- 學位論文
利用褐皮書預測美國聯準會升降息決策
Use Beige Book to Predict FOMC Monetary Policy
摘要
過去對美國聯準會(Fed)的研究主要聚焦於利率決策對資產價格的影響，例如對股市和債市造成的變化，預測未來利率決策的研究則較為稀少。利率決策預測方法可分為結構性資料預測和非結構性資料預測，結構性資料預測最著名的為利用美國聯邦基金利率期貨(Fed Fund Futures)推估升降息機率；非結構性資料則以分析Fed所發布的文本、官員談話和演講為主。本研究便是使用非結構性資料預測的方式，使用的資料為Fed發布的褐皮書。 吾人採用1983至2019的褐皮書樣本並使用Python的自然語言處理工具(Natural Language Tool Kit, NLTK)進行文字探勘(Text Mining)，篩選適合的文字變數建立升息模型和降息模型，並分別使用羅吉斯回歸(Logistic Regression)和支援向量分類(Support Vector Classification, SVC)模型進行訓練。研究結果顯示SVC在樣本內回測中幾乎可捕捉所有升降息且訊號準確率100%，但樣本外回測表現則有下修情形；Logistic樣本內回測和樣本外回測差異不大，表現不如SVC。同時，若在模型中加入執政黨變數，對Logistic模型表現有顯著改善。整體而言，褐皮書對未來利率決策具有解釋力。
並列摘要
Past research on Federal Reserve mainly focused on the impact of monetary policy on asset prices, such as stock market and bond market. However, research on predicting future monetary policy is relatively rare. Data used to predict monetary policy can be divided into structured data and unstructured data. Using Fed Fund Futures to predict future hike and cut probability is a well-known way to predict by structured data. For unstructured data, document released by Fed and speech of Board members are the most commonly used. This paper used unstructured data to predict future monetary policy, and the data we used is Beige Book released by Fed. We collected Beige book from 1983 to 2019 and used Python's Natural Language Tool Kit (NLTK) as textual analysis software. After conducting text mining, we extracted several text variables from Beige Book and established hike model and cut model. We used two machine learning method, Logistic Regression and Support Vector Classification (SVC), to train our models. The result shows that SVC could almost seize every hike and cut event and the precision rate achieves 100% in the in-sample test. However, the out-sample test of SVC is not as good as the in-sample test. For Logistic Regression, the result of in-sample test and out-sample test are similar. Overall, Logistic Regression underperforms against SVC. Besides, the addition of partisan variable significantly improves the performance of Logistic Regression. To conclude, Beige Book has explanatory power against future monetary policy.