本研究針對台灣 5G 產業公司之致股東報告書之文字情緒對股票市場影響。研究分成二階段，首先，本研究運用文字探勘方法中的 Bert (Bidirectional Encoder Representations from Transformers) 模型建立自然語言處理模型，以 Financial PhraseBank 資料集為訓練樣本將 Bert 進行微調。第二，運用經微調後的 Bert 模型剖析台灣 5G 公司之致股東報告書財務績效段落，研究致股東報告書財務績效段落對機構投資人買賣超張數以及個股股票成交量。本研究之實證結果如下，
(一) 本次模型之測試集準確率達 84%，驗證集準確率達 85%。
(二) 致股東報告書文字情緒與外資買賣超張數並無顯著相關。
(三) 致股東報告書文字情緒與投信買賣超張數並無顯著相關。
(四) 致股東報告書文字情緒與自營商買賣超張數呈顯著負相關。
(五) 致股東報告書文字情緒與成交量呈顯著正相關。
(六) 即使增加年度現金流量變化之控制變數，仍維持上述結論。

綜觀本研究之研究結果，本研究之實證結果顯示，機構投資人交易中，僅自營商買賣超張數與致股東報告書文字情緒呈顯著相關，推論其原因為外資之投資交易策略為長期投資策略，投信之投資交易策略為中期投資策略，故無法於本研究衡量期間財務報告發布當日至前五日中反應。然而，自營商買賣超張數與成交量與致股東報告書文字情緒呈現顯著相關，代表文字情緒確實影響市場中投資人投資交易。

This study focuses on the impact of the text sentiment of the report to shareholders to the stock markets. First, I apply a Bert (Bidirectional Encoder Representations from Transformers) model, which is a nature language process model, to construct the construct a text mining model. I use Financial PhraseBank which is a dataset involved lots of financial news and their text sentiment to fine-tune Bert. Moreover, after trying to deploy different hyperparameters, I choose a best model based on validation performance. Next, I analyze the text sentiment of the report to shareholders by the fine-tune Bert model. Finally, taking 5th generation mobile network companies in Taiwan as samples, this study conduct empirical analysis about the impact of the text sentiment in report to shareholders on the institution investors holding and trading volume. The empirical results of this study show that：
(1) The classification accuracy of the Bert model reaches up to 84%.
(2) There are no significant impact of sentiment in report to shareholders on foreign investments holding. The research result suggests that the investment strategy of foreign investments is long-term. As a result, we may not find the action of foreign investments in the measurement period.
(3) There are no significant impact of sentiment in report to shareholders on investment trusts holding. The research result suggests that the investment strategy of investment trusts is mid-term. As a result, we may not find the action of investment trusts in the measurement period.
(4) There are significant impact of sentiment in report to shareholders on dealer holdings.
(5) The more positive the report to shareholders are, the stock trading volume will increase.

Agarwal, S., Chen, V.Y.S., Zhang, W., 2016. The information value of credit rating action reports: A textual analysis. Management Science 62(8), 2218-2240

Ahmadi, Z., Martens, P., Koch, C., Gottron, T. and Kramer, S., 2018. Towards bankruptcy prediction: deep sentiment mining to detect financial distress from business management reports. 2018 IEEE 5th International Conference on Data Science and Advanced Analytics (DSAA)

Barber, B. M., Lee, Y.-T., Liu, Y.-J. and Odean, T., 2009. Just How Much Do Individual Investors Lose by Trading?. Review of Financial Studies 22(2), 609-632

Bhattacharya, N., Cho, Y. J. and Kim, J. B., 2018. Leveling the Playing Field between Large and Small Institutions: Evidence from the SEC's XBRL Mandate. The Accounting Review 93(5), 51-71

Building the AI bank of the future, 2021. McKinsey & Company

Das, S. R. and Chen, M. Y., 2007. Yahoo! for Amazon: Sentiment Extraction from Small Talk on the Web. Management Science 53(9), 1375-1388

Cecchini, M., Aytug, H., Koehler, G. J. and Pathak, P., 2010. Making words work: Using financial text as a predictor of financial events. Decision Support Systems 50(1), 164-175

Chuang, W.-I. and Susmel, R., 2011. Who is the more overconfident trader? Individual vs. institutional investors. Journal of Banking & Finance 35(7), 1626-1644

Chen, C. L., Liu, C. L., Chang, Y. C. and Tsai, H. P., 2013. Opinion mining for relating multiword subjective expressions and annual earnings in US financial statements. Journal of Information Science and Engineering 29(3), 743–764

Chouliaras, Andreas, 2015. Institutional Investors, Analysts’ Recommendations, Annual Reports, Textual Analysis and Stock Returns: Evidence from SEC EDGAR 10-K and 13-F Forms. Available at SSRN: 2638379

Cascino, S., Clatworthy, M., Garcia, B., Gassen, J., Imam, S. and Jeanjean, T., 2016. Professional Investors and the Decision Usefulness of Financial Reporting. ICAS and EFRAG

Chiang, Y. C. and Chan. M. H., 2017. Foreign ownership and firm level stock return volatility in Taiwan. Investment Management and Financial Innovations 14(3), 261-269.

Davis, A. K., Piger, J. M. and Sedor, L. M., 2012. Beyond the Numbers: Measuring the Information Content of Earnings Press Release Language. Contemporary Accounting Research 29(3), 845-868

Dong, W., Liao, S. and Zhang, Z., 2018. Leveraging Financial Social Media Data for Corporate Fraud Detection. Journal of Management Information Systems 35(2), 461-487.

