運用相似度分群演算法進行台灣股市開盤模式之知識發現

Abstract

自2007年發生美國次貸風暴以來，全球股市動盪，動輒每日以5%至10%下跌，投資人若不採取避險操作，則財富將快速下降，日內交易可以完全避免隔夜風險，故本研究將以股市開盤模式進行分群，以輔助投資人進行日內交易的決策擬定。 本研究以樣本特徵參數相似度做為分群的依據，以去除不同正規化的方法對分群效果的影響，並且以基因演算法對各特徵參數的權重進行調整，以得到較佳的權重設定後，再對樣本進行分群以得到交易策略的擬定依據。實驗組分為單純以本研究提出之演算法進行策略擬定及以本研究的方法學判定交易時間點，然後用自組織映射圖的分群結果進行多空的走勢研判。 由交易績效實證結果得知，以本研究所提出的演算法，分群後進行群體決策，可以有較佳的平均獲利；若以本研究的演算法進行交易時間點的研判後，再結合自組織映射圖進行策略的擬定，則可以有較佳的平均獲利及較高的準確率。可知藉由分群方法學來研判開盤模式，可以提高交易準確率及獲利，以提供一般投資大眾及專業人員做為交易決策的輔助系統。

Since 2007, the subprime mortgage crisis in USA has not only caused a severe recession but also stock markets worldwide have suffered severe falls, which can cause indices dropped by 5% to 10% a day, therefore investors may lose their investments quickly without hedging their portfolios. Day trading can have absolutely no overnight risk, thus this study is to create clusters of stock opening patterns, and support investors on day trading decisions. In this study, clusters are created based on the similarity of sample characteristic values, which reduces the impact of normalization on clusters, and uses genetic algorithm to adjust the weights of each characteristic value, and then repeat clustering procedures. With the algorithm and methodology proposed in this paper, we can determine the trading strategy and trading time, and use the cluster result of self-organizing map to forecast the index trend. The result of study shows that the proposed algorithm can determine the trading time, and lead to better profitability. Additionally, integrating self-organizing map with the proposed algorithm can help formulate strategies, and achieve better profitability and accuracy rate. The analysis of indices opening pattern with clustering can enhance profitability and accuracy rate, thus can be developed as trading support systems.