HOW WAVELETS AND GENETIC ALGORITHMS CAN ASSIST INTELLIGENT HYBRID METHODOLOGIES IN HANDLING DATA DRIVEN STOCK EXCHANGE DAILY TRADING
In this paper is explored the suitability of genetic algorithms for constructing fuzzy rule bases, as part of an hybrid decision support architecture, involving neural networks for wavelet-filtered daily stock rates of change. Specifically, the main structure of the suggested methodology combines a wavelet-based noise removal system, a "multilayer perceptron feedforward neural network" and finally a fuzzy system, which provides the trader with both, linguistic and numerical output, representing a buy/hold/sell strategy. The use of wavelet filtering in data pre-processing, improves the predictability of neural networks, however, it involves the selection of proper wavelet bases. Therefore, by applying genetic algorithms in fuzzy rule bases for optimizing the decision policy, the paper aims at offering a decision support, independent of the selection of the wavelet basis. It is also demonstrated how, based on the test results, the overall system is able to make successful trend prediction, which is then used to create an output similar to the policy that traders would apply if forward price movement was considered to be known.