Delvin, J., Chang, M. W., Lee, K. and Toutanova, K., 2019. BERT: Pre-training of deep bidirectional transformer for language understanding. aarrXiv:1810.04805

Duz Tan, S. and Tas, O., 2020. Social Media Sentiment in International Stock Returns and Trading Activity. Journal of Behavioral Finance 22(2), 221-234

Fang, H., Lu, Y.-C., Yau, H.-Y. and Lee, Y.-H., 2016. Causes and Impacts of Foreign and Domestic Institutional Investors’ Herding in the Taiwan Stock Market. Emerging Markets Finance and Trade 53(4), 727-745

Groß-Klußmann, A., König, S. and Ebner, M., 2019. Buzzwords build momentum: Global financial Twitter sentiment and the aggregate stock market. Expert Systems with Applications 136, 171-186

Hung, P.-H. and Lien, D., 2019. Trading aggressiveness, order execution quality, and stock price movements: Evidence from the Taiwan stock exchange. Journal of International Financial Markets, Institutions and Money 60, 231-251

Hiew, J.Z.G., Huang, X., Mou, H., Li, D., Wu, Q., and XU, Y., 2019. BERT-based Financial Sentiment Index and LSTM-based Stock Return Predictability. submitted to NeurIPS 2019, arXiv preprint arXiv:1906.09024

Hsieh, W.-l. G. and Lee, C.-S., 2021. Who reacts to what information in securities analyst reports? Direct evidence from the investor trade imbalance. Pacific-Basin Finance Journal 65

Investor Bulletin: Social Sentiment Investing Tools —Think Twice Before Trading Based on Social Media., 2019. United States Securities and Exchange Commission

Jung, M. J., Wong, M. H. F. and Zhang, X. F., 2018. Buy-Side Analysts and Earnings Conference Calls. Journal of Accounting Research 56(3), 913-952

Kim ,J. H. and Shamsuddin, A., 2008. Are Asian stock markets efficient? Evidence from new multiple variance ratio tests. Journal of Empirical Finance 15(3)3, 518-532

Kraaijeveld, O. and Smedt, J. De, 2020. The predictive power of public Twitter sentiment for forecasting cryptocurrency prices. Journal of International Financial Markets, Institutions and Money 65

Loughran, T., and McDonald, B., 2011. When is a liability not a liability? Textual analysis, dictionaries, and 10-Ks. Journal of Finance 66(1), 35–65.

Li, J., Chen, Y., Shen, Y., Wang, J., and Huang, Z., 2019. Measuring China’s stock market sentiment. Available at SSRN 3377684

Li, M., Li, W., Wang, F., Jia, X. and Rui, G., 2020. Applying bert to analyze investor sentiment in stock market. Neural Computing and Applications 33(10), 4663–4676

Li, X., Wu, P. and Wang, W., 2020. Incorporating stock prices and news sentiments for stock market prediction: A case of Hong Kong. Information Processing & Management 57(5)

Lien, D., Hung, P.-H. and Chen, H.-J., 2021. Who knows more and makes more? A perspective of order suBMission decisions across investor types. The Quarterly Review of Economics and Finance 79, 381-398

Malo, P., Sinha, A., Korhonen, P., Wallenius, J. and Takala, P., 2014. Good debt or bad debt: Detecting semantic orientations in economic texts. Journal of the Association for Information Science and Technology 65(4), 782–796
Mai, F., Tian, S., Lee, C. and Ma, L., 2019. Deep learning models for bankruptcy prediction using textual disclosures. European Journal of Operational Research 274(2), 743-758.

Nguyen, H., Calantone, R. and Krishnan, R., 2020. Influence of Social Media Emotional Word of Mouth on Institutional Investors’ Decisions and Firm Value. Management Science 66(2), 887-910

Price, S. M., Doran, J. S. Peterson, D. R. and Bliss, B. A., 2012. Earnings conference calls and stock returns: The incremental informativeness of textual tone. Journal of Banking & Finance 36(4), 92-1011

Sadique, S., In, F., Veeraraghavan, M., , 2008. The impact of spin and tone on stock returns and volatility: evidence from firm-issued earnings announcements and the related press coverage. Working paper, Monash University

Sousa, M. G., Sakiyama, K., Rodrigues, L. d. S., Moraes, P. H., Fernandes, E. R. and Matsubara, E. T., 2019. BERT for Stock Market Sentiment Analysis. 2019 IEEE 31st International Conference on Tools with Artificial Intelligence (ICTAI)

Sentiment, Signals & Surveillance: Natural Language Processing for Financial Markets, 2020. Greenwich Associates

Shapiro, A. H., Sudhof, M. and Wilson, D. J., 2020. Measuring news sentiment. Journal of Econometrics

Tetlock, P. C., 2011. All the News That’s Fit to Reprint: Do Investors React to Stale Information?. Review of Financial Studies 24(5), 1481–512

Why investors want alternative data, 2016. The Economist

Xiao Wu, D., Yao, X. and Luan Guo, J., 2021. Is Textual Tone Informative or Inflated for Firm’s Future Value? Evidence from Chinese Listed Firms. Economic Modelling 94, 513-525

Yang, B., Sun, J., Guo, J. and Fu, J., 2019. Can financial media sentiment predict merger and acquisition performance?. Economic Modelling 80, 121-129
Yang, Y., Uy, M. C. S., and Huang, A., 2020. Finbert: A pretrained language model for financial communications. arXiv: 2006.08097

Zheng, R., Zang, A. Y. and Huang, A. H., 2014. Evidence on the Information Content of Text in Analyst Reports. The Accounting Review 89(6), 2151-2180

今周刊 1196期