本論文是研究在台灣開放型的股票型共同基金，並且利用資料採掘的技術加以分析並分類所謂優異績效及劣質績效的共同基金。我們使用分類決策樹(Classification and regression trees, CART)的方法來進行共同基金績效的分析及預測。本篇論文，我們採用了13種重要的變數來建構樹並找出優質基金，此外更驗證CART對於我們進行台灣共同基金績效的分析是穩定且有效的。最後，我們利用cross-validation test進行兩個月的基金的選取及持有，並各透過一個月的持有來視其績效。我們特別發現利用此方法選取出來的基金，其平均績效將優於所有共同基金的績效，並且其中有一個月的平均報酬率高於僅投資於高科技股的共同基金平均報酬率。 / We study the performance of open-end mutual funds in Taiwan, and use the technology of data mining to classify the outperforming and underperforming mutual funds. Classification and regression tree (CART) is the method to evaluate and predict the performance of mutual funds. In this paper, we utilize thirteen crucial factors to build trees and pick mutual funds by its classification rules. Moreover, we will verify precision of each tree. We find that the CART is a good tool to evaluate the performance of mutual funds in Taiwan because of its stability in outperforming - underperforming spreads. Moreover, we use two kinds of learning sample to build two trees and pick mutual funds to compose of them into the fund of funds. The results are better than the total average returns monthly, and one of them is better than the mutual funds that its investing target is high-tech stocks.

共同基金是未來投資理財約主流!專業法人機構在股市之投資比重日益提高，對國內證券市場生態產生結構性改變。展望未來趨勢，隨著股市發展成熟化、上市上櫃公司遽增與金融市場情勢連動關係複雜化等因素，專業投資理財時代的來臨必將成為事實。「專業、分工理財」投資型態將逐漸取代以往投資人自行下海操作之「自助式理財」型態。故共同基金操作績效正是進行本評鑑研究的最大動機，其中辨認基金經理人操作策略與績效顯現最具代表性的指標--選股能力與擇時能力．乃本研究所欲深入探討的主題。本研究模型之構建乃根據 Lee-Rahman (1990) 所提出之改良評鑑模式，簡言之，該模型的發展歷經早期 Treynor & Mazuy (1966) 的理論雛形，並經過 Jensen 針對評鑑擇時能力方面之改進，然後再經由 Bhattacharya & Pfleiderer (1983) 的進一步修正，產生了較完善的整體架構，最後經由 Lee-Rahman (1990) 針對評鑑過程中評鑑迴歸式之殘餘項異質性問題，作一根本的解決，建構完整的基金選股與市場擇時能力績效之評鑑模式。同時並運用幾個傳統績效評鑑指標 (即 Treynor、 Sharpe 與 Jensen 指標三種)，作為基金整體性績效之輔助研究。 實證結果顯示:(一)在基金整體績效評估方面，得到的結果顯示不論研究期間的長短．基金表現擊敗大盤的比例皆高於五成，約介於 60% 至 70% 左右，而在 Treynor、Sharpe 與 Jensen 指標三種不同評鑑模式分析下，基金績效排名有相當顯著的一致性。(二)在選股與市場擇時能力評鑑( Lee-Rahman 修正模型)方面:1.同時兼具選股與市場擇時能力的基金皆有穩定的一致性;2.大多頭時期基金擇時能力普遍提昇、新投信基金績效表現較優異。(三)整合比較結論方面:1.本研究中不同的評鑑模式所得實證結果十分肯定基金績效的持續性與穩定性;2.基金顯著的擇時能力並非基金整體績效脫穎而出的決定性因素，績效傑出與否幾乎決定於選股功力因素:3.台灣股市投資 Bottom-up 的投資邏輯漸漸取代 Top-down ;4.新投信績效已有凌駕老投信的趨勢。 / Mutual Funds Investment will be very popular! While the professional institutionalist's participation being promoted, domestic stock market environment has beenchanged structurally. In the future, accompanied by several reasons such us the_maturationof stock market, dramatic increasing in listed company numbers and more complex correlation in international financial markets, the age stressing on professional financial adversary will come. "Professional financial investment and adversary" will substitute for"Self-adversary" investment. Therefore, measuring the mutual fund's performance is the mailmotivation of my study', and the insight in selectivity ability and market timing ability will bethe subject for further research. The research model is based on the Lee-Rahman's modified model (1990), that is, the model structure was first built by Treynor & Mazuy (1966), improved by Jensen in Timingability measurement, and then revised by Bhattacharya & Pfleiderer (1983), and throughout structured by solving the "heteroscedasticity" in error terms of regressions finally Therefore, the better integral research model on measuring selectivity ability a timing abilityhas been accomplished. Meanwhile, we also use three traditional performance measure(Treynor, Sharpe & Jensen index) to be the general performance research. Empirical results show that (1)By measuring the general performance of mutual funds, we know that about 60%-70% funds managers beat market at each research period, and there exists consistency of fund's performance ranking among those three measures. (2) By using Lee-Rahman modified model (1990), we got three conclusions. First, for those funds consist of selectivity and timing ability, there exists consistency. In bull market period, timing ability of most funds performed better generally, and new funds had better performance. (3) Integral comparative conclusions have four results. First, no matter what measures being used, empirical results confirm stability and consistency in performance ranking. Second, significant timing ability is not the determinate factor of significant performance, significant selectivity ability plays the determinate role. Third, in Taiwan, bottom-up investment strategy gradually substitutes for top-down investment strategy. Finally, new mutual funds investment co. had outperformed old mutual funds investment institutions.

