隨著計算智慧(Computational Intelligence)工具的發展日益成熟, 將這些工具應用在經濟或財務問題上的文章逐漸增多, 國際上對這塊跨科際的研究領域愈來愈重視, 本論文討論和使用的工具是演化計算中的遺傳規劃(Genetic Programming)。 很多應用遺傳規劃的文章,經常被人質疑的一個問題是參數的任意設定, 參數任意或故意的設定是否會影響搜尋的速度或結果,是很多使用者所關心的; 除了參數設定的問題之外,也會有人質疑應用強大的搜尋工具是否會發生過度學習, 關於這些質疑,本論文仔細討論應用此工具時應該注意的參數設定和相關問題, 同時也討論關於過度學習和學習不足的問題,並提出一些心得,可以作為遺傳規劃使用者的參考。 除了討論這些問題之外,本論文使用這些經驗和心得,實際應用在交易策略尋找的問題上, 其中一個應用範圍是關於跨國資金流動的問題,另一個則應用在股票市場。 由以前相關的文獻可以知道,使用遺傳規劃演化出來的交易策略通常是複雜而且不易閱讀, 如果希望遺傳規劃能夠真正對投資者有所幫助,除了交易策略是否能夠獲利之外, 如何改善策略不易閱讀是很重要的問題。 本論文也針對這個問題提出方法並且實際應用在股票市場, 結果發現尋找出來的交易策略不但有超額報酬而且策略簡單易懂。

本文應用遺傳規劃交易程式來檢驗八個國家的股票指數和外匯巿場的表現，採用移動視窗的方法，測試三組獨立的期間，重新檢驗較早期的研究結果，並繼續延申探討，包括交易報酬與交易行為。實證結果顯示，不論在股票還是外匯巿場，若訓練期間的資料選擇不當，遺傳規劃的獲利表現會不理想。資料形態不但會影響遺傳規劃交易程式的獲利性，同時也決定了程式本身的一些觀察特性。我們另外分析了交易程式的複雜度、演化時間、交易頻率和一致性。交易程式的複雜度和演化時間有正向的相關性，但複雜度和報酬、以及演化時間和報酬之間都只有很弱的關係。這些發現可以讓我們更了解遺傳規劃演化交易策略的過程，有助往後更進一步的研究。

軟體元件電子市集突現：以代理人為基礎之計算經濟研究途徑 摘 要 軟體發展與演進過程中，產生軟體危機問題，而軟體再用是解決軟體危機的重要因應之道。軟體元件電子市集的興起是軟體演進史上一個重要里程碑，提供軟體再用的核心基礎建設。 本文探討軟體元件電子市集突現的本質原因和信任關係的發展過程，以遺傳規劃法(Genetic Programming, GP)為主的代理人基礎的計算經濟 (Agent-based Computational Economics, ACE) 研究途徑，整合軟體元件特性、交易成本、滿意和信任關係建立模擬模式。藉以觀察和分析底層買賣雙方連續滿意交易與信任關係發展，和上層軟體元件電子市集行為突現(emerge)動態過程。 結果顯示：在市場力量下，具標準化軟體元件，電子市集行為突現過程中，謹慎型交易策略將會勝出，進而主導整個市場。當元件功能特殊性程度低時，電子市集行為的購買率將比元件功能特殊性程度高者更為顯著。如果考慮交易態度滿意與否，記憶型滿意者市集行為的購買率將顯著低於高滿意型，而顯著高於低滿意型。若考慮不同信任程度函數，高信任型電子市集購買率顯著高於低信任型，低信任型其電子市集購買率顯著高於不信任型，對於目錄型市集行為和忠誠目錄型市集行為，上述信任函數的形態亦依序顯著影響購買率的高低。 同時，在不同信任型之間，高信任型大多數有連續累積交易行為；而低信任型則同時採用連續和臨時交易行為；不信任型大多數是臨時交易行為，要花費更多時間的演化，以建立彼此信任關係才會出現連續交易乃至於連續累積交易行為。 關鍵字：軟體元件電子市集、交易成本、遺傳規劃法、代理人基礎計算經濟、信任、突現 / The Emergence of Software Component Electronic Marketplaces: Through An Agent-based Computing Economics Approach Abstract Software reuse plays a vital role in response to software crises in software evolution. An emergence of software component e-marketplace is one of the great milestones providing a core infrastructure for software reuse. The objective of this study involving features of s/w components, transaction costs and satisfaction-trust relations intends to understand why s/w component e-marketplaces emerge as well as demonstrate how they do. The model allows agents to develop their trust in the market as a function of continuation of a satisfied relation through an agent-based computational economics approach with genetic programming. The findings show that the agents with prudent strategies tend to dominate the market in evolution of e-marketplaces under the market power. In addition, the lower level the functional particularity of component is, the higher the buying rate is. As the satisfaction attitude is taken into consideration, the buying rate of recall-satisfied agents lies between that of low-satisfied agents and that of high-satisfied agents. Moreover, when the comparisons are made among the three types of trust function, the buying rate of the high-trust agent is higher than that of low-trust agents. And the buying rate of the low-trust agent is bigger than that of not-trust agents. Similarly, the sequences of the buying rate are strongly influenced by different type of trust function at the catalog market and the loyal catalog market. Meanwhile, almost all high-trust agents have continuous and loyal trade behavior. Either continuous or temporal trade behavior is usually found in the low-trust agents. The tentative trade behavior is seen among almost every not-trust agents. In another words, it is well obvious that it takes more time for the not-trust agents to accumulate trust from their possible trade partners. Keywords: Software component electronic marketplaces; Transaction costs; Genetic programming (GP); Agent-based computational economics (ACE); Trust, Emergence

