Financial Time Series Forecasting Using Soft Computing Techniques

Abstract

Analyzing and forecasting and#64257;nancial time series has remains a critical problem for economist, researchers, and and#64257;nancial managers. Variation occurs in the trend of and#64257;nancial time series data due to many reasons such as change in political scenario, and#64258;uctuation in oil and gold price, global economy scenario, and even psychology of investors. Such random and#64258;uctuations lead to sudden fall after a steady increase or a sudden rise after a gradual fall in the trend of and#64257;nancial time series, which are diand#64259;cult to predict with conventional methods. Hence the and#64257;nancial time series has been characterized with high volatility, chaotic, nonlinearity as well as randomicity. The underlying system models of time series prediction are complex and not known a priori, hence accurate and unbiased estimation cannot be always achieved using well known linear techniques. This complexity and dynamism makes forecasting and#64257;nancial time series intellectually challenging and attract the attention of researchers. Being a key area of data mining task, it has been taken for research in this thesis. It has been analyzed and experimented with various soft computing and their hybrid techniques to achieve better accuracy with least complex model architectures. Speciand#64257;cally, an Artiand#64257;cial Chemical Reaction of Neural Networks (ACRNN); a Fuzzy Neural Network (FNN); an Artiand#64257;cial Chemical Reaction based Functional Link Network (ACFLN); and a GA based Pi-Sigma Neural Network (PSNN-GA) have been proposed and applied to forecast some fast growing stock market future value(s). These models are adaptive, hybrid in nature and found to yield good solutions in terms of producing close to accurate future value(s). Usually the time series prediction is based on the observations of past trend over a pe- riod of time. In general the curve the time series data has a linear part and a non linear part. Prediction of the linear part is not a diand#64259;cult task, but the prediction of non linear segments is diand#64259;cult. Though diand#64256;erent non-linear prediction models are in use, but their predi