應用貝氏定理與鑑別分析建立金融危機預警系統

Abstract

本論文的目標在於開發二個預警系統用以預測金融危機的發生。從決策者的觀點，預警系統之所以舉足輕重在於它能夠偵測出潛藏的經濟疲弱，進而使決策者能夠在金融危機發生之前提早擬出對策，進而減輕金融危機所帶來的衝擊，甚至能夠制止此金融危機的發生。就預測能力而言，欲建立一個健全系統，需倚賴於鑑定出的領先指標是否可靠。 根據以往的經驗可以察覺，一個金融體系在金融危機發生之前，必有經濟疲弱的發生。因此，我們首先利用貝氏理論偵測金融變數是否發生體系變遷，並利用F統計量和斯皮爾曼相關分析去評估經由貝氏分析所得到的事後機率值和危機前導變數在時間序列的相似度，如此便可篩選出關鍵變數。爾後我們繼續利用因素分析以及鑑別分析，分別從這些關鍵變數之中鑑定出兩組不同的領先指標。 最後，我們再利用這兩組領先指標建立二個預警系統，較諸於之前已存在的預警系統模型，期能達成預測表現之增進與改善。

The objective of this thesis is to develop two new Early Warning Systems (EWS) for predicting financial crises. From the policy-makers' perspectives, an EWS plays an important role by allowing them to detect potential economic weaknesses and take pre-emptive strategies to lessen the impact of a crisis or even to prevent it. In terms of predictive power, building a robust system relies on identifying reliable leading indicators.

According to an observation of the past experience, before a financial crisis takes place in an economic system, there must be economic weaknesses. Therefore, we first used Bayesian Theorem to detect regime shifts of financial variables, and then used F statistics and Spearman correlation to estimate the degree of similarity between the obtained posterior probability and a forward crisis variable in a time series in order to extract key variables. We proceeded to identified two sets of leading indicators among the key variables with factor analysis and discriminant analysis.

Finally we employed the leading indicators to build two EWS, which achieve the expectancy of improving the performance upon existing models.