金融市場中的圖形分析(charting analysis)在實務界已存在多年，然而學術界中對於類似的研究仍相當稀少，本文的主要目的乃是對圖形分析進行系統性的研究，這樣的研究可分為兩部分來說明，首先是如何確認市場中的型態(pattern)，其次是型態確認後的市場異常行為分析。在型態確認方面，本研究利用類神經網路中的自我組織圖(self-organizing maps)，對金融市場中的交易軌跡進行計量建模，這樣的模型我們稱之為軌跡域模型(Trajectory-Domain Model, TDM)。藉由軌跡域模型的使用，市場中的型態可自動的被擷取出來，而非以往文獻中只是外生給定一個型態進行研究。研究發現在不同的市場中可找出類似的型態，另一方面也發現軌跡域模型在不同的市場有不同的配適能力。在型態確認後的異常行為分析上，我們首先將型態視為一個事件，接下來便可利用事件研究法(event study approach)進行異常報酬分析。研究發現某些型態被確認後，市場確實在未來出現異常報酬，而這些異常報酬是不能被許多財務議題所解釋，例如市場價差、非同步交易、報酬率的異質變異、短期動能效應等。 Using Kohonen's self-organizing maps (SOMs), this research takes a systematic and automatic approach to charting into consideration, or more generally, geometric pattern recognition. Such a model is referred to be as the trajectory-domain model (TDM). By applying trajectory-domain models to financial time series, financial patterns are automatically discovered. To see whether these patterns transmit signals, a rigorous analysis of the aftermath behavior of the pattern is conducted based on the event study approach. The procedure proposed in this research therefore starts a formal analysis of financial charts, which has long been used by financial analysts and perceived by econometricians, but has not been put into a closer examination. To some extent, we find that the SOM is an ideal tool to simulate human intelligence in finding or creating patterns that summarize and store useful aspects of our perceptions.
