Forecasting of stock market using soft computing techniques

Abstract

newlineFor hundreds of years speculators have tried to make a profit from the financial newlinemarkets by attempting the difficult task of predicting their future movements. To this end, newlinemany methods and techniques have been developed that purport to assist the market newlineparticipant in generating profits. This study examines the effectiveness of technical newlineanalysis and time series modelling to forecast the next day closing prices of some stock newlinemarkets. A number of research efforts has been devoted to forecast stock price based on newlinetechnical indicators which rely purely on historical stock price data. However, the newlineperformances of such technical indicators are not always satisfactory. newlineIn this thesis various swam and evolutionary based algorithms are incorporated newlinewith some prediction models to effectively predict the next day closing prices of different newlinemarkets. Initially, the prediction is started with conventional backpropagation neural newlinenetwork and least mean square models to predict the next day closing prices. Then newlineExtreme Learning Machine (ELM) model is combined with different metaheuristic newlineoptimization techniques such as Particle Swarm Optimization (PSO), Differential newlineEvolution (DE), Teaching Learning Based Optimization (TLBO) and Symbiotic Organism newlineSearch (SOS) to successfully predict the closing prices of three different markets. These newlinealgorithms helps in optimally designing the weights and biases of ELM network. Various newlinestatistical parameters such as mean square error (MSE), mean absolute percentage error newline(MAPE) and accuracy are used to verify the effectiveness of the proposed models. Paired newlinet-test is also carried out to hypothetically prove the closeness of predicted closing prices newlinewith that of the corresponding actual prices. It is observed that SOS-ELM prediction newlinemodel outperforms the other models in all aspects. newlineFurther, a novel quasi-oppositional based learning (QOBL) is integrated with newlineSymbiotic Organism Search to enhance the performance of ELM prediction model. The newlineconcept of opposite number helps in generation of weights and bias in a more effective newlinemanner. Various popular stock markets Dow Jones Industrial Average, Hang Seng Index, newlineInfosys Ltd., Nasdaq Composite, Euronext-100, Nifty 50, and Russell 2000 are considered newlinevi newlineas case studies to prove the efficacy of QSOS-ELM over conventional SOS-ELM newlineprediction model. newlineFinally, the concept of fuzzy logic is introduced to determine two important newlineparameters of SOS algorithm instead of taking them as either 1 or 2 as in case of SOS and newlineQSOS algorithm. Comparative performance analysis amongst fuzzy adaptive QSOS-ELM newline(FQSOS-ELM), QSOS-ELM and SOS-ELM is done to prove the supremacy of FQSOSELM newlineprediction model in comparison with QSOS-ELM and SOS-ELM prediction models