配對交易是投資理財中常見的操作行為，其交易的方式為選取兩兩相近的股票，並計算股票間歷史股價的價差平均與標準差，若兩股之新一期股價的價差大於由歷史股價的價差平均與標準差所形成的基準值，則對一支股票進行做多，並對另外一支進行放空。主要目的為消除市場變動造成的系統風險，但仍是會受到個別風險影響，如果能減少錯誤決策造成的損失，將可以提升投資報酬率。 本研究提出了以群集分析結合配對交易的新型態交易方式，命名為「群集交易」。這種交易方式突破傳統配對交易一次交易兩支股票(一對)的限制，可同時進行多支股票的操作，由於該方法同時考慮多支股票的資訊，使其在發生錯誤決策時，能以其他股票的獲利進行攤銷，降低損失，相較於舊有方法，群集交易將在獲利上有著更高的表現與穩健性。另外，由Gatev(2006)等人提出的相似度衡量方法中會遇到迭代的問題，使的相似度計算結果具有瑕疵，進而增加錯誤決策的機率。而本文將以指數平滑法改良其計算方式，同時加入趨勢走勢的資訊，使我們可以得到更精確的相似度計算結果，以提升投資報酬率。 我們利用近四年來台灣50成分股的資料依據產業分類並分別進行實驗，然後將大盤、配對交易、群集交易(原始相似度)和群集交易(新的相似度)進行比較，最後再依據各個產業找出投資報酬率最佳的參數設定。 ;Pairs trading is one of famous strategy in investment, the way of trading is being chosen by the two similar stocks, and being calculated on their historical price data’s gap of mean and standard deviation. If the gap of new price data is exceed the boundary, we will buy the stock and short selling the other stock. The purpose of it is removing the systemic risk by the market of changing, but it also be effected by idiosyncratic risk. If we can reduce the loss of making fault decision, we will get higher ROI. We propose a new trading way which combine clustering analysis and pairs trading is named “Cluster trading”. The limit of pairs trading which only can trade two stocks (one pair) at each time is broken. It can be traded with more than two stocks at each time. More stock’s information can be considered and the loss can be shared by other stocks when making the fault decision to let the loss be decreased. Thus, we will have higher and more robust in the profit of making decision by comparing with pairs trading. However, we also find out the famous similarity which is proposed by Gatev, Goetzmann, and Rouwenhorst (2006) in pairs trading has a drawback about iterating, and it makes the similarity inaccurate. We use EWMA and stock’s trend to improve the similarity to get the better results and get higher ROI. We use the stocks FTSE TWSE Taiwan 50 Index during recent four years and classify some of stocks into different industries. And then, compare TAIEX, pairs trading, cluster trading with original similarity and cluster trading with new similarity which is better. Finally, find the setting which have high ROI in cluster trading with new similarity in different industries.
