應用關連式規則分析於主力券商分行之操作策略

Applying Association Rule on Analysis of Equity Brokerage Branch on Trading Strategy

陳志萍(Chih-Ping Chen)

指導教授 ： 林博文

國立清華大學/科技管理學院/高階主管經營管理碩士在職專班/碩士(2009年)

摘要

從一千多支股票裡找出自己想要的個股，光靠人工之力，實在太累, 隨著網際網路等資訊科技的發達，這許多股市資料都被數位化,利用人工智慧技術應用整理、分析使得人們得以節省最多時間達到選股的目的, 排除人為或主觀的判斷模式已經是必然的趨勢，從許多的文獻以及實務上的應用可以發現傳統統計分析方法無法正確掌握到市場持續不斷變動的特性。

當投資人，不論法人或散戶，選擇一檔股票做為投資時，一定認為該股票的股價將會飆漲，本論文研究嘗試從籌碼面的角度切進，希望找出主力投資者買進的訊息，我們預期可以透過跟隨主力投資者的買賣行為以及避開一般散戶投資人的不理性買賣行為而獲得較佳的報酬，探討是否可由每次獲利金額較高的券商進出因子來判斷較佳的選股與選時策略。

本論文研究以金麗科(代碼3228)做為模型訓練標的，訓練期間的資料樣本為2005年3月2日至2008年10月30日共計950個交易日，本研究中採用量價關係做為找出每次獲利期間的工具，並利用大富資訊的券商進出分析系統5.1R版來模擬每日券商分行買賣進出的買賣平均成本與成交量，做為計算主力券商分行籌碼面的依據指標。

本研究所提出之關聯式規則的Apriori資料挖礦模式，透過可擴充的量價與籌碼架構的知識，來建立主力投資者之關聯式規則模型。迅速找出金麗科(代碼:3228) 14個主力券商分行，而其報酬金額總和佔前幾名報酬金額總和平均值為高達60.7%，而透過股價與籌碼的相關係數可以找出主力投資者進貨模型且具有極佳之投資跟隨效果。此外，為驗證所提模式之有效性本研究利用另外5檔上櫃股票來進行實證研究，其中3 支股票的可以由關聯式規則找出的主力券商分行。若透過跟進主力投資者的買賣行為而獲得較大盤的報酬率2-3倍的超額報酬。

關鍵字

資料挖礦 ； 關聯式規則

並列摘要

To identify individual stocks from thousands of available stocks is simply too exhaustive. Advancements in the internet and information technology have digitized most stock market data, allowing stock selection to be done in greatly reduced time through the application of artificial intelligence to manage and analyze the data, elimination of human factor or subjective judgment is an inevitable trend. Many literature and real world application have shown that traditional statistical analysis can no longer grasp the continuous variability of the market.

When investors, whether institutional or individual, selects a stock for investment, it must believe that stock has great potential for capital gain. This thesis attempts to analyze the market player profile to try and identify signals for buy position by major market players. We expect to obtain higher return on investment by following major market players buy and sell positions and avoid the general individual investors irrational buy and sell behavior, and explore whether we can formulate a better stock selection and market timing strategy based on security firms with higher return from each transaction.

This thesis use RDC Semiconductor (Code 3228) as simulation target and sampled data from 2 Mar 2005 to 30 Oct 2008, total 950 trading days. This research use quantity and price correlation to identify periods of positive return, and use security firm transaction analysis system version 5.1R from Fortek to simulate the daily average transaction cost and volume by each security firm as the indicator of a major market player.

The Apriori data mining model based on correlation rule proposed by this research can, through expandable volume/price and player profile knowledge framework, build a association rule model of major market players, and quickly identify the 14 major security firms associated with RDC Semiconductor (Code:3228) whose return on investment accounts for 60.7% of total returns. Establishing major market players transaction model through price and market profile correlation proves to be a good investment strategy. In addition, in order to verify the effectiveness of the correlation model, this research looked at 5 other OTC traded stocks, among them, we were able to identify the major market players through the correlation model for 3 stocks. One can earn up to 2 ~ 3 times the market return if he followed the transaction pattern of a major market player.

並列關鍵字

data mining ； association rule

參考文獻

 [李安邦，1997] 李安邦，「以遺傳演算法為基底的模糊專家系統於投資策略之應用」，元智管研所碩士論文，1997

 [簡禎富、李培瑞、彭誠湧，2003]簡禎富、李培瑞、彭誠湧，「半導體製程資料特徵萃取與資料挖礦之研究」，資訊管理學報，第十卷第一期，頁63-84，2003。

 [林萍珍，1998] 林萍珍，「遺傳演算法在使用者導向的投資組合選擇之應用」，中央資管所碩士論文，1998

 [Cios and Swiniarski，1998] Cios, K.J., Pedrycz, W. and Swiniarski, R. Data Mining Methods for Knowledge Discovery. ，1998.

 [Levy, 1966] Robert A. Levy. Conceptual Foundation of Technical Analysis. Financial Analysis Journal, 84-85, 1966