In this paper, a LLRBFN is proposed to forecast the nonlinear dynamics of stock price time-series. The LLRBFN provides
a parsimonious interpolation in high-dimension space and it is more efficient than RBFN. Also the local linear model gives more ability than a constant weight thus in the proposed LLRBFN, only a few of radial basis functions is necessary for time-series predicting. In other words, the LLRBFN has better results than RBFN with a same structure. This claim has been demonstrated in the simulation that two networks (LLRBFN and RBFN) with the same inputs have been utilized for the stock price forecasting and MSE of the LLRBFN is less than the RBFN