應用台灣期貨指數圖形分析於期貨市場趨勢行為研究

Abstract

市場結構提供了關於市場的明確資訊，其中價格的變動模式，可以視為是市場中供應與需求的狀態。當需求超過供應時，價格便會上漲，反之亦然。這些行為模式含有寶貴的資訊，有助於對未來市場動向的了解。 本研究欲了解台指期的價格變動模式與其趨勢行為的關連性，透過圖形分析及類神經網路，學習並預測市場趨勢方向，驗證台灣期貨指數是否遵從隨機漫步理論。本研究使用統計領域中的CPA(changepoint-analysis)及PIP(perceptually important points)理論進行研究。CPA尋找價值出現轉折的位置，PIP演算法執行模式特徵值的萃取，並運用這些特徵值作為類神經的輸入變數進而預測市場的趨勢。對照組根據即時環境找出轉折的位置作比較。 研究結果顯示，應用CPA及PIP方法預測市場的趨勢方向，最高能達到83.6%的準確度，證實透過歷史資料的分析，確實能夠預測台指期的市場方向，市場結構的確具有關於市場未來趨勢行為的隱藏資訊，證明台灣期貨指數並不遵從隨機漫步理論。

Market structure provides concrete information about the market. Price patterns can be imagined as the evidence of a supply and demand states in the market. Price shifts higher as the demands exceed the available supply and vice-versa. These patterns convey precious information about what is going to happen in the market. The purpose of this study is to investigate the underlying relation between price pattern in Taiwan Futures Exchange (TAIFEX) Futures Index Market and its following trend. Forecasting the directions of price shift following the pattern through supervised learning and testing with artificial neural network (ANN). This research implements changepoint-analysis (CPA) under statistics field, and perceptually important points (PIP) theory. CPA finds the locations where the shifts in value occur. Then, PIP algorithm performs the feature extraction of the pattern. Then, the PIP are then fed to ANN to forecast the following trends. Control group is built based on online time segmentation algorithm for comparison. The results of this research shows that robust patterns found by CPA have the ability to forecast market trend direction up to 83.6% accuracy. The result indicates that TAIFEX Futures market directions can be forecasted through its historical price robust patterns. Thus, rejecting that TAIFEX Futures Index Market follows random walk theory. In contrast, the control model which was built based on online time segmentation also has the ability to forecast but not as accurate as using the CPA method. In conclusion, analyzing the patterns reflected in the market effectively provide precious insights about its trends behavior.