMoney has developed a platform that fulfill the demand of professional. 政治大 investment institutions, and assist them in enhancing their investment decision quality. 立 CMoney include most information that investors require in making investment ‧. ‧ 國. 學. decisions. As the daily new volume is huge, news article that are firm-specific or containing useless content will be filtered out. Keywords related to major industries and equity markets all over the world will be used to extract useful and related news articles.. n. al. er. io. sit. y. Nat. The first step of the study is to setup a keyword dictionary identifying major industries in Taiwan and major equity markets all over the world. The major industries are identified according to the “Standard Industrial Classification of R.O.C.”, published by the Directorate General of Budget, Accounting & Statistics of Executive Yuan. The corporation classifies all Taiwanese listed companies into twenty-eight different categories, all categories are listed in Table 3-1.. Ch. engchi. i n U. v. We look through all categories, we believe that it is difficult to collect news related to other electronic industry and general and other industry. News article rarely mention the word “other” and firms categorized in other industry do not experience industrial fluctuations in the same way. Therefore, these two industries are not included in our keyword dictionary. Moreover, news articles often mentioned the industries in short, in most of the times, the full name of the industry will not be used in the news articles. Therefore, we read over pieces of financial news articles to identify the abbreviations for different industries. Both the full name and the abbreviations of industries are used to search for related news articles over the targeted period. The list of major industries and their abbreviations are shown in Appendix 1.. 27. DOI:10.6814/THE.NCCU.ACCT.025.2018.F07.

(37) Table 3-1 Standard Industrial Classification of R.O.C. Standard Industrial Classification Cement industry. Food industry. Plastic industry. Textile industry. Electric machinery. Electrical and cable. industry. industry. Biotechnology and. Glass and ceramic. medical care industry. industry. Paper and pulp industry. Iron and steel industry. Rubber industry. Automobile industry. Semiconductor industry. Computer and peripheral. Chemical industry. equipment industry Optoelectronic industry. Communications and. Electronic parts and. 政治大 components industry Information service Other electronic industry internet industry. Electronic products. 立. industry. 學. ‧ 國. distribution industry. construction industry. transportation industry. Financial and insurance. Trading and consumers'. Gas and electricity. industry. goods industry. industry. y er. io. sit. General industry and other industry. Tourism industry. ‧. Shipping and. Nat. Building material and. al. n. v i n C industries, Apart from the major local U that global macroeconomic h e n g cwehbelieve i events would also affect Taiwanese market overall sentiment. We list out major Asian and Global equity markets, we believe that movements in these stock markets would affect sentiment of Taiwanese investors. Major equity markets include Global, US, European, Chinese, Singaporean, Hong Kong, Korean and Japanese market. We carry out keyword search in the way similar to that of industrial news search. The name of major stock markets and their abbreviations are indicated in Appendix 2. According to the result of the keyword search, we notice that reporters rarely mention issues of the Singaporean market. The average number of Singaporean market news collected between 2007 and 2017 is twenty pieces per year. We believe that the information content related to the Singaporean market that Taiwanese investors received are very limited and would not impact general market investors’ sentiment. Therefore, we exclude data of the Singaporean market in our study. The news articles related to these major equity markets will be used for sentiment analysis in the next section. 28. DOI:10.6814/THE.NCCU.ACCT.025.2018.F07.

(38) The sample period of this study is January 2007 to December 2017. The total number of news articles available in CMoney is 1,170,498. However, news article published between November and December 2015 is missing from the database, therefore, the number of days without any news is 61 days. In this study, data mining technique is used to extract financial news that contain the name or abbreviations of major local industries or global stock markets. The total number of local industrial news and global stock news collected are 979,769 and 77,525 respectively. The quantity of news articles related to local industries and major stock markets are lower in nontrading days and public holidays. The number of days without any search result for local industrial news and global stock market news are 121 days and 422 days respectively. Observations with missing values will be excluded in our analysis. VIXTWN would only published on trading days, data of non-trading days are excluded as we are not able to observe market volatility. Moreover, we could not observe the relationship between news sentiment and market volatility on days without any. 政治大 keyword search. We exclude observations without search result. The summary of search 立 result is demonstrated in Table 3-2. The annual quantity of news collected for each Table 3-2 Result of News Search. y. a l 2007 – 2017 1,170,498 i v n C 2017 979,769 U 2007h–e ngchi. n. Local Industrial News. news collected. Number of days. without Search Result. er. io. Financial News. Number of. Period. sit. Nat. Type of News. ‧. ‧ 國. 學. industry and stock market between 2007 and 2017 is presented in Appendix 3 and Appendix 4 respectively.. Global Stock Market News 2007 – 2017. 77,525. 61 121 422. 29. DOI:10.6814/THE.NCCU.ACCT.025.2018.F07.