本研究探討2001年1月至2016年12月內所有以台股市場為標的之開放式股票型基金，透過多元迴歸模型與交易策略法深入的了解資金流量與過去和未來一期報酬率之間的關聯性，並從中探討台灣投資人的行為偏好。 透過多元迴歸模型與交易策略法可以發現代表台灣投資人投資偏好的資金淨流量變動率普遍有追逐過去績效表現優異之基金的傾向，接著探討資金淨流量變動率與未來一期報酬率的關聯後發現，台灣共同基金市場上當期資金淨流量變動率越高的基金，普遍在未來短期內所獲得的報酬率有較低的現象，然而隨著未來報酬期間的拉長，此現象便逐漸消失，最主要的解釋原因為台灣共同基金投資人普遍有追逐過去績效表現優異之基金的傾向，使過去績效表現較好的基金容易湧入過多的申購資金，而這些基金雖然在過去一期該基金經理團隊可以憑藉著自己所擅長的產業與個股經驗，挑選到具有成長潛力的投資標的，但隨著過去一期的優異表現，這些基金的投資組合持股價格已經來到相對高點，難以持續擁有良好的報酬表現，加上基金經理團隊手上仍握有許多等待投資的現金，最終可能迫使基金經理團隊必須開始涉入自己不熟悉的產業與個股，增加錯誤投資的機會而使績效表現變差，然而長期而言，該基金經理團隊仍可以憑藉著自己的專業投資能力，重新尋找到優良投資標的，消化過去湧入的投資資金，改善過去短期績效表現不佳的狀況。 / This study explored all open-ended equity funds targeting Taiwan’s stock market from January 2001 to December 2016. Through multiple regression model and trading strategy method, we got an in-depth understanding of the relationship between fund flows and both past and future returns, and the characteristics of the trading behavior of Taiwan’s investors were further investigated. By using multiple regression model and trading strategy method we found evidence that Taiwan’s investors have the tendency to chase mutual funds which had superior performance in the last period. Following this issue, we also found that funds with higher fund inflow generally had lower return in the short term time horizons, but the phenomenon would gradually disappear when the time horizons were extended. The main explanation of this phenomenon is that Taiwan’s investors generally have the tendency to buy mutual funds which gave superior return in the last period, so that funds with better performance in the past are prone to attract subscription. Although in the last period, these funds’ management team could rely on their own industrial and individual stock-picking experience, selecting those stocks with high growth potential. However, with an outstanding performance in previous period, stock prices in those fund’s portfolio had come to a relatively high point, so it’s hard to maintain good performance. With a vast sum of fund inflow, the management team may also be forced to invest in the industries or companies that they’re unfamiliar with, causing the possibility of wrong investment. However, when the time horizons were extended, the management team could digest the inflow of investment funds by rediscovering good investment targets and improve their fund performance.

本研究考慮市場、規模、淨值市價比及前期累積報酬，這四個影響股票報酬的因子，分別以個股特徵之持股比例變動法與四因子評估模型，對共同基金風險調整後的報酬作績效評比，不但可以評估基金的選股能力與擇時能力，並進一步瞭解報酬之風險來源。最後討論這兩種評比方式的適用性，並藉由基準投資組合將市場股票區分成不同的風險類別，根據基金在各類別股票的持有比例，引伸出對基金持股風格的另一種看法。現將本篇研究結果整理如下： 1. 四因子模型對於資產的解釋能力比資本資產評價模型（CAPM）好；並且透過規模、淨值市價比、前期累積報酬之風險溢酬因子，可以瞭解報酬之不同風險來源。 2. 依照個股特徵為基準之持股比例變動法，計算出實際績效、特徵擇時、特徵選股及平均持股型態的績效。結果顯示共同基金多具有正的選股能力，擇時能力，但經過檢定，並沒有顯著的超額報酬。 3. 以四因子評價模型對共同基金績效做評估。結果發現幾乎不具有顯著的超額報酬；兩種方法的評比結果相類似。但是部份基金在規模、與前期累積報酬項有顯著異於零的結果，顯示基金在規模、量能操作上有穩定的績效表現，因此使得檢定的結果顯著。 4. 而以持股類型風格上來看，顯示部份基金會高度持有大型股、以及過去表現良好的股票，持股風險類群明顯而集中，屬於穩健、偏重長期，並配合量能操作的投資策略。 最後根據本文的實證結果，分別對投資人與基金經理人提出建議。而從持股比例計算的過程，對持股風格分析提供一個更簡易明瞭的看法，並將研究中發現的問題，一併列在建議中，提供給後續研究者作為參考。 第一章 緒論…………………………………………1 第二章 文獻回顧……………………………………4 第一節 風險調整因素……………………………4 第二節 四因子評估模型…………………………7 第三節 依個股特徵之持股比例變動法…………9 第三章 研究設計……………………………………13 第一節 研究假說…………………………………13 第二節 研究架構…………………………………14 第三節 研究範圍與期間…………………………16 第四節 變數定義與資料處理……………………18 第四章 實證結果與分析……………………………22 第一節 四因子評估模型…………………………22 第二節 共同基金績效評估………………………27 第三節 基金之持股類型比例……………………36 第五章 結論與建議…………………………………40 第一節 結論………………………………………40 第二節 建議…………………………………………41 參考文獻……………………………………………45