Fractional Differencing Predictive Power in FOREX Market
22 Pages Posted: 9 Jan 2020
Date Written: December 3, 2019
Efficient Market Hypothesis (EMH) states that all available information is immediately reflected in the price of any asset or financial instrument, so that it is impossible to predict its future values making it follow a pure stochastic process. Among all financial markets FOREX is usually addressed as one of the most efficient. This paper proposes a methodology to test the efficiency of the EURUSD pair taking only into consideration the price itself. A novel categorical classification of all possible single candlestick patterns is presented based on adaptive criteria. Candlestick patterns predictive power is evaluated from a statistical inference approach, where the mean of the average returns of the strategies in out-of-sample historical data is taken as test statistics. No net positive average returns are found in any case after taking into account transaction costs. More complex candlestick patterns are considered feeding supervised learning systems with the information of past bars. Fractional differences applied to close prices are found to be more informative than integer differences for all supervised learning systems studied (decision trees, random forest and AdaBoost) that are used as classifiers for learning the conditions for which winning trades happen. AdaBoost is found to reach the highest learning score among all three different classifiers employed.
Keywords: FOREX, Efficient Market Hypothesis, Adaptive Candlestick Patterns, Decision Trees, Random Forest, AdaBoost, Fractional Differences, Finance
JEL Classification: G14
Suggested Citation: Suggested Citation