篇名 | 美國聯準會會議紀要的文字探勘與台灣經濟變數預測 |
---|---|
卷期 | 47:3 |
並列篇名 | Text Mining of the FOMC Minutes and Forecasts of Taiwan Economic Variables |
作者 | 黃裕烈 、 管中閔 |
頁次 | 363-392 |
關鍵字 | 文字探勘 、 美國聯邦公開市場委員 、 主題模型 、 情緒分析 、 federal Open Market Committee 、 sentiment analysis 、 text mining 、 topic models 、 EconLit 、 TSSCI |
出刊日期 | 201909 |
DOI | 10.6277/TER.201909_47(3).0002 |
本文利用文字探勘的技巧, 從美國聯邦公開市場委員會(Federal Open Market Committee, 簡稱FOMC) 所發表的正式官方文件:minutes of the FOMC 中萃取出重要的資訊, 再利用情緒分析來判斷FOMC 對三大經濟「使命」的正/負面看法, 並建立指標, 以預測台灣的財經相關變數。相較於之前的文獻, 本文統一FOMC 常用的專門術語, 並且考慮了複合字的情況; 這可以避免後續在計算文字字數時, 產生重覆計算以及語意錯誤的情況。我們也利用統計模型(MAP-PLSA) 將每一份文件中的句子依據FOMC 的經濟使命分成三個主題, 再利用情緒分析技巧, 建立各類主題的指標數列。最後, 我們以迴歸模型來分析情緒指標與台灣相關財經變數之間的關聯性。
In this paper we extract useful information from the minutes of the Federal Open Market Committee (FOMC) and examine how such information can help predict economic/financial variables. Based on the minutes during 1993–2016, we conduct sentiment analysis to determine the FOMC’s attitude towards different topics, i.e., the mandates of the FED. Our approach is different from related studies in the following respects. First, we identify compound words which carry more specific meaning than do single words. Second, we adopt the topic model, MAP-PLSA, for estimating the conditional probabilities of these words/terms, which in turn can be used to classify sentences in the minutes under different topics. Third, the attitude towards each topic is determined by the “tone” of its sentences. We then proceed to evaluate whether the FOMC’s attitude towards different topics can be used to improve economic forecasts.