技術分析之有效性檢定與資料探查誤差研究：道瓊工業指數之實證

Abstract

本研究目的在檢定財務市場中技術分析（Technical analysis）之可行性，並以美國道瓊工業指數十年之日資料（1987-1996）為實證。自1960年代以來，學術界對於在實務界中廣泛被使用的技術分析一直有所爭議，許多學者認為看似有利可圖的技術交易法則（Technical trading rules）所產生的利潤實際上是資料探查（Data snooping）的結果，並不具統計上顯著性。White（2000）提出了一「真實檢定」（White’s Reality Check），該方法可以在檢定大量模型之有效性時避免資料探查之誤差，協助研究者得到正確之檢定結果。 本研究共提出了4,0108項技術分析模型，其中包括過去文獻中的「簡單交易法則」，以及本研究自行發展出之「反向交易法則」與「投資人交易策略」。而經由White之真實檢定（White’s Reality Check），本研究主要發現如下：一、無論是「簡單交易法則」或是「反向交易法則」，其在實證資料中獲利能力均無法到達檢定上之顯著水準，意即投資人無法利用單一技術交易法則在股市中獲利。二、「投資人交易策略」在實證資料上的獲利程度仍無法到達檢定上之顯著水準，所以即使是考慮到投資者之策略與選擇性行為，投資人仍無法利用多種技術交易法則在股市中獲利。三、「反向交易法則」所獲得之平均獲利高於傳統上的「簡單交易法則」，意即事實上大多數傳統技術分析方法都誤判了市場的走勢。本研究之主要結論為技術分析在實證資料中無法使投資者獲利，並確認股票市場之弱效率（Weak efficient）假設。

The technical analysis, which has been widely used in financial markets for many decades, is believed to be profitable by many practitioners. However, some researchers considered such profitability is a result of data snooping, due to inappropriate tests and incomplete universe. Different from most researches that based their conclusions on only few technical rules and inappropriate tests, this study provides a more complete universe of technical trading rules and tests their profitability with data snooping correction. In this study, the universe consists of 40,108 trading rules, including the contrarian rules and the investor's strategies that were never analyzed in previous studies. With a more complete universe, the profitability of technical analysis is tested using the Reality Check of White (2000, Econometrica, 68, 1097–1126) to correct the data snooping bias. Based on Dow Jones Industrial Average (DJIA) Index data from 1987 to 1996, we find that the contrarian rules and investor's strategies outperform simple trading rules on average return, yet there exists neither profitable technical rule nor profitable investor's strategy. As a result, the efficiency of market is further supported by this study.