探討極端金融波動發生時距之研究-以ACD模型為研究方法

Abstract

鑑於金融危機對各國總體經濟、乃至全球景氣所帶來的影響甚鉅，相關之學術研究一直相當豐富且多樣化。而相較傳統文獻試圖由總體經濟基本面之模型來找出事前徵兆，以建立預警系統，本文以另一種角度切入。利用匯率及大盤指數為發生極端金融波動之代理變數，將十七個發生過金融危機國家之日資料，應用Engle and Russell (1998) 的自我相關條件時距(autoregressive conditional duration, ACD) 模型，估計連續兩次極端匯率波動及極端大盤指數波動之間的發生時距，以此預測未來可能發生金融危機的時間點。除了將所得之極端匯率波動時距及極端大盤指數波動時距套用三種常見之分配- Weibull，exponentional，generalized Gamma，還有經由Box-Cox轉換之ACD模型。實證結果顯示，ACD模型確實大幅解釋了極端金融波動時距間的自我相關性，而在三種常見分配情形下，exponentional 為分析極端金融波動時距之最適分配。整體而言，Box-Cox ACD模型表現最佳， 從其具體地降低Ljung-Box統計量即可得到印證。

Financial crisis has attracted a great amount of researches by scholars in recent years. This study we take a look at the occurrence of financial crises by econometrical viewpoint instead of traditional economical thoughts. Taking the exchange rates and equity indices raw data of 17 countries where financial crises have taken place to represent the consequences of extreme financial fluctuations, we use this data to apply the Autoregressive Conditional Duration (ACD) model which is proposed by Engle and Russell (1998) that seem particularly well suited for financial data. We try three often distributions - exponential, Weibull and generalized gamma distribution and an improved form - Box-Cox transformation on the empirical analysis. The results show that the ACD family models did a very good job of reducing excessively large Ljung-Box statistics, especially for Box-Cox ACD models that were associated with interim periods between extreme financial fluctuation events ranging from statistically insignificant to marginally significant level. Future studies may need to consider the sample quantities intentionally if they want to have a more accurate analysis.