Neural net as adaptive systems for the time series forecasting

The aim of adaptive methods for short-term forecasting is the construction of self-evaluation models. They need to reflect time varying conditions and take account of relative information value or give good estimation of the time series members in the future. A great deal of effort has been devoted to developing systems for modeling and forecasting in financial engineering. In short term forecasting the authors could find no significant improvements due to the following: in the financial sphere a large number of determinants take place at any one time, the classical statistical technique is restricted to number of varying parameters, sample sizes and nonlinearities in the data; the structural relationship between factors in the financial markets change over time; and many of the rules in this area have fuzzy character and are not susceptible to quantitative analysis. The authors present a method for using neural nets as adaptive nonlinearity systems for the approximation and forecasting of time series