| dc.contributor | 金融系 | |
| dc.creator (作者) | 江彌修 | |
| dc.date (日期) | 2019-10 | |
| dc.date.accessioned | 7-Apr-2026 13:17:24 (UTC+8) | - |
| dc.date.available | 7-Apr-2026 13:17:24 (UTC+8) | - |
| dc.date.issued (上傳時間) | 7-Apr-2026 13:17:24 (UTC+8) | - |
| dc.identifier.uri (URI) | https://ah.lib.nccu.edu.tw/item?item_id=181928 | - |
| dc.description.abstract (摘要) | 我們爭論人工智慧所學習的對象在於人，因而開發任何自動化程式交易的首要目的，需要先能夠釐清市場參與者本身的思維和邏輯在其交易的過程中是如何運作的。而在有限的認知資源之下，金融市場參與者的推理方式如何影響他們的交易買賣決策及資產配置？在本研究計劃中，驅動投資人買賣決定的是來自於他們基於實例的推理下(case-based reasoning)的情境相似度辨認(recognition by similarity/analogy)。我們探究如何藉由機器學習產生具有自適應性和類別化功用的情境相似度辨認機制，進而建構結合機器學習與行為財務意涵的技術交易策略。這樣的嘗試就我們的認知中是文獻中的首創，也呼應了一個人工智慧時代的來臨。在本研究計劃提案中，我們首先嘗試找出情境相似度辨認中最佳的不同距離定義和相似度函數的組合，進而延拓Chiu, Chiang, Kuo (2017) 所提出的基於相似度情境的技術交易策略(Similarity-based Trading Rule, SBTR)。特別是，我們在原始SBTR基礎上，探求得以產生最優化隨機平均預測(stochastic averaging returns)報酬之不同距離定義與相似度函數的最佳組合。第二，就情境相似度的辨認上我們將引入自應性機器學習。其中我們將考量自應性增強(Adaptive Booting)和隨機森林(Random Forest)來進行情境相似度的辨認與類別化。第三、為了避免以偏誤的結果進行分類器的學習，我們將在研究計劃最後一個階段引入支持向量器(Support Vector Machine)來進行相似度矩陣和分類器同時學習，以期在整體上能有效降低預測的錯誤率並獲取更高的分類準確率。SVM的加入將使得SBTR中的線性相似度衡量得以被拓展到非線性相似度之衡量。 | |
| dc.description.abstract (摘要) | Because the human is right at the center of focus in any form of machine-assisted artificial intelligence, the primal goal in the development of any kind of automated program trading is in need of a full understanding, or at least an attempt to understand, the rationality and logics behind the market participants’ trading mentality. Under constrained cognitive resource, how does the way that market participants reason during decision making would affect his/her trading and asset allocating decisions. In raising such question, we inevitably probe at the behavioral factors driving any buy-sell trading decisions. In this research proposal, by devising a mechanism that conveys how one recognizes past scenario resemblances of the present by similarity/analogy under case-based reasoning, we aim at constructing an automated technical trading rule that synergizes human behavioral motives with machine-assisted adaptive learning. This mechanism, via adaptive machine learning, shall incorporate the ability to adapt to changes and to classify/categorize one’s recognition of the resemblances between market-scenarios of the past and the present. Echoing an era of artificial intelligence to prevail, this research proposal, in the best of our knowledge, shall be the first in the literature to make such attempt. To meet our research objectives, we begin with extending the Similarity-based Trading Rule (SBTR) of Chiu, Chiang and Kuo (2010). In particular, we explore feasible specifications of similarity measures that are capable of producing optimal stochastic averaging returns. Second, we shall employ adaptive machine learning algorithms to devise the mechanism that governs trading decision making via recognition by similarity. In this regard, we use Adaptive Boosting and Random Forest to proceed with the recognition and categorization of market-scenario analogies/similarities. Finally, to avoid inputting biased estimates in one’s classifiers, and in hope of effectively reducing prediction errors while retaining higher categorization accuracy, we shall employ support vector machine to facilities simultaneous learning by the similarity matrices and the classifiers. With support vector machine, we intend to postulate a new measure that quantifies one’s recognition of non-linear similarities between scenario resemblances. | |
| dc.format.extent | 116 bytes | - |
| dc.format.mimetype | text/html | - |
| dc.relation (關聯) | 科技部, MOST107-2410-H004-071, 107.08-108.07 | |
| dc.subject (關鍵詞) | 實證相似度; 基於實例的推理; 機器學習 | |
| dc.subject (關鍵詞) | empirical similarity; case-based reasoning; machine learning | |
| dc.title (題名) | 自適應機器學習與情境相似度辨認於技術交易策略之建構 | |
| dc.title (題名) | Constructing Technical Trading Rules Based on Adaptive Machine Learning and Recognition by Similarity | |
| dc.type (資料類型) | report | |
