過去因子模型對於超額報酬之探討主要著重於公司營運與財務指標,然而,近年來隨著市場中的非結構化資料逐漸增加,如何透過分析各種不同的資料引入更多資訊成為重大議題,本研究為探討文字資料運用於財務領域的成效,使用情感分析技術VADER,分析美國公司10-K財報中的文字資訊,計算出財報情緒分數,並驗證若以該分數做為建構投資組合之標準能否獲取超額報酬,並建立基於財報情緒分數之情緒因子,加入Fama-French三因子模型,使用此四因子模型針對美國股市進行超額報酬橫斷面變異進行分析。實證結果顯示使用財報情緒分數做為建立投資組合之標準在Fama-French三因子模型下確實能帶來異常報酬,尤其以負面情緒分數做為標準會有更明顯的趨勢,而更進一步使用基於財報情緒分數做為基準建構的情緒因子,更能使因子模型的解釋力上升。
In the past, the discussion of excess returns in factor models mainly focused on company operations and financial indicators. However, in recent years, as unstructured data has gradually increased, how to introduce more information through analysis of various data has become a major issue. In order to discuss the effectiveness of the application of text data in the financial field, we use VADER to analyze the text information in the 10-K financial reports of American companies, calculate the financial report sentiment score, and verify whether the score can be used as the criterion for constructing an investment portfolio and obtain excess returns. Further, we establish an emotional factor based on financial report sentiment scores and combine with Fama-French three-factor model, and use this four-factor model to analyze the cross-sectional variation of excess returns for the US stock market. Empirical results show that using financial report sentiment scores as the criterion for establishing portfolios can indeed bring abnormal returns under the Fama-French three-factor model. Especially with negative sentiment, we can see a clear trend with excess returns. And using the form 10-K sentiment based factor can improve the explanatory power of the multi-factor model.