摘要 本文利用狀態變化之copula模型來刻畫因時間變化的雙變量相依結構。以馬可夫轉換(Markov switching)模型為基準,我們將copula函數之參數設定為依不同狀態而內生變化之動態模型,此模型設定適合用於刻畫雙變量相依結構的結構性轉折情形。 實證研究發現,道瓊工業股價指數與那斯達克股價指數之報酬率間的相依結構最適合利用馬可夫變換之混合常態的copula模型來描繪它們之間的相關性,亦即此二股價報酬率之間的共變性(concordant association)明顯的具有隨不同時點變換的現象。
ABSTRACTS In this thesis, we explore the time-variant bivariate dependence structure by a class of regime switching copula models. We consider a dynamic mixed copula in which the parameters are governed by a hidden Markov chain. In our empirical study, we apply a number of MS-mixed copulae to explore bivariate dependence between the daily returns of DJIA and NASDAQ. The structural change analysis indicates that the concordant association between these two return series is time-variant. Markov switching mixed normal copula model is suitable for interpreting the bivariate dependence of DJIA and NASDAQ returns. This result implies that the dependence may switch between the low and high concordance states.