應用馬可夫決策過程進行台股期貨日內交易策略之研究

Tunghai University Institutional Repository > 工學院 > 工業工程與經營資訊學系所 > 碩博士論文 > Item 310901/4324

Please use this identifier to cite or link to this item: http://140.128.103.80:8080/handle/310901/4324

Title:	應用馬可夫決策過程進行台股期貨日內交易策略之研究
Other Titles:	On the Optimal Intraday Trading Strategy Analysis for TAIMEX Index Future Using Markov Decision Process
Authors:	楊士賢 Yang, Shih-Hsien
Contributors:	姚銘忠;胡坤德 Yao, Ming-Jong;Hu, Kun-Te 東海大學工業工程與經營資訊學系
Keywords:	日內交易;馬可夫決策過程;門檻值;移動平均線 Intraday;Markov Decision Process;Threshold;Moving Averages Curve
Date:	2003
Issue Date:	2011-05-19T05:59:47Z (UTC)
Abstract:	本研究試圖運用「馬可夫決策過程」，建立面對日內期貨價格波動時，最佳動態決策的數學模式。亦即提供期貨市場日內交易者一個決策支援系統，協助其判斷價格變動時，交易者應該採取的最佳策略。如：買進、賣出或是觀望等。從文獻中本研究發現：探討日內交易之文獻以研究其交易型態居多，鮮少針對日內交易策略提出探討者。再者運用馬可夫決策過程為決策輔助系統的研究，並沒有以實際「日內交易資料」當作其狀態變數者。本研究根據日內交易型態呈現U字型態之特性和敘述統計分析資料，運用日內每五分鐘交易價格的相對變動比例為基準，定義出10個狀態變數；本研究再依歷史資料建立移轉機率矩陣，求算各個狀態變數下其各個決策之期望獲得，再運用線性規劃求解馬可夫決策過程數學模式的最佳動態決策。為驗證該模式的效率，本研究以台指期貨為研究標的物，以實際資料進行模擬操作；模擬驗證所涵蓋的期間為2002年五月至2003年四月。在第一階段的數據實驗中，本研究發現有過度頻繁的交易次數，導致獲利均被交易時的保證金與手續費稀釋的現象；因此本研究針對最佳策略加入不同門檻值（Threshold）和移動平均線(Moving Averages Curve)的改善方法，來探討減少交易次數而獲得增進報酬的可行性。在加入改善方法後的第二階段數據實驗中，顯示該模式的效率優於其他常見的交易策略；故本研究結論：馬可夫決策過程數學模式可以提供日內期貨交易者有效的決策支援。 In this study, we demonstrate how to derive a decision support system to maximize the dealer’s expected profit in Taiwan stock index (TX) futures markets by solving a mathematical model based on Markov Decision Process (MDP). The MDP decision support system assists the intraday dealers to take an optimal trading strategy, e.g., offset, holding, buy-in, or sell-out, as the price of future changes. From the research of reference, we find that the most of investors much consider in the different kind of trading systems using in daily trading, they seldom discuss on the studying about the strategy of daily trading systems. The data of daily trading are not actual state variables in the use of decision procedure on the research of trading systems by MDP. We define the ten state variables according to the intraday pattern appearing as a U-shaped pattern and the analytic data of descriptive statistical and basis on the daily five minutes trading price report of proportional changing. We establish the transition probability matrix by the historical data to calculate each trading decision under every state variable, and then we obtain the expected reward. In the use of linear program, we get the best mathematic model of the optimal moving decision. In order to evaluate the effectiveness of the optimal trading strategy from our decision support system, we simulate the MDP using the historical data of the future market (i.e., from May of 2002 to April of 2003). In the first phase of our experiments, we discover that the optimal trading strategy has a problem of being too nervous in trading the future to earn reasonable profit. Therefore, we add the threshold rule and the moving averages curve to improve the efficiency of the MDP model. It shows that our optimal trading strategy out-performs other popular trading strategies after adding the improvement. Therefore, we conclude that our decision support system based on Markov Decision Process provides an efficient tool for dealers of daily trading in future market.
Appears in Collections:	[工業工程與經營資訊學系所] 碩博士論文

Files in This Item:

File	Size	Format
091THU00030007-001.pdf	277Kb	Adobe PDF	1464	View/Open

Loading...