We discuss some questions on the nonlinear Bayesian Dynamic Models in this article. First, some random simulations are introduced and the important sample is improved, which supplies the method and theory for the choice of the important sample. An interpolation rule as a method for the nonlinear models is implied. Neural Net as an intelligent machine has more superiority in dealing with the sophisticated nonlinear system. In this article, the neural net theory is introduced and solved three nonlinear models, which gives the intelligence for the models forecasting. In the forth chapter, the Combination of forecast is discussed. Further more, we discuss the posterior mode estimation for nonlinear and non-Gaussian state space models with the Quadratic Hill-climbing Method (Goldfeld, Quandt, Trotter, 1966). At last, we discuss the models choice in detail including the neural net models choice and the Bayes factors for model choice, and generalize the fractional Bayes factors.
|