新聞文字隱含資訊與投資人情緒及實際波動度之關係

Abstract

本文參考Manela和Morierla (2017)所提出的論點：認為新聞資訊能當成投資人情緒之代理變數，進而透過文字探勘與機器學習演算法去探討新聞文字資訊與恐慌指數(VIX)和實際波動度(RV)之關聯性；本文擴充了Manela和Moreira (2017)的論點，除了將資料頻率由月資料改成日資料來探討之外，也將單一新聞資料源擴充成多個新聞資訊源，更改變了計算投資人情緒之方法，利用Loughran和McDonal於2014所提出的修正版分類辭典(LM words lists)將字詞的正負面性與其他三類情緒納入考量，再透過機器學習演算法建構出新聞文字隱含之波動度(NVIX)，藉此希望能捕捉到更精確的投資人情緒、改善新聞文字資訊解釋VIX與預測RV之績效。實證結果顯示：當日新聞資訊利用LM分類詞典所形成的各類情緒變數對於當日VIX確實有顯著的解釋能力、不同的新聞資料源對於VIX的解釋能力也有顯著差異、多家新聞資料源對於VIX的解釋能力普遍高於單家新聞資料源；然而，無論是利用單一或多家新聞資料所建構成的NVIX，比起VIX，對於實際波動度(RV)的預測績效並沒有額外顯著的提升。

Referring to Manela and Morierla (2017), with the help of text mining techniques and machine learning algorithms, news articles can be linked to market investors’ sentiment; furthermore, we can use the news information as the proxy for investors’ sentiment, and use it to explain movements of the VIX index or to forecast the future realized volatility (RV). We further expand Manela and Morierla’s (2017) idea, adopting multiple sources of news articles instead of single source of data. Also, we modify the way to quantify investors’ sentiment, we utilize LM word lists provided by Loughran and McDonald in 2014, trying to precisely seize the market investors’ fear (i.e., VIX) and better predict the future RV. Finally, our empirical results show that: (1) Sentimental words derived from daily news articles do have significant power to explain VIX. (2) Multi-sources of daily news data generally have stronger explanatory power of VIX compared with the single-source of news data. (3) Neither single nor multi-sources of news data have additional forecasting power for the future RV compared with the VIX index.