在目前的企業環境中，很多企業致力於管理和應用組織知識，來維持他們的核心能力和創造競爭優勢。有效率的管理組織知識，能減少解決問題的時間和成本，並增加組織學習和創新能力。並且，由於累積知識資源的需求，很多企業開始發展知識庫，以儲存組織及個人的知識，用來增加組織整體的效率、支援日常的運作以及企業策略的操作。 知識管理是現代的典範，可用來有效管理組織知識，進而改善組織績效。知識管理的目的是強調管理知識的流動及流程。在知識管理流程方面，主要區分為知識擷取、整合、儲存/歸類、散播和應用知識等程序。另外，資訊技術可用來協助知識管理，並能使知識管理更有效率。知識管理的主要議題之ㄧ是知識的擷取，由於目前知識來源的提供，主要是透過知識工作者，可是它對於知識工作者而言，是一種額外的負擔。因此，設計一個有效的方法來自動產生組織知識，以減輕他們的額外負擔，將是一個很重要的課題。第二個相當重要的議題是知識整合，由於不同來源的知識，可能造成組織知識的衝突，因此設計一個知識整合方法，把不同來源的知識整合成一個完整的知識，組織將會逐漸增加這方面的需求。 分類在很多應用中是常遭遇的問題，例如財務預測、疾病診斷等。在過去，分類規則常藉由決策樹的方法所產生，並用於解決分類的問題。在本論文中，提出兩個以遺傳規劃為基礎的知識擷取方法和兩個以遺傳規劃為基礎的知識整合方法，分別支援知識管理流程中的知識擷取和知識整合。 在兩個所提的知識擷取方法中，第一個方法是著重在快速和容易地找到想要的分類樹，但是，此方法可能會產生結構較複雜的分類樹。第二個方法是修正第一個方法，產生一個較精簡和應用性高的分類樹。這些所獲得的分類樹，能被轉換成規則集合，並匯入知識庫中，幫助企業決策的制定和日常的運作。 此外，在兩個所提的知識整合方法中，第一個方法，能自動結合多重的知識來源成為一個整合的知識，並可匯入知識庫中，但是此方法只考慮到單一時間點的整合。第二個方法則是可以解決不同時間點的知識整合問題。另外，本論文提出三個新的遺傳運算子，在演化過程中，可用來解決規則集合中有重複、包含和衝突等常見的問題，因而可以產生較精簡及一致性高的分類規則。最後，本論文採用信用卡資料及乳癌資料來驗證所提方法的可行性，實驗結果皆獲得良好的成效。 / In today’s business environment, many enterprises make efforts in managing and applying organizational knowledge to sustain their core competence and create competitive advantage. The effective management of organizational knowledge can reduce the time and cost of solving problems, improve organizational performance, and increase organizational learning as well as innovative competence. Moreover, due to the need to accumulate knowledge resources, many enterprises have devoted to developing their knowledge repositories. These repositories store organizational and individual knowledge that are used to increase overall organization efficiency, support daily operations, and implement business strategies. Knowledge management (KM) is the modern paradigm for effective management of organizational knowledge to improve organizational performance. The intent of KM is to emphasize knowledge flows and the main process of acquisition, integration, storage/categorization, dissemination, and application. Furthermore, extant information technologies can provide a way of enabling more effective knowledge management. One of the important issues in knowledge management is knowledge acquisition. It is an extra burden for knowledge workers to contribute their knowledge into repositories, such that designing an effective method for abating an extra burden to automatically generate organizational knowledge will play a critical role in knowledge management. A second rather important issue in knowledge management is knowledge integration from different knowledge sources. Designing a knowledge-integration method to combine multiple knowledge sources will gradually become a necessity for enterprises. Classification problems, such as financial prediction and disease diagnosis, are often encountered in many applications. In the past, classification trees were often generated by decision-tree methods and commonly used to solve classification problems. In this dissertation, we propose two GP-based knowledge-acquisition methods and two GP-based knowledge-integration methods to support knowledge acquisition and knowledge integration respectively in the knowledge management processes for classification tasks. In the two proposed knowledge-acquisition methods, the first