本研究提出一個結合自適應共振理論Ⅱ(Adaptive Resonance Theory Ⅱ；ART2)與K-平均數集群(General K-Means網Method)的改良式二階段集群分析法，利用ART2判斷分群組數並搜尋初始群心代入K-means中進行國內基金績效的分類（本研究定義為ART2+K），並對國內基金績效的持續性進行Spearman等級相關檢定。經實證研究後發現：ART2+K對基金績效分群的結果優於單獨使用ART2或K-means的分群結果。ART2+K對國內基金績效的分群相當完善，並且可以根據其他相關的變數：報酬率、β係數、Sharpe指標、Jensen指標與Treynor指標將所分類的群組予以分別命名為「高報酬高績效--股票型基金」、「穩定收益型基金」、「中報酬中績效--股票型基金」、「中報酬高績效--跨國投資型基金」、「低報酬低績效--債券股票平衡型基金」。國內基金績效經Spearman等級相關檢定後，不論是針對年度報酬率或Sharpe績效指標而言，短期內均具績效持續性，長期則不具績效持續性。

This research supply a new two-stage clustering method which integration of adaptive resonance theory II (ART2) and K-means method. By using ART2 neural network to determine the number of clusters and the staring points and then employing the K-means method to find the final solution, can provide very good solution of data clust4ering. We apply this two-stage clustering method ART2+K to cluster the performance of mutual funds and use Spearman rank-order correlation to study the performance persistence of mutual funds. Data is collected from January 2001 to May 2006, and the evaluation indexes of mutual funds include return. Beta coefficient, Sharp Index, Jensen index, and Treynor index. This research obtains the following conclusion after empirical study: For classification, this two-stage clustering method ART2+K is better than ART2 or K-means method in the performance clustering of mutual funds. For persistence, in a short term, there is persistency for performance of mutual funds by Spearman rank-order correlation. But there is not persistency for performance of mutual funds in a long term.