(39) 3.3 Text Analytic Approach on News This study uses the linear regression to test the association between different sentiment variables extracted from news and VIXTWN. Text mining technique is used to extract useful information from financial news articles. News contents are unstructured data. In this study, we try to investigate in what extent that news volume, mood of news and news tone would affect market volatility. Carretta et al. (2011) suggested the use of a sentiment dictionary to measure the quantity of positive and negative words appeared in the news articles and used the dictionary approach to further analyze the investors’ sentiment. In order to extract useful information from local industrial and global equity market news, we adopt the dictionary approach to score news tone and other sentiment variables.. 政治大 read news articles written in Chinese and their sentiment are mainly impacted by 立 Chinese-written news published by local media. Therefore, we extract news written by. The test region of our study is the Taiwanese market. Taiwanese investors usually. ‧. ‧ 國. 學. Taiwanese press media. We construct a Chinese sentiment dictionary to count positive and negative words and score news tone on news articles available on Taiwanese press media.. n. al. er. io. sit. y. Nat. The sentiment dictionary that we use in this study is based on the dictionary built by the Innovative and Mobile Financial Service Technologies, Modeling and Applications project supported by the Ministry of Science and Technology. Words included in the dictionary are mostly adjectives describing investors’ physiological status (e.g. pessimistic, optimistic) and other sentiment-related words. In this study, we consider the effect of financial warning on VIXTWN, words related to financial warning and risk are important to our study. We made reference to the sentiment dictionaries proposed by Lin (2013) and Chang (2009), their sentiment dictionaries include words related to financial warning and risk, and those words are added to our sentiment dictionary. Words included in the dictionary are classified into two categories,. Ch. engchi. i n U. v. positive words and negative words. Classifying words into different categories are essential for further analysis. The sentiment dictionary contains thousands of words, the partial positive and negative sentiment dictionary are demonstrated in Appendix 5 and Appendix 6 respectively.. 30. DOI:10.6814/THE.NCCU.ACCT.025.2018.F07.

(40) After constructing our sentiment dictionary, we use the text analytic approach to read through all related news collected in the previous step. We measure the frequency of positive words and negative words that appeared on the collected news and count the total number of words in the news articles. According to Li et al. (2014), optimistic (pessimistic) mood can be calculated by dividing number of positive (negative) words by the total number of words appeared in the news article. Ferguson et al. (2012) has also recognized the percentage of positive (negative) content as one of the variables, its calculation formula is the same as that of the optimistic (pessimistic) mood. The formula of optimistic mood and pessimistic mood are shown as follows: Optimistic mood =. 𝑛𝑢𝑚𝑏𝑒𝑟 𝑜𝑓 𝑝𝑜𝑠𝑖𝑡𝑖𝑣𝑒 𝑤𝑜𝑟𝑑𝑠 𝑡𝑜𝑡𝑎𝑙 𝑛𝑢𝑚𝑏𝑒𝑟 𝑜𝑓 𝑤𝑜𝑟𝑑𝑠. 𝑜𝑓 𝑛𝑒𝑔𝑎𝑡𝑖𝑣𝑒 𝑤𝑜𝑟𝑑𝑠 治政𝑛𝑢𝑚𝑏𝑒𝑟 𝑡𝑜𝑡𝑎𝑙 𝑛𝑢𝑚𝑏𝑒𝑟 大𝑜𝑓 𝑤𝑜𝑟𝑑𝑠. Pessimistic mood =. 立. ‧. ‧ 國. 學. Apart from optimistic and pessimistic mood, we use news tone to measure investors’ sentiment, we calculate news tone using the tone-formula suggested by Carretta et al. (2011). Zhang and Skiena (2011) identified media polarity as one of the sentiment variables, the formula of calculating media polarity is identical to the toneformula. Tone is calculated by dividing the difference between number of positive and. sit. er. io. 𝑛𝑢𝑚𝑏𝑒𝑟 𝑜𝑓 𝑝𝑜𝑠𝑖𝑡𝑖𝑣𝑒 𝑤𝑜𝑟𝑑𝑠 − 𝑛𝑢𝑚𝑏𝑒𝑟 𝑜𝑓 𝑛𝑒𝑔𝑎𝑡𝑖𝑣𝑒 𝑤𝑜𝑟𝑑𝑠 𝑛𝑢𝑚𝑏𝑒𝑟 𝑜𝑓 𝑝𝑜𝑠𝑖𝑡𝑖𝑣𝑒 𝑤𝑜𝑟𝑑𝑠 + 𝑛𝑢𝑚𝑏𝑒𝑟 𝑜𝑓 𝑛𝑒𝑔𝑎𝑡𝑖𝑣𝑒 𝑤𝑜𝑟𝑑𝑠. al. n. Tone =. y. Nat. negative words by the sum of frequencies of positive and negative words, the formula is displayed as follows:. Ch. engchi. i n U. v. The value of tone lies between 1 and -1, positive value will be obtained for news with more positive wordings and negative value will be obtained for news with a more negative tone. Value of tone will be close to 0 for news expressing neutral tone, tone value equals to 1 (-1) means that the particular news pieces express a completely positive (negative) findings. We measure industrial and market specific sentiment variables. We believe that issues of different industries would impact investors’ sentiment in different extent, as the market size of each industry is different. We calculate weighted average overall sentiment variables by multiplying the industrial sentiment variables with its relative weighting. The weighting that we use is the same as the weighting used by the Taiwan Stock Exchange Capitalization Weighted Stock Index (TAIEX). The weighting that TAIEX used is calculated by dividing the industrial market value by total market value 31. DOI:10.6814/THE.NCCU.ACCT.025.2018.F07.

(41) of all listed companies. We believe that the impulse brought by industry related news are correlated to its relative market value. As the market value of companies change over time, we use the weighting of TAIEX obtained on 15 May 2018. The detailed weighting of different industries is shown in Appendix 7. As the influence of market movements in different market on Taiwanese investors’ sentiment is hard to quantify, the equity market variables are calculated by averaging variables of different equity market. The sentiment variables that we use in our analysis is the overall industrial and equity market sentiment. We have mentioned many different types of data that we use in Section 3.2 and 3.3. The data that we use in this study is summarized in Table 3-3. The table listed out the sources of data and their features.. 立. 政治大. ‧. ‧ 國. 學. n. er. io. sit. y. Nat. al. Ch. engchi. i n U. v. 32. DOI:10.6814/THE.NCCU.ACCT.025.2018.F07.