本研究採用Blumen et al.(1955)的「漂移者－停駐者」模型(the mover-stayer model)探討台灣開放式股票型基金績效持續性的動態特性，以改善過去文獻所採用的靜態分析模型，以期獲得有關基金績效持續性更詳細的資訊。我們利用 Frydman(1984)所發展的最大概似法來估計此模型，並以概度比檢定檢驗「漂移者－停駐者」模型相較於單純馬可夫鏈模型的資料配適能力。主要的實證結果包括：(1)基本上我們的實證結果支持基金具有某種程度的績效持續性，只是該持續性並不很明顯。原因之一是大部分的基金乃屬於績效表現不穩定的漂移基金，績效穩定的停駐基金佔相對少數。有趣的是，表現最佳與最差兩個組別之基金其績效持續性都比中等績效之基金高。換言之，表現中等的基金表現最不穩定，顯現出這些基金的經理人對於改善下一期績效具有強烈企圖心，但此企圖心並非百分之百成功。另外，表現最佳與最差的基金發生績效反轉的機率也不低，而績效最差組別裡停駐基金所佔的比率最高，反映出這個組別的績效持續性較其他組別顯著。雖然整體基金表現出某種程度的績效持續性，唯此持續性似乎並不特別顯著。(2)與相關文獻的發現一致的是，基金績效持續性會因為績效指標的不同而有所差異。但主要的差異通常只反映在各績效組別裡停駐基金比率的估計，對漂移基金轉換機率矩陣的估計影響較少。(3)根據不同樣本期間下停駐基金比率的估計值可知，本研究的實證結果難免受到「存活偏誤」的影響，但此偏誤似乎集中於停駐基金比率的估計，而非漂移基金的轉換機率，加上財務學界目前對「存活偏誤」是否影響績效持續性仍存有歧見，因此我們認為「存活偏誤」應不至於改變本研究的重要結論。(4)我們利用概度比檢定比較單純馬可夫鏈模型與「漂移者－停駐者」模型的資料配適程度，發現概度比統計檢定量均能在百分之一的顯著水準下拒絕虛無假設的簡單馬可夫鏈模型。換言之，「漂移者－停駐者」模型較適用於台灣開放式股票型基金的績效持續性研究上。

We employ the mover-stayer model developed by Blumen et al. (1955) to study the dynamics of performance persistence of mutual funds in Taiwan. In comparison with the static analysis adopted by the literature, this model provides us with more detailed information about and helps us further understand the nature of mutual fund performance persistence. We also use the maximum likelihood methodology suggested by Frydman (1984) to estimate the mover-stayer model and test the data-fitting ability of this model against that of a pure Markov chain based on the likelihood ratio test. We find that: (1) There exists a certain degree of persistence in mutual fund performance. Such persistence is, however, not very significant. It is because most funds are mover funds with unstable performance rather than stayer funds with consistence performance. More interestingly, funds within the best and the worst performance groups have more persistent performance than those within the middle performance group. This implies that in view of the previous mediocre performance, fund managers within the middle group have strong intention to improve their future performance. Nevertheless, rewards for such intentions are never guaranteed. In addition, the fact that the worst performance group has the highest proportion of stayer funds implies that losers are more persistent than winners in Taiwan's mutual fund industry. Overall, mutual funds in Taiwan have only weak performance persistence. (2) Consistent with the literature, the degree of persistence in performance is dependent on the performance evaluation criteria. It seems that this difference of degree of persistence is reflected in the estimation of stayer fund proportion, not in the estimation of the transition probability matrix of mover funds. (3) There exists survivorship bias in our study. It mainly influences the estimation of stayer funds proportion, not that of the transition probability matrix of mover funds. Having said that, we believe that this bias will not alter the important conclusions of this article. (4) According to the likelihood ratio test, we are able to strongly reject the pure Markov chain model at the 1% significance level. This result supports our application of the mover-stayer model to study the performance persistence of mutual funds in Taiwan. Surely, it will be valuable to apply this model to investigate the same issue in other countries.