one is fast and easy to find the desired classification tree. It may, however, generate a complicated classification tree. The second method then further modifies the first method and produces a more concise and applicatory classification tree than the first one. The classification tree obtained can be transferred into a rule set, which can then be fed into a knowledge base to support decision making and facilitate daily operations. Furthermore, in the two proposed knowledge-integration methods, the former method can automatically combine multiple knowledge sources into one integrated knowledge base; nevertheless, it focuses on a single time point to deal with such knowledge-integration problems. The latter method then extends the former one to handle integrating situations properly with different time points. Additionally, three new genetic operators are designed in the evolving process to remove redundancy, subsumption and contradiction, thus producing more concise and consistent classification rules than those without using them. Finally, the proposed methods are applied to credit card data and breast cancer data for evaluating their effectiveness. They are also compared with several well-known classification methods. The experimental results show the good performance and feasibility of the proposed approaches.

我們把經濟體視為一個複雜適應系統(complex adaptive system), 強調系統中異質性(heterogeneous)agent的學習適應行為與agent之間的互動性交互作用, 此時主流經濟學裡的分析架構, 如:代表性個人模型(represesentive agent model)、理性預期(rational expectation)、固定點均衡分析(fixed-point equilibrium analysis)等將不再適用, 取而代之的是演化經濟學(evolutionary economics)的研究典範, 這樣的研究架構下, 並沒有適當的數學分析工具可資運用, 因此我們改以agent-based建模(agent-based modelng)的社會模擬(social simulation)來建構一個人工的經濟體(artificial economy), 以此為主要研究方法, 這就是agent-based計算經濟學(agent-based computational economics)或稱人工經濟生命(artificial economic life)。 本文中以股票市場為主要的研究課題, 我們以遺傳規劃(genetic programming)的人工智慧(artificial intelligence)方法來模擬股市中有限理性(bounded rational)異質交易者的交易策略學習行為, 建構出一個人工股票市場(artificial stock market), 在這樣的架構下, 我們成功地產生出類似真實股票市場的股價時間序列特性, 我們同時也檢定了人工股票市場中價量的因果關係, 說明了在沒有外生因素之下, 人工股票市場的複雜系統可自發地產生出雙向的價量因果關係, 進一步地, 我們研究下層agent(交易者)行為與上層股價時間序列行為的關聯性, 我們也發現個體的行為並不能直接加總或推論出複雜適應系統的總體行為, 這就是突現性質(emergent property)的發生, 最後, 本文描述了agent-based計算經濟學研究架構的優勢與缺點, 再附帶介紹一個用以進行agent-based建模相關研究的軟體程式庫-SWARM。

知訊者與非知訊者資訊不對稱之議題在學術殿堂一直廣為學者所研究討論，且各國證管機關為了維持證券市場公平性、保護非知訊者權益並且維持股票市場的穩定運作，適度的資訊揭露以維持證券市場的公平性一直都是各國證券交易所重視的政策目標。 是故，本研究利用代理人基人工股票市場來探討資訊揭露對於金融市場之影響。在此架構下之交易者皆已有限理性方式來呈現。他們是以遺傳規劃(genetic programming)之方式來學習並修正他們對於未來之金融市場之預期。在透過即時的模擬價格之資訊揭露，我們嘗試探討此資訊揭露之金融政策措施對於市場之波動性、市場之流動性之影響。 / The topic of asymmetric information between the informed traders and uninformed traders has been widely discussed by researchers in academics. To maintain the fairness of securities market, an appropriate information disclosure is quite important for authorities of securities regulation to protect the rights and interests of uninformed traders, and to maintain the operations of securities market stable. Based on these reasons, we construct an agent-based artificial stock market to investigate how information disclosure affects a financial market. In this framework of artificial stock market, all traders are characterized by bounded rationality. The traders are able to learn and adjust their predictions of financial market by means of a genetic programming algorithm. We try to understand how market transparency affects the volatility and the liquidity of a securities market.

