Integrating High Volume Financial Datasets to Achieve Profitable and Interpretable Short Term Trading with the FTSE100 Index

摘要

During the financial crisis of 2009 traditional models have failed to provide satisfactory results. Lately many techniques have been proposed to overcome the deficiencies of traditional models but most of them deal with the examined financial indices as they are cut off from the rest global market. However, many late studies are indicating that such dependencies exist. The enormous number of the potential financial time series which could be integrated to trade a single financial index enables the characterization of this problem as a "big data" problem and raises the need for advanced dimensionality reduction techniques which should additionally be interpretable in order to extract meaningful conclusions. In the present paper, ESVM-Fuzzy Inference Trader is introduced. This technique is based on the hybrid methodology ESVM Fuzzy Inference which combines genetic algorithms and some deterministic methods to extract interpretable fuzzy rules from SVM classification models. The ESVM-Fuzzy Inference Trader was applied to the task of modeling and trading the FTSE100 index using a plethora of inputs including the closing prices of various European indexes. Its experimental results were compared with a state of the art hybrid technique which combines genetic algorithm with Multilayer Perceptron Neural Networks and indicated the superiority of ESVM-Fuzzy Inference Trader. Moreover, the proposed method extracted a compact set of fuzzy trading rules which among others can be utilized to describe the dependencies between other financial indices and FTSE100 index.