以模糊概念選擇高效能投資股

Abstract

由於資訊科技的發達，使得投資人可以便捷地取得上市公司公開揭露的財務報表，藉由分析這些財務報表，可以幫助我們選擇投資標的與擬定投資策略。在本篇論文中，我們提出一個方法以選取高效能投資股。我們所提出的方法包括四個階段。第一階段，我們從每一篇財務報表中，萃取財務活動片語(financial activity phrase)與財務比率(financial ratio)，並將它轉換成一個特徵向量。第二階段，我們利用affinity propagation (AP) 演算法，將所有特徵向量依相似性分成數個群集，並找出每一個群集內的代表特徵向量。第三階段，我們利用fuzzy k-nearest neighbors (FKNN)演算法，將每個特徵向量模糊化，以計算出每一個特徵向量對不同股價變動趨勢的隸屬程度。第四階段，我們利用這些模糊化後的特徵向量，計算新的特徵向量的隸屬程度，並對股票進行排序，然後，從排序中選出高效能投資股。由於，我們利用AP演算法將特徵向量分群，可降低FKNN選到不好的參考特徵向量，因此，我們所提出的方法可以提供一個不錯的管道，選取高效能投資股。實驗結果顯示，在中長期股價趨勢預測的投資平均獲利，我們所提出的方法優於支援向量機方法。

With advance in information technology, a large amount of financial reports can be accessed easily. Analyzing those financial reports can help investors to select investment targets and plan their investment strategies. In this thesis, we propose an effective method to select top performing stocks. The proposed method consists of four phases. First, we extract financial activity phrases and financial ratios from each financial report and transform it into a feature vector. Second, we utilize the affinity propagation (AP) algorithm to group similar feature vectors into clusters, and identify exemplar of each cluster. Third, we use the fuzzy k-nearest neighbors (FKNN) algorithm to compute membership degrees towards each class for a feature vector. Finally, we use these fuzzified feature vectors as references to rank new feature vectors and select top performing stocks from the ranked list. Since we utilize the AP algorithm to reduce the chance for the FKNN algorithm to choose bad references when fuzzifying the membership degrees of a feature vector, the proposed method provides a good channel to select top performing stocks. The experimental results show that the proposed method outperforms the SVM method in terms of average trading profit.