Abstract:

The improved structural optimization algorithm of the flexible neural tree model is employed to select the parameters for effecting stock market production. With higher accuracy and shorter time, important parameters which affect the risk of the stock market are found. In the period of learning of the flexible neural tree model, the evolution generation of the algorithm is not a fixed value and the mean error rate is utilized to control the evolution generation. The structure and parameters of the flexible neural tree model are optimized by probabilistic incremental program evolution and simulation annealing, respectively. It has been demonstrated that the method is very effective for forecasting stock market risk.

Key words: stock market; flexible neural tree model; error rate; probabilistic incremental program evolution; simulation annealing