本論文提出了貝氏在線變點檢測方法基於Clayton copula 且邊際分布為混合 常態分布的copula-Markov 模型。在模擬研究中，我們研究了相關時間序列數據 的結構變化，包括單峰數據和雙峰數據之間的轉換以及雙峰數據內部的結構變 化，並將其與邊際分布為常態分布的copula-Markov 模型進行比較，模擬的結果 指出在相關性資料由單峰改變成雙峰時，我們所使用的模型在準確率、平均絕對 誤差、真陽個數、偽陽個數皆勝過對比模型。我們應用這種方法來檢測加密貨幣 市場從2017 年到2018 年初的變點，其中包括ICO 繁榮的開始和結束，我們選用 BTC、ETH、NEO 的每日報酬率進行偵測。結果顯示了我們的方法在檢測相關時 間序列數據的變化方面的有效性，並提供了關於虛擬貨幣市場在這ICO 繁榮-蕭條 期間行為的見解。 關鍵字：在線改變點檢測，Clayton copula，馬可夫模型，貝氏推論，ICO boom;In this paper, we propose an online Bayesian changepoint detection approach for dependent time series data under the copula-based Markov model with the marginal distributions being mixture normal distributions. Simulation studies examine structural changes in sequential time series data with dependency, including transformation between unimodal and bimodal data and structural changes within bimodal data. For comparison, we consider the Clayton copula-based Markov model with normal marginal distributions as the benchmark model. The results show that the proposed model outperforms the benchmark model in detecting change when the correlation structure of the changes from unimodal data to bimodal data. In the empirical analysis, we use the daily returns of BTC, ETH, and NEO to identify change points in the cryptocurrency market from 2017 to early 2018. The results demonstrate the effectiveness of our approach in detecting changes and providing insight into cryptocurrency market behavior during ICO booms and busts. Keywords: online change point detection, Clayton copula, Markov model, Bayesian inference, ICO boom
