Title: Computational intelligence methods for data mining of causality extent in the time series
Authors: Lukáš Pichl; Taisei Kaizoji
Addresses: International Christian University, Osawa 3-10-2 Mitaka Tokyo 181-8585, Japan ' International Christian University, Osawa 3-10-2 Mitaka Tokyo 181-8585, Japan
Abstract: We adopt the support vector machine (SVM) and artificial neural network (ANN) for causality rate extraction. The dataset records all details of the futures contracts on the commodity of gasoline traded in Japan. By sampling the tick data at 1 min, 5 min, 10 min, 30 min, 1 hour and 1 day scales, we derive time series of varying causal degree. Trend predictions are computed by using the SVM binary classifier trained on 66.6% of the data using a five-step-back moving window which samples the log returns as the predictor data. From the testing data, we extract varying rates of causality degree, starting from the borderline of 50% up to the order of 60% in rare cases. The trend prediction analysis is complemented by the ANN method with four hidden layers. Overall, the market of the gasoline futures in Japan is found to be rather close to the efficient market hypothesis in comparison with other commodities markets.
Keywords: commodity market; artificial neural network; ANN; support vector machine; SVM; trend prediction; causality extraction.
DOI: 10.1504/IJCSE.2018.093782
International Journal of Computational Science and Engineering, 2018 Vol.16 No.4, pp.411 - 418
Received: 07 Jun 2016
Accepted: 30 Nov 2016
Published online: 06 Aug 2018 *