Prediction of financial time series with recurrent LoLiMot (locally linear model tree)

The tolerance and non-stability in financial indexes make changes to other sub-systems like human resources, economics, factory productions and etc. Having underling knowledge and a model to simulate such systems obtains a fine vision to estimate further and calculate hard-decision making tasks before execution like: dept from banks, cash injecting and insurance services. Using Neuro-fuzzy networks are one of the most powerful tools for this estimation. The particular locally linear model type of these networks called LoLiMot are in interest because of their linear training and construction optimization. These network can be much efficient when be a recurrent network why can better capture the dynamism´s order of dynamic processes. The Locally linear Neuro-Fuzzy model (LoLiMot) here is as basis for making recurrent. In this paper this network with a global state feedback is implemented and the accuracy and the results of this recurrent network on Dow Jones index as a financial time series are compared with the static LoLiMot. The obtained results were better.