| 摘要: | 本計晝預計發展多重約略法則為核心的創新兩極式決策模型，結合直觀式模糊绩效評估，應 用在金融市場的投資選股問題。在多評準決策科學（MCDM)的領域，過去主要可分為多屬性決策 (MADM)以及多目標決策（MODM)兩大類；MADM大多應用於選擇、排序決策，而MODM 則用於最佳化資源分配。然而，自從國際知名的IDSS約略集合理論研究團隊（主要成員為Greco, Slowinski , and Matarazzo 教授）在 2001 年提出透過 dominance-based rough set approach (DRSA)來解 決MCDM問題，開啟了以多重約略法則為基礎的決策研究，在此計晝稱為multiple rough-rule-based decision making (MRDM)。面對充滿複雜因素、交互影響並具有不明確關係的金融市場，本計晝透 過解析歷史財務绩效與未來股票報酬的關係，開發基於多重約略法則的創新兩極式決策模型。此模 型將可擷取財務表現與股票報酬間的複雜因果關係，轉化為正、負兩極約略知識法則群。投資者可 依據本身選定的門檻值，融合直觀式模糊（intuitionistic fuzzy)绩效評估；應用此模型，投資人的 選股決策過程能加透明，建立在可理解的約略知識法則上。研究對象為台灣特有的金控產業，預計 涵蓋2006-2016的期間，將發展滾動式單期以及多重期間的輔助投資決策模型，並將近年期的選股 結果分群比較，也將分別與大盤指數、金融類股價指數的同期報酬比較，以驗證此創新投資決策模 型的有效性。
This project attempts to develop a novel bipolar decision model based on multiple rough rules (i.e., rough knowledge), infused with intuitionistic fuzzy performance evaluation, for supporting equity investment (i.e., stock selection) decision. In conventional multiple criteria decision making (MCDM) research, it could be mainly divided as MADM (for ranking or selection) and MODM (for resource optimization problems). Nevertheless, since the famous IDSS rough set research lab proposed a new approach—dominance-based rough set approach (DRSA)—for solving MCDM problems, there has been a rising trend in MCDM based on multiple rough decision rules; in the present project, it is termed as multiple rough-rule-based decision making (MRDM). In the presence of complex, imprecise, and interrelated relationships among the multiple criteria and the subsequent returns in security market, this project intends to develop a novel bipolar decision model by analyzing the relationship between complicated historical financial performance and the subsequent stock returns. The proposed model may retrieve/induct the implicit complex patterns (rough knowledge) from historical data, and transform the analytical results into two groups of positive and negative rough decision rules; decision maker could form his/her preferred bipolar model by setting a threshold value to select the covered rules in a bipolar model, which will be further infused with the intuitionistic fuzzy performance evaluation for gaining investment insights. The research target will be the public-listed financial holding companies (stocks) in Taiwan, and the covered research period will range from 2006 to 2016 (11 years). The rolling single period (i.e., annual) model and multiple periods (includes quarterly, biannual, and annual) time-lagged models will be developed; the stock selection results will be examined and compared with the major market index (also the financial industry composite index) to examine the effectiveness of the proposed novel bipolar decision model. |