Black-Scholes選擇權定價模型是各種選擇定價的開山始祖，無論在理論或實務上均獲致許多的便利及好評，美中不足的是，這種既定模型下結構化參數的估計問題，在真實體系的結構訊息未知或是不明朗時，或是模式錯誤，亦或政治結構或金融環境不知時，該模型在實證資料的評價上會面臨價格偏誤的窘境。是故，許多的數值演算法（numerical algorithms）便因應而生，這些方法一則源於對此基本模型的修正，一則是屬於逼近的數值解。 評價選擇權的方法雖不一而足，然所有的這些理論或模型可分為二大類即模型驅動的理論（model-drive approach）及資料驅動的理論（data-driven approach）。前者是建構在許多重要的假設，當這些假設成立時，則選擇權的價格可用如Black-Scholes偏微分方程來表示，而後再用數值解法求算出，許多的數值方法即屬於此類的範疇；而資料驅動的理論（data-driven approach），其理論的特色是它的有效性（validity）不像前者是依其假設，職是之故，他在處理現實世界的財務資料時更顯見其具有極大的彈性。這些以計算智慧（computation intelligence）為主的財務計量方法，如人工神經網路（ANNs），遺傳演算法（GAs），遺傳規劃（GP）已在財務工程（financial engineering）領域上萌芽，並有日趨蓬勃的態勢，而將機器學習技術（machine learning techniques）應用在衍生性商品的定價，應是目前財務應用上最複雜及困難，亦是最富挑戰性的問題。 本文除了對現有文獻的整理評析外，在人工神經網路方面，除用於S&P 500的實證外，並用於台灣剛推行不久的認購構證評價之實證研究；而遺傳規劃在計算智慧發展的領域中，算是較年輕的一員，但發展卻相當的快速，雖目前在經濟及財務上已有一些文獻，但就目前所知的二篇文獻選擇權定價理論的文獻中，仍是試圖學習Black-Scholes選擇權定價模型，而本文則提出修正模型，使之成為完全以資料驅動的模型，應用於S&P 500實證，亦證實可行。最後，本文結合計算智慧中的遺傳演算法（ genetic algorithms）及數學上的加權殘差法（weight-residual method）來建構一條除二項式定價模型，人工神經網路定價模型，遺傳規劃定價模型等資料驅動模型之外的另一種具適應性學習能力的選擇權定價模式。 / The option pricing development rapid in recent years. However, the recent rapid development of theory and the application can be traced to the pathbreaking paper by Fischer Black and Myron Scholes(1973). In that pioneer paper, they provided the first explicit general equilibrium solution to the option pricing problem for simple calls and puts and formed a basis for the contingent claim asset pricing and many subsequent academic studies. Although the Black-Scholes option pricing model has enjoyed tremendous success both in practice and research, Nevertheless, it produce biased price estimates. So, many numerical algorithms have advanced to modify the basic model. I classified these traditional numerical algorithms and computational intelligence methods into two categories. Namely, the model-driven approach and the data-driven approach. The model-driven approach is built on several major assumptions. When these assumption hold, the option price usually can be described as a partial differential equation such as the Black-Scholes formula and can be solved numerically. Several numerical methods can be regarded as a member of this category. There are the Galerkin method, finite-difference method, Monte-Carlo method, etc. Another is the data-driven approach. The validity of this approach does not rests on the assumptions usually made for the model-driven one, and hence has a great flexibility in handling real world financial data. Artificial neural networks, genetic algorithms and genetic programming are a member of this approach. In my dissertation, I take a literature review about option pricing. I use artificial neural networks in S & P 500 index option and Taiwan stock call warrant pricing empirical study. On the other hand, genetic programming development rapid in recent three years, I modified the past model and contruct a data-driven genetic programming model. andThen, I usd it to S & P 500 index option empirical study. In the last, I combined genetic algorithms and weight-residual method to develop a option pricing model.